Now [genes] swarm in huge colonies, safe inside gigantic lumbering robots, sealed off from the outside world, communicating with it by tortuous indirect routes, manipulating it by remote control.
—Richard Dawkins (1976)
This famous sentence from The Selfish Gene is often interpreted to mean that organisms are puppets of their genes. Genes pull the strings. They are in control. At the same time, Dawkins subtly undercuts genetic autonomy. Genes are remote and sealed off from an outside world with which their communication is tortuous and indirect. Do genes make the decisive inputs to choice, or is control primarily an attribute of the robotic structures constructed by genes? Can the robot decide when to consult its genes? Can genes override robotic decisions? Are genes merely interested bystanders of a play performed for their benefit?
We construct robots to free ourselves from mundane decision making and to perform actions that we are unable to achieve without prosthetic aid. Some decisions in the control of robotic action are reserved for human actors but others are delegated to the robot. Consider a robot built to explore Mars. The robot interacts directly with the Martian environment, but its earthbound controllers experience Mars only vicariously through the robot’s sensors. The robot and its controllers use inputs from these sensors to modify the robot’s behavior. Not everything sensed by the robot need be communicated to Earth. The robot has other sensors that receive input from ground control. For some decisions, the robot is told what to do by its earthbound controllers; but for other decisions, it is on its own, because, at crucial moments, it must respond to events more rapidly than signals can be exchanged with Earth (Dennett 1984, 55).
One can imagine a spectrum of robot autonomy. At one end of the spectrum, robots are simple mechanical prostheses with all important decisions taken by humans. At the other end, robots are designed by humans but make all decisions on their own. Where do Dawkins’s lumbering robots reside on this continuum? Are the robots on a short leash, with all important decisions taken by their genes, or are the robots fully autonomous, exploring and exploiting their environment without consulting their genes?
“Lumbering” has connotations of awkwardness, but some robots perform actions with a delicacy and precision that are beyond the powers of unaided human actors. The aim of robotic design is to produce supple rather than clumsy machines and the same is surely true of the “designs” of natural selection. An organism is an agile automaton (from the Greek, meaning “acting of itself”) designed by natural selection to function effectively in a complex world. Simple automata (genes and proteins) interact with each other in complex networks to create a hierarchy of higher-level automata (cells, organs, organisms). Simple automata have only a small number of possible states, but as the number of simple automata that constitute a larger system increases, so does the number of possible states of the system, in a combinatorial explosion. As a result, higher-level automata can express more flexible behavior and possess more sophisticated information about their environment than can lower-level automata. Genes may be less nimble than the robots they construct.
An automaton “detects” a property of the environment when this property causes a change in the automaton’s state. This environment may contain other automata. “Communication” takes place when one automaton (the sender) causes a change in state of another automaton (the receiver). A receiver detects a change of state of the sender, either directly, by physical contact, or indirectly, by the detection of a change in the environment induced by the sender’s change of state. Genes and proteins can potentially fill the roles of sender and receiver, but so too can cells and other higher-level automata.
Genes encode instructions for the construction of protein automata but genes are themselves automata. A gene’s states are determined by the binding of transcription factors and other proteins, by interactions with RNAs and with other DNA sequences, and by chemical modifications such as cytosine methylation. These factors determine where and when the gene is expressed. A gene may possess information about past environments as well as information about the current environment. An imprinted gene, for example, “remembers” whether it was present in a male or female body in the previous generation. A common criticism of the selfish-gene approach is that it assigns too much agency to genes. A gene is not a homunculus aware of everything that is happening to the organism and making plans accordingly, but one can also err on the side of underestimating the strategic options available to genes.
The principal way that a gene interacts with the world is by the production of RNA transcripts, some of which are translated into proteins. Many proteins have multiple states and function as simple automata. The factors that induce changes of a protein’s state are what the protein “knows” about the world. A protein communicates with another protein when it induces a change in the other protein’s state. Each protein’s repertoire of functional states may include changes of conformation caused by interaction with things in its cellular or extracellular environment, as well as modifications of its chemical structure.
