Origin of biomolecular games: deception and molecular evolution

Abstract

Biological macromolecules encode information: some of it to endow the molecule with structural flexibility, some of it to enable molecular actions as a catalyst or a substrate, but a residual part can be used to communicate with other macromolecules. Thus, macromolecules do not need to possess information only to survive in an environment, but also to strategically interact with others by sending signals to a receiving macromolecule that can properly interpret the signal and act suitably. These sender–receiver signalling games are sustained by the information asymmetry that exists among the macromolecules. In both biochemistry and molecular evolution, the important role of information asymmetry remains largely unaddressed. Here, we provide a new unifying perspective on the impact of information symmetry between macromolecules on molecular evolutionary processes, while focusing on molecular deception. Biomolecular games arise from the ability of biological macromolecules to exert precise recognition, and their role as units of selection, meaning that they are subject to competition and cooperation with other macromolecules. Thus, signalling game theory can be used to better understand fundamental features of living systems such as molecular recognition, molecular mimicry, selfish elements and ‘junk’ DNA. We show how deceptive behaviour at the molecular level indicates a conflict of interest, and so provides evidence of genetic conflict. This model proposes that molecular deception is diagnostic of selfish behaviour, helping to explain the evasive behaviour of transposable elements in ‘junk’ DNA, for example. Additionally, in this broad review, a range of major evolutionary transitions are shown to be associated with the establishment of signalling conventions, many of which are susceptible to molecular deception. These perspectives allow us to assign rudimentary behaviour to macromolecules, and show how participation in signalling games differentiates biochemistry from abiotic chemistry.

1. Biochemistry: it's just a game

The field of biochemistry entails the characterization of the reactions occurring in living systems, as well as the macromolecules and cellular apparatus responsible. Biochemistry's offspring, molecular evolution, was initiated with the molecular characterization of inheritance via the DNA molecule, and the minutiae of translation. To quote Crick, protein sequences are ‘the most delicate expression possible of the phenotype of an organism and … vast amounts of evolutionary information may be hidden away within them’ [1]. However, classical molecular evolutionary approaches do not address the ‘hidden strategies’ of macromolecules, which is the focus of this work.

Game theory involves the study of the strategies followed by individuals, and organizations, in situations of conflict and cooperation. A Nash equilibrium refers to a certain mixture of strategies where a unilateral change in strategy by one player will not bring any benefit to it [2,3]. Maynard Smith pioneered the use of game theory in evolutionary biology, developing the concept of the evolutionarily stable strategy (ESS) [4]. An ESS is a form of Nash equilibrium in a population where a mutant with a variant strategy cannot successfully invade. Replicator dynamics addresses the dynamics of fitter players (which possess superior utility) that preferentially replicate within a population [5]. An important contribution in these types of evolutionary games was the recognition that there is no need for epistemologically aware agents given that the players are non-human organisms which do not consciously adopt strategies.

Signalling game theory is a branch of game theory that was developed concurrently in both economics and organismal evolutionary biology in the 1970s, and it involves the sending of signals, honest or deceptive, from a sender to a receiver [6,7]. Information asymmetry occurs when the sender possesses information about its type, that is not available to the receiver, thus the sender can choose whether or not to reveal its true type to the receiver. In comparison to organismal evolutionary biology, molecular evolution has made lesser use of concepts from game theory, but with a growing number of contributions, for example [8–16]. In addition, microbial ecology has made use of evolutionary game theory to explain cooperative interactions where metabolites are public goods shared between microbes [17–19]. In this work, we intend to demonstrate that signalling game theory has great explanatory power for a range of molecular evolutionary processes, by pinpointing the ‘strategies’ of macromolecules in their interactions with other macromolecules. In doing so, we also wish to highlight commonalities between signalling behaviour at the molecular, organismal and human levels.

We propose that biochemical reactions also include signalling games played between pairs of macromolecules, which may be regarded as the players. Some general principles may be outlined as follows. A key observation is that macromolecules are able to exert selectivity. This allows recognition of a macromolecular binding partner (the ‘signal’), encoded by a gene (the ‘sender’), by another macromolecule (the ‘receiver’). The ‘signal’ is the macromolecular three-dimensional conformation and physicochemical properties¹ of the sender gene product, i.e. its expressed phenotype. If the gene product is a protein, then the amino acid sequence of the protein determines its conformation, which constitutes the signal. The resulting behaviour of the receiver macromolecule upon binding to and recognizing the signal macromolecule is an ‘action’ that produces ‘utility’, which benefits the fitness of the common genetic system via individual utilities. Information asymmetry exists at the molecular level given that the sender possesses private information consisting of a ‘type’, the identity of which need not be revealed to the receiver.

Macromolecules that are encoded at the genetic level possess inheritance, which introduces replicator dynamics into the game. This process implies that in a game where the players have perfect common interest, the utility of an action results in increased fitness of both sender and receiver. An increase in fitness is observed as a concomitant increase in frequency of both the sender allele and the allele encoding the receiver, in a population. Here, the unit of selection is the sender (gene), but the selected phenotype is the signal, which consists of the conformation and physicochemical properties of the macromolecule.

If the genes encoding two macromolecular binding partners are co-inherited, then they will have perfect common interest and no conflict of interest. Such cooperative signalling games where the players have perfect common interest are termed ‘Lewis’ signalling games [20]. An example of a biochemical Lewis signalling game is provided by thermotolerance via the RheA-HSP18 system: RheA is a thermosensor, thus an informed agent; but HSP18 is needed to modulate thermotolerance, thus an uninformed agent, which nonetheless acts in response to the signal consisting of RheA's conversion to a non-DNA-binding form. They enable the cell to survive spikes in temperature in the environment, thus improving the utility of all the macromolecules contained in the cells—including RheA and HSP18 [21].

