Simplification, innateness, and the absorption of meaning from context: how novelty arises from gradual network evolution
SSimplification, innateness, and the absorption ofmeaning from context: how novelty arises fromgradual network evolution
Adi LivnatDepartment of Evolutionary and Environmental Biology and Institute ofEvolution, University of Haifa, Mount Carmel, Haifa 3498838, [email protected] a r X i v : . [ q - b i o . P E ] M a y bstract The theory of interaction-based evolution argues that, at the most basic level of analysis, there is athird alternative for how adaptive evolution works besides a ) accidental mutation and natural se-lection and b ) Lamarckism, namely, c ) information provided by natural selection on the fit betweenthe organism and its environment is absorbed by non-accidental mutation. This non-accidental mu-tation is non-Lamarckian yet useful for evolution, and is due to evolved and continually evolvingmutational mechanisms operating in the germ cells. However, this theory has left a fundamentalproblem open: If mutational mechanisms are not Lamarckian—if they are not “aware” of the en-vironment and the macroscale phenotype—then how could heritable novelty be due to anythingother than accidental mutation? This paper aims to address this question by arguing the following.Mutational mechanisms can be broadly construed as enacting local simplification operations on theDNA in germ cells, along with gene duplication. The joint action of these mutational operations andnatural selection provides simplification under performance pressure. This joint action creates frompreexisting biological interactions new elements that have the inherent capacity to come togetherinto unexpected useful interactions with other such elements, thus explaining nature’s tendencyfor cooption. Novelty thus arises not from a local genetic accident but from gradual network-levelevolution. Many empirical observations are explained from this perspective, from cooption andgene fusion at the molecular level, to the evolution of behavior and instinct at the organismal level.Finally, the nature of mutational mechanisms and the need to study them in detail are described,and a connection is drawn between evolution and learning. Keywords:
Evolvability, learning, instinct, stereotypy, genetic assimilation, evolution of language,parsimony. 2 he problem is not to choose the correct scale of description, butrather to recognize that change is taking place on many scales atthe same time, and that it is the interaction among phenomenaon different scales that must occupy our attention. —Simon A. Levin, 1992.
The theory of interaction-based evolution [93] argues that the mutations that drive adaptive evolu-tion under selection are not local accidents occurring to the genome. Instead, they result from theaction of evolved and continually evolving complex biological mechanisms [93] and are thereforeaffected by genetic interactions across loci. It follows that mutation combines information fromalleles across loci and writes the result of the combination into one locus—the locus of the mu-tation [93]. The schematic figure that describes this nature of mutation (Figure 1a) is much likethat which would represent gene interaction and regulation, except that the outcome of the actionin this case is genetic change. “Mutation” here is broadly construed to encompass not only DNAmutations but also epigenetic changes.Moving to the population level, we see that the outcome of a mutational event in one generation—namely the mutation itself—can serve as an input into mutational events at later generations [93].Therefore, mutations create a network of information flow across the genome and through the gen-erations (Figure 2) [93]. This suggests at the outset a process by which the genome can evolve asa cohesive whole [109, 93].This view immediately affects how we conceptualize fundamental questions in evolution, suchas the question of the role of sex in evolution [44]. A layman’s intuition has been that, since naturalselection acts on individual variation, the vast number of different genetic combinations generatedby sex facilitates adaptive evolution. However, this answer has been incomplete from a theoreticalperspective because, just as sex puts together these combinations, it also breaks them down: theyare not heritable. However, if mutation is not simply a local accident, but instead encapsulatesa flow of information across loci, then although individual genotypes are transient, they can haveeffects on future generations through the mutations that are derived from them (Figure 1b) [93, 94],and the original intuition holds in some sense. Such information flow through mutation enables a3ituation where selection evaluates each individual as a complex whole , and information from thatindividual as a complex whole is passed on by mutations precisely in accord with the individual’sfitness [93] .Another such question is the nature of mutation. Recently, evidence has been accumulatingshowing that mutational events are complex and involve genetic information and biological mech-anisms [93]. From a traditional standpoint, these complex influences on mutation are seen ashappenstantial and do not command attention. In contrast, interaction-based evolution arguesthat they are at the heart of the evolutionary process.By putting together the problem of the role of sex in evolution, the question of the nature ofmutation and more, interaction-based evolution has put together many questions and observationspreviously disconnected and has raised multiple predictions and directions for future research [93].However, it has left a fundamental problem open. The traditional view takes random mutationto be the ultimate source of heritable innovation and creativity in evolution: random mutationinvents, and natural selection selects . However, if mutation is not accidental and never was, thenwhat is the ultimate source of heritable novelty?In particular, interaction-based evolution does not admit Lamarckism—it does not admit amechanism that senses a phenotypic need in multicellulars through interaction with the environmentand translates that need into the required genetic change. But if the influences on mutation arenot “aware” of the environment and the phenotypic need, then how could the ultimate source ofheritable novelty in evolution be anything other than random mutation? This paper will proposean answer, thus completing the replacement to random mutation at a conceptual level that startedwith the first interaction-based evolution paper [93].Inspiring, long-term efforts by Wagner and colleagues have shown that network-level evolutionis key to innovation (e.g., [106, 37, 38, 154]). To answer the question above, I will continue theseefforts in the direction of interaction-based evolution. I will propose here the following. i ) Noveltyarises from gradual network-level evolution. ii ) The phenotypic meaning of a genetic element is gradually absorbed from the network in the course of network-level evolution and is not bequeathed We are no longer restricted to the effective transmission of additive genetic effects. Even the evolvability approach [80, 155], which allows for the evolution of mechanisms affecting mutation [74,88, 1], still relies either implicitly or explicitly on accidental mutation at the origin of things, and assumes thatevolvability mechanisms are merely later add-ons to the core process of random mutation and natural selection, onesthat play a facilitatory but not a fundamentally necessary role.
4o it by a local genetic accident. iii ) Molecular cooption—i.e., the case where a preexisting geneticelement comes to be used in a new context—is not simply an outcome of stochastic events but is anoutcome of a gradual process of network-level evolution, where non-accidental mutations pave theway and predispose the genome to cooption. iv ) This process of gradual network-level evolutionand the fact that the phenotypic meaning of a mutation comes from context rather than arisesanew based on a specific function per se also explain the evolution of innateness, previously knownas the problem of the “inheritance of acquired characters.” In this connection, we will see thatautomatization is at the essence of the evolutionary process. v ) Simplification and complexity areconnected: While selection puts a pressure for organismal level performance, there exists in additiongenetic simplification pressure due to mutational and recombinational mechanisms. Together, thepressures for performance and simplification drive the evolution of complexity and novelty, surpris-ingly connecting simplicity and complexity. In particular, elements simplified under performancepressure are expectedly unexpectedly useful: they have the inherent capacity to come togetherin interaction with other such elements and thus become useful in unexpected, novel ways. Thisinherent ability, which accounts for cooption, is the source of novelty in evolution. vi ) Evolutionis driven at the molecular level by evolved and continually evolving mutational mechanisms thatimplement useful operations, much like Hebbian learning and other non-random operations arethought to be useful in learning. A search for these mutational mechanisms, both empirical andtheoretical, needs to begin.The paper is organized as follows. The next section will describe the nature of network-levelevolution. Section 3 will introduce the idea of simplification under performance pressure. Together,these two sections will propose how non-accidental mutations could be useful for evolution yet benon-Lamarckian, and how novelty arises. Section 4 will then bring a large number of empiricalobservations in support of the view proposed here. These will be observations on the evolutionof behavior at the organismal scale. Of particular importance will be subsection 4.12.2, whereall the concepts developed will come together in an empirical example with an emphasis on theevolution of novelty. Finally, section 5 will revisit the molecular level in light of the conceptsdeveloped, discuss the nature of mutational mechanisms and draw a connection between evolutionand learning, including machine learning, thus underscoring the importance of the algorithmic lens[120, 73] for our understanding of evolution. 5
A contextual view of genetics
Due to the molecular biological revolution, it has become clear that the same or similar geneticelement can be seen in two or more different genetic contexts within the same species or in differentspecies [76]. This means that, over evolutionary time, a molecule can change the context in which itserves—it can be “coopted.” For example, the frog toxin caerulein has been independently cooptedfrom the homologous gastrointestinal peptide hormones cholecystokinin and gastrin, with whoseaction it interferes in the affected animals [124, 12]. And proteins involved in cellular stress response,like the small heat shock proteins [68], have often been coopted as light refracting crystallins inthe lens of the eye, an avascular tissue presenting harsh biophysical conditions [141]. Indeed,“Cooption,” “opportunism,” or “tinkering” [69, 59] is so important that it has been called “theparadigm of molecular evolution” [59]. But how does cooption happen? Does a genetic sequencejust jump one day by accident from one locus to another and acquires a new use?Traditional discussions admit but do not explain shifts in the context of usage of a geneticelement or a phenotype. In them, natural selection is limited to building up one independent oradditive contribution to fitness on top of another toward advancement in the same adaptation.This provides no explanation for cases where an element is first used in one context and then inanother, beyond saying that they are due to chance. This paper will begin to fill this gap, bydelving into the question of what makes it so that evolution is capable of producing building blocksthat, combined with other elements in a network, produce novel functionality.As will become relevant soon, we often see that fusion accompanies cooption. For example,members of the cyclophilin family, which have been found in bacteria, fungi, plants and animals[140], have a peptidyl-prolyl cis-trans isomerase activity which allows them to participate in diversebiological processes in all subcellular compartments, from protein translocation across membranes,to mitochondrial function, to control of transcription, and more (see [23] and references therein);and it is the presence of different additional domains in the different family members that specifiestheir unique localizations and interactants [23]. The next section will examine a particular fusioninvolving a cyclophilin family member, cyclophilin A.6 .1 Cooption at the molecular level is due to a gradual process
We will now see two motivating examples, one from molecular evolution and one from phenotypicevolution.Cyclophilin A (CypA) is a highly abundant cytosolic protein [60] that, among its various ac-tivities, potently binds several retroviral capsids, including HIV-1 [71]. TRIM5 is a restrictionfactor that recognizes and inactivates incoming retroviral capsids [146]. A copy of the
CypA gene has retroposed into the
TRIM5 gene independently in at least two different simian lineages[146, 116, 128, 92, 13, 168, 115], and the resulting TRIM5-CypA fusion protein appears to providestrong protection against certain lentiviruses [116, 128]. The curious nature of these independentfusions has been noted [146, 93]: not only is a repeated fusion event even more surprising from atraditional perspective than a repeated point mutation (there are many more possibilities of fusion,making repeated fusion by chance even less likely), there are many other
TRIM genes, and testsof artificial fusions of the CypA domain to some TRIM motifs have shown that they too can pro-vide retrovirus protection [172, 170, 171], yet
TRIM5 specifically repeats in both fusions mentionedabove [146]. Furthermore, it has been suggested that genetic factors have influenced the probabilityof the fusion, such as the extensive transcription of
CypA in the germline [71, 70, 146, 93].The current theory argues that this fusion (and others like it) was not due to a sudden, chanceevent, but rather was the culmination of a gradual genetic and phenotypic evolutionary process thatled to it. Minor genetic changes have accumulated, predisposing the genome to the appearanceof the fusion, and thus accounting for the fact that it appeared independently multiple times.Furthermore, I argue that
TRIM5 and
CypA interacted with each other prior to their fusion.Thus, the fusion did not cause
TRIM5 and
CypA to interact to begin with, but rather was led bytheir preexisting interaction.Furthermore, I hypothesize a specific mechanism that promotes such fusions. Two genes thatwork together in the soma in a particular context likely are transcribed at the same time. Becausethey may share cis elements and transcription factors that activate them, information indicatingthat they work together in the soma is likely present in the DNA and accessible in the germline,in particular to the transcriptional machinery. The two genes may be transcribed in the germlineat the same time, making it so that the chromatin will be open at both loci at the same time.7nd since reverse transcription occurs in the germline [15], it will be more likely to land a copyof one of these genes next to the other in the DNA. Other steps may further facilitate the fusion,such as trans-splicing prior to reverse transcription. Interestingly, the fact that transcription ispromiscuous in the germline allows any genes—somatic as well as germline genes—to participatein this mechanism [93].One may think that it just so happens that the genetic system allows for such mechanisms,or that they are fortuitous “accidents.” However, following [93], I argue that, rather than beinghappenstantial, mechanisms of this sort are of much significance. In particular, the mechanismabovementioned is reminiscent in a certain respect of Hebbian learning in neuroscience (StephenPacala, personal communications) . According to Hebbian learning [62], when one neuron per-sistently participates in causing another to fire, the strength of the connection between them isincreased, making it so that neurons “wire together if they fire together” [104]. Similarly, here, Iargue that copies of genes that are used together are fused together . Or, to be more precise, copiesof genes that are persistently used together in a new context are more likely to be fused. Note thatthis Hebbian-learning–like genetic operation is implemented by the mutational mechanism itself.This contrasts with a recent proposal involving Hebbian learning in evolution without invoking non-accidental mutation [160] and accords with the principle of interaction-based evolution, accordingto which the mutations relevant for adaptive evolution are non-accidental. Examples of cooption and fusion are also apparent at the phenotypic level. Consider the incitingceremony in ducks [101, 98, 100]. In the European common shelduck (
Tadorna tadorna ), when thefemale is standing near her mate, her aggression instinct is triggered by the presence of neighbors,and she may run toward them with her neck stretched, which is the threat posture in ducks [100].As she approaches them she naturally becomes fearful, turns around and flees back toward herdrake . Approaching her drake, the former instinct is triggered again. In those cases where herbreast is still facing him, she turns her neck back to threaten the neighbors over her shoulder. Thisbehavior by the female can incite her mate to attack the neighbors. Note that the angle between the The connection between evolution and learning will be further elaborated on in section 5.6 This to-and-fro movement is not surprising, as it is very common in territorial disputes across species of birds,fish and mammals.
Tadorna ferruginea ), the neck and body orientations arestill controlled separately, but in most of the cases the female stands with her breast to the drake andher neck pointing backwards (and very rarely this behavior may be performed without a neighborpresent) [100]. And in the mallard (
Anas platyrhynchos ), the same breast-to-the-male-and-pointing-backwards is observed, but now this posture is compulsory and, at high excitation, which turnsthe instinct on (the same relationship between excitation and activation of instinct exists for manyother instincts), the female is compelled to turn her neck over her shoulder even if that means thatthe neck moves away from the neighbor [98]. Thus, two elements of behavior, previously triggeredseparately by two separate environmental triggers, have become welded together and triggered asone. Finally, in the golden-eye (
Bucephala ), where the movement is highly ritualized (see below),the presence of a conspecific is not even required [100].Interestingly, along with the evolutionary change of form of the behavior, there has been alsoan evolutionary change of meaning. In the species with the less-ritualized form, the behavior hasthe effect of inciting and is related to territorial behavior. However, note that it already has in itan element of pair-bonding, or team work. In the more ritualized cases, this pair-bonding meaninghas moved to the fore: in the mallard, though it sometimes still elicits a demonstration of attack bythe male, inciting serves mostly as an invitation to pair-bond; and in the golden-eye, the incitinghas become almost entirely independent of the presence of neighbors, and takes a highly ritualized,exaggerated and rhythmic form of neck movements over one shoulder and then the other (andrhythmic movement is indicative of highly ritualized behaviors in general).It is due to the highly surprising nature of this example and others that Lorenz has been accused9f Lamarckian thinking. However, many examples of this sort exist, and we will see that they areexplained not by Lamarckism but by network-level evolution (sections 2.3, 4). What is importantto notice in the two examples discussed so far is as follows. In both of them, we see a gradualprocess arising from preexisting interactions . A novel phenotype (the fused protein in one case,the ritualized display in the other) arises from the change in context in which preexisting elements(preexisting genes, movements) are embedded. In fact, what was once an interaction has nowbecome an object: in the case of
TRIM5-CypA , a hypothesized interaction between two separategenes is succeeded by a gene fusion; and in the evolution of the inciting ceremony, two separatebehavioral responses to two separate environmental triggers (orienting the body toward the drakeand threatening the neighbors over the shoulder) has now become fused into a new instinct. Thus,the source of novelty is in system-level changes. In both cases, novelty arises not from a point-wisechange, not suddenly and not out of thin air.Among else, we also see local simplification in both cases: in simians, what previously requiredthe separate transcription of two genes now requires the transcription of one, and in ducks, aroundabout to-and-fro behavior has now turned into a stationary clear display. These aspects andmore will be explored in-depth in this paper, leading to novel insights on the fundamental natureof evolution and to a macroscale-view of the theory of interaction-based evolution [93].
I will now propose a verbal model that ties shifts in context to network-level evolution. The modelis purposely described at a high level because its role is to elucidate concepts, not to providemechanistic detail.Consider that in the course of genetic evolution, the network of genetic interactions graduallychanges as a whole. Many changes take place over the genome and over time, and these changesinteract. This process involves regulatory changes that can rewire the genetic network [18], suchas movements of transposable elements carrying with them cryptic enhancer/promoter sites andmultiple mutations activating those sites, for example [106]. But even a regulatory change thatat first sight appears only to change the strength of an already existing connection between twonodes—e.g., to increase the effect of a regulator on its target—can effectively cause rewiring; becausethere is no sharp boundary between the case where the regulator has a negligible effect on its target10in which case the two nodes can be said to be effectively disconnected) and the case where it hasa non-negligible effect (where the two nodes can be considered to be connected).Rewiring means that, in the course of evolution, the connections between some nodes on thenetwork become tighter and the connections between other nodes become weaker, and recognizingit is important. When the connections between nodes become tighter, they come to be regulatedmore and more as one unit, and a new module arises. What in the beginning may be two separateelements regulated by two separate lines of control can gradually come under one line of control. Aswill be understood later, this change represents the arrival of a new automatic unit. Furthermore,when this coming together of genes is preceded by the duplication of those genes and their regulatoryelements, this new module does not arise at the expense of previous ones, but represents a totalincrease in the number of modules; and together with this increase in the number of modules comesan increase in the extent of higher-level interactions between modules (since all the modules mustultimately come together into one organism, and now there are more of them .)While the term “module” usually refers to a set of tightly interacting genes, a rather basicmodule or unit is an exon; and since exons in separate loci may interact through trans-splicing, orthrough protein-protein interactions, etc., the same kind of process can cause the coming togetherof two previously interacting exons into a gene, or gene fusion. Such a fusion may be long in themaking. This shows us a case where a new elementary unit evolves from an interaction—from aprocess—and where a process can become an object—a gene. And as an object, it begins to acceptthe kind of operations that the system can apply to other objects. It is now interacting directlyand indirectly with many other units.A critical point in the above now calls for reflection. It takes time for two elements to undergoseparate regulation and transcription in order to come together later into a functional unit orinteraction. But when they come together evolutionarily into one genetic unit, regulated as oneand performing through one product, this time is cut to zero. Previously, the joint effect of thesetwo elements came into being as developmental interactions do; now it is “innate”—it is a gene. It “Number” of modules and “more” modules could be put in quotations because modules do not have a precisenumber, as they ultimately grade into each other, indeed because they have to be connected to each other. Thedefinition of a module used in the literature is a fuzzy one and rightly so: it is a set of genes that interact more closelywith each other than with other genes, even though to interact with the “outside,” at least some of its members haveto have just as strong a connection to members outside of the module. However, the fact that we cannot perfectlycount the total number of modules is an inherent characteristic of the process: it allows new modules to graduallyform.
11o longer needs to be constructed from more elementary units, and it exerts its effect in interactionwith other (now-peer) elementary units, in whose context it has phenotypic meaning. The emphasishere is not on the actual amount of time cut, but on the local simplification of the network.Thus, in the gradual fusion of two elements into one, we see a sense of evolutionary accelerationof developmental interactions; and if this fusion is preceded by the copying of those two elements,we see at the same time an increase in the “genetic vocabulary,” which comes together with anincrease in the extent of higher-level interactions—an increase in complexity.Having thus formed a clear view of acceleration and the arising of new interactions with the helpof the gene fusion case, it is important to step back again and observe these two aspects in the bigpicture. It is enough to consider the copying of modules and the changing of regulatory connections(prior to considering actual gene fusion) in order to notice that these changes of connections canbe seen from two angles: When we look at the lower levels of organization—at the tightening ofconnections between nodes—we see an increase in innate abilities. When we look at the higherlevels of organization—at the increase in the extent of interactions between modules due to theappearance of new modules—we see an increase in the complexity of the life-form, the phenotype.Importantly, these are two facets of one integrated process: the new parts observed at the lowerlevels (which are due to constriction) and the new whole (which is due to the increase in the extentof high-level interactions) coevolve . The novelty comes from a network-level change, not from asequence of independent, atomistic changes. And, as will be discussed, adaptation comes togetherwith innateness—with automatization.Notice also that there are useful operators in the evolution of networks: The copying of nodesalong with their connections adds syntactic material to the network from the inside, which serves asa basis for increasing complexity. The chunking of nodes and the severing of connections betweennodes allows nodes to separate from their previous context and join new contexts gradually.One thing that is important about this section, and that will become clearer later, is the senseof an Archimedes-screw–like operation of network-level evolution. An Archimedes screw is a helicalsurface wrapped around a central shaft inside a pipe that is designed to carry water up from oneside of the pipe to another as the screw rotates. Each point rotates at its own level, yet due tothat rotation, water flows up. Likewise, in network-level evolution, when genetic interaction isreplaced by a gene in the course of evolution, or when a behavioral sequence with environmental12riggers is replaced by an instinct, there is a sense of a transfer of meaning from phenotype togenotype—from higher to lower levels of organization—despite the fact that materialistic changeslike movements of genes are confined to their respective levels (the phenotype does not actuallybecome a genotype). This will help us replace the notion of novelty from a local genetic accidentwith the idea that novelty arises at the system level and is then crystallized in an evolutionaryprocess based on mutational operators working under natural selection. It also addresses, from anunexpected direction, the fundamental question articulated by Levin of how the different scalesof biological organization are connected [90]. As Levin wrote: “change is taking place on manyscales at the same time, and... it is the interaction among phenomena on different scales that mustoccupy our attention” [90].