Most housekeeping functions of organisms are performed by proteins. Genes and proteins are equally mindless aperiodic polymers. Why should we privilege genes, rather than proteins, as the evolutionary actors on an ecological stage? We think differently about genes because genes are a very special kind of automaton. They are, to use Dawkins’s term, replicators. Chemical changes to their structure are transmitted to their descendants, if descendants they have. By contrast, structural changes to proteins are not transmitted to their copies because proteins are not copied. Informational genes perform the role of evolutionary repositories of heritable information used to construct organisms. Although informational genes specify the construction of organisms, it does not necessarily follow that material genes control organisms.
Rhodopsin is a gene expressed in rod cells of the retina but not in other cell types. Thus, Rhodopsin must, in some sense, sense when it resides in a rod cell and use this datum to switch between active and silent states. In its active state, Rhodopsin produces an mRNA that is translated at the endoplasmic reticulum to produce a protein that forms a covalent linkage with the chromophore 11-cis-retinal to form the visual pigment rhodopsin. The receipt of a photon of an appropriate wave-length causes isomerization of 11-cis-retinal to all-trans-retinal. This causes a conformational change in rhodopsin that is propagated to other proteins, triggering a complex biochemical cascade that culminates in hyperpolarization of the plasma membrane and discharge of a higher-level automaton, the rod cell (Okada et al. 2001; Ridge et al. 2003). Rhodopsin (the gene) encodes rhodopsin (the protein). The common use of the same name for gene and protein is an example of metonymy. (Geneticists’ standard convention, used in this paragraph and the remainder of this chapter, is that genes are italicized whereas their protein products are not.)
Two important points can be made about this example. First, the photon is detected by the protein, not by the gene. The gene neither detects photons nor signals their presence. Rather, the gene directs the construction of protein automata that detect the presence of photons and signal their presence to other protein automata. Second, a single state of the gene produces an automaton that switches among multiple states in response to “environmental” cues. A gene can produce a protein with a “behavioral flexibility” that the gene itself lacks. There is no simple relation between the number of states of a gene and the number of states of the protein automata it constructs.
Consider a quintessential example of social communication in which rhodopsin plays a part. Multitudinous photons are received by the rod and cone cells of an infant’s retina. The pattern of discharge of retinal cells initiates complex processes in the infant’s brain that result in the recognition of the infant’s mother and the coordination of a motor response: a smile. By an equally complex process, the baby’s smile is detected by its mother and elicits a smile in return. The entire chain of events—from receipt of photons at the baby’s retina to the contraction of muscles in its mother’s face—takes place without the causal intervention of changes in gene state.
An exchange of smiles is possible because countless gene copies specify the production of innumerable protein automata in an untold number of higher-level automata (nerve cells and muscle fibers). These cellular automata are organized into two very high-level automata (mother and child) who are able to respond to each other’s facial gestures. The limber robots communicate without consulting their lumbering genes. The development and maintenance of organism-level automata clearly involve coordinated changes in gene state but the exchange of smiles is too rapid for transcription and translation to play a role. Higher-level automata acquire and act on information that is unavailable to lower-level automata. No gene in the infant’s genome perceives its mother’s face.
What is true of the interpretation of photons is also true of the interpretation of other sources of information about an organism’s external and internal environment. Another mammalian example reaffirms the substantial autonomy of protein (and cellular) automata from direct genetic control. Information about external and internal temperature is detected by temperature-sensitive neurons, some located peripherally and others centrally. These inputs are integrated in the hypothalamus and other brain regions to coordinate thermoregulatory responses (Morrison 2004; Romanovsky 2007). One such response is the activation of nonshivering thermogenesis in brown adipose tissue. Uncoupling protein 1 (UCP1) resides in the inner membrane of the mitochondria of brown adipocytes. Activation of UCP1 causes a proton leak that uncouples mitochondrial respiration from oxidative phosphorylation. As a result, organic substrates are burned with the release of heat. The entire chain of events from stimulus (skin-cooling) to response (activation of nonshivering thermogenesis) may take place without the direct intervention of genes.