1.1. Road-map

The paper is organized as follows. Sections 2 and 3 elaborate the details of the game and the pay-offs specified in an extensive form. Section 4 describes molecular deception (formalized in the electronic supplementary material), and sets the stage for various examples of deception (e.g. junk DNA in §5, molecular mimicry in §6) and mechanisms that evolved to tame them (e.g. molecular sanctioning in §7). Sections 8 and 9 focus on an example illustrating codon evolution, marked by its stability, optimality and universality (Crick's Frozen Accident scenario)—notwithstanding its conventionality, which is exploited by pathogens such as retroviruses. Sections 10 and 11 conclude with a discussion of the Darwinian role of information asymmetry in biological evolution, progressing—undeterred by the selective disadvantages due to deceptive strategies of certain macromolecules.

2. Elaborations of the model

Elaborating this basic model, a number of biochemical and molecular evolutionary processes can be viewed in a new light. If there is no action by the receiver, for example when the sender gene fails to send a signal, perhaps due to pseudogenization, then there will be no subsequent utility, and both sender and receiver will drift neutrally. While this will be observed as drift through sequence space, more accurately the sender is drifting through signal space and the receiver is drifting through action space.

Typically, a single action may be expected from a receiver protein on receipt of a certain signal. This relation arises because most proteins have a single major binding partner, and usually only one outcome from the interaction. However, there might be multiple senders, for example for a receiver protein that binds multiple proteins, and in a similar manner there might be multiple receivers for a single signal protein, if that protein possesses multiple binding partners. Metabolites that bind to receiver proteins can be viewed as signals produced by the enzymes responsible for their biosynthesis, and the genes encoding the enzymes are the ultimate sender. Thus, promiscuous proteins with multiple metabolites as substrates receive multiple signals from multiple senders. The framework also encompasses other macromolecular interactions, for example protein-to-nucleic acid, and nucleic acid-to-nucleic acid. The key feature is that the receiving macromolecule is able to distinguish signals, and that the molecular signals are ultimately derived from a sender gene.

The expression level of a gene may be thought of as constituting the signal intensity, usually considered as proportional to the signal cost [22]. Posttranslational modification of a signal protein is then a form of expressional control, and so results in a modification of the signal intensity, via modification of its three-dimensional conformation. Posttranslational modifications that abolish the binding of a receiver protein to a ligand result in dismissal of the receiver from the game. The complexity of the signal contributes to its intrinsic cost as it increases the difficulty of copying or mimicking the signal; this may be represented by the tertiary structure of the expressed gene product.

This reasoning leads to a novel explanation for the similarity in the physicochemical properties of several amino acids in the genetic code (isoleucine and leucine, for example). Such apparent functional redundancy may hide a subtle purpose, that of ensuring sufficient complexity in tertiary structure, thus helping to ensure the integrity of signalling. Subtle differences in physicochemical properties have utility in ant cuticular hydrocarbons (CHCs). These are pheromones in the ant cuticle responsible for the regulation of complex behaviours associated with eusociality, including the assignment of ant colony group identity [23]. For example, changes in CHC stereochemistry can lead to altered signal recognition [24].

Another means of maintaining signal integrity is that of ‘misinteraction avoidance’, which refers to the need to avoid detrimental misrecognition of a protein by other proteins [25], and has some cost in terms of increased sequence constraint. Costly signalling is one mechanism to avoid deception, discussed further below. The integrity of a signal is also affected by the quantity of noise present. In biochemical signalling games, signal noise results from the occurrence of phenotypic mutations during the expression of a gene (a ‘phenotypic mutation’ is a transcriptional or translational error [26]). Finally, within a genome there are multiple senders (genes), and the number of senders correlates with genome size. Genome expansion is typically accompanied by gene duplications [27]. Sender gene duplication leads to increased signal dosage, and subsequent signal ‘neofunctionalization’. Sender and receiver gene duplication is discussed in detail in electronic supplementary material, S1.

3. The rise of molecular complexity brought the first games

To have a living system biochemistry is necessary, but biochemistry by definition only occurs within living systems. This apparent logical circularity presents the classic chicken in the egg (in the chicken, etc. ad infinitum) scenario regarding the origin of life; which came first: biochemistry or life, or did they both co-emerge concurrently? Most origin-of-life scenarios envisage the early emergence of self-reproducing macromolecules, which also possessed the ability to selectively bind other molecules. With the large size and complexity of macromolecules comes specificity, and with specificity comes choice, consisting of the ability to ‘choose’ which molecule to bind to, and to be able to distinguish or recognize one molecule from another. Specific physical interactions between two macromolecular gene products allow identity information to be transferred and so is a form of communication; in this manner the first signalling games would have occurred.

‘Behaviour’ may be broadly defined as ‘the way in which one conducts oneself, especially towards others’ (Oxford English Dictionary), and is most commonly used in connection with organisms, human beings and complex systems. The origin of behaviour has been attributed to the rise of complex macromolecules [28], and we extend this idea by emphasizing that a key feature of early life was the playing of signalling games by these replicating macromolecules, resulting from their ability to exert selectivity and engage in molecular recognition, which would have given rise to identity information and information asymmetry between sender and receiver. We propose that these features, coupled with replication dynamics, distinguish biochemistry from abiotic chemistry. From the origin of the earliest replicator macromolecules, the evolution of life has involved a series of major transitions [29]. Each of these has involved an increase in complexity, resulting from new cooperative relationships, promoted by the establishment of signalling conventions and cellularization to promote a Shapley value. The Shapley value describes the utility (or benefit) produced in a cooperative game and its distribution among players [30]. These are discussed in more detail in the second part of the paper.