As an example, the above bears on the evolution of chimeric genes. Traditional discussions on theevolution of chimeric genes seem to assume that they arise by sudden fortuitous events. In contrast,I argue that, as further molecular evolutionary details are uncovered, we will see that such genesare generated by a gradual process. The difference between these views is striking in the case ofthe evolution of alternative splicing patterns, and here, it brings together various aspects of thepresent view.“Exon shuffling” refers to the fact that homologous exons can appear in different genetic contextsin different species or even the same species. “Alternative splicing” refers to the fact that, ineukaryotes, multiple products can be generated from different combinations of exons, whether theexons are taken from nearby as in the case of cis-splicing, or from different loci as in the caseof trans-splicing. The former implies a process in evolutionary time. The latter is a process indevelopmental time. Now, we know that there are cases where the same exons are being trans-spliced in one species or strain but cis-spliced in another [79], such as the exons of the separate eri-6 and eri-7 in C. elegans strain N2 and their fused homologs in
C. briggsae and in other strainsof
C. elegans [48]. Likewise, we know that some functions are achieved by multiple single-moduleproteins in one species but by a single, multi-module protein in another, where the genetic sequencesencoding these modules are fused [59]. For example, the activities required for the synthesis of fatty13cids from acetyl-CoA are carried on by discrete monofunctional proteins in most bacteria, and areencoded by two unlinked genes in fungi [21, 113] and by a single multi-exon gene in animals [2] (see[59]). While a connection between exon shuffling and alternative splicing was suggested as soonas the latter was discovered [54], I offer to sharpen the nature of this connection as follows: exonshuffling is the gradually evolved innate state of alternative splicing. Namely, what is constructed indevelopmental time is gradually replaced in evolutionary time with new innate elements and a newdevelopmental construction. Specifically, when two exons previously spliced together at the RNAlevel are now fused at the DNA level, it is a case where a process in developmental time—a splicingpattern affected by various factors—has become an innate object—a gene fusion, emancipated fromthe influence of those factors.Accidental mutation and natural selection are not suitable for explaining this gradual evolutionof innateness of an alternative splicing pattern because it is a long term process that requiresmultiple changes that interact with each other, each of which is hard to justify by a short-termadaptive value. However, it can occur by mutational mechanisms operating under selection, asdiscussed in section 2.1 and in [93]. One may hypothesize that alleles evolving at multiple locigradually change the regulation of the alternative splicing pattern in the focal gene as well asin other, coevolving genes. Genetic information from these loci can then be gradually collectedby non-random mutation [93], setting the new genetic sequences as well as the new alternativesplicing patterns that we see today. In other words, many mutation-writing events, in each ofmany individuals, in each of many generations, under natural selection, gradually pave the way fornetwork evolution at the gene level. Evolution is a process where many interacting changes happenin parallel over long periods of time [93].Two noteworthy precedents to the above are these. First, Stone and Schwartz hypothesized thatseparate genes whose products first aggregated in the cytosol to form a functioning enzyme couldlater become fused at the DNA level [133]. They suggested, as an example, that the different lobes ofan enzyme such as glyceraldehyde-3-phosphate dehydrogenase may have come from separate genesfar in the past, before those genes became genetically fused; and that this could also explain theexistence of a family of dehydrogenases, each of which has fused the same gene encoding the NADbinding protein with differently mutated copies of the gene encoding the substrate binding domain.Second, West-Eberhard [163] predicted that the connection between evolution and development will14e found in the connection between exon shuffling and alternative splicing and in other phenomena;and that somehow what undergoes genetic change during development is also more likely to undergoevolutionary change [163]. In this paper, I agree with the above and add that the gene-fusion caseis merely an example of a more general principle, where meaning is absorbed from context by thegradual change of strength of connections between nodes in a network.
In developing his ideas on evolution, Darwin drew inspiration, among else, from the evolution oflanguage. In
The Descent of Man , he wrote: “The formation of different languages and of distinctspecies, and the proofs that both have been developed through a gradual process, are curiouslythe same... We find in distinct languages striking homologies due to community of descent, andanalogies due to a similar process of formation... We have in both cases the reduplication of parts,the effects of long-continued use, and so forth” [26, pp.59-60]. Had Darwin known what we knowtoday about the evolution of language and molecular evolution, he would have been able to takehis analogy further, and show that principles analogous to those proposed above are essential notonly for biological evolution but also for the evolution of language.Reminiscent of the ubiquity of cooption in biology, in the course of the evolution of language,words change their meanings as well as adopt multiple meanings. For example, words for “sharp”in different languages are related by descent to words for “tooth” or “shard” of clay, among else;and third person pronouns like “he” or “she” across different languages are generally related topointing words for distant objects [32]. The change of meaning is pervasive and the principle ofcooption appears to account essentially for all of language [32].Furthermore, the meanings of words generally change gradually, as the following example bylinguist Guy Deutscher demonstrates [32]. The word pair “going to,” in general and specifically inthe shorthand form “going [to some place in order] to [do something],” originally meant movement.Gradually, the movement meaning was relegated to the background, while the implication thatsomething was soon about to happen has come to the fore, until “going to” has become a futuremarker, independent of movement [32]. For example, a sentence from the mid 1400s tells of a travelto some place: “As they were goynge to bringe hym there.” A later example reads: “was goyngto be brought into helle,” where the passive form “to be brought” begins to shift the focus to the15emporal realm [32]. Finally, after further such changes, an example from 1642 spoken by KingCharles I shows the phrase to mean specifically that something was soon going to happen, withoutany implication of travel by the subject: ”My Magazine [arms] is going to be taken from Me”. Atthat point it was recognized by a linguist as a future marker [32].Note that it was not a sudden change in the words themselves that gave rise to the futuremarker, but rather a gradual change of context of usage: the more people used the word-pair toemphasize that an activity was soon to take place, the more it has come to be conceptualized in thisnew meaning. The novelty arose at the system level. Note that there was a hint of the final meaningalready in the beginning—when we go somewhere in order to do something, it implies that we willbe doing it soon. This meaning was sharpened and gradually released from the previous usage,leading at the end to an abstract concept that applies more broadly than before—to inanimate aswell as animate objects.Note also that in this fusion of “going to,” “going” and “to” are in some sense duplicates of“going” in the original sense of movement, as in “going to the store,” and of the “to” that is in “inorder to,” respectively—the latter are the source copies. In fact, in the slang word “gonna,” thetwo words have actually fused in the sense that the space between the words as well as some soundshave dropped. But it is important to notice that an essential part of the fusion had already happenbefore these local changes, which demonstrates that we must attend to the gradually changingcontext of usage as leading the process.Indeed, not only do new words commonly arise from fusions, they often start with a metaphorthat, in the course of the evolution of language, gradually becomes routinized with its own stand-alone meaning. For example, the Old English “hlaf weard” (loaf warden; i.e., bread keeper) hasgone through the stages of “hlaford,” “laferd,” “lowerd,” finally providing us the abstract “lord”[32]. The Latin de-caedere, or “cut off,” has evolved into “decide” [32]. (Note the metaphorbetween the literal meaning of “bread” and “keeper” on the one hand, and “lord” on the other, forexample.) Indeed, metaphor is a metaphor of itself, because it literally means carry across from onecontext to another (meta: across; phor: carry) [32], which is our topic—cooption. Furthermore,it is a common occurrence that when two words are used frequently and obligatorily together inan emerging context, their independent existence in that context becomes irrelevant, and they areshortened and fused into one word—which is reminiscent of the
TRIM5-CypA fusion mechanism.16he above provides also an analogy to innateness. In the beginning of the use of a pair ofwords that are to become a word fusion, the pair is constructed using the ability of speakers to puttogether previously learned words into combinations with their own meanings, and is understoodusing the ability of listeners to analyze combinations in terms of the words they are made of. Butas the two words come to be used more and more frequently together in a certain emerging context,they come to be perceived less and less as a constructed phrase and more and more as a word inits own sake. That is, increasingly the new fusion is learned by children directly from the contextof its usage at the same time as other words are learned, rather than being constructed figurativelyduring speech. Eventually it is hanging by its context alone. It is no longer constructed from unitsmore elementary than itself, but is a new elementary unit with its own literal meaning. In thisquickening of the construction of the new fusion until it becomes an elementary unit there is ametaphor for the evolution of innateness.In summary, the new elements of language are not invented out of thin air. Rather, the sourcefor their creation preexists at the system level. One may say that an essential point about humanlanguage is that it allows us to put together words into phrases and sentences that communicatenovel meaning; but note also that from these phrases and their contexts of usage, new words arise.The vocabulary grows in a manner connected to word usage. And as this vocabulary grows, ourability to express meanings is refined. Whereas previously “going to” had the explicit meaning oftravel and an implicit meaning of “soon,” now we have both, including a clear, separate meaningof “soon” that is applicable in new situations. Thus, from the ambiguous that can play multipleroles, come the distinct, diversified and specialized. The process “starts” at the system level.
Now, gene fusion may be discussed as one topic, and cooption as another. But they are actuallytwo sides of the same coin. In both cases we see elements or copies thereof leaving their previouscontext and moving to a new context. But although fusion and cooption are parts of the sameprocess, the case of fusion is especially grabbing to the eye, because it shows the creation of a newelementary unit in a manner that traditional theory has not prepared us for. In traditional theory,there is point mutation and presumed novelty from it, and there is gene duplication followed bypoint mutations in the duplicates [105], but there is no evolutionary process where a process can ecome an object —where a new elementary unit is created from something that previously was aninteraction (indeed, this new elementary unit absorbs new meaning from its gradually changingcontext).Two remarks are important. First, this manner of creating a new elementary unit requiresthe existence of a hierarchical structure of organization—a network—where, by a gradual changein the network, such a process can happen. Since this hierarchical structure does exist and is afundamental aspect of nature, it is an advantage of the present theory that it engages this structure .Second, the gradual creation of a new elementary unit from what was previously an interaction isimportant because it shows us that the barrier between “unit” and “interaction” has been broken .There is no sharp dividing line between elementary units and higher-level interactions. As withthe fact that the phrase “going to” never needs to become the fusion “gonna” in order to becomea word for all intents and purposes—a unified concept, automated and regulated as one—and asthere is no particular point in time where it suddenly turned from two words into one, so there isno clear line telling us when two exons that interact need to be considered as making up one geneas opposed to belonging to different genes [53]. The collapse of the gene concept as a well-definedunit is supportive of this absence of a sharp division between process and object [53] and fits witha gradual process of gene formation.Indeed, the view proposed here is importantly different from the traditional one. Not only doesthe traditional view focus on object minus context and claim that novelty arises in the object by alocal genetic accident that emanates this novelty “upward” to the complex system—novelty from apoint—but in addition, this point-like change is considered to be an error akin to a “misspelling.” Ifwe let genes be words, metaphorically speaking, and let the phenotype be the technology that theydescribe, then the traditional notion of mutation can be exemplified by misspelling unintentionallythe word “incubator” while making all effort to copy the word “incubate” accurately, and thus sud- In contrast, in traditional theory, genes are often perceived as independent actors, and mutation is perceived as alocal genetic accident that brings new phenotypic meaning on its own. Traditional models do not have a representationof the phenotype—of biological structure—and therefore treat genes more as beads on a string than as nodes in anetwork. They are oblivious to what is happening above the bottom level of the biological hierarchy, and to thepossibility that from higher up comes a force that changes something at the bottom level. They simply assume thatthe bottom level of the hierarchy is in control all on its own of what is happening evolutionarily, by means of randommutation. We now know that genetic elements we previously thought to participate in “one” gene actually form productstogether with elements that we previously thought to belong exclusively to “other” genes, and so the boundariesbetween genes have been much blurred.
Considering all the above, we can now describe a main point of this paper. Evolution is a “bot-tomless system .” One cannot define all words in the dictionary in terms of other words withoutgetting into a circularity. Ultimately, the meaning of words comes from the context of their usage;that is how language is learned and even how it evolves. The genes are similar in this regard.Their meaning comes from their context of usage. They themselves are nodes in a network, indevelopment as well as in evolution. The upshot of this is that the bottom of the hierarchy ofbiological interactions—the genetic sequence—is not a stable ground upward of which life is built.Mutation is not a local accident that brings innovation all on its own as though there is no livingnetwork that it needs to connect to. The process of genetic change is a complex one where theconnections between nodes in the network become stronger and weaker as they form modules that absorb meaning from context .With this key, we will begin to replace the source of novelty in evolution. Traditionally, we havebeen thinking about an accident, disconnected from the living network, as an event that creates newinformation. This was conceived of as a point-like event, which then emanates the novelty that itbrings about to the phenotypic level. I argue instead that novelty arises from network-level change,not from a point. This involves a mutation-writing phenotype that executes network change in asyntactic and evolving fashion [93]. This term, which aptly describes one of the most important points of this paper, was proposed by Nick Pippenger. Simplification and novelty
For Darwin as well as for Fisher [49], complexity evolved in cases where an increase in it was neededfor an increase in fitness. However, the question of why complexity evolves has never been resolved[154, p. 11]. I argue here that simplification under performance pressure leads to both complexityand novelty.This section will be entirely devoted to discussing the concepts. Once they are discussed, thenumerous empirical examples given in section 4 can be understood.
Several points in the present theory may be organized under the heading of “simplification,” eachof which comes with its own corresponding increase in complexity. • Simplification and modularity are tightly connected concepts. A module serves multiplecontexts—in fact it is defined by them—and in the case where one serves the many, as in thecase where one explains the many, there is frugality, parsimony, or simplification. • I discussed above the gradual appearance of modules in networks. A key example of theappearance of a module was the fusion of two genetic elements. Here, the developmentalprocess originally putting them together is simplified away in the course of evolution. Moregenerally, the gradual arising of new modules from a previously complex, interconnected massof nodes is the evolutionary streamlining, or simplification, of development. Elements insidea module are emancipated from the complex influence of elements that are now outside of itand are no longer connected to it. • As will be shown soon, an extension of the last point is the evolution of innateness, which in-volves evolved independence from environmental triggers. During evolution, an evolving traitcan become emancipated from complex environmental influence involved in the development20f an adaptive ancestral phenotype as a more orderly, simplified and compartmentalized de-velopmental process evolves. Thus, the evolution of innateness involves simplification: whatconsumed developmental (and sometimes learning) time is simplified away in the course ofevolution.Now, the cases of simplification described above come together with an increase in complexity.As argued earlier, due to the duplication of genes, the formation of a new module need not comeat the expense of old modules. The increase in the number of modules or elementary units comestogether with an increase in the number of interactions between such modules or units, which rep-resents an increase in complexity. Simplification is what we see when we look at the modularizationof an interconnected mass, and complexity is what we see when we look at emerging interactionsinvolving newly formed modules.
Local simplification leads to a global increase in complexity.
There are observations that show the development of organs or tissues taking ever straighter pathsover evolutionary time [114]. For example, in cetacean embryos (e.g., whales and dolphins), hindlimb buds still appear fleetingly in development and grow to a small size before they are removed[129]. In such cases, it is evident that, over evolutionary time, the developmental process graduallycomes to spend less and less time and energy on developing structure that is slated to be supercededby another or to be removed later in development.How does it happen that evolution straightens up developmental paths? A neo-Darwiniananswer is that the savings of time and energy are directly favored by natural selection, so that awhale that acquires by chance a mutation that reduces the development of the useless bones by evena small amount gains a slight benefit in terms of survival and reproduction, and thus accidentalmutations of this sort are passed on preferentially. We must ask, however, whether it is reasonablethat a slight straightening of the developmental path of useless, internal small bones is truly enoughto make such an impact on differential survival and reproduction that would be noticeable, whenpresumably many and much more important other individual differences contribute to differentialsuccess. Indeed, the problem of the obliteration of rudimentary organs is a very old one [25],and was discussed by Weismann hand in hand with that of the final touch of perfection—how21daptations become perfected beyond what seems to be possible by traditional means [162]. Darwinhimself agreed that it was not possible to explain the removal of rudimentary organs as an outcomeof natural selection alone based on minute economic considerations [27]. Indeed, in light of thesections to follow on innateness, it is remarkable that he held steadfastly to the Lamarckian “lawsof use and disuse” to explain them. And if Darwin is not neo-Darwinian enough to defend thelatter, then one may consider the father of neo-Darwinism, August Weismann—who is responsiblefor the rejection of Lamarckism: How did he explain the final touch of perfection? By suggesting aprinciple of “momentum” or “inertia,” where a mutation in a certain direction will be followed byothers in the same direction, so that noticeable, selected improvements of economy will be followedup by minute, unselected ones [162]—a point which is completely outside of the view based onrandom mutation and natural selection, a view which traces its ideological origins to Weismann.It seems that no serious explanation was found for these phenomena within neo-Darwinism, andindeed those who were supposed to be the two greatest pillars of it went to great lengths to lookfor alternative explanations.The theory proposed here tackles this old, unresolved problem head on. It argues that it isnot accidental mutation, but simplification, that explains the final touch of perfection, both inthe complete obliteration of a trait and in the crystallization of adaptation (see section 4.12.2).In addition, Weismann’s idea of inertia is not beyond the pale for a theory where the writing ofmutations has evolved under the influence of past selection.Now, notice again the connection between simplification and complexity: the intriguing straight-ening of developmental paths demonstrated by the unexplained old observations is tied to the “finaltouch of perfection”—a honing in on an optimum in the evolution of a complex adaptation.
According to [93], the writing of mutations over the generations combines information from differentloci and from different individuals that succeeded in survival and reproduction. Alleles at differentloci concomitantly spreading in the population do not each bring an independent piece of thephenotype to all individuals, but rather interact with each other. Thus, an adaptation evolves atthe level of the population as a whole, at the same time as it becomes more genetically stable [93].This process slowly gives rise to the true, common reason for success shared by individuals , as the22nitially many and highly variable ways by which different individuals approximate the adaptationonly roughly at first are gradually superseded by an adaptation, uniform across individuals (seenest-digging by sand wasps, discussed in section 4.12.1 and in [93]). We may now note that in thisreplacement of many by one—of the different rough approximations by one uniform adaptation—there is simplification. At the same time, this one that replaces the many is a complex adaptation—apoint of optimality. Therefore, simplification and complexity again come together: the complexitythat is in the different ways of approaching an adaptation has been converted into the complexityof the adaptation itself.
We have seen that each of the above connections to simplification comes together with an increasein complexity. Could simplification under performance pressure (e.g., under selection) be the causeof the evolution of complexity? This question is best answered together with another, relatedquestion, discussed next: What is the source of novelty in evolution?
Lamarckian or “adaptive” mutation has been the only alternative so far to accidental mutation ,but it has fundamental problems. First, it does not apply to multicellulars: there is no intra-organismal mechanism that senses that the hawk needs sharper vision and then makes the geneticchanges in the germ cells needed to bring about that phenotypic change. Second, hypotheticallyspeaking, even if there were mutational mechanisms that knew what would have been favored bynatural selection in a particular organism at a particular point in time and how to produce it, thiswould not have solved the problem of how novelty arises, because the novelty would have been inhow such supposed mechanisms acquired that particular knowledge to begin with. Indeed, it is easyto erroneously think that, if there is knowledge of the thing to be produced, there is no novelty,and if it is to be produced without knowledge, it must be produced by accident. Thus, we canunderstand the immense attraction of accidental mutation from a traditional perspective: First,it requires no impossible mechanism transferring knowledge from the macroscale to the genotype. As noted, the evolvability approach implicitly or explicitly relies on accidental mutation as the ultimate cause ofheritable novelty.