The efferent arm of this response (from brain to UCP1) involves a complex signaling cascade involving multiple protein automata. Brown adipose tissue is innervated by noradrenergic neurons of the sympathetic nervous system. When these neurons receive appropriate input from the brain, they release norepinephrine, which binds to β3-adrenergic receptors (β3ARs) at the cell surface of brown adipocytes, causing the release of a small protein (Gαs) within the cell. Gαs stimulates another protein, adenylyl cyclase, to produce cyclic adenosine monophosphate (cAMP). Via a further series of protein automata, increased cAMP causes activation of UCP1 in the inner mitochondrial membrane (Cannon and Nedergaard 2004; Nakamura and Morrison 2007; Romanovsky 2007).
Although changes of gene state do not play a direct role in the acute response to cold, such changes play important roles in modulating responses over longer timescales. The processes by which cold-exposure increases the thermogenic capacity of brown adipocytes are instructive. Noradrenergic signaling via β3AR not only activates UCP1 (the protein) but also promotes transcription of Ucp1 (the gene), promoting the translation of more UCP1 protein. Noradrenergic neurons also form synapses on preadipocytes that express β1-adrenergic receptors. Cold-induced activation of β1ARs promotes the differentiation of preadipocytes into mature brown adipocytes, a process that involves the transcription of many genes that are inactive in preadipocytes. As a result of these changes in gene expression, the inner mitochondrial membranes of brown adipocytes contain more copies of UCP1, and the total number of brown adipocytes is increased, in anticipation of the next cold exposure (Cannon and Nedergaard 2004).
This book is an attempt by one organism-level automaton to communicate to other organism-level automata, your good selves. Each of us has a set of genetic controllers “back on Earth” trying to pull the strings, but we make most decisions on our own. Our pains and our pleasures are the sticks and carrots that genes use to influence our decisions in pursuit of their ends. Genes do not care for us, know little about our world, and cannot agree among themselves. We should respect their suggestions, but not too much. If this book serves my genes’ interests, it will be by a very indirect route. Our genes’ purposes are not our own.
We think of machines as integrated wholes in which all parts work together to achieve common goals. One of the recurring themes of this book is that genes may have divergent ends even within an individual organism. Thus, the parts of the organismal “machine” may sometimes act at cross purposes in pursuit of conflicting agendas. This is not how we usually think about machines. Such internal conflicts suggest an alternative metaphor of a society of actors that must work together to achieve common goals but do not always act with unanimity. These alternative metaphors, of organisms as machines or societies, yield different insights into the basic processes within cells.
Cell biologists and behavioral ecologists both use the language of “communication” and “signals” but make different implicit assumptions about how signals evolve. Cell biologists are usually interested in signals that are transmitted within cells or between cells of a single body. Signaler and receiver are implicitly assumed to have identical interests. The question of whether signals are credible does not arise because signalers do not have incentives to deceive. In behavioral ecology, on the other hand, signaler and receiver are different individuals, potentially with conflicting interests. Receivers must decide whether signals can be trusted. Although behavioral ecologists recognize the possibility of conflicts between individuals, they usually assume that individuals have well-defined, unitary interests.
These two areas of inquiry are, of course, intimately linked. A behavioral signal is usually the external output of a complicated process of signaling among and within cells of the sender. The perception and interpretation of the signal usually involves an equally complex process of signaling among and within cells of the receiver. Despite these intimate connections, different kinds of questions are typically asked about signaling within individuals and signaling between individuals. Communication within individuals is usually viewed as a problem in signal engineering within the paradigm of the organism as machine. Relevant questions are how to send signals efficiently, how to cope with noise and interference from other signaling pathways, and how to correct errors. Questions of signal efficacy also arise for communication between individuals, but behavioral ecologists usually focus on question of signal credibility. Can a signal be trusted? What is the sender’s motive? Does the sender have something to hide? Communication is seen as a social rather than a mechanical activity.