In signalling games, if both players have perfect common interest, meaning that they both prefer the same outcomes in all situations, then honest signalling is at an equilibrium [22]. However, if the players do not possess perfect common interest, then maintaining informative signalling is a problem. Deception is expected to occur where there is information asymmetry, in the absence of perfect common interest [31], and the level of deception in signalling games is expected to increase as common interest diminishes [32]. This situation arises because it is often in the interest of the sender to emit a less than informative signal to the receiver in order to elicit an action that will benefit the sender, but not necessarily the receiver: these are deceptive signals. When macromolecules are not co-inherited in concert, such as those encoded by selfish elements or pathogens, then there will be a degree of competition and a conflict of interest. In these cases, the macromolecules engage in competition without being perfectly cooperative, as each ‘strategizes’ to maximize its utility at the expense of the other. Here, a level of molecular deception will be expected as a utility maximizing strategy on the part of the sender. An extensive form decision tree illustrating honest signalling when players have perfect common interest (Lewis signalling game), and the occurrence of deception where there is a conflict of interest between sender and receiver is shown in figure 1. The mathematical formalism behind deception is described in electronic supplementary material, S2. Signalling games where there is a conflict of interest between gene players within an individual genome are discussed next.

Figure 1.

Extensive form decision tree for a macromolecular signalling game. The sender (S) has two potential types assigned by nature (open circle): cooperative (T_C) and defective (T_D). Two signal molecules may potentially be sent by the sender, c (sender type is cooperative) or d (sender type is defective). The receiver (R) has two potential actions, trust (R will bind the signal molecule, and consequently undertake some biochemical process) and reject (R does not bind the signal molecule). The utility payoffs are in brackets (S,R). Dotted lines indicate two nodes that R may arrive at; however, R is unable to distinguish between them due to the information asymmetry between S and R. The Nash equilibrium for the sender of type T_C and R is indicated with an asterisk. Costs for sending a signal are not included. An illustrative example is provided by the epidermal growth factor gene (sender), which expresses epidermal growth factor (EGF, the signal), which interacts with epidermal growth factor receptor (EGFR, the receiver). The payoff from the separating equilibrium describing the interaction of EGF with EGFR (i.e. signal with receiver) is provided by the loss of fitness when the EGFR gene is deleted. This could be indirectly measured by determining the level of evolutionary constraint on the gene, using a measure of negative selection such as the K_a/K_s ratio. Alternatively, a direct measure would involve assessing the fitness loss associated with entire deletion of the gene. Deceptive signals to the EGFR receiver may be sent by some viruses, in the form of virokines: some of which mimic the structure of EGF [33]. The payoff to the virus is the fitness gained by the expression of virokine. The cost to the host is the fitness loss experienced by expression of the virokine. In this scenario, it is assumed that the virokine binds only to one receiver, EGFR. In addition, potential therapeutic drugs are designed to specifically interact with EGFR by mimicking the structure of EGF [34,35]; these also constitute deceptive signals. Here, the ultimate sender is the physician.

4. Intragenomic conflict, conflict of interest between sender and receiver, and molecular deception

Lewis signalling games are simple, because they entail perfect common interest between macromolecules and their genes, i.e. players. However, there exist more complex scenarios where genetic elements within the genome may be in competition, and to an extent possess interests conflicting with the host genome [36]. Such elements are described as ‘selfish elements’, and are characterized by an ability to self-replicate [37], thus foiling perfect reproductive coordination with the rest of the genome. The suggestion has been made that selfish elements may be involved in evolutionary games [13], and in particular meiotic drive has been examined from an evolutionary game theoretic perspective [38]. These ideas may be extended by proposing that many selfish elements are engaged in deceptive signalling games with the host. Signalling games can be used to understand why molecular deception is important for selfish elements, and in turn we suggest that the occurrence of molecular deception is diagnostic of selfishness and intragenomic conflict.

The suggestion has been made that if transposons are selfish, then they would be expected to engage in a game of ‘cat and mouse’ with the host, attempting to disguise themselves in order to avoid detection [39,40]. Electronic supplementary material, table S1, contains a list of selfish elements, and the molecular mechanisms that they use to deceive the host genome. For a number of selfish elements such as transposons, there are multiple lines of evidence supporting their selfish nature, while for other systems their selfishness is not so clear, for example toxin–antitoxin systems [41] and restriction–modification systems [42]. Demonstrating the presence of molecular deception mechanisms in these systems could be taken as evidence for their selfish nature.

Mitochondria have been proposed to engage in intragenomic conflict with the nucleus (in game theory terms, the ‘resource holder’), due to their asymmetric maternal mode of inheritance, a manifestation of this being cytoplasmic male sterility (CMS) in plants [43]. CMS refers to the failure of male plants to produce functioning anthers, pollen or gametes, and is caused by a maternally inherited extranuclear genome (mitochondrion or chloroplast). Our theory of biochemical signalling games suggests that if the mitochondrial genome is truly in conflict with the nuclear genome, then macromolecules encoded by the mitochondrial genome should use some level of molecular deception in their interactions with nuclear-encoded macromolecules.

Consistent with this hypothesis, the prediction has been made that the unknown mitochondrial-encoded RNA targets of the nuclear-encoded fertility restorer (Rf) plant genes should use sequence variability to escape recognition. Such a strategy would promote CMS, which is in the interests of the maternally inherited mitochondrial genome [44]. A popular hypothesis is that the formation of the nucleus was stimulated by endosymbiosis of the mitochondrial progenitor during the process of eukaryogenesis [45]. In this scenario, nuclealization could be seen as a response to genetic conflict. This approach to improving the Shapley value would involve mitigation of deceptive signalling by the transfer of endosymbiont genes to the nucleus, thus cementing common interest, given the resulting synchronization of replication. A question then is why the mitochondrion would retain a vestigial genome; a potential functional explanation is that there is a regulatory advantage in local expression of genes encoding components of the electron transport system [45].