To try to answer this question, let us allow ourselves to step outside of evolution and look at hownovelty arises in other creative processes.Consider the development of scientific theories. It has two fundamental principles. First,theories need to fit the data—they need to perform . Second, they must be parsimonious. Whenwe take disconnected facts and find a theory that explains them all in one, we create a moreparsimonious picture of reality than existed before. It is a fortunate fact of nature that when wedo so we often obtain a model of reality that will hold better when new and unexpected data laterarises and that will lead to findings not previously expected.A well known example of the use of parsimony in science is the Copernican revolution—theplacing of the sun instead of the earth at the center of the solar system. Copernicus proposed thismodel not because it allowed him to make better predictions of the movements of the planets, butbecause it was simpler on an essential point [131]. This simpler model paved the way to futurescience, generally fitting with major later findings by Kepler and Galileo, like the phases of Venus.From this and many other examples we see that the pursuit of parsimony does not merelyprovide elegance per se . Parsimony expectedly brings the unexpected —useful things that were notinitially predicted and were not the goal of the work, yet commonly appear as a result of work .By simplifying under performance pressure we do not act randomly. Rather, we put work in, andget novelty out: a new, useful prediction or connection emerges that was not originally expected.Thus, it is not the case that either one knows one’s goal and there is no novelty in getting there,or one does not know it and the only way to get there is by accident. Rather, there is a third way24o novelty.Several important comments follow. First, we need not explain why simplification under per-formance pressure leads to novel, useful things in science that were not directly sought. For now,we may simply take it as a grand fact.Second, importantly, this simplification does not make science as a whole simpler but rathermore complex. As new theories connect between previously unconnected facts, new predictions andnew questions arise. The more knowns there are, the more they interact and expand our ability toask yet new questions. Thus, I argue that simplification under performance pressure leads to bothnovelty and complexity.Third, simplification and performance function together. As statisticians or investigators inmachine learning know, it is useless to make a model that predicts a given set of data pointsperfectly if the model is overly complicated, as it is useless to set up a model that is very simplebut has nothing to do with the data. A balance must be maintained between fit to data and modelelegance, and to maintain it is an art.Indeed, the desires for simplicity and for performance are conflicting: at the time when Galileooriginally favored the Copernican over the Ptolemaic system, he did it despite the fact that theformer fit the data a little worse, and because of the fact that it was much more parsimonious.Indeed, later scientific research showed that the more parsimonious model was far more improvable .The development of mathematics gives us a similar picture. It happened once and again in his-tory that pure mathematicians working on the principles of aesthetics or parsimony have producedthings that years later were found to have unexpected utilitarian value [166, 61, 16]. Indeed, thepower of operations other than the test of performance in the growth of mathematical and scien-tific knowledge has been amply demonstrated. We see it in simplification or parsimony, elegance oraesthetics, symmetry, pattern completion and analogy [166, 61, 16]. I use the word “simplification”in a very broad sense to refer to all these variants and the creative force they represent. Notealso that in both mathematics and science, we operate with a network of concepts. We connectbetween ideas to create a fuzzy, new idea, distill a fuzzy new idea to its essence, and pursue theconsequences of a distilled idea to new connections (Christos Papadimitriou and Umesh Vazirani,personal communications). Thus, novelty arises from the network, not from random, point-likechanges. This network change is driven by both simplification and performance, and we can see25hat it leads to complexity, novelty and improvement.The evolution of technology is also illustrative. What is simple appears in many differenttechnologies. The concept of a disc appears in the potter’s wheel, in wheels for transportation, ina round table, and in the cross section of a tree trunk. The concept of a sharp edge appears in astone tool, a peg, and even a shingle roof. Once we generate a functional but elegant object in onecontext, it is going to have the inherent capacity of working well in future, different contexts.I argue here that, also in biological evolution, simplification and performance pressure, andnot accidental mutation and performance pressure, drive complexity, novelty and advancement.This new theory has an advantage over the previous one. When we rely on simplification underperformance pressure, we rely on something that we can see to be central to other creative processes.A key aspect of simplification is that it allows us to circumvent the problem posed earlier: howmutation can do anything useful, how it can be anything besides accidental, without “awareness”of the environment and the macroscale phenotype. The solution is that biochemical work that sim-plifies local connections in the genetic network requires no knowledge of the macroscale phenotypeand the environment, and can take place in the germ cells. That is, while local simplification andgene duplication operations take place in the germ cells, natural selection evaluates the organismas a complex whole, and together these two forces lead to novelty. This allows us to replace theconcept of accidental mutation with a concept of non-accidental mutation that is useful yet notLamarckian , and thus to replace the traditional notion of random mutation as the ultimate sourceof novelty in evolution.
In addition to simplification pressure at the genetic level and performance pressure at the organismallevel, each of the two may have, at its own level, the other on the other side of the coin. For example,in an ecological community, each species is pressing to produce more of itself and at the same timeis undoing the growth of others, thus pressing to simplify the ecological network. The same could besaid of a gene that comes to replace another in the course of evolution by usurping the other’s role,a phenomenon called “genetic piracy” by Roth [126] (see also [154]). The ecological example aboveclarifies that the implementation of simplification can be as basic and follow as naturally from thesituation as differential survival. In fact, here they are two sides of the same coin: inasmuch as the26aking more of one entity means making less of another, the performance of any one entity putssimplification pressure on the network, and this principle may apply both to the ecological networkand to the genetic network. It is also noteworthy in this regard that a gene that is extensively used(performs well) and therefore highly expressed may, due to mutational mechanisms, be more likelyto be duplicated. This will be relevant in section 5.4.
It is time to substantiate the ideas proposed in this paper with many examples from the phenotypiclevel. This section will do so with the help of empirical observations relating to one of the oldestand most mysterious problems in evolution—the problem of the evolution of innateness. Althoughthe observations to be discussed are each known and available in the literature, here I will argue fortheir fundamental importance in evolution through a connection with interaction-based evolution. Iwill first cover innateness from multiple angles in sections 4.1–4.11, and then discuss the emergenceof novelty in detail (section 4.12.2). Readers interested in the molecular level may note that it willbe revisited in section 5.
The ability of pointer dogs to point at the prey in a statuesque manner (among other abilities) is to alarge degree innate [4]. How did this instinct evolve? To argue that a sequence of random mutationsof small effects has built up the behavior from scratch such that it has always been instinctive andnever learned is unappealing: Would breeders have recognized slight inborn tendencies to pointat the beginning of the evolutionary process involved and, without regard for the outcome of anytraining, base their artificial selection on these differences? And if training was important in theevolution of pointing, the highly evolved abilities of the animal to learn would have masked outpresumed mutations of slight effect for an independently developed instinct. All would be moreunderstandable if we consider that a trait that previously required learning through reward and/orpunishment has become emancipated in the course of evolution from these external cues.Consider the evolution of migration. In an instinctive and automatic manner, a young com-mon cuckoo (
Cuculus canorus ) takes off in the fall from its breeding grounds in Scandinavia, flies27housands of miles to its wintering site in Central Africa and then returns in the spring [167]. Howdid this complex suite of instincts get started in evolution? Both Darwin [125] and Wallace [156]hypothesized that the breeding and wintering grounds gradually became separated and the distancebetween them increased; that originally, the animals were tracking seasonal changes in resourcesover short distances as a direct response to the environment; and that in time this behavior hasbecome habitual and instinctive [125, 156] (see also [169]). To assume that the migratory instinctevolved afresh, independently of the behavior that came before it, brings up the same problem asin the case of the pointer dogs: the pre-existence of an evolved, general-level mechanism (in thiscase, the brain) that is able to respond adaptively to environmental changes and was presumablyinvolved in the original phenotype.In an experiment designed to capture the evolution of innateness [148] (see also [149, 153, 151,9, 8]), Waddington took
Drosophila melanogaster flies and exposed their pupae to a heat shock.As a result, a fair number of the flies that developed showed a particular vein pattern on theirwings—an absence of or a gap in the posterior crossvein and sometimes the anterior one too—called“crossveinless.” He then bred the crossveinless flies to form the next generation of the experimentand repeated this procedure of heat shock and selective breeding over the generations. As a result,the percentage of crossveinless flies increased over the generations and, beginning at generation 14, asmall percentage of flies started showing the new vein pattern without exposure to heat shock, thatis, innately [148]. The fact that this trait became innate, when no selection for such innatenesshad been performed , is an intriguing experimental outcome called “genetic assimilation” [148].To explain genetic assimilation, Stern [132] (see also [41]) proposed a model based on traditionalprinciples. The model assumes the preexistence of alleles that make independent contributionstoward a certain sum, such that if the sum surpasses a certain threshold, the trait of interestis exhibited. Furthermore it makes certain assumptions about the initial frequencies of allelesand the normal conditions and experimental conditions thresholds that make it so that, prior toselection, the trait of interest (e.g., crossveinless) is exhibited in practice only under experimentalconditions (e.g., heat shock), whereas post selection it is exhibited under both experimental andnormal conditions, and thus the trait can be said to have become innate. However, despite its In order to observe this, the experimenters took at each generation a certain sample of flies and raised themwithout heat shock. . Indeed, Waddington himself rejected this model [150, 152], because it did not applyto the complex cases that motivated the problem. Here, I will provide another explanation forinnateness based on network evolution. As Waddington alluded to [147], when an emerging module is released from the influence of anelement inside the organism, the result is seen as modularization; and when it is released from theinfluence of an environmental factor, the result is seen as the evolution of innateness. Earlier Iargued that simplification is connected to modularity and innateness: the formation of modulesstreamlines the developmental process and involves emancipation of an emerging module fromcomplex influences, both internal and external. Indeed, simplification leads to modularity andinnateness.Approaching the topic of innateness equipped with the theory of gradual network change pre-sented here, it is useful to distinguish between two important phenomena that I will call “eman-cipation” and “acceleration.” Emancipation refers to the fact that nodes (modules or elements)can be copied and the connections between nodes can gradually evolve such that a node can besubjected to a different regulation than that of its source copy. Acceleration refers to the ideathat the coming together of nodes under one control simplifies development locally while absorbingnovel phenotypic meaning from the changing context. Both these aspects of network level evolu-tion, discussed in section 2.3, will be clarified with the help of examples, and both figure into theexplanation of innateness to be given in the following sections. In fact, once we assume that Stern’s model taken at face value is the relevant method of explanation, it wouldhave been easier to assume that complex instincts in nature evolve afresh, without relation to a preexisting behaviormodulated by a brain and modified by the environment, because the model does not describe a world where such arelation is biologically reasonable—it requires a brain that affects independently threshold expediently assumed foreach particular trait that is to become innate, each with its own expediently assumed set of additive alleles. .3 The evolution of innateness is more common than we realize because theinnate, derived phenotype is usually not identical with its non-innate, an-cestral source In an idealized view of the crossveinless experiment, we can think of the crossveinless trait asqualitative (present or absent) and assume that it is the same in the beginning of the experimentas it is at the end. The only thing that evolves under this assumption is the propensity to produceit. In this case, we may simply use the word “emancipation” to describe what happens to thecrossveinless trait when it comes to appear without the environmental trigger. But crossveinless isan extreme, chosen for its simplicity. In nature, when the evolution of innateness or emancipationtakes place, the trait that is to become innate also evolves at the same time. For example, in casesof ritualization, a non-signaling behavior is gradually released from its context and becomes usedas a signal (e.g., an egg-fanning movement becomes a showing-the-nest signal [136]; see section4.6) [67, 165, 5, 137]. As Tinbergen noted, those ritualized traits that are emancipated are usuallytraits that have already changed much from their original form; and we would not have been ableto make a connection between the signal and its origin if it were not for the fact that, at least insome cases, there happened to be a transitional series betraying the connection between the two,such as in the threat posture of the Manchurian crane (
Grus japonensis ) [96, 137]. In other words,there exists a continuum of differences between the non-innate ancestral and the innate derivedtraits, that ranges from no difference, to a great difference that obscures the connection betweenthe ancestral and the derived; and cases at the former end of the spectrum are rare.I argue that this is precisely the problem with observing innateness. Darwin and other earlynaturalists believed that what is habitually performed due to environmental triggers over the gen-erations gradually impresses itself on the hereditary constitution of the species and becomes innateand emancipated from the environment, and that this is explained by the laws of use and disuse,or Lamarckism [125, 25]. I argue that such automatization happens in general but is often hard tosee because of the difference between the ancestral and derived traits past the point of cooption,and that it is network-level evolution and not Lamarckism that is responsible for it.A spectrum of differences between the new innate and the old non-innate is predicted byinteraction-based evolution. If we do not recognize this spectrum, we are liable to notice only30he easily visible cases at one end of it and then falsely argue that because of their rarity wecan continue business as usual. However, it is better to recognize that what we easily see of thisspectrum is its extreme, which is the tip of the iceberg.
The fact that a trait changes as it becomes innate allows us to examine the evolution of a complexwhole, while involving not only emancipation but also welding and acceleration.While I have been using the term “innate” without qualification so far, it is useful to notethat there is no strict separation between the innate and the non-innate. Learning itself is enabledby instinct [99, 55]. No trait develops in a manner that is independent of the innate nature ofthe organism, and no trait develops entirely independently of the environment, when the latter isbroadly construed [87]. Therefore, rather than speaking of “innate” and “non-innate,” we realizethat there is a continuum between things developed more directly and quickly (“innate”) and thingsthat require more unfolding that involves more interactions with the environment.When we consider this continuum as it applies to a given organism, we should consider that itevolves as a whole—the process of development evolves as a whole. Then, we can bring the ideasof gradual network evolution to bear on it. I argue here that the evolution of innateness arises asa result of the “chunking” or modularization of a previously complex part of the network—whatwere previously independent elements each under a different control have now become simplified orcombined into a singe unit. Although it may seem that this simplification accelerates developmentin the course of evolution toward the final trait, in general, development is not accelerating towardthe final trait as it was before, but rather toward what that trait has in the meantime changedinto, and therefore nothing is being accelerated strictly speaking. Therefore, it is the signature ofthe previously less innate that we generally see in the current more innate, rather than a directfacsimile. There is no Lamarckian transmission that takes a developed or a learned trait and makesit innate. As argued in section 2.3, to use a metaphor, in an Archimedes screw, water is movedalong the shaft even though each point in the screw only rotates at its own level. So in evolution, thenon-innate does not itself become innate—the phenotypic does not become genotypic—but ratherevolutionary action at each level of biological organization remains at that level, while acceleratingdevelopment. 31e know that the adult form influences the evolution of the earlier stages of development: thereis selection on the adult form, and therefore there is selection on earlier stages of development tolead to a well-performing adult. When considered from a traditional standpoint, this trivial pointturns into a problem because development is a complex process , and traditional evolutionary theoryis not equipped to deal with a complex process. Traditional evolutionary theory cannot conceiveof a complex evolutionary change that modifies the developmental process as a complex whole: itdoes not have a sense of acceleration or an emphasis on emancipation, and therefore when a traitappears earlier in development that seems to relate to one that used to come later in development(e.g., innate migration relates to earlier, learned migration), it absurdly has to invoke an evolutionof that trait afresh, absent any connection to that which it obviously relates to. While Gouldattempted to address this problem, he did so by breaking the whole again into parts and arguingthat the timing of appearance of one part or a developmental process in and of itself can be movedearlier or later in development [56], which is a very limited explanation that does not address therange of phenomena discussed here.I have presented, in contrast, a view of the evolution of the whole as a whole. Instead of theevolution “afresh” idea that arises from a traditional perspective, this view raises the notion ofacceleration as described. The quicker arriving at an evolving developmental outcome has theappearance of the evolution of innateness, thus involving interaction-based, network-level evolutionin innateness. We have seen this in the case of the TRIM5-CypA fusion and the evolution ofalternative splicing patterns at the molecular level, and will now see it in many examples at thephenotypic level. Pointing in pointer dogs will serve as an example of the importance of the evolution of the whole asa whole in innateness. I propose that selection has operated on the outcome of the training, favoringhunting dogs whose behavior following training was more pleasing to their owners, specifically instopping upon discovery of the prey instead of chasing it further. However, since innate tendenciesguide the learning, this selection has operated indirectly on innate tendencies, favoring dogs with but see the lively debate in the 19 th century on it [56]. per se : pointing is a system-level phenomenon that emerges from a suite of interacting instinctsand learning.In an exceptionally inspiring chapter, Papaj has already argued that what is first learned cancome over evolutionary time to be learned more quickly until it eventually becomes innate [121].In this respect, his hypothesis is similar to the above. However, lacking the ideas of interaction-based evolution, and treating instinct and learning as separate elements, he tried to create a modelof a traditional kind, and admitted that the model failed to provide an explanation, because theevolution of innateness came out of an artificiality built into it [121]. In contrast, the hypothesispresented here allows us to preserve Papaj’s intuition but in a natural way: it holds that selec-tion has affected interacting instincts that guide the complex process of development and learningthrough a process of network-based evolution, and network evolution involves acceleration andemancipation—an increase in the innate abilities. In addition, interaction-based evolution also ex-33lains why innateness and stereotypy are deeply intertwined (see sections 4.10, 4.11), which Papaj’smodel does not [121].To reiterate, I propose that the process of evolution toward a better outcome of the learningleads at the same time to a quickening of the learning and that ultimately, a new innate traitappears, because what enables the organism to reach a better outcome through learning is that itis naturally inclined in the right direction. The organism “gets it” more, naturally and inherently,because underlying interacting instincts are being shaped. Thus, improvement comes together withinnateness. Importantly, the evolutionary process that shapes the network of underlying instincts can beseen as uncovering better “principles” that guide the learning (and more generally, development)—emerging underlying elements that organize a preexisting complex more simply. Consistent withsection 3 and with later sections (see section 4.12.2), viewing things in terms of such principlesleads us to a new prediction regarding novelty: the evolution of the new innate and the new andimproved adult form will come together with the production of new, beneficial things that werenot selected for in and of themselves but arose as corollaries or windfalls of figuring out the rightprinciples. Improvement, innateness, and generalization —or the emergence of useful novelty notselected for—come together. As an example, backing in pointer dogs may have evolved as a“corollary”—an unintended but desirable outcome. As noted, ritualization is an evolutionary process that occurs when a behavioral element is graduallyemancipated from its original use as it becomes coopted for use as a signal in the course of evolution(see, among else, [67, 165, 5, 137] and further references below). For example, when a bird is aboutto hop or take flight, it bends its legs, lowers its breast, raises its hind parts and sometimes itstail, folds its neck and brings its head back almost to the shoulders, while slightly expanding itswings, so that the whole body is like a tight spring ready to be released for jumping, at whichinstant the legs straighten, the breast and hind parts line up with the direction of the jump, andthe neck is stretched forward. In an ethology classic, Daanje [24] argued that, from this movement, “Backing” refers to the fact that these dogs copy the posture of another dog who is on point; even this behavior,which is of use to the hunters, is strongly innate. Gasterosteusaculeatus ) shows the nest entrance to the female. According to Tinbergen, this movement wasderived from the egg fanning movement [136], which again demonstrates a shift from one contextto another.As we have just seen, ritualization requires emancipation from one context and cooption toanother. Critically, these are operations of network evolution. The gradual release of an elementfrom one context concomitant with the subjecting of it to another context involves two aspects ofthe organism at once and is inherently an interactive operation, well-described by modules movingin a network. It is not well described by the traditional notion of evolution as a process affecting“one thing at a time.”Baerends’s work on nest building, egg laying and offspring provisioning in the digger wasp
Ammophila adriaansei ( campestris ) [6] demonstrates clearly that behavior is underlain by a networkof modules. A normal behavioral sequence of the wasps is as follows: Build a nest; close the entrancetemporarily with soil; fly away and hunt a caterpillar; carry the paralyzed caterpillar back; reopenthe nest; put the caterpillar in; lay an egg; close the entrance again, this time with greater care.Now build another nest and repeat the entire process so far. Now return to the first nest; open theclosure; make an inspection visit. If the egg has hatched and the nest is in order, close the entrance,and now bring 1-3 caterpillars in succession. Repeat this second phase for the second nest. Nowreturn to the first nest, open the closure and make an inspection visit. If all is in order, bring 3-7caterpillars in succession; then make an especially careful final closure of the nest entrance. Repeatthis third phase for the second nest. Now build another nest, and repeat all from the beginning.Furthermore, if, in the first inspection described above, the egg has not hatched, the wasp maybuild another nest at that time. It can manage 4 nests at a time, with offspring at different ages at35ach nest requiring different amounts of provisioning (based on information obtained in inspectionvisits). If a nest has been disturbed, the wasp may abandon it.A computer programmer would instantly recognize that the digger wasp’s behavior is an al-gorithm with subroutines (see flow chart in Figure 3). The most parsimonious description of thisbehavior involves activation of the same behavioral modules or subroutines, such as “carry a cater-pillar to nest” or “perform an inspection visit” in different contexts, and the different contexts,namely the different stages of laying and provisioning, themselves consist of different combinationsof lower-level behavioral modules [102].Interestingly, Tinbergen wrote that the process underlying emancipation was not known, thoughit must somehow involve natural selection [137]. The present theory highlights how correct he wasto emphasize that unknown. At once we can understand the inability of traditional evolutionarytheory to explain empirical observations from ethology: A network is defined by interactions. Theevolution of a network is the evolution of a complex whole. The conceptualization of evolutionbased on traditional theory encouraged a one-trait-at-a-time type of thinking and was not suit-able for discussing network evolution and the transfer of an element from one context to another.Importantly, Tinbergen also noted that there is no point during emancipation at which a behav-ioral element stops belonging to its original function and starts belonging to a new function [137].Rather, as in the evolution of language and in the verbal model of network evolution discussedearlier, the change of context and meaning is gradual.Let us now think about the evolution of a network such as described by Baerends. Obviously,elements were not added to it in the form in which they exist today. For example, the constructionof a well-shaped nest with a cell at the end had been preceded by a less involved modificationof the environment. Also, elements were not appended in the course of evolution at the end ofthe behavioral sequence. That is, if “build nest”, “make closure,” “hunt caterpillar,” etc., aredenoted a , b , c , etc., then it is patently obvious that the stages of evolution did not proceed in thefollowing sequence: a , ab , abc , etc., or else absurdities arise such as not laying eggs until a certainpoint in evolution, performing an inspection visit before the existence of a foraging stage whereinformation from this visit is used, etc. This means that the behavioral sequence was reorganized inthe course of evolution and/or new elements were added at internal spots in the sequence. It followsthat elements that came after spots into which new or preexisting elements were inserted, or from36hich preexisting elements were removed or translocated, must have been emancipated from theirprevious triggers (namely the completion of the behavioral steps that used to come before them)and subjected to new triggers (the completion of the behavioral steps that come before them now).Finally, we would not assume that each repeating element or subroutine evolved afresh in its entiretyfor each instance in the sequence in which it is used. This means that there has been a copying ofroutine calls, or, to use more generic terms, copying and differentiation of modules . Thus, operatorsof network evolution—emancipation, cooption, copying and differentiation of modules—have beeninvolved in the evolution of digger wasp behavior.Another example showing the insertion of elements at internal points in a sequence and sequencereordering is Lorenz’s study of display sequences in surface feeding ducks [98]. Lorenz found about20 behavioral elements, of which different combinations make different display sequences in differentspecies and even within the same species. Lorenz [98] describes the study of three different species,the mallard, the European teal ( Anas crecca ) and the gadwall (
Anas strepera ), which share thefollowing 10 elements:1. Initial bill-shake2. Head-flick3. Tail-shake4. Grunt-whistle5. Head-up-tail-up6. Turn toward the female7. Nod-swimming8. Turning the back of the head9. Bridling10. Down-up movementHe then presents some display sequences (where one element follows another in quick succession)for each of the three species. For the mallard: 37 • • • • • • • the picture that we obtain by lookingclosely at evolution at the phenotypic level mirrors what the molecular biological and genomic rev-olutions have taught us about the genetic level: both at the molecular and at the phenotypic levels,network-level evolution is key. And network-level evolution is much better understood with thehelp of the principles of interaction-based evolution, including cooption, emancipation, accelerationand simplification.Welding, like emancipation, is also a network-evolution operation. While Baerends’s andLorenz’s examples above demonstrate it at the level of sequences of FAPs, welding can also generateelements at a lower level, namely the FAP itself; though—critically—and in accord with our earlierdiscussion of network evolution at the molecular level—there is no sharp boundary between the38AP and sequences thereof. One telling example was the inciting ceremony in ducks [101, 98, 100]described in section 2.2: To-and-fro movements of the female duck, originally triggered by sepa-rate environmental stimuli, have gradually fused in evolution and have become triggered as one,while becoming emancipated from the presence of neighbors. These movements originally were aterritorial behavior, with an indirect, implied meaning of pair-bonding and team work, and as theyfused, the pair-bonding meaning crystallized and moved to the fore [101, 98, 100]. In fact, manyrelated examples exist; for instance, the territorial marking in the fire-mouth cyclid, Cichlasomameeki . The tendency to attack a neighbor when in one’s own territory and flee from the neighborwhen in the neighbor’s territory is indeed a very general one, spanning birds, fish and mammals.In some fish, the neighbors exchange chase and be-chased turns, coincident with crossing the terri-torial boundary [102]. In the fire-mouth cyclid, this chase and be-chased movement has become ahighly rhythmic oscillation—it has become stereotyped. The welding of the previously separatelytriggered back-and-forth movements in this species is revealed when one fish suddenly loses interestand disengages yet the other continues oscillating [102].It is due to the highly surprising nature of these examples that Lorenz has been accused ofLamarckian thinking, even though he rejected it. The problem is that traditional evolutionarytheory is not network based, and thus it has been impossible to properly conceptualize theseexamples from its perspective.