Neither cell biologists nor behavioral ecologists have given much thought to the implications of conflicts within the genomes of individual organisms. Current theory does not predict how an organism should behave if its responses are influenced by genes with conflicting interests. I suspect there is no general answer to this question and that answers for specific cases will require detailed knowledge of intricate molecular processes within cells. The genome has aspects of a fractious and poorly informed committee attempting to set policy, but with most decisions on how to implement policy taken elsewhere. This perspective raises new and unsettling questions. Is deception possible within an individual? Could different parts of an individual disagree over whether to send a signal to another individual? Would such a signal be sent?
There are many sources of intragenomic conflict, but I will focus on antagonism between genes of maternal and paternal origin. This conflict can be illustrated with a minor modification of a famous thought experiment. J. B. S. Haldane wrote:
Let us suppose that you carry a rare gene which affects your behaviour so that you jump into a flooded river and save a child, but you have one chance in ten of being drowned, while I do not possess the gene, and stand on the bank and watch the child drown. If the child is your own child or your brother or sister, there is an even chance that the child will also have this gene, so five such genes will be saved in children for one lost in an adult. If you save a grandchild or nephew the advantage is only two and a half to one. If you only save a first cousin, the effect is very slight. If you try to save your first cousin once removed the population is more likely to lose this valuable gene than to gain it. (1955, 44)
Haldane’s logic is simple. Your first-degree relatives, including your children and full siblings, have one chance in two of carrying a copy of a rare gene present in your body. Therefore, your genes would increase their number of surviving copies if you were to sacrifice your life for more than two first-degree relatives. A second-degree relative, such as a grandchild or nephew, has one chance in four of carrying a copy of the gene. You need to rescue five second-degree relatives at the cost of your own life to increase the number of surviving copies of your genes. A third-degree relative, such as a cousin, has one chance in eight, and a fourth-degree relative only one chance in sixteen. You need to rescue nine third-degree relatives or seventeen third-degree relatives to be marginally ahead in the genetic accounting. This calculus was formalized in W. D. Hamilton’s (1964) theory of inclusive fitness in which an actor values the fitness of other individuals in proportion to their probability of sharing a randomly chosen gene of the actor.
Genomic imprinting refers to molecular modifications of genes, in either the mother’s germline or father’s germline, that provide a historical record (in offspring) of whether a gene resided in a male or female body in the previous generation. An intragenomic conflict lay hidden in Hamilton’s theory because imprinted genes have effects that differ depending on the gene’s parental origin. Consider the question whether your genes would be prepared to sacrifice your life to save three drowning maternal half-siblings. Haldane and Hamilton would probably have answered no, because they assumed that a gene’s effects were independent of its parental origin and a randomly chosen gene in you would have one chance in four of being present in each half-sibling. But this answer averages a one-in-two chance that one of your maternal genes has copies present in each half-sibling and a zero chance that one of your paternal genes has copies in your maternal half-siblings. Your maternal genes would gain a substantial benefit from the nepotistic sacrifice of your life for three first-degree relatives, but your paternal genes would suffer a major cost from the rescue of nonrelatives.
In the presence of genomic imprinting, genes possess information about their parental origin and can employ conditional strategies to do one thing when inherited from a mother and something else when inherited from a father (Haig 1997). This means that internal genetic conflicts are possible in many of our interactions with kin because most relatives have different probabilities of carrying copies of genes we inherited from our mother and our father. The major exceptions are our descendants, who have equal probabilities of inheriting the genes we inherited from our mothers and fathers. Full siblings are another category of kin with equal probabilities of carrying copies of our maternal and paternal genes. However, in our evolutionary past the category of “siblings” would have contained a mixture of full and half-siblings. Therefore, sibling interactions are also predicted to be subject to some degree of internal genetic conflict.