Genetic conflict between two parents is termed ‘sexual conflict’ and often concerns reproduction [46]. Given only partial common interest between a pair of mates, usually due to the greater resources devoted to parental care by the female, then we would expect deceptive signalling to occur at the organismal level; and indeed this is well described in animals [47]. Intragenomic sexual conflict occurs when there is an interaction between antagonistic loci, termed ‘interlocus sexual conflict’. In these scenarios, molecular deception would be expected to occur between the gene products of the conflicting loci. If the gene products are proteins, a feature may be an elevated K_a/K_s ratio, characteristic of positive selection and molecular arms races. A classic example of a molecule engaged in a molecular arms race is the IgG molecule which often shows signatures of positive selection in its antigen binding domain, with an elevated proportion of non-synonymous substitutions compared to the rest of the molecule [48]. This is indicative of competition with pathogen antigen molecules, which co-evolve to evade molecular recognition by the IgG molecule: this scenario provides a classic example of molecular deception and its sequence signature.

Parent–offspring conflict results from misaligned interests between a parent and offspring, which in mammals typically concerns resource allocation by the mother to its offspring [49]. One illustration is the proposed conflict between maternal genes and fetal genes for resources during pregnancy. Its resolution has been linked to gene imprinting, which the mother may use to ‘stifle’ fetal genes [50,51]. Our biochemical signalling theory would lead to a hypothesis that the fetus should engage in molecular deception with the mother, in order to obtain more resources. In particular, the expressed macromolecules of the imprinted fetal genes should use a degree of molecular deception in their interactions with maternal macromolecules. Sure enough, there is evidence for evasion of the maternal immune system by the fetus [51]. In addition, an offspring's manipulative behaviour towards the mother in order to extract more resources (such as vocal begging behaviour, i.e. bleating) may exhibit conflict between maternal and paternal genes acting within the nervous system [51]. Here again, an element of molecular deception would be expected if there is a physical interaction between the gene products.

Allelic exclusion refers to the suppression of an allele by epigenetic mechanisms [52], and is often involved in genetic conflict. A signalling game interpretation of allelic exclusion is discussed in more detail in electronic supplementary material, S4. Finally, cancers are selfish entities whose interests are in direct conflict with the soma. Therefore, we should expect that cancer progression should exhibit elements of molecular deception. One example of this is the evasion of immune surveillance by tumours. There are many mechanisms by which this can occur; one is by the overexpression of the protein PD-L1 on the tumour cell surface, which binds to the PD1 receptor on T cells when the T cell is bound to the tumour, signalling to the T cell to inhibit proliferation [53].

5. ‘Junk’ DNA is selfish if it deceives

There has been much discussion over the role of ‘junk’ DNA, as to whether it serves an adaptive benefit to the organism, provides no discernable effect on fitness, or is largely selfish [54] (selfishness implies harm to the host [55,56]). While independent replication of a genetic unit is a pre-requisite for selfishness, its occurrence does not automatically imply a loss of fitness in the host genome in which it resides and additional criteria are necessary to determine if a selfish element is truly harmful. We propose that the occurrence of molecular deception indicates a conflict of interest with the host genome, and that such conflicts can be used as a criterion for judging the selfishness of portions of ‘junk’ DNA.

A large proportion of ‘junk’ DNA in metazoans is composed of transposable elements (TEs), including transposons and retrotransposons [57], and have been considered classic examples of selfish elements [37]. However, it has been unclear how much harm occurs to the host, or if they have little effect on host fitness [58]. A good example of a deceptive strategy used by TEs is found in one method by which they propagate themselves—namely, by homologous recombination. This strategy involves inducing mismatches between homologous chromosomes at the flanking TE repeat sequences during meiosis, and is an example of subversion of sexual reproduction, one of the major evolutionary transitions, discussed further below.

TEs display a range of additional molecular deception strategies, listed in electronic supplementary material, table S1, as would be expected if they are harmful to the host genome. Given the deceptive behaviour of these TEs, they can be viewed as truly selfish, as their deceptive strategies indicate a conflict of interest with host fitness. These examples provide additional evidence that TEs are truly parasitic, and harmful to the host. Using the lack of sequence conservation in whole genome alignments between species as an indication of the lack of ‘functionality’ of ‘junk’ DNA [59] may therefore be something of a misconception, as many TEs evolve and proliferate rapidly within lineages, and are usually excluded from such analyses. Recent discussion of the functionality or otherwise of ‘junk’ DNA seems to envisage ‘functionality’ as something that only contributes to the fitness of the host (e.g. [54]), while neglecting the possibility that the ‘functionality’ of a significant proportion of junk DNA may actually be dedicated to the welfare of TEs and other selfish elements, at the expense of the host. The monotonic view of ‘functionality’ that it is purely for the benefit of the host genome ignores the widespread occurrence of intragenomic conflict, of which the large TE component in the human genome may be evidence.

6. Molecular mimicry

A characteristic of macromolecular mimicry is that while there is conformational similarity between the mimic molecule and the model molecule, there is no common ancestor between the two, which suggests that the molecular mimic has evolved its conformation independently. There are a number of examples within the cell, often involving informational processes such as transcription and translation [60]. Molecular mimicry that occurs between cooperating components of the genome (cellular molecular mimicry) could be considered a form of self-deception given that the macromolecular mimic and receiver molecule (dupe) typically have perfect common interest. This strategy implies that cellular molecular mimicry must bring some advantage to the genome, indicating that it is a more efficient system than alternative molecular solutions. For example, elongation factor P appears to have evolved tRNA mimicry in order to optimize interactions with the ribosome [61]. In this case, the description of the receiver, the ribosome, as a ‘dupe’ per se is something of a misnomer.