The following examples not only show that welding and other elements of network evolution extendbeyond ritualization but also demonstrate the automatic nature of instinct. Consider, for example,the pecking instinct in domestic chicks. This FAP is present at birth and consists of three mainelements: lunging the head, opening and closing the beak, and swallowing [87]. Since we would notassume that these three elements of the fixed action pattern have each evolved from scratch in thecontext of this FAP, we are forced to assume that they have been welded.The classic example of a FAP—egg rolling in the greylag goose (
Anser anser )—also showswelding. Upon seeing an egg placed by the side of its nest, the goose stretches its neck in aparticular fashion, places its beak over the egg, and then slowly rolls the egg back into the nestwhile performing balancing sideways motions with the beak to prevent the egg from slipping from39he side [103]. This seems like an insightful sequence of operations, but in fact, when the egg isquickly pulled out from under the beak while in motion, the goose will continue to roll the remainingnothingness all the way to completion and tuck it under [103], again showing the automatic natureof instinct.Indeed, it is implicit in Barlow’s definition of the FAP (fixed action pattern) that the FAP isa welding of elements in general [7]. Barlow’s definition of the FAP, which he renamed “modalaction pattern” (MAP), is that it consists of a behavioral module usually indivisible but made ofelements that appear individually elsewhere.
Relatedly, Lorenz had suggested [98] that “perhaps allbehavioral patterns” arise from welding such as seen in the inciting example.
Interestingly, a single case of ritualization often exemplifies multiple or even all of the followingcharacteristics: emancipation (also: routinization, autonomization, or evolution of innateness),cooption, chunking (or welding), increased efficiency, exaggeration (or caricaturization), schema-tization, simplification, stereotypy, automatization and rigidification [67, 165, 5, 137]. Notably,traditional theory has only offered to explain one or another of these phenomena in separate fromthe others. For example, Maynard-Smith and Harper [108] suggested that stereotypy evolved be-cause it standardizes competition, which not only ignores the co-occurrence of the many elementsabove-mentioned, but also ignores the fact that stereotypy exists also in non-signaling instincts. Incontrast, it is striking that interaction-based evolution unifies all of these observations under oneumbrella, as outlined below: • Emancipation.
Emancipation (the release of a module from previous influences), or theevolution of innateness, is clearly demonstrated by the examples above, and has been ad-dressed here as a part of interaction-based (or network-based) evolution. The same is truefor chunking—the combining or welding of modules—which is also a part of network-basedevolution. • Simplification.
A fundamental concept in ethology is that of the “sign stimulus”—thestimulus that elicits a fixed action pattern. Here, “sign” means “simple”: the sign stimulusobtained its name from the fact that the animal attends only to a very limited part of the40ituation that we know it to be capable of perceiving through its senses. Namely, it attends toa parsimonious summary of the situation—a key. Yet the simplicity of the key is only relative:it is still a complex whole, a pattern involving relations between elements [139, 97, 82, 136].For example, the gaping response of nestling
Turdus merula as soon as they open their eyes canbe directed at a model consisting of a mere three discs that touch each other. However, it ispreferentially directed toward one of the discs that bears the right size-relation to another disc,such that the two together can be interpreted as head and body [139]. As another example,an abstract cross-like model (including symmetrical anterior and posterior “wing” edges, andcentral short and long protrusions perpendicular to them) elicited an escape response fromyoung birds, but only when it is moved in the direction of the short end of the cross, as onlyin this case the short end can be interpreted as a short neck, which is the case for birds ofprey [97, 82, 136]. In other words, an abstract combination of elements is the evolved key.Now, Tinbergen argued that evolved rituals, which are themselves stimuli eliciting behaviorin others, have been “schematized” through evolution and are evolved sign stimuli [137].Thus, he implied that rituals (and I will add the reception of signals, the “innate releasingmechanism,” or IRM [102]) have been evolutionarily simplified to their complex essence. • Exaggeration.
Ritualized signals are often exaggerated, as in the case of throwing theneck over one shoulder and then the other during inciting in the golden-eye (section 2.2).Although it has been suggested that exaggeration has evolved under natural selection forvisual clarity, it is questionable that organisms would need such a degree of clarity . Iargue that exaggeration is related to “caricaturization” or “schematization” (terms used inthe literature) and the final touch of perfection (section 3.1.3), and evolves by simplificationunder performance pressure (section 3). • Stereotypy.
Stereotypy—or the lack of variation between individuals in a certain trait, oreven between different instances of the behavior in the same individual—is another prominentaspect of rituals. According to interaction-based evolution, stereotypy is an inherent aspectof evolution, as will be discussed in section 4.10. As an example of the animals’ acute discriminatory abilities, a herring gull can recognize its mate among a groupof other gulls from 30 yards away [136] Cooption.
Cooption is inherent to Tinbergen’s definition of ritualization, as noted (a non-signaling behavior is coopted as a signal) and is also a crucial part of network-level evolution.Thus, interaction-based evolution provides a much more parsimonious view of ritualization thantraditional theory, which provides both additional support for the present theory and an improvedconceptual understanding of ritualization.
In the model of network evolution (section 2.3), I argued that two genetic elements that previouslywere regulated by two separate lines of controls and had to be separately expressed before comingtogether in an interaction can, over evolutionary time, gradually come together under one controland even fuse to form a new gene. In this process, there is not only emancipation (one or both ofthese elements is emancipated from what previously controlled it) but also a sense of automatization,innateness and acceleration, as the emerging unit is no longer constructed from its elements bydevelopmental interactions but rather has been evolutionarily accelerated into a ready-made unitor gene.We can now see that the phenotypic-level examples from the previous sections that demon-strate emancipation and cooption also demonstrate the evolution of innateness, automatization,and acceleration, as expected. In the pecking instinct of domestic chicks, for instance, the lungingof the head, the opening and closing of the beak, and the swallowing, have been welded together.Therefore, the last two elements follow the first one now automatically, even though they must havebeen originally triggered separately by the environment. Furthermore, this welded instinct appearssoon after hatching, and perhaps to some degree in the embryo [87, 83, 84], demonstrating theevolution of innateness and acceleration. In the case of egg rolling in the greylag goose, the initialstimulus from the egg suffices to trigger the entire motion of the beak that performs rolling all theway back to the nest even if the egg is removed during its journey. And in the inciting ceremony,the evolutionary process has emancipated the to and fro movements from environmental triggers,welded them together and put them under the control of one trigger, resulting in the evolution ofinnateness, and has brought the pair-bonding meaning to the fore (section 2.2).Finally, in the evolution of language, we saw that a pair of words can gradually acquire new42eaning from the gradual change of the context of its usage, and at the same time can begin tobe perceived and learned directly as a new word (word fusion) or concept in and of itself; whereaspreviously the emerging meaning of it had to be constructed from other words that had to belearned earlier. In this process, there is not only emancipation of the word fusion from its previouscontext but also a sense of acceleration of the learning of the meaning of the word fusion. Thisacceleration amounts to automatization of the new concept, which is no longer constructed fromother, more elementary units. Thus, the concepts of network-based evolution and the absorptionof meaning from context are central to genetic evolution, phenotypic evolution and more.
Interaction-based evolution holds that selection continually operates on complex interactions be-tween alleles across loci [93]. Thus, mutations do not generally bring independent pieces of thephenotype from the individuals in which they originated to all, as they are required to under tra-ditional natural selection. Instead, they interact, and the phenotype evolves at the level of thepopulation as a whole [93, 95]. It follows that, as some genetic variation across loci graduallydisappears (even while new genetic variation appears elsewhere), the phenotypic variation that itcaused due to the sexual shuffling of the genes disappears, and thus, over the generations, par-ents and offspring gradually become more similar to each other. In other words, interaction-basedevolution necessitates that a trait gradually becomes stabilized at the level of the population asa whole—it becomes “fixed” at the phenotypic level. This concept, called “convergence” in [93],shows us that if evolution is based on interactions between alleles across loci, then it must involvestabilization and therefore stereotypy as an inherent part of the process.If traits are gradually stabilized in such manner, then it follows that we should see a continuumof phenotypic fixedness corresponding to the formation of traits, with different traits lying at presentat different points along that continuum. Some traits, still early in the process of formation, willappear as less stereotyped, and others, at later stages in the process, will appear as more so. Thus,through this lens, stereotypy can be viewed as an indication of the degree of evolutionary progressof a trait.
An example is provided again by the pointers. Darwin wrote that the hunting behaviors that43haracterize the pointers are basically innate, and that the only difference between them and “trueinstinct” is that they are “less strictly inherited” in that there is variation in the individuals’ “degreeof inborn perfection” and therefore in the extent to which they require training [125, p.237]. Indeed,pointing in a statuesque manner, backing other dogs and other hunting-relevant behaviors have allbeen observed to occur often in pups that have not had the opportunity for learning by instruction,imitation or experience, and they are not exhibited immediately or to the same degree in all pups[4]. When training is required, the amount of training required is small: the trainer only guides thedog toward expressing what it has already a strong natural tendency to express [4]. The existence ofvariation in pointing behavior becomes eminently natural from the perspective of interaction-basedevolution: if traits evolve from fuzzy to sharp, if they gradually become stabilized, as discussedin [93], then this simply means that pointing is still in the process of formation and has not yetbecome perfectly innate.Those who previously tried to explain the fixedness that is stereotypy tended to argue thatsignals must be clear, that stereotypy makes them clearer, and that they are selected for this extraclarity in a traditional process of selection, hence stereotypy [7, 110]. However, the early ethologistsbegan by studying an extreme—the fixed action pattern—and later, Barlow noted that even whatwere previously called FAPs are not uniformly uniform, but rather some FAPs vary more thanothers; they are not all completely “fixed” [7]. To explain this continuum of stereotypy, it hasbeen proposed that signals that need to be clearer are more stereotyped, and others are less so [7].However, I argue that the clarity-based approach lacked parsimony from the beginning, becausestereotypy is a property of instinct in general, not just of signaling behavior specifically; and thateven if clarity plays some role in the evolution of stereotypy, uniformity varies first and foremostbecause of the temporal nature of the process. That is, the degree of stereotypy is in generalassociated with the point that the trait has reached along the spectrum of formation, and weobserve that it varies across traits because we are witnessing traits at different stages of formation.In other words, even if we can imagine reasons why uniformity per se would also be of value, thetraditional focus on uniformity as a separate end obscures the general point of interest: stabilizationis an inherent part of interaction-based evolution . This is not to say that all traits must inevitably cover the whole spectrum and reach the extreme, nor thatthey all move along the spectrum at the same rate; but it is to say that the degree of fixedness comes from thenature of the process of network-based evolution, and is not an independent element traditionally selected for, as “Innate behavior” or “instinct” has been used to mean different things in the literature [121]:1. Independence: It has been used to refer to behavior that is independent of interactions withthe external environment like learning or experience. Such independence is especially clearwhen a behavior is present from birth, as for example in the pecking of domestic chicks [87] .2. Stereotypy: Innate behavior has been referred to as stereotyped (also, “fixed,” “constant,”“rigid,” [121]—“robotic” ).3. Sharedness: Innate behaviors have been found to be homologizable between species andtherefore useful for taxonomy. In other words, they are shared between species , with some Maynard-Smith and Harper [108] and Mayr [110] have argued. Indeed, whether the evolving trait is a signal ornot, the evolutionary process is converging on an adaptation: along with the decrease in variance and disorder, itconverges on highly efficient structure and behavior. Clarity, which is part of the effectiveness of the signal, andefficiency in other adaptations, are outcomes of the evolutionary process, and uniformity is an inherent concomitantof the process in both cases. “Environment” and “experience” need to be qualified: No behavior is entirely independent from the “environ-ment” or from “experience” when these notions are so broadly construed as to include such factors as the flow offood materials during development or the “source of experience” that one body tissue provides a neighboring onein the course of development. Even what we more normally call “experience,” that is interaction with the outsideduring growth and learning, can figure into and change aspects of behaviors that are otherwise innate. For example,the tendency of the rat to build a nest is innate, yet it will not be able to carry it out if it is deprived of experi-encing the carrying and manipulating of objects during development [87]. Despite this important qualification, theconceptualization of innateness as independence from the environment is still true and useful in an important sense. (Lorenz called it “an epoch making discovery”[102, p.103]). Second, there is an agreement that, empirically, what is innate in the first senseabove (1)—what is automatic—appears also to be stereotyped, or “fixed;” it appears “robotic”(2) [7, 121].But why are the different aspects of innateness connected? Tinbergen’s tone regarding the firstconnection outlined above (between points 2 and 3) was that of surprise [136, p.191]. And Barlowdiscussed stereotypy in the context of the clarity of signals [7], which neither addressed stereotypyin non-signaling behavior nor explained the connection between stereotypy and independence (eventhough his continuum of FAPs ran both from variable to stereotyped and from dependent on toindependent of the environment at the same time [7]). Finally, Papaj’s model was unable to connectstereotypy and independence [121].However, from the point of view of interaction-based evolution, the different aspects of innate-ness are connected. First, traits evolve by a process of convergence as defined (see section 4.10 and[93])—a process of stabilization which begins at a state of high variance and eventually leads tothe evolution of uniformity and therefore stereotypy [93]. Since this process takes time, it makes itso that stereotyped elements are older and more widely shared than elements not yet stereotyped,while implying that stereotypy evolves in parallel. This point connects the sharing of a trait amongspecies with stereotypy (aspects 3 and 2 above, respectively).Second, I have argued that interaction-based evolution works in the long-term through sim-plification under performance pressure. This can not only make one intra-organismal moduleindependent of another, but also make it independent of environmental factors (see section 4.2).Therefore, besides leading to stereotypy, the process of interaction-based evolution also leads to in- One must note, however, that parallel evolution of stereotyped traits in related species could also lead to “homol-ogizable” traits in this sense even though in this case there is no common origin, strictly speaking, a scenario whichis in fact expected from interaction-based evolution. Even though there is variation in the degree of stereotypy, there is also variation in the degree of independencefrom the environment, and biologists agree that what is strongly fixed tends to be strongly uninfluenced by theenvironment, or “innate” in the first sense [121].
This ties together independence, phylogeneticsharedness and stereotypy.
Two additional aspects of innateness also make sense in light of interaction-based evolution.One is that welded elements often and perhaps always [98] are present in the fixed action pattern(see sections 4.6 and 4.7), which aligns with the fact that welding is an outcome of network-level evolution. It also sheds light on the issue of circuitous vs. accelerated development, as will bediscussed below (see section 4.13), which will clarify the connection between evolution and learning.Traditional evolutionary theory has had difficulty explaining and reconciling the various aspectsof innateness. This has resulted in a call by Bateson to give up the use of the term “innate” alto-gether and specify instead what particular meaning of innateness one is referring to [10]. In contrast,by connecting the different aspects of innateness with recourse to one parsimonious mechanism,interaction-based evolution fits with the fact that the term “innate” has been used intuitively as aunifying concept for a long time. Furthermore, it shows how these different aspects are related toeach other.