Your feelings toward your children and grandchildren should not be associated with major internal conflicts, because your descendants have equal probabilities of inheriting the copy of a gene you inherited from your mother or the copy you inherited from your father. But the reverse is not true: your genes of maternal origin necessarily have copies in your mother and your genes of paternal origin are absent from your mother. Therefore, strong internal conflicts are predicted in an offspring’s relations to its mother. Readers who are parents might reflect on the different flavor of their relations to their own parents (predicted to be associated with internal conflicts) and their relations to their own children (predicted to lack major internal conflicts).
Knowledge of proximate mechanisms is required to understand how internal genetic conflicts will be resolved. The simplest resolution of the internal conflict between genes of maternal and paternal origin occurs if the relevant genes are unimprinted and therefore lack information about their parental origin. In the absence of imprinting, a gene is constrained to exhibit the same behavior when it is transmitted via eggs or sperm. Consider an imprinted gene for which natural selection has favored a higher level of gene product when the gene is maternally derived than when it is paternally derived. A “resolution” of this conflict occurs when the gene is silent when paternally derived but expressed at the maternal optimum when maternally derived. The nature of the resolution is reversed at a locus where natural selection favors higher expression when a gene is paternally derived. For such a gene, the unbeatable strategy is to be silent when maternally derived but produce the paternal optimum when paternally derived. I have called this the loudest-voice-prevails principle (Haig 1997).
The loudest-voice-prevails is a simple form of “conflict resolution.” Whichever allele favors more product produces that amount, and the other allele produces none. Silencing of one of the two alleles at a diploid locus has a number of important consequences, of which I will mention two. First, alternative alleles at the locus have phenotypic effects when inherited from one sex but are without effect when inherited from the other sex. Therefore, alleles at a maternally silent locus will be selected solely for their effects on patrilineal fitness, whereas alleles at a paternally silent locus will be selected solely for their effects on matrilineal fitness (Haig 1997, 2000). Second, the loudest-voice-prevails principle reveals a sender’s identity to the recipient. If both alleles are transcribed, a recipient of a signal (gene product) has no way of telling whether the sender is a maternal or paternal allele. If one of the two potential sources of a signal is reliably silent, then the signal reveals the identity of the signaler.
At a single locus, the loudest-voice-prevails principle suggests that whichever allele favors the larger amount “wins.” However, most organismic outcomes are influenced by many genes. For example, maternal and paternal alleles may disagree over how much investment an offspring extracts from its mother. Paternal alleles at a demand-enhancing locus may produce their favored amount of a demand enhancer, but these effects may be countered by maternal alleles at another locus producing their favored amount of a demand inhibitor (Haig and Graham 1991). A “resolution” of this conflict is possible with maternal silencing of the demand enhancer and paternal silencing of the demand inhibitor (Haig and Wilkins 2000; Wilkins and Haig 2001). This resolution of the conflict has the form of a stalemate: the marginal cost of an increment of demand enhancer balances any benefit paternal alleles would gain from increased demand; likewise, the marginal cost of an increment of demand inhibitor balances any benefit maternal alleles would gain from reduced demand. In general, neither matrilineal nor patrilineal fitness is optimized at the stalemate.
An example will illustrate how such a stalemate can be expressed in a signaling system. Insulin-like growth factor 2 (IGF2) is a paternally expressed gene that promotes fetal growth. Its protein product, IGF2, binds to two receptors (IGF1R and IGF2R). IGF1R mediates the growth-promoting effects of IGF2. IGF2R is a decoy receptor (a deceptor) that binds IGF2 and transports it to lysosomes for degradation (Filson et al. 1993). In most eutherian mammals IGF2R is paternally-silent (Killian et al. 2001). Thus, a paternal gene produces a growth factor (IGF2) that is degraded by the product (IGF2R) of a maternal gene (Haig and Graham 1991). This can be considered a simple form of deception: IGF2 sends a signal to IGF1R, but the message is intercepted by IGF2R before it reaches the intended recipient. In this example, there is no transfer of information between IGF2 and IGF2R (the respective genes). Rather, the message from IGF2 is intercepted by a protein produced by IGF2R.