Pathogens also commonly use molecular mimicry, to deceive the host [62]; this strategy may be referred to as true molecular deception, as opposed to self-deception, as it exerts a cost on the host. Likewise, while there is currently a lack of clear examples of selfish elements that use molecular mimicry, from a signalling games perspective they would be expected to also use this strategy. True molecular deception exhibited by pathogens is a form of mimicry where there is a ‘model’, ‘mimic’ and ‘dupe’ molecule, with the model and dupe molecules being expressed by the host. In the case of true molecular mimicry, biochemistry can allow the quantification of the deception via binding constants. The affinity of the molecular mimic for the binding site of the receiver molecule, usually occupied by the model molecule, can be used as a measure of the strength and apparent integrity of the signal. This viewpoint allows us to consider binding constants as measures of molecular ‘dishonesty’. As is the case in classic organismal mimicry [63], a selection pressure would be expected on the model macromolecule to evolve differences with the mimic macromolecule.

Cellular molecular mimicry bears similarity to Müllerian mimicry. Müllerian mimicry classically involves two species (senders) that evolve similar warning markings to indicate toxicity to potential predators (the receiver). The receiver is not truly a ‘dupe’, as the signal of toxicity is an honest one, thus both sender and receiver have a common interest in avoiding predation. By contrast, in Batesian mimicry, a non-toxic mimic species will copy the warning markings of a toxic species, deceiving potential predators into supposing that it is also toxic. Here, the predators are true dupes, as they are deceived by the sender resulting in a loss of fitness. Because the components of the cell generally share perfect common interest, each macromolecule involved in cellular molecular mimicry gains by converging on the same signal (i.e. three-dimensional structure), as in Müllerian mimicry. The ‘two-step’ hypothesis of Müllerian mimicry proposes that an initial species with an existing signal is then mimicked by a second species via a large mutational leap, and from there more incremental change [64]. In cellular molecular mimicry, a similar process may be envisioned whereby an initial molecule with a unique structure (e.g. a tRNA) was then imitated by a second molecule that assumes a similar structure by chance initially, with both structures then converging onto each other.

In classical Müllerian mimicry, ‘mimicry rings’ may arise: these consist of numerous co-mimics, an example being the bee–wasp mimicry ring, whereby diverse species of bee and wasp (which all have a venomous sting) have converged on the same warning signal of harmfulness, black and yellow body markings (figure 2a). Likewise, at the molecular level, taking the example of tRNA, there are a number of molecules within the cell with tRNA-like structures. In prokaryotes, these include elongation factor P, ribosome recycling factor and release factors 1 and 2. These constitute a ‘molecular mimicry ring’, as they share a common signal, an L-shaped three-dimensional structure, and they all interact with the same receiver, the ribosome (figure 2b), sharing common interest. Determining the original model molecule in the tRNA molecular mimicry ring (whether it was tRNA itself, or one of the tRNA-like protein molecules), and the order of addition of members of the ring, would reveal insights into the evolution of protein translation. An analogy to Batesian mimicry is provided by the turnip yellow mosaic virus (TYMV) tRNA-like structure, which binds to the ribosome (receiver; figure 2b), apparently promoting viral fitness [65]. This behaviour represents true molecular deception as it results in a loss of fitness of the receiver. For comparison, an example of classical Batesian mimicry is provided by species that mimic the black and yellow warning markings of bees and wasps, even though they are not toxic themselves, such as the drone fly (figure 2a).

Figure 2.

Animal and molecular mimicry rings. Two mimicry rings are shown: (a) an animal mimicry ring, the bee–wasp mimicry ring; (b) a molecular mimicry ring, the tRNA mimicry ring. In the bee–wasp mimicry ring, the signal consists of black and yellow body markings, and the receiver is a potential predator such as a bird. Both bees and wasps have converged on the black and yellow signal, which they share as a warning to the predator, both bees and wasps being toxic. The predator only needs to learn the one warning signal, and consequently avoid both bees and wasps. Sharing a single signal is also more efficient for the bees and wasps, as predators have less to learn and so are less likely to make a mistake. The signal is exploited by Batesian mimics, which benefit from deterring potential predators, but at a cost to both bees and wasps, and the potential predator, by diluting the impact of the signal. In the tRNA molecular mimicry ring, the signal is the L-shaped tertiary structure, and the receiver is the ribosome. When the signal is shared for common benefit, this is Müllerian-type mimicry (indicated with blue arrows), but when a signal is used for deceptive purposes, then this is Batesian-type mimicry (indicated with red arrows). Müllerian-type molecular mimicry benefits all tRNAs and tRNA mimics encoded by the cellular genome, as it results in efficient protein translation, which increases the common good (or common wealth). However, Batesian-type molecular mimicry by pathogens is harmful to the host cellular genome as it reduces the efficiency of protein translation through competition for the receiver ribosome. The image details may be found in the electronic supplementary material.

Trivers' theory of self-deception in humans proposes that individuals deceive themselves in order to more effectively mask the signs of deception from other individuals who are in competition [66]. The cellular molecular mimicry described here differs as deception occurs within the cell between cooperating elements, and so bears more similarity to models of self-deception that confer an intrapersonal benefit and lower cognitive costs [67]. This process succeeds because cellular molecular mimicry may reduce costs by presenting a molecular shortcut. While Trivers' theory, like some evolutionary ideas in psychology, remains to be widely adopted, we use it to illustrate a potential and logical connection between human and molecular behaviour, and expect additional parallels between the two types of behaviour to be drawn in the future.