Because of their ostensible simplicity, examples like the evolution of malaria resistance due to theHbS mutation have served the traditional notion of natural selection and random mutation. Infact, fascinating molecular-level details are available that question the accidental nature of thismutation and others [93]. However, putting aside this mounting evidence, another fundamentalissue is that examples of the evolution of complex phenotypes have been very much underplayedin the evolutionary theory literature. In this paper, we have seen that these examples fit with theview of interaction-based evolution: they demonstrate cooption and emancipation, stabilizationand stereotypy, and evolution from fuzzy to sharp. Most intriguingly, they speak to the arising ofnovelty in a way that is consistent with network-level evolution .For the reader who is interested in the detail, I will first discuss the example of decoy constructionin sand wasps. Other readers may skip to the final and most important phenotypic-level example—that of egg retrieval by backward walking in the nightjar (section 4.12.2)—where I will demonstrateall the elements discussed in this paper in one, with an emphasis on the central point of novelty.47 .12.1 Elements of network evolution in the construction of decoys in sand wasps
This example is taken from Evans’s classic work on the comparative ethology of the sand wasps,Bembicinae (previously called Nyssoninae; [40]). To make a nest, the sand wasps dig a tunnelof many body lengths, at the end of which they build a cell or a complex of cells, where theyplace their offspring and the prey on which the offspring feed. They have natural enemies—twotaxonomic groups of flies (the bee flies, Bombyliidae, and miltogrammine flies, Sarcophagidae)and two taxonomic groups of parasitic wasps (the cuckoo wasps, Chrysididae, and “velvet ants,Mutillidae)— that parasitize their nests by laying there, and whose larvae either takes up valuableresources or destroy the larvae of the sand wasp. Parasites of the former two groups seek the nestsof their hosts by sight, and of the latter two groups by touch and odor—they fly over the ground,tapping the soil with their antennae in search of their target. Across the sand wasps we find varioustechniques of hiding and concealment as well as a decoy construction technique.The example concerns the decoy construction [40, 39]): Some species of sand wasp dig a falseburrow (or multiple such burrows) next to the real one and leave it (or them) open, while leavingthe real nest burrow closed. Various parasites have been observed to either lay their eggs or lingerin the decoys. Evans hypothesized that false burrows originated in a behavior that had a differentpurpose—that its origin is in cooption—and that the process of the evolution of digging falseburrows was a process of improvement through stereotypy, together with emancipation (see also[142]).He based his hypothesis on the following facts. Very commonly across the Bembicinae, thewasps close the entrance to their nest from the outside before leaving it, either temporarily with asmall amount of soil before leaving temporarily for provisioning, or with a large amount of soil at thefinal closure before leaving for good; and both within and outside of the Bembicinae, species havebeen observed where individuals obtain soil for nest closure mostly from several or one particularspot/s around the entrance. In this case they leave behind a small pit or pits of a size that dependson how much a particular spot was used. The tendency to take soil from a particular spot orspots appears to relate to environmental conditions, where individuals quarry soil for closure whenloose soil is not as easily available [39], though it also has a genetic component [39]. The moresoil was quarried from a particular spot, the bigger the pit left behind. In
Bembecinus neglectus ,48or example, most individuals took the soil for closure from several particular spots around theentrance, so that a ring of small depressions was left behind. But some took it mostly from oneparticular spot, which formed a “depression or short false burrow up to 1 cm deep” [40, p.137].Now, in the genus
Bembix , which typifies the advanced behaviors of sand wasps, we see specieswith decided false burrows. Three species are particularly telling:
Bembix amoena , B. texana and
B. sayi . In
B. amoena , false burrows are of irregular occurrence and spatial pattern and arerelatively short, and the burrowing is still almost always associated with quarrying soil for closure.Most individuals obtain soil for the initial closure by scraping a small amount of soil from eachside of the nest entrance, and occasionally from a particular spot or spots, creating short falseburrows that stretch in a direction of 90 degrees or a bit less left or right of the direction of thetrue nest, obliquely into the ground. These short false side burrows varied in length from barelyperceptible to 2 cm long, with two exceptional cases observed at 3 cm and 5 cm long [40]. Evans[40] noted that, in one population, about half of the nests had no such false side burrows, about aquarter had one, and the rest had two such false side burrows at one time or another. In addition,individuals also collected soil from opposite of the nest entrance, most often, but not always, forthe final closure (which requires more soil), which resulted either in a furrow going through themound of soil opposite the true nest (a mound resulting from the excavation of the true nest),or in a burrow running under that mound obliquely through the ground. The former appeared aconsiderable number of times, [40, p.280] and were of varying length, between 1 and 7 cm long.The latter appeared rarely [40], with only three cases noted, at 1.5, 2 and 3 cm long. Evans [40,p.281] noted that two of these were made following final closure in a manner similar to that of
Bembix sayi (see below), which seems to suggest that these two rare instances occurred not in theservice of obtaining soil for closure.Besides the burrows themselves, the manner and timing of construction of the burrows wasalso variable of course (Evans 1966a,b). The burrows on the sides, when they occurred, occurredsometimes along with the initial closure and sometimes along with later closures. The back burrowsand furrows, when they appeared, appeared often along with the final closure but sometimes alongwith earlier closures, and the final closure did not always involve them. While burrows weresometimes revisited and expanded, they were sometimes accidentally filled while making a closure.Likewise, the spatial pattern resulting at the end was variable, with 0 or 1 or 2 side burrows and 049r 1 back burrow or furrow [40, 39]. Note that, although the soil was generally taken for closures,parasites were distracted by the false burrows that resulted [39].In
B. texana , construction of false burrows is more or less regular, and the soil is not used forclosure. Typically, individuals construct one short but relatively persistent false burrow on eachside of the entrance, right after the initial closure is made [40]. The method of construction is notyet entirely stereotyped, with some individuals digging one burrow and then the other, and othersalternating in digging both [40, p.325].In
B. sayi , construction is invariable and emancipated: all females dig one strong back burrow4–22 cm long under the mound after completion of final closure (which also means that the soil isnot used for closure) [40, 39].The three species above exemplify certain general trends [40, 39]. The primitive cases of falseburrows, where the burrow is but a small pit, are unreliable and irregular in appearance. The“transitional” case of
B. amoena shows burrows appearing in a notable number of cases but stillrather irregularly. They are often longer than “small pits” but are still relatively short and varygreatly in length. In the advanced cases, the burrows appear with greater regularity and aresubstantial. In
B. sayi , which makes the longest burrows, the burrows appear invariably in allindividuals and at a regular time. There is an association between stereotypy and completeness ofthe burrow [40, 39].In addition, there is an association between stereotypy and emancipation of burrow-making fromits previous cause [40, 39]. That is, limited quarrying in the service of obtaining soil for closure isa widespread phenomenon, and tends to be irregular in occurrence; whereas, in contrast, regularburrows are the result of burrowing for its own sake, an operation not used for closure, and areconstructed at regular times before or after closure, depending on the particular species concerned.In some of the advanced species where burrow-making is thus emancipated, the wasps refresh orfix burrows that have been destroyed [142], which further clarifies that they are programmed tomaintain a certain pattern of false burrows. The above characteristics are also associated withincreased regularity of the spatial pattern of the burrows, with each of the different emancipatedspecies having its own idiosyncratic characteristics of construction [40, 39].Furthermore, not only have emancipated burrows never been observed in species that lackclosures, but in addition, a careful reading of Evans [40] shows that, even though they are no50onger used for closure, emancipated back burrows are temporally associated with final closures,and emancipated side burrows are temporally associated with initial closures, which seems to cross-validate the fact that the origin of emancipated burrows is in closure-making.Thus, evidence clearly supports the predictions of interaction-based evolution. The evolutionof false burrows originated in cooption—in emerging high-level interactions between preevolved ele-ments like digging, quarrying and making closures, and environmental elements like sand conditions .That was a state of high variance in behavior and outcome within and between individuals. Evolu-tion then proceeded from fuzzy to sharp: through a process of convergence and gradual stabilizationof the trait as a whole toward a stable, emancipated and clock-work–like state . The process wasthat of improvement together with and at the same time as stereotypy and emancipation. Notethat it is not the case that complete but irregular burrows evolved first, and then were stabilized.That is, stereotypy, or uniformity, is not an outcome of a force of stabilizing selection separate fromthe selection for the adaptation itself. Rather, stabilization and improvement evolve together astwo aspects of the same coin—as inherent concomitants of the adaptive evolution of the whole as awhole , as predicted by interaction-based evolution.
I will now discuss the final and most important phenotypic-level example that puts all of theelements of the theory discussed here together, while emphasizing the central point of the emergenceof novelty. This is the example of the evolution of egg retrieval by backward walking in the nightjar(
Caprimulgus europaeus ) and other species, which applies to eggs that have rolled far outside thenest. Before we can understand it, I must first explain what the shifting motion in birds is.The shifting motion in birds is ancient and involves rolling an egg with the beak until it reachesunder the body. The egg may have thus gotten in between other eggs and stirred them, and theegg sides that are pointing up are thus changed [138]. Shifting may be needed to ensure eventemperature distribution to the eggs [17], and is performed upon arrival at the nest, or when thetactile stimulus provided by the eggs while brooding is not satisfying, or spontaneously after a longspell of quiet brooding [138]. In terns, the shifting motion will move an egg about 2–3 inches.Coming back to our case of egg retrieval, in terns (e.g.,
Onychoprion fuscatus ), the generalsituation is as follows [158, 138]: If they notice an egg lying several inches outside of the nest, they51eave the nest right away to it. However, they have an aversion toward being far from the nest,induced by their brooding state, and as they move away from the nest, they slow down, sometimesturning around and returning to the nest without having reached the external egg. But sometimesthey do get to the egg, stopping short of it just close enough that they can reach it with the beakand apply the shifting motion to it, which rolls the egg until it is under the breast.As they shift the egg, they sit down on it to incubate it, but only for a short time (indeed theymay at this point be dissatisfied with the tactile stimulus and/or with being outside of the nest).The moment they notice the nest again they stand up and walk to it. In the process, the egg hasmoved about 2-3 inches toward the nest due to the shifting motion.Having returned to the nest and started brooding the eggs there, they soon notice the externalegg again, venture out toward it again, and repeat the process, and the egg moves 2-3 inches againtoward the nest. Thus, after several trips, the egg finds its way back to the nest.The behavior that results in the egg being moved back to the nest is clearly unstructured. Thebrooding of the external egg outside of the nest and the back-and-forth trips show the lack of insightor “analysis of the situation as a whole” [158, p.83], as the different actions taken in the situationare under the proximate control of different preevolved instincts. In accordance with Tinbergen[138], these instincts are competing with each other for expression: the desire not to leave the nest,the desire to return to the nest, the desire to brood eggs, and the desire and ability to shift an egg.Also, as Marshall noted for the common tern (
Sterna hirundo ) [107], there is much variance in thebehavior and its outcome, with eggs sometimes being rolled back into the nest and sometimes not,and this variance is thought to reflect both individual variance and situational factors [107].In other birds, however, such as greylag geese (
Anser anser ), black-headed gulls (
Chroico-cephalus ridibundus ) and nightjars, the bird walks straight up to the egg, puts the beak over it asit would in shifting, but instead of incubating the egg there, it then walks backwards all the wayto the nest in one shot while shifting and dragging the egg under its beak [75, 138].According to Tinbergen, this egg rolling observed in nightjars and other birds evolved fromshifting and other elements of the situation [138]. Indeed, the fact that the birds are using ashifting motion while walking backward (even though rolling with the wing would have been muchmore efficient) together with the fact that shifting is ancient, supports this hypothesis [138]. In fact,Tinbergen notes that the very controversy about whether egg retrieval is an independent adaptation52r a by-product of a confluence of instincts in different species shows its route of evolution [138].The argument that this backward walking behavior appearing in nightjars and other birdsevolved from a situation akin to that of the terns exemplifies several elements in one: The traithas evolved from fuzzy to sharp; from unstructured and inefficient to structured and efficient; fromvariable and unstable to stable, stereotyped and “rigorous.” In addition, we also see emancipationin it: the return back to the nest originally required the visual stimulus of seeing the nest, but nowis triggered automatically as soon as the shifting motion begins and requires no turning-around tothe nest. We also see welding: the going-to-the-egg and the coming-back-to-the-nest legs have beenwelded together in one sequence unleashed by the stimulus of seeing the external egg for the firsttime, whereas previously they were two separate legs each triggered by its own visual stimulus. Thewhole situation has been simplified , the path has been straightened up. In fact, the simplificationhas been the creation of a method from a previously non-methodical occurrence, when all the whilethe whole evolved as a whole, not by the addition of independent elements one at a time.On top of all of the above, one topic deserves a special emphasis: novelty. The example of eggrolling shows clearly that different instincts or elements have the inherent ability to come togetherinto new, useful interactions that together can achieve what had been unachievable before by anyone of those instincts alone. Twice we see that this coming together of pre-evolved elements intouseful, high-level interactions, which is outside of the purview of the random mutation and naturalselection view, breaks a barrier in terms of being able to do something that could not have beendone before. First, the confluence of instincts for shifting, brooding and returning to the nesteffectively allows the egg to be returned to the nest after several trips, even if in a haphazard way,when none of these instincts by itself is capable of achieving this, nor did any of them originate dueto pressure for such egg retrieval. The second barrier broken was this: the invention of backwardwalking while shifting allows retrieval of the egg in time that is proportional simply to the distanceto the egg, whereas the haphazard way only allows retrieval in time that is quadratic in distance.This improvement allows nightjars to retrieve eggs from many yards away, which would not havebeen effectively possible in the case of terns (indeed terns retrieve eggs from only several inchesaway). Thus, emancipation and welding have created a behavior that now applies to a broaderrange of situations than the ancestral traits applied to. Some of these birds now dwell in beacheswhere eggs can indeed be blown away by wind a great distance.53hese breakings of barriers in the formation of new traits exemplify novelty. The novelty isin the inherent ability of elements to come together into new and useful high-level interactions.These elements come together first in a haphazard state. Their complex interaction then serves asa substrate for simplification under performance pressure, where new such elements will be formed.I propose that this cycle is the heart of the evolutionary process. I have furthermore proposed thatit is simplification under performance pressure that is responsible for this inherent usefulness ofelements—for their propensity to come together into new, useful interactions that they have notbeen directly selected for.This point provides an understanding of novelty in evolution that is completely different fromand not reducible to traditional random mutation and natural selection. The novelty that drivesevolution here arises from the coming together of high-level “modules,” i.e., from the network as awhole. It is not a local, “misspelling”-like change at the genetic level. It is not accidental, or randommutation that invents in evolution. Rather, non-accidental mutation and natural selection togetherprocess information gradually, and the source of novelty in evolution is the resulting inherent abilityof elements to come together in useful high-level interactions, an ability due to simplification underperformance pressure.Watson and Lashley [158] saw that the outcome of the haphazard mode of egg retrieval was notintentional. They noted that the egg rolls in the direction of the nest simply because the bird isoriented directly away from the nest as it reaches the egg, so that the shifting motion happens tobring the egg a bit closer to the nest each time. From this they concluded that the egg finds its wayto the nest by lucky happenstance. But this lucky happenstance is a far cry from the traditionalnotion of novelty in random mutation. First, it is not a local accidental mutation that invents,but rather the process starts with high-level interactions, and gradual network evolution createssomething new from this source. Second, an extraordinarily deep new question arises. Should wecall this source of novelty “randomness” or “lucky happenstance” at the phenotypic level, and sayno more? There is logic to the present situation that goes beyond the purely coincidental. That is,although the egg rolls to the nest only because of the bird’s orientation, could the bird have beenoriented any other way? Indeed, it is oriented in the way that it is because it is walking straightup to the egg from the nest. Thus, the efficiency of this movement is integrated with the feasibilityof the retrieval system. Indeed, the whole situation, while haphazard, is not purely random, but54an be seen as a fuzzy sort of “ shifting in an extended nest .” Thus, a confluence of instincts, eachuseful in and of itself, together give rise to something useful that is different from each of them, butwhich at first can only appear in a roundabout, highly variable, even though not purely random,fashion. As such, it serves as material for evolutionary simplification and streamlining, which endsup creating something that can be useful in contexts that go beyond the one that originated it.It is intriguing that an unqualified notion of the accidental does not sufficiently explain noveltyhere. When we apply this way of thinking to other cases of cooption, we will see that, individu-ally, they may appear more or less accidental than the above case; but they are not, in general,“pure coincidences.” This, together with the question of how exactly simplification under perfor-mance pressure leads to inherently useful elements, opens up an intriguing new area for scientificinvestigation—a science of novelty.While others have discussed the possibility of cooption being a source of novelty [57], it hasbeen discussed within the random mutation view—cooption has been treated as a random event and another source of novelty in addition to random mutation. In contrast, interaction-based evolutionargues that cooption is neither a random event nor another source of novelty in addition to randommutation. Rather, non-accidental mutation and natural selection gradually pave the way to bothgenetic and phenotypic cooption at the macroscale through network-level evolution. Thus, neithermutation nor cooption are random in the traditional sense even though they produce surprisingthings, and they are not separate sources of novelty but come together as two inseparable aspectsof one process. Cooption is at the heart of the process of interaction-based evolution and is builtinto this process.Thus, while in the traditional view, novelty arises by accident at a specific point in space andtime, according to interaction-based evolution, novelty is an outcome that arises over time at thenetwork level from the coevolutionary change of many elements. While the drivers of these localchanges are not random, these changes still interact with each other globally in a surprising way.Surprise, or novelty, exists, but it is not a mere direct effect of dice rolling.It is noteworthy that the tern situation is based on conflict, or competition between tendencies.The bird, on the one hand, acts as though it wants to reach up to the egg and incubate it, buton the other hand as though it wants to remain in the nest. It is also noteworthy that there isindividual variation in the overall behavior, and indeed, there may be different ways of increasing55he probability of success. One way may be to approach the egg without hesitation. Anothermay be to get back to the nest without delay once the egg has been shifted. There is an inherentconflict in the situation. Both tendencies have something to contribute, but they are conflicting.To strengthen one at the expense of the other may be harmful. Evolution may need to take amodest though complex step: to find a solution for returning to the nest immediately and onlyafter reaching the egg while engaging it with the beak. This can be achieved, for example, byovercoming the tendency to incubate but only while standing outside of the nest. The relevantrule to evolve, “incubate in the presence of eggs AND when standing in the nest” is simple butnon-linear. In performing this evolutionary step, the convergence process described in [93] maylead to the crystallization of the commonality between successful individuals while resolving theconflict inherent in the situation, making evolution a process of conflict resolution.The example also shows us, of course, that the whole is greater than the sum of its parts andthat the organism evolves as a whole. Conceptualizing evolution in this way provides an answer tothe many inconsistencies that arise from the accidental mutation framework, such as the fact thatit often leads us to surmise difficult evolutionary sequences leading to complex adaptations wherethe intervening steps are not adaptive in and of themselves.Finally, the example also shows us a connection between punctualism and gradualism. Supposethat, as the underlying instincts evolve and are being emancipated and adjusted, the balance oftendencies gradually changes such that the tendency to incubate the external egg while outsideof the nest falls below the tendency to return to the nest, while the tendency to return remainsbalanced with shifting, so that returning and shifting are expressed together. In that case, thesetendencies may evolve gradually, while the new trait may arise punctually: it may be possible fora bird species to evolve backward-walking retrieval as a whole and rather rapidly, causing a “phasetransition” at the level of the observed behavior. Backward walking will then appear first in oneindividual, then in another, then more and more—it will appear “like the rain.” Thus, punctualismis better understood when we start thinking in terms of network evolution, as an outcome of gradualtrends in the change of a network that interact with each other.56 .13 How evolution learns: circuitous vs. accelerated development
The example of the evolution of egg retrieval highlights a fifth aspect of innateness. The terns arenot learning to retrieve the egg in the same way that humans learn a task. They have a set ofinborn tendencies that, in the situation, result in egg retrieval. At the same time the haphazardbehavior which results in egg retrieval does not seem to fit the term “innate.” There is anotheraspect to innateness, and that is the degree to which a behavior develops straightforwardly andquickly in an endogenously driven fashion. Two opposing examples will demonstrate this point.After an emerged butterfly finishes its preparations for flight (like drying its wings), it takes tothe air and flies in search of food and mates. While the behavior of taking off is instigated byhaving just completed preparations (so it too is dependent on “experience” in some sense), it isnot driven by their completion, much like a car is not pushed forward by the gas pedal. It isendogenous for all intents and purposes and is a true instinct. In contrast, the behavior whichresults in egg retrieval in the terns arises circuitously and at a high level, from a meeting ofdifferent inborn behavioral elements as well as environmental factors which play a more inherentrole in inducing the behavior: the visual stimuli and conflicting instincts do cause the retrievalof the egg. This high level meeting of modules, both internal and external, now serves as thesource of evolution from fuzzy to sharp, at the end of which a new innate module, consisting of acombination of the previously independent elements, will arise (that of backward walking). Thisaspect, namely, how quickly, straightforwardly and endogenously a behavior (or a trait in general,including morphological traits) arises in development is important for acceleration: In the initialcircuity there is a potential for “straightening up,” there is a potential for simplification that willlead to the further breaking of barriers (e.g., substantially faster egg retrieval). The environmentalfactors play a more inherent role in inducing the behavioral outcome in the tern situation than inthe butterfly situation (they are less like the gas pedal and more a part of the engine), which meansthat, in becoming emancipated from them, the life form can “learn” from the environment throughevolutionary change. That is, it now produces more endogenously what the environment helped toproduce before.