The loci that influence an organismal outcome may have more than two sets of interests. In a 2006 study, I explored interactions among multiple “factions” with respect to a single organismal trait (“demand”). I found that the factions tended to align into two “parties”: one favoring increased demand and the other favoring reduced demand. More theoretical work is needed to see whether this prediction can be generalized to conflicts over multiple traits.
The control of nonshivering thermogenesis is a potential arena of conflict between genes of maternal and paternal origin in species that huddle together for warmth. Heat generation by one individual reduces the heating costs of other individuals in its huddle and creates an evolutionary temptation of free-riding (not paying a fair share of the communal heating bill). If the members of a huddle include half-siblings, maternal and paternal alleles can disagree over how much heat to contribute to the common good. Specifically, when members of a multiple-paternity litter huddle together, paternal alleles are predicted to favor a lower set-point for the brown-adipose thermostat than that favored by maternal alleles (Haig 2004a, 2008a).
Genomic imprinting influences at least one step in the signaling pathway that activates nonshivering thermogenesis in brown adipocytes. Gαs is one of several protein products of the complex GNAS locus (Abramowitz et al. 2004). Both alleles of GNAS produce Gαs in most tissues of the body, but only the maternal allele produces Gαs in brown adipose tissue (Yu et al. 1998). A second gene product, XLαs (“extra large” αs), is produced by paternal GNAS and antagonizes the effects of Gαs in brown adipose tissue (Plagge et al. 2004). Gαs and XLαs mRNAs are transcribed from different GNAS promoters, and use alternative first exons, but share their remaining twelve exons. Thus, Gαs is produced by maternal GNAS and promotes nonshivering thermogenesis whereas XLαs is produced by paternal GNAS and inhibits nonshivering thermogenesis. Imprinted genes also influence the recruitment of extra heating units. Two paternally expressed genes, Preadipocyte factor-1 and Necdin, produce proteins that inhibit the differentiation of preadipocytes into brown adipocytes (Tseng et al. 2005; Haig 2008a, 2010a).
Brown adipocytes are heat-generating automata. Their level of heat production is determined by the combined effects of unimprinted genes, maternally expressed imprinted genes, and paternally expressed imprinted genes. The loudest-voice-prevails principle predicts that imprinted genes that increase heat production in multiple paternity huddles will be maternally expressed whereas imprinted genes that reduce heat production will be paternally expressed. However, current theory has little to say about why some genes in a pathway are imprinted whereas others are not. Why is GNAS is imprinted in brown adipocytes but UCP1 is not?
Imprinting can only make a phenotypic difference at loci for which gene dosage matters. If one active allele is as good as two, then silencing one allele makes no selective difference. Some effects of Gαs are dosage sensitive. For example, loss of one functional copy of Gαs causes osteodystrophy despite the expression of Gαs transcripts from both alleles in bone (Mantovani et al. 2004). The effects of Gαs may be particularly dosage-sensitive because many G protein–coupled receptors activate multiple signaling pathways via alternative G proteins with different α subunits. Thus, the precise stoichiometry of α subunits may determine the balance of signaling among pathways and the nature of the cellular response. For example, the β3AR of brown adipocytes signals via both Gαs and Gαi (Chaudhry et al. 1994). However, it seems unlikely that Gαs is the only dosage-sensitive step in the pathway from the detection of norepinephrine to the activation of nonshivering thermogenesis. Should a brown adipocyte generate heat only when heat serves the interests of maternal genes, of paternal genes, of unimprinted genes, or of something else?
The paucity of spontaneous movement in infants with Prader-Willi syndrome, and their described placid nature, may result in decreased interaction with care-givers.