7. Molecular sanctioning

There are two ways of maintaining honest signalling when there is a conflict of interest: costly signalling, and the use of penalties and threats (credible or otherwise) [68]. In human society, informal penalties for dishonesty may include mechanisms of social opprobrium, such as shaming, shunning, gossip and stigmatization. More systematic mechanisms for maintaining honesty (in contracts, for example) include legally based sanctioning. The ubiquity of deceivers necessitates the implementation of scanning or probing systems to maintain probity. Thus, ‘probing’ is a behaviour predicated by the presence of cheaters in a signalling system [69], and is analogous to the presence of scanning mechanisms at the molecular level.

Penalties at the molecular level include a range of sophisticated mechanisms dedicated to the detection (probing) and subsequent destruction (sanctioning) of non-self macromolecules which may be associated with pathogens, but also aberrantly replicating cells, such as tumour cells. ‘Non-self’ may be defined as a genetic element that does not replicate in concert with the main genome, and so acts selfishly, at the expense of the main genome, but still rationally in order to increase its own fitness. A genetic element that cooperates with the main genome is also acting selfishly, but will improve its fitness more by cooperation than by acting individualistically and inflicting cost on the main genome. Such cooperating elements will be regarded as ‘self’ by the main genome.

In addition to the ability to identify non-self macromolecules, an added advantage would be the capacity to distinguish those non-self macromolecules that present credible threats, as opposed to non-credible threats. An example lies in the ability of toll-like receptors of the immune system to identify categories of macromolecules that are widely shared by pathogens [70]. Biochemical signalling games can present some intriguing parallels with the signalling games that occur in human society. For instance, the concept of molecular self and non-self has parallels in the phenomena of human homophily and tribalism, where a myriad of signals and cues are used to identify adherence to a particular in-group, the function of which appears to be to cement cooperation between individuals [71]. This role is similar to that of the molecular differentiation of self from non-self.

If a deceptive signal is used frequently, then it will encourage the evolution of detection mechanisms by the receiver, and if there is a high rate of detection of a deceptive signal, then it ameliorates the value of the deceptive signal to the sender [72]. Within the cell, there are a range of scanning mechanisms to ensure fidelity and to prevent subversion. Mechanisms for ensuring fidelity, of the genetic code for example, most likely evolved as a way of maintaining signal integrity, thus improving fitness. Such mechanisms include aminoacyl-tRNA synthetase proofreading, ribosome tRNA selection and nonsense-mediated decay of mRNA. These mechanisms may also have the added benefit of helping prevent subversion by selfish elements or pathogens. Molecular proofreading mechanisms improve the efficiency of receiver signal assessment (binding). In addition, there are mechanisms whose major function is that of scanning for selfish elements and pathogens. One example is that of DNA methylation, which functions to suppress both transposon [73] and retrovirus [74] expression. Another is that of the immune system, which monitors the body for pathogens, but also selfish entities such as tumours (i.e. tumour immune surveillance), destroying or inhibiting anything identified as ‘non-self’. Such inhibition or destruction may be regarded as a form of punishment or molecular sanctioning.

8. The genetic code is a signalling convention

Conventions are regularities that have become accepted in a society, such as the meaning of different traffic light colours, or the arrangement of letters on a keyboard. The genetic code can be considered a signalling convention between genes (senders), their mRNAs (the signal) and the set of tRNAs in the cell (receiver). The signal encapsulated by a particular mRNA consists of its ribonucleotide sequence, but also to some extent its conformation, as mRNAs can assume tertiary structures that may influence their translation. It has been postulated that the game's origin might have been in the primordial RNA world taking place among mRNA and tRNA precursors in response to invasion by prebiotic amino acids; they might have created codon–anticodon signalling conventions to string the amino acids in a way that avoids toxic forms such as homopolymers or random concatenation. The resulting action of the receiver is the production of a peptide, and the utility of the signal lies in its benefit to fitness of the whole organism, to both sender and receiver [75].

The genetic code is a signalling convention because a particular signal (mRNA) is always associated with a particular action (protein; figure 3). This convention leads to a so-called separating equilibrium and indicates common interest between sender and receiver, which is unavoidable once they have been cellularized. The genetic code is near universal, and stable (with a limited number of minor deviations), these are also characteristics of conventions. By contrast, a non-separating (pooling or babbling) equilibrium refers to scenarios where signals are random (cheap talk), and the action is the same no matter what the signal, while a semi-separating or partial pooling equilibrium is where a proportion of senders choose the same message.

Figure 3.

The genetic code as a signalling convention. The mapping of amino acids to codons represents a signalling convention, whereby the sender is a gene, the signal is an mRNA, and the receiver is the tRNA set (each tRNA being charged with an amino acid). The tRNA set receives the mRNA signal, and determines an action depending on the sequence of the mRNA, which involves the synthesis of a peptide chain. Multiple senders occur in genomes as there are multiple protein coding genes, but there is only one receiver, the tRNA set specific to that organism.

The process of establishing a separating equilibrium requires honest signalling [22]. The handicap principle proposes that when there is a conflict of interest, an investment of resources into a signal by a sender assures the receiver of the honesty of the signal [7,76–78]; this state of affairs is termed ‘costly’ signalling. At the molecular level, one can imagine a scenario where a receptor protein evolves low affinity for a ligand (the signal), so that it can only be effective in high concentration, thus exerting a cost on the production of the signal. This strategy would constitute a costly signal, ensuring its ‘honesty’. A potential mechanism to generate a costly signal is presented by epidermal growth factor receptor, overexpression of which is often linked with cancer functionalizing oncogenes. The receptor is a monomer that dimerizes upon binding of a ligand to one receptor, allowing the binding of a second ligand to the second binding site of the dimer, but with reduced affinity, known as negative cooperativity. Negative cooperativity, whose function remains to be established [79], could potentially act as a form of signal verification system to avoid false activation of the receptor, establishing signal honesty.