I argue that this emancipation is the intergenerational “learning” that is done byevolution , drawing an analogy between evolution and learning (see more in section 5.6).It is interesting that those who have tried to define innateness often seemed to mean that, in57ontrast with learned behavior, innate behavior is in a sense “predetermined” [121]. In other words,in innate traits, the fit with the environment is predetermined, as opposed to learned behavior ormorphological plasticity, where the fit is “acquired.” However, notice that this predetermined fit is adaptation. The evolution of innateness, or automatization, is the evolution of adaptation.And, according to interaction-based evolution, the evolution of adaptation involves network-levelevolution and the acquisition of a new phenotypic meaning as a result of the changing context inwhich modules are embedded. Interaction-based evolution shares with neo-Darwinism the relianceon natural selection for evolution of adaptations; it shares with Lamarckism the appearance of theinheritance of acquired characters (though it relies on a completely different mechanism); but itshares with neither the new idea that evolution is network-based and interactions-based. Interaction-based evolution argues that the process whereby a population converges [93] on anadaptation is a process that converts information from a less orderly to a more orderly state. Itproceeds from a fuzzy to a sharp, well-working and stereotyped state. However, evolution is not only a fuzzy-to-sharp process, in that the fuzzy source must first arise. The progress from fuzzy tosharp is therefore only a half of a cycle of the “engine of creativity” that is evolution. The otherhalf is that previously made sharp elements come together at a high level to make the new fuzzysource (e.g., the different instincts in the tern situation come together into a disorderly form ofegg-retrieval), from which new sharp elements can be made (e.g., backward walking) . I arguedthat simplification under performance pressure connects the two parts of the cycle. The simpleelements it creates not only are improvements but also come together in new complex interactionswhich serve as the raw material for the next round of simplification. Thus, novelty arises not fromaccident, but from evolutionary work. In this section I will revisit the molecular level from the perspective of interaction-based evolutionin light of the concepts learned so far. I will clarify the nature of mutation and raise directions for Of course, these cycles do not occur in a sequence one at a time. At any time point in the course of the evolutionof a given life form we may expect many co-occurring cycles, each at a different phase.
While [93] developed the micro-scale view of network-based evolution, the current paper developsthe macroscale view of it. In both, there is a sense of chunking: On the macroscale, genes as wellas phenotypes can become welded in the long-term. On the microscale, information from multipleloci comes together in each of many mutational events (including epigenetic changes) in each ofmany individuals in each of many generations.An important remark may now be made. One might think that, according to [93], non-accidentalmutation combines information from alleles at multiple loci into one locus in a way that recreatesin the mutated locus the combination of alleles as it was. However, this is not what is meant in [93].According to interaction-based evolution, there is a flow of information from the combination ofalleles across loci into one locus, which generates a hereditary effect; but this effect does not replicatethe combination as is. The situation is analogous to that of a neuron (or a logical gate), whoseoutput does not replicate its inputs yet represents a lasting effect of this combination of inputs thatis transferred to the next layer in the network. In other words, interaction-based evolution arguesthat many non-accidental mutational events over many generations and at many loci come togetherinto a network of information flow across the genome and through the generations, from many lociinto one and from one locus to many, and this information flow gradually leads to phenotypicchunking at the macroscale, among else. That is, the information flow in Figure 2 is the moment-to-moment workings of evolution; cooption and chunking at the macroscale (e.g., gene fusion orphenotypic fusion) are among the long-term consequences of it.This point may also help us understand better the gradual manner of occurrence of a gene fusionor of a splicing pattern which, according to interaction-based evolution, are not sudden stochasticevents but the results of long-term processes. As discussed earlier, alleles at multiple loci affect theregulation of an alternative splicing pattern, and the information they represent is processed andstabilized in the long term through convergence as defined in [93], thus setting the new alternativesplicing patterns and new contiguous sequences that we see today. Many writing events, in manyindividuals, over many generations, gradually pave the way for network evolution at the gene level(see the example of the fusion of
TRIM5 and
CypA in section 2.1).59 .2 The two ecologies working together: the ecology of energy and the ecologyof information
I will now attempt to put the various arguments of [93] and of this paper into one philosophicalpicture. A machine has several aspects: First, it is a finite, unchanging structure that repeatsits operation over and over again, performing the same “trick.” Second, we tend to think of amachine as something that operates harmoniously and whose parts have been conceived to fit eachother harmoniously. Third, novelty or “out of the box” thinking is the antithesis of machine-likebehavior.Now, the traditional idea of natural selection and random mutation is machine-like in the firstsense: it is one trick that repeats itself indefinitely without changing its own fundamental nature.That is, random mutation occurs either as an error during replication or for another accidentalreason, and natural selection either accepts it or rejects it. The repetition of this operation istraditionally supposed to be responsible for all of life and every innovation in it—a belief that Ihave argued against.Here, I will draw the distinction that the writing of mutation postulated by interaction-basedevolution [93] is not machine-like in any of the above senses. First, the writing itself evolves and itsevolution is fundamental to its operation—its operation is not repetitive [93]. Second, the workingsof evolution are not devoid of internal conflict but rather based on it, as will be discussed shortly.Third, the production of novelty is at the essence of evolution (Notice, however, that while evolutionis not machine-like, its products are machine-like: evolution is a process of automatization).What is the nature of the writing of mutations then? As discussed in [93], the mutation writingphenotype has the same meta-structure as that of the performing phenotype in the following sense.Take locomotion for example: we share with bears the fact that we have four limbs; but unlike bearswe are habitual bipeds; and each of us may have a specific leg length and muscular details slightlydifferent from those of others. In other words, a trait consists of widely shared and generally de-fined characteristics along with more specific and more narrowly shared characteristics, up to andincluding individual differences. According to interaction-based evolution, the mutation writingphenotype is the same in this regard [93]. It consists of generally defined and widely shared char-acteristics (for example, the long-term trend of the movement of genes out of the X chromosome in60 rosophila [145]), along with more specific and narrowly shared characteristics up to and includingindividual differences in mutational tendencies [93]. This meta-structure explains the observationson genetic relatedness in mutational tendencies discussed in [93]. Furthermore, it implies that thenature of mutational mechanisms can be conceptualized by analogy to ecological interactions: Thewriting of mutations happens not according to a fixed “rule” but by the ever evolving “rules of thejungle.” This “jungle” is a complex one consisting of DNA and other biomolecules. The actorsin it—the genetic influences on mutation—meet in an individual due to sexual reproduction, andgenetic changes happen in accordance with the usual tendencies of the actors as well as in accor-dance with their individual characteristics and the particular combination they appear in in thegiven situation. All the while, the actors themselves slowly evolve in the long-term. Thus, when wetalk about the workings of mutation, we are not talking about a harmonious, repetitive operationof a single mechanism. Instead, we are talking about the workings of an “ecology,” except that theoutcome is not remembered in terms of energy transfers such as food-web interactions but in termsof symbolic changes in genomes. It is an ecology of information.According to this picture, the biological world has two facets to it, two “forces”: one that isdue to biological interactions that make their mark through differential survival and reproduction;and one that is due to biological interactions that make their mark through the writing of geneticchanges [93]. These latter biological interactions are not limited to molecular mechanisms operatinginside the germ cells, but involve also everything else that affects the writing of mutations, such asmechanisms of mate choice and of the sexual shuffling of the genes [93].These two forces come together in the individual: the selection of individuals determines whichalleles will be passed on, and the writing of alleles determines which alleles will be there in the firstplace. Thus, selection and writing are equally influential opponents, and they both participate inchanging the genetic and phenotypic nature of the organism and thus of themselves. While eachof these forces has some long-term (phylogenetically shared) tendencies, each is oblivious to thepresent, immediate workings of the other: the intra-organismal writing of mutations that takesplace at the present moment is shielded from the external workings of natural selection that takesplace at the present moment, and likewise the latter is unaffected by the former, even thoughthe consequences of each will eventually affect the nature of the other. In that sense, evolutionarises from a conflict, or a process of negotiation, between these two fundamental forces, and what61appens in the long-term must be more or less congruent with both.
I argued above that the writing of mutations is analogous to an ecology. An ecology is a systemof conflicting forces, where each species presses to produce more of itself while at the same timeundoing the growth of others. And indeed, when we look at molecular evolutionary changes, weoften see long-term processes that are the result of a balance of forces in the short-term.As an example, consider tandem gene duplication. Due to the nature of the mutational mech-anisms of tandem duplication and deletion, a gene that is duplicated at tandem experiences anincreased chance not only of being further duplicated but also of losing a copy [59]. At the sametime, mutations that arise in the copies in the course of evolution push toward evolutionary di-vergence of the copies and thus toward the cessation of duplication/deletion (because homology isrequired for tandem duplication/deletion), while gene conversion events push to make the copiesthe same again, a situation where copies are more likely to disappear. Evolution here is a reversibleprocess where the long-term outcome depends on a balance of forces. Note also that, in this case,gene conversion may be seen as simplification, and diversification as complexification, and that theopposing tendencies to duplicate and specialize on the one hand and to equalize and collapse onthe other may be part of maintaining a balance between over-specialization (or “over-fitting”) andover-simplification, showing the importance of the “ecology of information” for evolution.As another example, consider the evolution of CpG content, which plays a role in gene regulationand therefore development [134, 31]. The cytosine in CpG dinucleotides mutates into thymine at ahigh rate after it is methylated [65], causing CpG-poor islands to lose their CpGs [112]. Importantly,this cytosine is methylated by complex enzymatic processes [77], which means that the locationsof these mutations are determined biologically, not by accident [93]. At the same time, anothermutational force—that of biased gene conversion (BGC)—adds cytosine to some CpG-poor islands[52, 36], and it has been shown that the balance of such forces determines the direction of theevolution of CpG content [112]. We have here a balance of mutational forces that in the long termaffects functional, adaptive structure [134, 31], which fits with interaction-based evolution [93] but62s hard to explain from the traditional view of evolution . Indeed, CpG mutations are not a rareanomaly, but have been estimated to account for nearly 25% of all point mutations in humans [51].As yet another example, based on an analysis of short open reading frames in yeast, Carvuninset al. have suggested that the evolution of new genes is gradual and reversible [19]: that a newgene does not arise suddenly as a complete whole, but gradually through forms more and moreresembling a complete gene; and that at each point in time, the gene can make a step toward oraway from completion. I argue that this process too is driven by a balance of forces. In mammals,for example, it would involve the evolution of CpG content.Finally, consider the proliferation vs. silencing and removal of transposable elements (TEs).There has been a well-known divide between those who think that TEs are serviceable to theorganism (e.g., [106, 43, 111, 14, 130]) and those who see them as “selfish-elements” [29, 117, 35] :on the one hand it is now clear that TEs play an immense role in adaptive evolution [11, 127, 42, 106],and on the other hand the evolutionary “benefit” they bring resides in a timescale too long to allowthem to fit comfortably in the traditional conceptualization of evolution except as selfish elements.However, the writing of mutations is an ecology; it is not a machine-like. TEs may well act asthough they are propelled to replicate and insert themselves wherever they can, and yet, in thecontext of the rules of the evolving information ecology, they may be serving the evolution of theorganism in the long-term . Indeed, giving contra-pressure to TE proliferation is an extensive andphylogenetically deep system of regulation, involving methylation and TE removal, active in thegermline [135]. I argue that this extensive system is part of the ecology of mutation writing and,like Fedoroff [43], I argue that it does not merely act as an “immune defense.”The four examples above clarify the view of genetic change as an ecology of information. It isa view of conflicting forces pushing against each other, including long-term reactions that may belocally reversible. This process computes in the long-term and involves the evolution of the network Indeed, as Duret and Galtier argued, CpGs work en masse, and the impact of any one particular CpG mutationis insignificant and could not be explained by traditional natural selection [36]. This is the case even though CpG dinucleotides account for only about 1% of the human genome. Further-more, they are often accompanied by mutations in nearby bases [123, 157], further compounding the involvement ofmutational mechanisms in their origination. along with some attempts to reconcile the opposing views [33, 34, 93]. This is the case even if they have the appearance of “selfish players” that has been attributed to them, and evenif they occasionally cause accidents in the short-term in the form of genetic disease [93].
63s a whole: the network gradually changes as it finds where it can give way under this complex setof forces. Thus, through mutational writing, the network processes a large amount of informationunder natural selection.
Interaction-based evolution opens up the search for non-Lamarckian yet useful mutational mecha-nisms. Earlier I proposed a gene-fusion mechanism that may play a role in evolution reminiscent ofthe role that Hebbian learning plays in neural networks (section 2.1; see also section 5.6), accordingto which copies of genes that are used together are more likely to be fused together. This typeof mechanism would not cause accidental changes but rather would produce evolutionarily usefulgenetic variation, without violating the principle that mutation does not respond to the immediateenvironment .Another example of a type of mechanism that would make sense in light of the current theory isas follows. Since information about the pattern and extent of expression of a gene is present in theDNA and accessible in the germ cells in principle, a gene that is highly expressed and is thereforeextensively used may be more likely to be duplicated by transcription-coupled mutational mecha-nisms. As in the case of the gene-fusion mechanism mentioned above, transcriptional promiscuityin the germ cells may be involved in such mechanisms (see section 2.1). When operating in uni-cellular organisms, such mechanisms could explain, among else, cases of rapid adaptive evolutionin response to environmental pressures such as extreme temperatures, extreme salinity, or toxins,where a gene whose product is in demand is duplicated/amplified (see [78] for review) and thus,at first glance, may seem to imply Lamarckism. However, mechanisms of this sort, coupling geneexpression level to gene duplication, may serve evolution in general in a non-Lamarckian fashion.Namely, increasing the probability of germline duplication of a gene that is extensively used in thesoma may be useful because such a gene may have a greater potential to beneficially specializeevolutionarily into different functions. Thus, while in unicellulars, environmental pressures may Interestingly, while the mutational fusion mechanism hypothesized earlier is based on putting together empiricalobservations (namely the nature of transcription, chromatin states, reverse transcription and gene fusion), its gen-eral nature was predicted based on theoretical considerations of interaction-based evolution [93]: in the latter, theconceptual connection between the problem of sexual recombination and mutational mechanisms required that genesplay a dual role: one of performance under natural selection, and another of influencing mutation in the germline[93]. Thus, the coupling of germline mutation and somatic performance through transcription as in the mechanismhypothesized earlier allows for a convergence of ideas and empirically based considerations.
CypA and
Trim5 serves as an example here as well, since this fusioninvolved duplication of
CypA through retrotranscription, and extensive transcription of
CypA inthe germline may have facilitated this evolutionary event [71].Also of interest in this regard are cases such as the evolution of insecticide resistance in themosquito
Culex pipiens due to the amplification of the genes coding for two non-specific esterasesas well as for the acetylcholinesterase that is the main target of the applied insecticides, which isactive in the central nervous system [89, 85]. These duplications may have originated not by accidentbut by a gradual albeit rapid process of evolution involving natural selection and non-accidentalmutation, which has created the genetic conditions under which the duplication mechanisms aremore likely to be activated .The current theory draws our attention to the fact that mechanisms such as the above can existand puts them front and center. The examples above demonstrate that the evolving organism canreceive feedback on what genetic changes would be useful to attempt—for example, what genesmay be beneficially chunked together. Furthermore, this feedback comes not from the immediateenvironment, but from the population’s past successes—the information is in the genome; there isno Lamarckism here. By accepting that mutation is not random, we can see that many findingsregarding genetic activity that have been thought of before as separate phenomena may actuallybe working together toward a larger goal—allowing evolution to be a smart process, but one thatrelies on defensible principles. This view opens the door to examining future research questionsthat would not have come to light otherwise.Indeed, we may expect a diversity of mutation-writing mechanisms in nature, and the above aremerely two examples of these. While some of these mechanisms may be well known phenomena thathave not yet been placed within a theoretical framework—for example the fact that recombinationalmechanisms interact with DNA sequences in such manner that enables whole gene duplication anddeletion—many others may remain to be discovered. Such a process would enable selection on multiple or many loci to be funneled by mutational mechanisms toinfluence the probability of duplication of a particular gene or genes. .5 The intimate relationship between useful change and error repair It is so often said that mutation is a replication error that one might think that this is a well-knownscientific fact. However, the only fact that has actually been established is the basic observationitself—that while some genes are duplicated, others undergo genetic change. To say that thesechanges represent nothing more than “replication errors” is to provide merely one interpretationto this fact, and it may be a prejudiced one. This interpretation has led among else to the terms“error-repair mechanisms” and “error-prone” repair mechanisms, which, according to the theorypresented here, may end up detracting from our understanding of evolution.Error is often a deviation from a pattern. By noticing a deviation from a pattern we can findand fix a typographical error: a word with a typo slightly differs from many instances of the wordseen before, which are all identical to each other; and it can be fixed by making it identical tothose many other instances. Then again, by noticing a deviation from a pattern, we can also avoidpicking a rotten apple in the store, even if we have never seen rot or an apple before. Taking onestep further, by noticing a deviation from a pattern we can also spot an error of thought. Take forexample
Scala Naturae , according to which all organisms fall into a linear order from the simplestto the most advanced. From that perspective, the fact that many organisms are hard to classifyas more or less advanced in relation to each other is a deviation from a pattern. By replacing
Scala Naturae with Darwin’s concept of common descent, this difficulty of classification becomesnot a deviation from a pattern but a part of a larger pattern involving other facts. Thus, botherrors of typing and errors of thought can be corrected by pattern completion at different levels.At the same time, pattern completion is a form of simplification—the fewer exceptions we need tohave, the smaller the amount of information needed to describe the entire system and its parts, asinformation theory makes clear [91]. Thus, if pattern-completion operations can be implemented bymutation, we may see the same genetic mechanisms operating both in “typographical corrections”and in the kind of mutational writing that leads to progressive evolution. As an example, repeatedevents of gene conversion have the potential to correct “typos,” but they also have the potentialto implement simplification pressure opposing the complexification pressure of diversifying in thecase of duplication and deletion discussed in section 5.3.The insufficiency of our current jargon is made particularly clear by the phrase “error-prone66epair.” Suppose that, in some cases where so-called “error-prone repair” is activated, the biologicalsystem is actually pushing for a change rather than a restoration of the genetic state, and thatthis change is a part of a pattern-completion or other progressive evolution of the network asa whole. Then what we have heretofore thought of as “error-prone” is actually an attempt at“error-correction,” where the “error” is of a different, deeper kind than we use to think about.