—S. B. Cassidy (1988)
Deletion of the paternal copy of a cluster of imprinted genes at human chromosome 15q11–q13 causes Prader-Willi syndrome, whereas deletion of the maternal copy causes Angelman syndrome. Thus, Prader-Willi syndrome is caused by an absence of expression of paternal genes and Angelman syndrome by an absence of expression of maternal genes. Therefore, the former syndrome is predicted to exaggerate behaviors that benefit the mother at a cost to the child, and the latter syndrome to exaggerate behaviors that benefit the child at a cost to the mother (Haig and Wharton 2003). These syndromes’ complex phenotypes suggest aspects of development and behavior that are sources of contention between genes of maternal and paternal origin (Holm et al. 1993; C. A. Williams et al. 2005). Together, these syndromes provide clues about a struggle for control that takes place in typically developing children who inherit copies of the imprinted gene cluster from both parents and whose behavior is determined by the balance of effects of paternally expressed and maternally expressed genes.
Infants with Prader-Willi syndrome are disinterested in feeding, suck very poorly, and are often fed directly through a tube to the stomach to ensure adequate nutrition (Cassidy 1988). Their cry is described as feeble, weak, squeaky, peculiar, or not sustained (Aughton and Cassidy 1990; Butler 1990; Miller, Riley, and Shevell 1999; Õiglane-Shlik et al. 2006). Infants with Angelman syndrome, on the other hand, obtain adequate nutrition and do not require tube feeding. Another contrast is found in sleep patterns. Infants with Prader-Willi syndrome exhibit excessive sleepiness, whereas infants with Angelman syndrome exhibit excessive wakefulness. This suggests that sleep in typically developing infants is disordered because the child is engaged in an internal struggle between genes of maternal origin that are endeavoring to put the child to sleep and genes of paternal origin that are endeavoring to keep it awake. An evolutionary interpretation is that paternally expressed genes (absent in Prader-Willi syndrome) have been selected to favor intense suckling and frequent waking to prolong maternal infertility after birth and delay the arrival of a younger sibling who would compete with the older child for the mother’s care and attention. By contrast, maternally expressed genes (absent in Angelman syndrome) have been selected to favor less frequent waking and earlier weaning. When new mothers complain of exhaustion, their fatigue can be considered an adaptation (or extended phenotype) of genes that their baby inherited from its father (Haig 2014; Kotler and Haig 2018).
Angelman syndrome is characterized by positive affect and a smiling demeanor with frequent laughter (Horsler and Oliver 2006a). This contrasts with the negative affect of Prader-Willi syndrome (Isles, Davies, and Wilkinson 2006). Laughter in Angelman syndrome has been described as inappropriate and unprovoked, but careful behavioral studies suggest laughter is rare in nonsocial contexts and is particularly pronounced after eye contact (Oliver, Demetriades, and Hall 2002; Horsler and Oliver 2006b). In a comparison of thirteen children with Angelman syndrome with a matched group of children with other forms of intellectual disability, the children with Angelman syndrome smiled more, were more likely to reach toward or touch adults before smiling, and their smiles were more effective at eliciting adult smiles in return (Oliver et al. 2007). The frequency of laughing and smiling decreases as children with Angelman syndrome grow older (Adams, Horsler, and Oliver 2011). Thus, young children with Angelman syndrome have been proposed to exhibit exaggeration of behaviors, laughing and smiling, that normally function to elicit maternal care, attention, and attachment (Brown and Consedine 2004; Isles, Davies, and Wilkinson 2006). Past social environments have evolutionarily shaped responses to smiles and have thereby shaped communication within cells.