Interestingly, work by Bergstrom et al. [22] shows that low cost signalling equilibria can become established despite a conflict of interest between sender and receiver. A signalling games perspective of molecular evolution asks how the separating equilibrium represented by (amino acid ⇒ codon) mappings of the genetic code became established [75]. A pre-existing cooperation between senders (genes) and the receiver (tRNA set) would facilitate the establishment of the separating equilibrium because honest signalling facilitates the establishment of separating equilibria, and honesty is at an equilibrium where there is perfect common interest between sender and receiver [75]. While the regularity of a signalling convention may be arbitrary, they appear to become established from a precedent often deriving from salience, and from there the convention is self-sustaining, as unilateral deviation by either sender or receiver is detrimental [20].

In a parallel, the Frozen Accident theory of Crick asserts that the amino acid codon mapping of the genetic code is essentially arbitrary (an ‘accident’), but once it had become fixed, it has become essentially immutable and self-sustaining (‘frozen’); this hypothesis holds because any change to the (amino acid ⇒ codon) mappings of the genetic code is typically catastrophic to fitness [80]. In a Lewis signalling game, it is in the interest of both sender and receiver to ensure fidelity, as their utility resulting from the action of the receiver is equal. This discussion helps to explain the rarity of deviations to the amino acid codon mapping of the standard genetic code in cellular genes, such as programmed frameshifting and stop codon read through. However, as soon as an independently replicating unit such as a selfish element or pathogen arises, then perfect common interest is no longer present and subversion of the genetic code signalling convention would be expected to occur, as follows.

9. Pathogens and selfish elements subvert the genetic code signalling convention

Given the view of protein translation as a signalling game with almost perfectly aligned common interest between senders and receivers, it follows that when there is a conflict of interest, the signalling convention could be expected to become subverted, via molecular deception. Some transposable elements subvert the genetic code signalling convention, by stop codon readthrough (retrotransposons, bacterial insertion sequences) and frameshifting (DNA transposons, retrotransposons, bacterial insertion sequences) (electronic supplementary material, table S1). When such a deception occurs, due to its subversive nature, it can be taken as an indication of selfishness, i.e. a conflict of interest with the host. Subversion of cell signalling by pathogens is common [81], and deception by the pathogen is predicted from the non-overlapping interests of host and pathogen, from a signalling games perspective. The many examples of viral hijacking of the translation apparatus involve molecular deception, whereby the genetic code signalling convention is subverted. Examples of subversion of the genetic code by RNA viruses include stop codon read through, frameshifting and the use of alternative translation start sites (reviewed by Firth & Brierly [82]). By contrast, there are only a few examples of genetic code subversion by DNA viruses: frameshifting has been reported in DNA tumour viruses [83], and a potential case of stop codon read through in cytomegalovirus [84].

Given the parasitism–mutualism continuum, whereby microbial pathogens with high initial virulence are expected to gradually attenuate their harmful effects on the host, and in some cases establish mutualistic relationships, over evolutionary time [85], this kind of symbiosis would be expected to be accompanied by a reduction in molecular deception by the microbe, as a higher level of common interest is established. Thus, whether a symbiotic microbe is truly a parasite or not (as in the case of Wolbachia, for example) [86] may be tested by assessing the presence and level of molecular deception used by the microbe. Likewise, it has been proposed that a pathogen's virulence correlates with the degree of horizontal transmission it experiences, which decouples its interests from that of the host [87]. If true, this should also be correlated with the level of molecular deception it uses.

10. The major evolutionary transitions necessitated new signalling conventions

We have discussed how the origin of the genetic code involved the establishment of a separating equilibrium. However, the onset of coding was only one of several major evolutionary transitions that led to present day organisms (table 1). Several of these transitions involved information storage and transmission [29]. In a parallel, code biology addresses the establishment of a range of molecular codes, many tied to evolutionary transitions [88]. In each of the transitions, independent units pooled themselves with other units. Typically, this acted to synchronize replication, inducing common interest, and so cooperative behaviour. An example discussed above is the slow transfer over evolutionary time of mitochondrial-encoded genes to the nucleus, which would have the effect of reducing genetic conflict.

Table 1.

The major evolutionary transitions and their signalling conventions. The major evolutionary transitions are outlined by Szathmary & Maynard Smith [29]. Here, the signalling convention(s) associated with each transition is identified, and the types of subversion that may occur.

evolutionary transition	type of cooperation	signalling convention	type of subversion
replicators	between monomer subunits, forming polymers that promote common interest of the monomer constituents	the first replicators are unknown, but would have used molecular specificity	parasitism of replicator function may have occurred by other polymers
protein translation	between mRNAs and the ribosome	genetic code	‘deceiver’ tRNAs ('Batesian' tRNA mimics) and ‘deceiver’ mRNAs (mRNAs that benefit the sender gene, but not the host genome)
eukaryogenesis	between nucleus and mitochondrion	nuclear targeting signals, mitochondrial targeting signals	bacteria can use nuclear localization signals to gain entry to the nucleus (a Trojan horse strategy) [92]
	between mRNA and the spliceosome	intron splice sites	selfish elements can hide in or mimic introns (see electronic supplementary material, table S1), viruses sequester the splicing machinery to regulate their gene expression [93,94]
	between DNA and histones	histone code	some bacteria appear to modify the histone code [95]
sexual reproduction	between two genders	species specific chemical, visual and auditory signalling	many examples of deception due to sexual conflict [96]
	between two gametes	gamete fusion involves the HAP2 protein. Sperm-egg recognition is species specific	undescribed
	between two homologous chromosomes	homologous recombination is initiated by Spo11	B chromosomes mimic sex chromosomes leading to chromosomal drive (electronic supplementary material, table S1). Some examples of meiotic drive use molecular deception (electronic supplementary material, table S1)
multicellularity	between cells	has arisen a number of times independently facilitated by the evolution of cell adhesion and signalling, and immune systems	cancers use a variety of molecular mechanisms that disrupt normal cell–cell recognition [97], and evade the immune system [98]
eusociality	between related individuals	has arisen a number of times independently. Each has established different signalling conventions based on acoustic, visual and chemical signals	mimicry of acoustic signals [99] and pheromones [100]
humanity	between unrelated individuals	spoken language	lying