From Paley to Dawkins [29], there is universal agreement that adaptations are incredibly impres-sive and complex pieces of “natural technology.” While Paley used this observation to make thenon-scientific point that, much like a watch has an intelligent watchmaker, life was created by asupernatural intelligence [118], Dawkins argued that the process responsible for life is a very simpleprocess of random mutation and natural selection that is fully understood at its essence. In thepreface to
The Blind Watchmaker [30], he wrote that “[t]his book is written in the conviction thatour own existence once presented the greatest of all mysteries, but that it is a mystery no longer...Darwin and Wallace solved it, though we shall continue to add footnotes to their solution for a whileyet.” Let us revisit the question, but from a strictly scientific perspective, and without assumingthat all that is important in principle was revealed at Darwin’s time: Could the process generatinglife forms and the process generating artificial technology be similar in some respects?Interestingly, according to Papaj [121], it is a curious historical fact that the earliest ideas onevolution, i.e., Lamarckism, revolved around observations on automatism in behavior: observationsshowing that instinct is similar to well-learned behavior —an evolved phenomenon is similar to alearned phenomenon—in that both can be carried out automatically and independently of externalinfluences, and both are stereotyped, or robotic, repeating with a high degree of uniformity. Theseobservations fostered the idea that what is repeated many times over the generations graduallyimpresses itself upon the hereditary makeup of the organism, which then led to the additional buterroneous idea of Lamarckian transmissionism. Until now, Lamarckism has been the only alterna-tive to natural selection at the most basic level of analysis. And even though it has been rightfullyrejected as a general-level explanation for evolution, the observations it was supposed to explainare still here (and have been discussed in section 4). That is, the controversy was never over the observations but rather over the mechanism of evolution. The current paper provides a new inter-67retation of these original observations and suggests that there is a connection between evolutionand learning: network-level evolution and automatization are key to both. This connection is free ofLamarckian transmissionism and requires a process based on non-accidental mutation and naturalselection.Not only in evolution, but also in the study of brain and behavior, the notion of randomgeneration and filtering was once used. For instance, Skinner had suggested a mechanism of randomgeneration of ideas and filtering for how learning by the brain works [50]. However, more recently,Fodor and Piattelli-Palmarini argued that such a mechanism applies neither to evolution nor tothe brain [50] . In this paper and in [93], I argued that the mechanism of evolution is not that ofrandom generation and filtering, and that the causes of mutation are critical for our understandingof evolution. This may also inform our understanding of learning.In this context of connecting evolution and learning, Valiant’s [143, 144] recent work attemptingto connect evolution and machine learning (see also [45, 47, 46, 72, 3] in the same line, and [20])signifies a methodological turning point: unlike classical population genetics, it provides rigorousmathematical techniques that capture analytically a complex phenotypic structure and allow usto quantify and study the evolution of complexity [143, 144]. Thus, with respect to theoreticalmethodology, it is a grand vision and, in principle, it allows mutation to be non-accidental .However, while Valiant’s framework allows for, but has not yet substantially pursued, non-accidentalmutation, interaction-based evolution argues that mutation is non-accidental and that this is crucialfor evolution. And while Valiant’s work may be an inspiring step in the right direction, according tothe present paper, there are elements that are not yet included in it that are essential for biologicalevolution based on non-accidental mutation. These include cooption; the idea that simplificationunder performance pressure produces elements that have the inherent capacity to become usefulin new contexts, which leads to cooption; the idea that learning through evolutionary change isa learning from the environment by emancipation and acceleration (see section 4.13), i.e., by theevolution of automatization and innateness; and the concept of the absorption of meaning fromcontext under gradual network-level change. Indeed, the importance of cooption in evolution cannot Likewise, Lakatosh asked whether there is nothing more to intelligence than randomization as generating ideasand selection sifting among them [86]. It allows mutation to be an outcome of any implementable randomized algorithm—an algorithm that is allowedto use random bits among else , which is different from mutations that are nothing but random changes anywhere inthe genome
68e overestimated, and has been demonstrated here at both the molecular and phenotypic levels(sections 2 and 4). Furthermore, cooption is analogous to an analogy or metaphor, which are crucialin the evolution of language as well as in human intelligence. It may be of much interest to explorethese missing elements from a computational perspective.Since its inception [93], interaction-based evolution has been deeply connected to the computa-tional worldview [120, 73], because it proposed that mutation is an event of information flow andcomputation: the inputs into a mutational event are the alleles at the loci affecting the mutationthrough genetic interaction, and the output is the mutation itself (by “mutation” I mean not only achange in the DNA but any heritable change, such as an epigenetic change) [93]. Furthermore, thefact that the output of a mutational event at one generation, namely the mutation itself, can serveas an input into mutational events at later generations means that the mutation-writing phenotypecreates a network of information flow through the generations, from many genes into any one geneand from any one gene to many (see Figure 2) [93]. Other examples of networks of informationflow and computation include the brain, and what computer scientists call a circuit [119, 161] (oneinstance of which is an artificial neural network [66]). Thus, according to interaction-based evolu-tion, genetic evolution can be seen as the result of the workings of a network, itself evolving overtime.Interestingly, in artificial neural networks, local computational elements are used such as Heb-bian learning (e.g., [66]). In the latter, when one neuron persistently participates in causing anotherto fire, the strength of the synapse between them is increased [62]. Hebbian learning is an exampleof a local simplification operation that, in the context of the gradual change of a complex network,is useful. Now, elements of this sort can play a role in the network of information flow generatedby sex and non-accidental mutation proposed by interaction-based evolution [93]; indeed, the mu-tational fusion mechanism in section 2.1 is one such case. Thus, we see in multiple ways that,according to interaction-based evolution, evolution and “thinking processes” have more to do witheach other than previously thought, even though no Lamarckism and no “foresight” or “adaptivemutation” as traditionally defined are involved. Thus, the study of evolution could inform thestudy of learning and vice-versa.Recently, a connection between evolution and learning was drawn by Watson and Szathm´ary[160] and Watson et al. [159]. While this connection shares with the theory of interaction-based69volution as proposed in [93] and here the idea that evolution is network-based, and that thechange of connections between the nodes of the network is key, there are also some fundamentaldifferences between the two. Watson and Szathm´ary [160] and Watson et al. [159] did not arguefor non-accidental mutation, and all that follows from it.For example, it follows from non-accidental mutation that Hebbian-learning–like mechanismscan be implemented directly by the mutational mechanisms themselves (as opposed to needing toarise from random mutation and natural selection), as discussed in section 2.1. There, I argued thatgenes that are used together are fused together . More generally, I argued that simplification canbe implemented by mutational mechanisms. In fact, the very concept of non-accidental mutationitself represents a vast network, as discussed here and in [93]. Conceptualizing mutation not as alocal accident disconnected from its genetic environment, but rather as the outcome of network-based processes, provides a far more involved network-based view of evolution than otherwise. Italso greatly strengthens the connection between computer science and evolution [120, 73].In addition, borrowing from knowledge in machine learning, the above-mentioned authors men-tion that, among other things, imposing parsimony pressure by imposing a connection cost inmodels of genotype-phenotype maps can facilitate evolution in these models [160, 159, 22, 81].However, they do not put simplification pressure front and center in biological evolution, as donehere. The present paper established the importance of simplification in biological evolution byproviding both the rationale and many empirical examples from both the molecular and pheno-typic levels behind this point. On this foundation, it argued that biological evolution is driven bytwo forces—the pressure for performance and the pressure for simplification. A cycle in evolutionbegins at a fuzzy state from the emergent interactions between preexisting elements. From theseinteractions, simplification under performance pressure creates new elements that have the inherentcapacity to come together into unexpected, useful interactions with other such elements. This leadsto cooption, and to the beginning of another cycle in the process. Thus, putting simplification frontand center in biological evolution also puts cooption at the heart of the evolutionary process. Inaddition, from this cycle we also obtain the idea that simplification leads to biological complexity In contrast, Watson et al. argue that “genes that are selected together are wired together.” There is a fundamentaldifference between the two statements, because Watson et al. imply a process of random mutation and naturalselection. Namely, they base their statement on a pioneering theoretical model [122], but one that is constructedwithin the random-mutation view. per se if traditional selection is to simplify a genetic network based on random mutation, accordingto interaction-based evolution, simplification is inherent to biological evolution, and can be im-plemented by mutational mechanisms. Indeed, interaction-based evolution argues that mutationalmechanisms, mixing of the hereditary material (which has evolved into sex), simplification andselection have all existed from the “beginning” of life, and that they did not evolve from an asexualworld with random mutation, since that world never existed [93]. Thus, they are not elements thatevolved by random mutation and natural selection based on different costs and benefits imposedby selection, but rather are primary elements that are original and inherent to the process of evo-lution. Thus, interaction-based evolution is different from the evolvability view present in previousbiological literature.Indeed, interaction-based evolution provides a complete, biologically motivated, conceptualframework for evolution with non-accidental mutation at its center. By arguing that novelty arisesfrom emergent interactions, it places the source of novelty at the system level. This in turn replacesthe notion of accidental mutation as the ultimate source of heritable novelty, which in turn connectsback to the center-piece of non-accidental mutation. This entire framework and all of its elements,including cooption, novelty and non-accidental mutation, as well as the idea that simplificationleads to complexity, and the idea that evolutionary learning occurs through automatization andinnateness, have not been discussed in previous papers on evolution and learning.
How does novelty arise? Traditional evolutionary thinking relies on random mutation and naturalselection. The idea is that radiation, or a copying error, or oxidative stress, “goes zap,” and a newmutation appears that, on rare occasions, provides a beneficial phenotypic change. All that remainsfor natural selection to do is to check whether this mutation on its own is good or bad—to playthe role of a filter. Where does the novelty, the new genetic information, come from? Presumably,in that view, it comes from the accident itself—from out of thin air—and there is nothing more toinquire regarding the source of it. 71nteraction-based evolution proposes an alternative to this view. The mutations that are rele-vant for adaptive evolution under selection are due to mutational mechanisms that are continuallyevolving, and that do not in and of themselves invent things. Rather, novelty arises from the systemlevel—from the macroscale—from gradual network-level evolution, as these mechanisms absorb in-formation from selection. In brief, mutation mechanisms perform simplification operations on thegenetic network, as well as gene duplication, in a heritable mode. These mechanisms work togetherwith natural selection which acts on the organism as a complex whole, so that adaptive evolution isa process of simplification under performance pressure. A cycle in this process begins with complexhigh-level interactions between preexisting elements. Simplification under performance pressuretakes these preexisting interactions and, gradually, in the course of evolution, creates from themnew elements—new adaptations. Because these new elements are created in a process of simplifica-tion under performance pressure, they have the inherent capacity of coming together in new, usefuland unexpected interactions at higher levels, thus initiating another cycle in the process. Thiscapacity to come together in useful high-level interactions that have not been pursued in advanceis the source of novelty in evolution. In short, mutations do not in and of themselves invent things,but rather are a key activity that takes part in turning the wheel of evolution. Interestingly, it issimplification that explains complexity: local simplification leads to a global increase in complexity.Thus, while traditional theory is based on the idea that random mutation invents—where thissupposed random mutation is a remote, presumed event that cannot be seen or confirmed—thetheory presented here is based on the empirically evident fact that preexisting elements cometogether into new, useful high-level interactions as the source of novelty in evolution. Note that itmatters not whether the novelty involved in the transitions from the genes
TRIM5 and
CypA totheir fusion, or from haphazard egg retrieval to backward walking, is small or great in and of itself.Rather, these transitions exemplify the steps that tie together the process of evolution, which inthe long term lead from the progenote to humans.We have a tendency to look for “foundations” from which everything else can be derived. Inparticular, it is convenient to assume that the causes of mutation are random, because it puts an endto all of our questions. The philosophical move that is required from the perspective of interaction-based evolution is to let go of the notion that random mutation and novelty from a point are at thebottom of things—that they provide a stable ground upward from which a conceptual edifice can72e built; and to accept instead that the action is at the network level: that both the meaning andorigin of genetic and phenotypic elements comes from the higher levels of organization—it comesfrom the network—from “above.” This move opens up the study of evolution substantially; becausewhile the notion of random mutation means that there is nothing of importance to be studied aboutthe causes of mutation from an evolutionary perspective, the concept of non-accidental mutationprovided by interaction-based evolution implies instead a whole world of biological mechanismsopen to investigation.Before Darwin, people used to think that different species were each created separately in aninstant. While Darwin made an immense contribution by showing that this was not the case, andthat species are generated gradually, a notion of creation in an instant has been maintained in neo-Darwinism in other areas: the origin of life, the origin of mutations, and cooption. While [93] arguedamong else against the origin of life as an instant, this paper argues against the other two. Noveltyarises not suddenly from a point but from gradual network-level evolution. Indeed, if evolutionaccording to random mutation and natural selection is a sequence of independent points, eachrepresenting a local accidental mutation disconnected from the rest, interaction-based evolutiondraws the lines between these points (see [93]) while fundamentally altering their interpretation.
I would like to thank Marc Feldman, Avraham Korol, Simon Levin, Amos Livnat, Daniel Melamed,Steve Pacala, Christos Papadimitriou, Nick Pippenger, Umesh Vazirani and Kim Weaver for in-valuable conversations during the course of the study. I would like to thank the Department ofEvolutionary and Environmental Biology and the Institute of Evolution at the University of Haifafor providing an intellectual environment conducive to the pursuit of this work, and for stimulatingconversations. I would like to acknowledge financial support from the Miller Institute for BasicResearch in Science and from NSF grant 0964033 to Christos Papadimitriou, Division of ComputerScience, UC Berkeley, for support during a formative part of the work in the years 2006–2011.73 eferences [1] L. Altenberg and M. W. Feldman. Selection, generalized transmission and the evolution ofmodifier genes. I. The reduction principle.
Genetics , 117:559–572, 1987.[2] C. M. Amy, B. Williams-Ahlf, J. Naggert, and S. Smith. Intron-exon organization of the genefor the multifunctional animal fatty acid synthase.
Proc Natl Acad Sci USA , 89(3):1105–1108,1992.[3] E. Angelino and V. Kanade. Attribute-efficient evolvability of linear functions. In
Proceedingsof the 5th conference on Innovations in Theoretical Computer Science , pages 287–300. ACM,2014.[4] W. Arkwright.
The Pointer and his Predecessors: An Illustrated History of the Pointing Dogfrom the Earliest Times . Arthur L. Humphreys, London, 1902.[5] E. A. Armstrong. The nature and function of displacement activities. In
Symposia of theSociety for Experimental Biology , volume 4, pages 361–384, 1950.[6] G. Baerends. On the life-history of Ammophila campestris jur. In
Proc K Ned Akad Wet ,volume 44, pages 483–488, 1941.[7] G. W. Barlow. Modal action patterns. In T. Sebeok, editor,
How Animals Communicate ,pages 98–134. Bloomington, Indiana, 1977.[8] K. Bateman. The genetic assimilation of four venation phenocopies.
J Genet , 56(3):443–474,1959.[9] K. Bateman. The genetic assimilation of the dumpy phenocopy.
J Genet , 56(3):341–351,1959.[10] P. Bateson. Are there principles of behavioural development. In
The Development and Inte-gration of Behaviour: Essays in honour of Robert Hinde , pages 19–40. Cambridge UniversityPress, Cambridge, 1991. 7411] G. Bourque, B. Leong, V. Vega, X. Che, Y. Lee, K. Srinivasan, J.-L. Chew, Y. Ruan, C.-L.Wei, H. Ng, and E. Liu. Evolution of the mammalian transcription factor binding repertoirevia transposable elements.
Genome Res , 18(11):1752–1762, 2008.[12] J. H. Bowie and M. J. Tyler. Host defense peptides from Australian amphibians: Caeruleinand other neuropeptides. In A. Kastin, editor,
Handbook of Biologically Active Peptides ,pages 283–289. Academic Press, San Diego, CA, 2006.[13] G. Brennan, Y. Kozyrev, and S.-L. Hu. TRIMCyp expression in old world primates
Macacanemestrina and
Macaca fascicularis . P Natl Acad Sci USA , 105:3569–3574, 2008.[14] R. J. Britten and E. H. Davidson. Gene regulation for higher cells: a theory.
Science ,165:349–357, 1969.[15] J. Brosius and H. Tiedge. Reverse transcriptase: mediator of genomic plasticity.
Virus Genes ,11:163–179, 1996.[16] W. Byers.
How Mathematicians Think: Using Ambiguity, Contradiction, and Paradox toCreate Mathematics . Princeton University Press, 2010.[17] P. J. Caldwell and G. W. Cornwell. Incubation behavior and temperatures of the mallardduck.
The Auk , 92(4):706–731, 1975.[18] S. Carroll.
Endless Forms Most Beautiful: The New Science of Evo Devo and the Making ofthe Animal Kingdom . W. W. Norton & Company, 2005.[19] A.-R. Carvunis, T. Rolland, M. Wapinski, I ad Calderwood, M. Yildirim, N. Simonis, B. Char-loteaux, C. Hidalgo, J. Barbette, B. Santhanam, G. Brar, J. Weissman, A. Regev, N. Thierry-Miueg, M. Cusick, and M. Vidal. Proto-genes and de novo gene birth.
Nature , 487(7407):370–374, 2012.[20] E. Chastain, A. Livnat, C. Papadimitriou, and U. Vazirani. Algorithms, games, and evolution.
Proc Natl Acad Sci USA , 111(29):10620–10623, 2014.[21] S. S. Chirala, M. Kuziora, D. Spector, and S. Wakil. Complementation of mutations and75ucleotide sequence of
FAS1 gene encoding beta subunit of yeast fatty acid synthase.
J BiolChem , 262(9):4231–4240, 1987.[22] J. Clune, J.-B. Mouret, and H. Lipson. The evolutionary origins of modularity.
Proceedingsof the Royal Society of London B: Biological Sciences , 280(1755), 2013.[23] J. Colgan, M. Asmal, and J. Luban. Isolation, characterization and targeted disruption ofmouse ppia: cyclophilin A is not essential for mammalian cell viability.
Genomics , 68(2):167–178, 2000.[24] A. Daanje. On locomotory movements in birds and the intention movements derived fromthem.
Behaviour , 3(1):48–98, 1951.[25] C. Darwin.
On the Origin of Species by Means of Natural Selection, or the Preservation ofFavoured Races in the Struggle for Life . Murray, London, 1859.[26] C. Darwin.
The Descent of Man, and Selection in Relation to Sex. Volume 1 . John Murray,London, 1871.[27] C. Darwin.
The Origin of Species by Means of Natural Selection, Or The Preservation ofFavoured Races in the Struggle for Life . John Murray, London, 6th edition, 1876.[28] F. Darwin, editor.
The Foundations of the Origin of Species: Two Essays Written in 1842and 1844 . CUP Archive, 1909.[29] R. Dawkins.
The Selfish Gene . Oxford University Press, 1976.[30] R. Dawkins.
The Blind Watchmaker: Why the Evidence of Evolution Reveals a UniverseWithout Design . WW Norton & Company, New York, 1986.[31] A. M. Deaton and A. Bird. CpG islands and the regulation of transcription.
Genes Dev ,25:1010–1022, 2011.[32] G. Deutscher.
The Unfolding of Language . Random House, 2010.[33] W. F. Doolittle. What introns have to tell us: hierarchy in genome evolution.
Cold SpringHarb Symp Quant Biol , 52:907–913, 1987.7634] W. F. Doolittle. Is junk DNA bunk? A critique of ENCODE.
P Natl Acad Sci USA ,110:5294–5300, 2013.[35] W. F. Doolittle and C. Sapienza. Selfish genes, the phenotype paradigm and genome evolu-tion.
Nature , 284:601–603, 1980.[36] L. Duret and N. Galtier. Biased gene conversion and the evolution of mammalian genomiclandscapes.
Annu Rev Genomics Hum Genet , 10:285–311, 2009.[37] D. Emera and G. P. Wagner. Transformation of a transposon into a derived prolactin promoterwith function during human pregnancy.
P Natl Acad Sci USA , 109:11246–11251, 2012.[38] D. Emera and G. P. Wagner. Transposable element recruitments in the mammalian placenta:impacts and mechanisms.
Briefings in Functional Genomics , 11:267–276, 2012.[39] H. Evans. The accessory burrows of digger wasps.
Science , 152:465–471, 1966.[40] H. Evans.
The Comparative Ethology and Evolution of the Sand Wasps . Harvard UniversityPress, Cambridge, Massachusetts, 1966.[41] D. Falconer.
Introduction to quantitative genetics . Oliver and Boyd, 1960.[42] C. Fechotte. Transposable elements and the evolution of regulatory networks.
Nature ReviewsGenetics , 9(5):397–405, 2008.[43] N. V. Fedoroff. Presidential address. transposable elements, epigenetics, and genome evolu-tion.
Science , 338:758–767, 2012.[44] M. W. Feldman, S. P. Otto, and F. B. Christiansen. Population genetic perspectives on theevolution of recombination.
Annu Rev Genet , 30:261–295, 1997.[45] V. Feldman. Evolvability from learning algorithms. In
Proceedings of the 40th Annual ACMSymposium on Theory of Computing , pages 619–628. ACM, 2008.[46] V. Feldman. A complete characterization of statistical query learning with applications toevolvability. In
Proceedings of the 50th Annual Symposium on Foundations of ComputerScience , pages 375–384. IEEE, 2009. 7747] V. Feldman. Robustness of evolvability. In
Proceedings of the 22nd Conference on LearningTheory , 2009.[48] S. E. Fischer, M. D. Butler, Q. Pan, and G. Ruvkun. Trans-splicing in
C. elegans generatesthe negative RNAi regulator ERI-6/7.
Nature , 455(7212):491–496, 2008.[49] R. A. Fisher.
The Genetical Theory of Natural Selection . The Clarendon Press, Oxford, 1930.[50] J. Fodor and M. Piattelli-Palmarini.
What Darwin Got Wrong . Profile Books, 2011.[51] K. J. Fryxell and W.-J. Moon. CpG mutation rates in the human genome are highly dependenton local GC content.
Mol Biol Evol , 22:650–658, 2005.[52] N. Galtier, G. Piganeau, D. Mouchiroud, and L. Duret. GC-content evolution in mammaliangenomes: the biased gene conversion hypothesis.
Genetics , 159:907–911, 2001.[53] M. B. Gerstein, C. Bruce, J. S. Rozowsky, D. Zheng, J. Du, J. O. Korbel, O. Emanuelsson,Z. D. Zhang, S. Weissman, and M. Snyder. What is a gene, post-ENCODE? History andupdated definition.
Genome Res , 17(6):669–681, 2007.[54] W. Gilbert. Why genes in pieces?
Nature , 271(5645):501, 1978.[55] J. Gould and P. Marler. Learning by instinct.
Scientific American , 256(1):74–85, 1986.[56] S. J. Gould.
Ontogeny and Phylogeny . Harvard University Press, 1977.[57] S. J. Gould and E. S. Vrba. Exaptation—a missing term in the science of form.
Paleobiology ,8(01):4–15, 1982.[58] T. Grandin and M. Deesing. Genetics and animal welfare. In T. Grandin, editor,
Geneticsand the Behaviour of Domestic Animals , pages 319–341. Academic Press, San Diego, 1998.[59] D. Graur and W.-H. Li.
Fundamentals of Molecular Evolution . Sinauer Associates, Sunder-land, MA, 2000.[60] B. Haendler and E. Hofer. Characterization of the human cyclophilin gene and of relatedprocessed pseudogenes.