The exuberant personality of children with Angelman syndrome is combined with profound deficits in verbal and nonverbal communication. Infants possess an abnormal high-pitched cry (Clayton Smith 1993) and babbling is delayed or absent (Yamada and Volpe 1990; Penner et al. 1993). Nonverbal communication is primarily used for making requests and rejecting offers (Didden et al. 2004). Such communication usually involves direct manipulation of the other person—pushing a hand away, leading by the hand, touching to gain attention—rather than gesture or pointing (Jolleff and Ryan 1993). Joint attention, joint action, and taking turns are poorly developed (Penner et al. 1993). Affected children never learn to speak (Clarke and Marston 2000; C. A. Williams et al. 1995). The absence of speech appears out of proportion to the underlying level of cognitive impairment (Alvares & Downing 1998; Pembrey 1996; Penner et al. 1993). Children with Angelman syndrome have poor motor imitation skills, and most fail to imitate verbal behavior (Didden et al. 2004; Duker, van Driel, and van de Bercken 2002; Jolleff and Ryan 1993; Penner et al. 1993). “Motor theories” of the evolution of human language posit that language is based on the perception and imitation of gestures of the vocal tract (Galantucci, Fowler, and Turvey 2006; Gentilucci and Corballis 2006). An interesting hypothesis is that the ataxia and absence of speech in Angelman syndrome have a common etiology in defects in the neural representation of motor actions.
The absence of speech with the absence of expression of maternally derived genes is intriguing. Badcock and Crespi (2006) have suggested that genes of maternal origin have been selected to act in the language centers of the child’s brain to promote attentiveness to maternal instruction and maternal example, coordinating maternal and child needs for the benefit of the matriline. Verbal communication appears to be arrested at a very early stage of development (Grieco et al. 2018). Perhaps earlier onset of language in children reduced costs to mothers and the genes whose expression is disrupted in Angelman syndrome normally function to initiate language development.
Let us adopt, for a moment, the perspective of developmental systems theory, and turn again to the question whether genes control organismal automata. Ontogeny involves necessary interactions between the material parts of the developing organism and its environment. Genes are important parts of this process but cannot act on their own. Nucleic acids, the material stuff of genes, are molecular components of the organism, no different from proteins, fats, carbohydrates, and minerals. These molecular components are organized into higher-level components—muscles, nerves, bones, glands—that work together to act purposefully in the environment. Changes of gene expression play little role in the moment-by-moment control of organismal behavior. At longer timescales of development, genes are tools used to remodel the organism in response to environmental inputs. The sense of gene invoked in this account of organismal function is close to the gene token or material gene. Material genes do not control organismal behavior. The organism, in its manifest complexity, controls itself. This is the metaphor of the autonomous robot.
How should we think about control if the whole can be divided against itself? The metaphor of a society suggests more flexible ways of thinking about agency. Consider the agency of nations and their citizens. Nations act in the world, declare war, sign treaties, invest in infrastructure, resolve conflicts among and discipline their citizens. These actions of nations are partially determined by actions of citizens whose choices and preferences are shaped by actions of nations. One can believe in the independent agency of citizens without denying the collective agency of nations. But the problem of understanding the causal relations between societies and their members is not simply the hermeneutic circle of making sense of the whole by reference to the parts and making sense of the parts from their place in the whole. Human individuals are not simply subsidiary parts of social groups. Groups may have overlapping memberships. Some individuals, for example, may be citizens of more than one nation. Dual allegiance can be a source of conflict within nations and a facilitator of cooperation between nations.
I recognize two kinds of actors on the timescale of organismal behavior: the historical individuals we identify as organisms and the historical individuals I have called strategic genes. Their relations resemble, in some respects, the relations between nations and citizens. Organisms act in the world in ways that are determined by the aggregate actions of their strategic genes, but individual genetic actions are determined by information processing at the level of the organism. Strategic genes are not merely parts of an organismal machine, because coteries of strategic genes within organisms may subvert the good of the whole for partisan advantage, as may coteries of strategic genes distributed across organisms. Strategic genes and organisms have different kinds of agency. They are different ways of carving nature at the joints.
Let us now turn our attention from the synchronic axis of development to the diachronic axis of evolution. On an evolutionary timescale, informational genes are texts. They are repositories of information about what has worked in past environments and what is anticipated to work in future environments. These texts have been written and revised by past environments and are inscribed in the medium of nucleic acid sequences. They include specifications for the construction of organisms that are present-time interpreters of the environment and readers of the genetic text. The present environment provides the context for interpretation of the material text over the course of present development.