The establishment of signalling conventions and separating equilibria was critical in each of these transitions (table 1). This process implies that there was a reduction in the frequency and degree of deceptive signalling, and the resulting need for costly signalling and/ or sanctioning to maintain honest signalling. We note that the evolution of organismal signalling conventions has been linked to the concomitant reduction in the need for costly signalling [69]. Efficiency benefits have been posited as a driving force in the major evolutionary transitions [89]. The establishment of such separating equilibria in the evolutionary transitions would have brought the specific benefit of the utility derived from the newly established cooperative signalling game, with the concomitant promotion of a Shapley value, in addition to the reduction in the need for sanctioning/costly signalling. Implicitly, improvements in cooperation itself are linked to increased division of labour, which is viewed as leading to greater efficiency of the group [90]. Thus, a sharp increase in fitness would be expected to result from the establishment of a novel separating equilibrium allied with a new cooperative relationship. It has not escaped our notice that this would result in a sudden burst in evolutionary success, which may have explanatory power for some cases of punctuated equilibrium [91].

However, the establishment of signalling conventions also brought the possibility of subversion. We have discussed this possibility by using the example of the genetic code, and how the occurrence of subversion can indicate a conflict of interest. Table 1 lists examples of subversion of the signalling conventions that are associated with the major evolutionary transitions. Many (but not all) of the deceptive strategies listed here are molecular in nature. Thus, when we assess the occurrence of deceptive signalling in the natural world, molecular deception becomes an important category that has hitherto been overlooked in the context of signalling game theory. An example of the breakdown of a molecular signalling convention is provided by cancer, the occurrence of which is a consequence of multicellularity (table 1), discussed in electronic supplementary material, S6.

11. Signalling by macromolecules confers rudimentary behaviour

Signalling game theory has been widely used in evolutionary biology in order to better understand organismal behaviour. While most often applied to animals with a nervous system, it has also been applied to cancer [101], individual cells [102] and microbes [103,104]. Signalling can occur by vision, touch, sound, temperature, odour and other types of chemical signalling. Most relevant to this work, the basis of odour recognition is mediated by macromolecular olfactory receptors which exert molecular selectivity on volatile ligands. Molecular signalling mediated by odour can occur within or between species, and odours may present themselves as honest or dishonest signals. An example of an honest signal mediated by odour is the pungent smell emitted by porcupines, which is a warning as to the consequences of attacking the spiny animal, which would have harmful consequences for both the porcupine and the predator. An example of a deceptive signal mediated by odour lies in the Rafflesia flower, which emits an aroma of rotten meat, thus attracting insects that normally feed on carrion, promoting pollination. There are many examples of molecular trickery and deceptive signalling that occur between species, such as between hosts and pathogens, and predator and prey. Deceptive signalling can be also within a species, between competing individuals. However, here we have focused on molecular signalling games that occur below the level of the cell.

Clearly, the type of behaviour and response exhibited by macromolecules is expected to be simpler than that of the entire organism, as the range of potential responses exhibited by an organism are larger and more complex. In particular, learning is important for more advanced signalling games than those described here. Memory is derived from modification, and is a basis of cooperation between individual organisms. Learning is difficult for an individual molecule to display, although direct chemical modification or gene methylation may offer a ‘learned’ modification of response, of an individual genetically encoded element, within the lifetime of the cell. However, learning can occur on a global level via an increase in numbers of genetic elements in a population that possess a successful strategy [105]. A key difference with animal learning is that this would not occur during the lifetime of the molecule, but would operate through the progeny of the sender gene, in a similar process to how organisms converge on strategies in evolutionary games [4], so producing a model of the world.

Rather than a form of anthropomorphism, we note that descriptors associated with behaviour are frequently attributed to macromolecules. For example, ‘hijacking’ [106], ‘moonlighting’ [107], ‘scavenger’ [108], ‘promiscuous’ [109] and ‘enforcer’ [110]. We suggest an extension of such terminology to indicate the behaviour of macromolecular signals from sender genes within signalling games as ‘honest’ and ‘dishonest’. This is distinct from deceptive molecular signals sent on behalf of the entire organism, as in the example of Rafflesia; in this work, we describe how individual macromolecules and their genes are players. The description of ‘honest’ or ‘dishonest’ does not imply epistemological awareness by the players, but the strategies and outcomes are the same as if they were; this terminological convention is commonly adopted when describing signalling games at the organismal level. In this work, we have attempted to promote a unified theory for the role of deception in evolution, and so would expect it to have generative power in the form of testable hypotheses. Mathematics, which demystifies signalling game theory, describes universal forms, and so it is no surprise that hidden in their shadows lie multiple levels of elegant biological organization.

Data accessibility

Additional data are included in the electronic supplementary material.

Authors' contributions

S.E.M. and B.M. designed the research and wrote the paper.

Competing interests

We declare we have no competing interests.

Funding

B.M. was supported by National Cancer Institute Physical Sciences-Oncology Center grant no. U54 CA193313-01.