Eur J Biochem , 190(3):477–482, 1990.7861] R. W. Hamming. The unreasonable effectiveness of mathematics.
Am Math Mon , 87(2):81–90,1980.[62] D. Hebb.
The Organization of Behavior . Wiley & Sons, New York, 1949.[63] O. Heinroth.
Beitr¨age zur Biologie, namentlich Ethologie und Psychologie der Anatiden . Verh.V. Int. Ornothol. Kongr., Berlin, 1911.[64] R. Hinde. A comparative study of the courtship of certain finches (fringillidae).
Ibis ,97(4):706–745, 1955–1956.[65] A. Hodgkinson and A. Eyre-Walker. Variation in the mutation rate across mammaliangenomes.
Nat Rev Genet , 12:756–766, 2011.[66] J. J. Hopfield. Neural networks and physical systems with emergent collective computationalabilities.
Proc Natl Acad Sci USA , 79(8):2554–2558, 1982.[67] J. Huxley. The courtship habits of the great crested grebe (
Podiceps cristatus ); with anaddition to the theory of sexual selection.
Zool. Soc. (London) Proc. , 2:491–562, 1914.[68] T. D. Ingolia and E. A. Craig. Four small drosophila heat shock proteins are related to eachother and to mammalian alpha-crystallin.
Proc Natl Acad Sci USA , 79(7):2360–2364, 1982.[69] F. Jacob. Evolution and tinkering.
Science , 196:1161–1166, 1977.[70] W. E. Johnson and S. L. Sawyer. Molecular evolution of the antiretroviral
TRIM5 gene.
Immunogenetics , 61:163–176, 2009.[71] H. Kaessmann, N. Vinckenbosch, and M. Long. RNA-based gene duplication: mechanisticand evolutionary insights.
Nat Rev Genet , 10:19–31, 2009.[72] V. Kanade. Evolution with recombination. In
Proceedings of the 52nd Annual Symposiumon Foundations of Computer Science , pages 837–846. IEEE, 2011.[73] R. M. Karp. Understanding science through the computational lens.
Journal of ComputerScience and Technology , 26(4):569–577, 2011.7974] M. Kimura. On the evolutionary adjustment of spontaneous mutation rates.
Genet Res ,9:23–34, 1967.[75] F. B. Kirkman.
Bird Behavior: A Contribution Based Chiefly on a Study of the Black-HeadedGull . T. Nelson & Sons, Ltd., London, 1937.[76] M. W. Kirschner, J. C. Gerhart, and J. Norton.
The Plausibility of Life: Resolving Darwin’sDilemma . Yale University Press, 2006.[77] R. J. Klose and A. P. Bird. Genomic DNA methylation: the mark and its mediators.
TrendsBiochem Sci , 31(2):89–97, 2006.[78] F. Kondrashov. Gene duplication as a mechanism of genomic adaptation to a changingenvironment.
Proc R Soc B , 279:5048–5057, 2012.[79] Y. Kong, H. Zhou, Y. Yu, L. Chen, P. Hao, and X. Li. The evolutionary landscape ofintergenic trans-splicing events in insects.
Nat Commun , 6:8734, 2015.[80] E. V. Koonin.
The Logic of Chance: The Nature and Origin of Biological Evolution . FTPress, Upper Saddle River, NJ, 2011.[81] K. Kouvaris, J. Clune, M. Kounios, and R. Watson. How evolution learns to generalise:Principles of under-fitting, over-fitting and induction in the evolution of developmental or-ganisation. arXiv , 1508.06854v1, 2015.[82] H. Kr¨atzig. Untersuchungen zur Lebensweise des Moorschneehuhns
Lagopus l. lagopus ,w¨ahrend der Jugendentwicklung.
J Ornithol , 88(1):139–166, 1940.[83] Z. Kuo. Ontogeny of embryonic behavior in aves. I. The chronology and general nature ofthe behavior of the chick embryo.
J Exp Zool , 61:395–430, 1932.[84] Z. Kuo. Ontogeny of embryonic behavior in aves. IV. The influence of embryonic movementsupon the behavior after hatching.
J Comp Psychol , 14:109–122, 1932.[85] P. Labb´e, A. Bethomieu, C. Berticat, H. Alout, T. Raymond, M annd Lenormand, andM. Weill. Independent duplications of the acetylcholinesterase gene conferring insecticideresistance in the mosquito
Culex pipiens . Mol Biol Evol , 24:1056–1067, 2007.8086] I. Lakatosh.
Proofs and Refutations: The Logic of Mathematical Discovery . CambridgeUniversity Press, Cambridge, 1976.[87] D. S. Lehrman. A critique of Konrad Lorenz’s theory of instinctive behavior.
QuarterlyReview of Biology , 28:337–363, 1953.[88] E. G. Leigh Jr. Natural selection and mutability.
Am Nat , 104:301–305, 1970.[89] T. Lenormand, T. Guillemaud, D. Bourguet, and M. Raymond. Appearance and sweep of agene duplication: adaptive response and potential for new functions in the mosquito
Culexpipiens . Evolution , 52:1705–1712, 1998.[90] S. Levin. The problem of pattern and scale in ecology: The Robert H. MacArthur AwardLecture.
Ecology , 73(6):1943–1967, 1992.[91] M. Li and P. Vit´anyi.
An Introduction to Kolmogorov Complexity and Its Applications, 2ndedition . Springer-Verlag, New York, 1997.[92] C.-H. Liao, Y.-Q. Kuang, H.-L. Liu, Y.-T. Zheng, and B. Su. A novel fusion gene, TRIM5-Cyclophilin A in the pig-tailed macaque determines its susceptibility to HIV-1 infection.
Aids ,21(Suppl 8):S19–S26, 2007.[93] A. Livnat. Interaction-based evolution: how natural selection and nonrandom mutation worktogether.
Biology Direct , 8(1):24, 2013.[94] A. Livnat and C. Papadimitriou. Sex as an algorithm: The theory of evolution under thelens of computation.
Journal of the ACM, In press , 2016.[95] A. Livnat, C. Papadimitriou, A. Rubinstein, G. Valiant, and A. Wan. Satisfiability andevolution. In
Proceedings of the 55th Annual Symposium on Foundations of Computer Science ,pages 524–530. IEEE, 2014.[96] K. Lorenz. Der Kumpan in der Umwelt des Vogels.
J Ornithol , 83:137–215, 289–413, 1935.[97] K. Lorenz. Vergleichende Verhaltensforschung.
Zool. Anz. Suppl. Bd. , 12:69–102, 1939.[98] K. Lorenz. The evolution of behavior.
Scientific American , 199(6):67–78, 1958.8199] K. Lorenz.
Evolution and Modification of Behavior . University of Chicago Press, Chicago,1965.[100] K. Lorenz.
On Aggression . Harcourt, Brace & World, New York, 1966.[101] K. Lorenz. Comparative studies of the motor patterns of Anatinae. In K. Lorenz, editor,
Studies in Animal and Human Behavior , volume II. Butler & Tanner Ltd, Frome and London,1971 (1941).[102] K. Lorenz.
The Foundations of Ethology . Springer-Verlab, New York, 1981.[103] K. Lorenz and N. Tinbergen. Taxis and instinctive behavior pattern in egg-rolling by theGreylag goose. In K. Lorenz, editor,
Studies in Animal and Human Behavior , volume I.Butler & Tanner Ltd., Frome and London, 1970 (1938).[104] S. L¨owel and W. Singer. Selection of intrinsic horizontal connections in the visual cortex bycorrelated neuronal activity.
Science , 255(5041):209–212, 1992.[105] M. Lynch.
The Origins of Genome Architecture . Sinauer Associates, Inc., Sunderland, MA,2007.[106] V. J. Lynch, R. D. Leclerc, G. May, and G. P. Wagner. Transposon-mediated rewiring ofgene regulatory networks contributed to the evolution of pregnancy in mammals.
Nat Genet ,43:1154–1159, 2011.[107] N. Marshall. Factors in the incubation behavior of the common tern.
The Auk , 60:574–588,1943.[108] J. Maynard-Smith and D. Harper.
Animal Signals. Oxford Series in Ecology and Evolution .Oxford University Press, 2004.[109] E. Mayr.
Animal Species and Evolution . Belknap Press, 1963.[110] E. Mayr. Behavior programs and evolutionary strategies: natural selection sometimes favorsa genetically “closed” behavior program, sometimes an “open” one.
Am Sci , 62(6):650–659,1974. 82111] B. McClintock. Components of action of the regulators
Spm and Ac . Carnegie Inst WashYearbook , 64:527–534, 1965.[112] N. Mendelson-Cohen, E. Kenigsberg, and A. Tanay. Primate CpG islands are maintained byheterogeneous evolutionary regimes involving minimal selection.
Cell , 145:773–786, 2011.[113] A. H. Mohamed, S. S. Chirala, N. H. Mody, W.-Y. Huang, and S. Wakil. Primary structureof the multifunctional alpha subunit protein of yeast fatty acid synthase derived from
FAS2 gene sequence.
J Biol Chem , 263(25):12315–12325, 1988.[114] F. M¨uller.
Facts and Arguments for Darwin . John Murray, London, 1869.[115] R. M. Newman, L. Hall, A. Kirmaier, L.-A. Pozzi, E. Pery, M. Farzan, S. P. O’Neil, andW. Johnson. Evolution of a TRIM5-CypA splice isoform in old world monkeys.
PLoS Pathog ,4:e1000003, 2008.[116] S. Nisole, C. Lynch, J. P. Stoye, and M. W. Yap. A Trim5-cyclophilin A fusion protein foundin owl monkey kidney cells can restrict HIV-1.
P Natl Acad Sci USA , 101:13324–13328, 2004.[117] L. E. Orgel and F. H. Crick. Selfish DNA: the ultimate parasite.
Nature , 284:604–607, 1980.[118] W. Paley.
Natural Theology: Or, Evidences of the Existence and Attributes of the Deity . JohnMorgan, Philadelphia, 1802.[119] C. H. Papadimitriou.
Computational Complexity . Addison–Wesley, Reading, MA, 2003.[120] C. H. Papadimitriou. The algorithmic lens: How the computational perspective is trans-forming the sciences. In , 2007.[121] D. R. Papaj. Automatic behavior and the evolution of instinct: lessons from learning inparasitoids. In
Insect Learning , pages 243–272. Springer, 1993.[122] M. Pavlicev, J. M. Cheverud, and G. P. Wagner. Evolution of adaptive phenotypic variationpatterns by direct selection for evolvability.
Proceedings of the Royal Society of London B:Biological Sciences , 278:1903–1912, 2011. 83123] W. Qu, S. Hashimoto, A. Shimada, Y. Nakatani, K. Ichikawa, T. L. Saito, K. Ogoshi, K. Mat-sushima, Y. Suzuki, S. Sugano, H. Takeda, and S. Morishita. Genome-wide genetic variationsare highly correlated with proximal DNA methylation patterns.
Genome Res , 22:1419–1425,2012.[124] K. Roelants, B. G. Fry, J. A. Norman, E. Clynen, L. Schoofs, and F. Bossuyt. Identical skintoxins by convergent molecular adaptation in frogs.
Current Biology , 20(2):125–130, 2010.[125] G. J. Romains.
Mental evolution in animals. With a posthumous essay on instinct by CharlesDarwin . Appleton, New York, 1883.[126] V. Roth. The biological basis of homology. In Humphries, editor,
Ontogeny and Systematics ,pages 1–26. Columbia University Press, New York, 1988.[127] T. Sasaki, M. Nishihara, H ad Hirakawa, K. Fujimura, N. Tanaka, M annd Kokubo,C. Kimura-Yoshida, I. Matsuo, K. Sumiyama, N. Saitou, T. Shimogory, and N. Okada.Possible involvement of SINEs in mammalian-specific brain formation.
Proc Nat Acad Sci ,105(11):4220–4225, 2008.[128] D. M. Sayah, E. Sokolskaja, L. Berthoux, and J. Luban. Cyclophilin A retrotranspositioninto TRIM5 explains owl monkey resistance to HIV-1.
Nature , 430:569–573, 2004.[129] D. Sedmera, I. Misek, and M. Klima. On the development of Cetacean extremities: I. Hindlimb rudimentation in the Spotted dolphin (
Stenella attenuata) . Eur J Morphol , 35(1):25–30,1997.[130] J. A. Shapiro.
Evolution: A View from the 21st Century . FT Press, Upper Saddle River, NJ,2011.[131] S. Singh.
The Big Bag: The Origin of the Universe . Harper Collins Publishers, Ney York,2004.[132] C. Stern. Selection for subthreshold differences and the origin of pseudoexogenous adapta-tions.
Am Nat , 92:313–316, 1958. 84133] E. Stone and R. Schwartz. Intron-dependent evolution of progenotic enzymes. In E. Stoneand R. Schwartz, editors,
Intervening sequences in evolution and development , pages 63–91.Oxford University Press, New York, 1990.[134] M. M. Suzuki and A. Bird. DNA methylation landscapes: provocative insights from epige-nomics.
Nature Rev Genet , 9:465–476, 2008.[135] T. Thomson and H. Lin. The biogenesis and function of piwi proteins and pirnas: Progressand prospect.
Annu Rev Cell Dev Biol , 25:355–376, 2009.[136] N. Tinbergen.
The Study of Instinct . Clarendon Press, Oxford, 1951.[137] N. Tinbergen. “derived” activities; their causation, biological significance, origin, and eman-cipation during evolution.
Quarterly Review of Biology , 27:1–32, 1952.[138] N. Tinbergen.
The Herring Gull’s World: a Study of the Social Behaviour of Birds . BasicBooks, Inc., New York, 1960.[139] N. Tinbergen and D. Kuenen. ¨Uber die ausl¨osenden und die richtunggebenden Reizsituationender Sperrbewegung von jungen Drosseln (
Turdus m. merula
L. und
T. e. ericetorum
Turton).
Z Tierpsychol , 3(1):37–60, 1939.[140] C. C. Trandinh, G. Pao, and M. Saier. Structural and evolutionary relationships among theimmunophilins: two ubiquitous families of peptidyl-prolyl cis-trans isomerases.
FASEB J ,6(15):3410–3420, 1992.[141] J. R. True and S. B. Carroll. Gene co-option in physiological and morphological evolution.
Annu Rev Cell Dev Bio , 18(1):53–80, 2002.[142] K. Tsuneki. Comparative studies on the nesting biology of the genus
Sphex (s. l.) in EastAsia (Hymenoptera, Sphecidae).
Mem. Fac. Lib. Arts, Fukui Univ. , 2(13):13–78, 1963.[143] L. G. Valiant. Evolvability.
Journal of the ACM , 56(1):3, 2009.[144] L. G. Valiant.
Probably Approximately Correct: Nature’s Algorithms for Learning and Pros-pering in a Complex World . Basic Books, 2013.85145] M. Vibranovski, Y. Zhang, and M. Long. General gene movement off the x chromosome inthe
Drosophila genus . Genome Res , 19(5):897–903, 2009.[146] C. A. Virgen, Z. Kratovac, P. D. Bieniasz, and T. Hatziioannou. Independent genesis ofchimeric TRIM5-cyclophilin proteins in two primate species.
P Natl Acad Sci USA , 105:3563–3568, 2008.[147] C. H. Waddington. Canalization of development and the inheritance of acquired characters.
Nature , 150(3811):563–565, 1942.[148] C. H. Waddington. The genetic assimilation of an acquired character.
Evolution , 7:118–126,1953.[149] C. H. Waddington. Genetic assimilation of the bithorax phenotype.
Evolution , 10:1–13, 1956.[150] C. H. Waddington. Comment on Professor Stern’s letter.
Am Nat , 92:375–376, 1958.[151] C. H. Waddington. Canalization of development and genetic assimilation of acquired charac-ters.
Nature , 183(4676):1654–1655, 1959.[152] C. H. Waddington. Genetic assimilation.
Adv Genet , 10:257–293, 1961.[153] C. H. Waddington. The genetic basis of the assimilated bithorax stock.
J Genet , 85(2):101–105, 2006.[154] G. P. Wagner.
Homology, Genes, and Evolutionary Innovation . Princeton University Press,2014.[155] G. P. Wagner and L. Altenberg. Perspective: complex adaptations and the evolution ofevolvability.
Evolution , pages 967–976, 1996.[156] A. R. Wallace. Migration of birds.
Nature , 10:459, 1874.[157] J.-C. Walser, L. Ponger, and A. V. Furano. CpG dinucleotides and the mutation rate ofnon-CpG DNA.
Genome Res , 18:1403–1414, 2008.86158] J. Watson and K. Lashley. Homing and related activities of birds. In
Papers from theDepartment of Marine Biology of the Carnegie Institution of Washington , volume VII, pages1–104. The Carnegie Institution of Washington, 1915.[159] R. Watson, R. Mills, K. Buckley, CL ad Kouvaris, A. Jackson, S. Powers, C. Cox, S. Tudge,P. Ryan, A. Davies, L. Kounios, and D. Power. Evolutionary connectionism: algorithmicprinciples underlying the evolution of biological organisation in evo-devo, evo-eco and evolu-tionary transitions.
Evolutionary Biology , pages 1–29, 2016.[160] R. Watson and E. Szathm´ary. How can evolution learn?
Trends Ecol Evol , 31(2):147–157,2016.[161] I. Wegener.
The Complexity of Boolean Functions . Wiley, New York, 1987.[162] A. Weismann.
On Germinal Selection as a Source of Definite Variation . The Open CourtPublishing Company, Chicago, 1902 (1896).[163] M. J. West-Eberhard.
Developmental Plasticity and Evolution . Oxford University Press,2003.[164] C. O. Whitman. Animal behavior. In
Biol Lect Mar Biol Lab , pages 285–338. Woods Hole,Massachusetts, 1898.[165] C. O. Whitman.
The Behavior of Pigeons (Posthumous works of Charles Otis Whitman) .Carnegie Institution of Washington, 1919.[166] E. P. Wigner. The unreasonable effectiveness of mathematics in the natural sciences. RichardCourant lecture in mathematical sciences delivered at New York University, May 11, 1959.
Commun Pur Appl Math , 13(1):1–14, 1960.[167] R. Willemoes, M ad Strandberg, R. Klaassen, A. Tøttrup, Y. Vardanis, P. Howey, K. Thorup,M. Wikelski, and T. Alerstam. Narrow-front loop migration in a population of the commoncuckoo
Cuculus canorus , as revealed by sattellite telemetry.
PLoS ONE , 9(1):e83515, 2014.[168] S. J. Wilson, B. L. Webb, L. M. Ylinen, E. Verschoor, J. L. Heeney, and G. J. Towers.87ndependent evolution of an antiviral TRIMCyp in rhesus macaques.
P Natl Acad Sci USA ,105:3557–3562, 2008.[169] A. M. Woodbury. Animal migration: periodic-response theory.
The Auk , 58:463–505, 1941.[170] M. W. Yap, M. P. Dodding, and J. P. Stoye. Trim-cyclophilin A fusion proteins can restricthuman immunodeficiency virus type 1 infection at two distinct phases in the viral life cycle.
J Virol , 80(8):4061–4067, 2006.[171] M. W. Yap, G. B. Mortuza, I. A. Taylor, and J. P. Stoye. The design of artificial retroviralrestriction factors.
Virology , 365(2):302–314, 2007.[172] F. Zhang, T. Hatziioannou, D. Perez-Caballero, D. Derse, and P. D. Bieniasz. Antiretroviralpotential of human tripartite motif-5 and related proteins.
Virology , 353(2):396–409, 2006.88 is-Elements
Interaction
A B* CA B C
Gene product Gene productMutation (a)(b)
Figure 1: Mutation as a biological process, from [93]. a) In this figure we see three loci comingtogether in a biological interaction through gene products and cis elements. This part of thefigure merely represents schematically the gene regulation and interaction that are key to ourunderstanding of molecular and cellular biology. What is new about the figure is that we have notyet fully considered the possibility that there could be a mutation arrow too, i.e., that mutation is anoutcome of genetic interactions in a heritable mode; i.e., that much like genes interact in influencinga classical trait, like the eye or the ear, they also interact in influencing genetic change. Note thatthe figure purposely leaves open the particulars of the biochemical mechanisms involved, as theremay be many such mechanisms, and that “mutation” is broadly construed to mean any heritablegenetic change. These may involve not only DNA changes but also epigenetic changes. b) Mutationas an event of information flow and computation changes many things in our conceptualizationof evolution. Particularly, the biological process of mutation creates from the combination ofinteracting alleles across loci a new heritable piece of information—a new mutation—a new allele,B*. Even though the particular combination of interacting alleles will sooner or later disappear dueto the sexual shuffling of the genes, information from it can be transmitted to future generationsthrough the mutation. In this manner, the problems of the role of sexual recombination and of thenature of mutation may be tied together. 89
B C D E F G A B C D E F G
Figure 2: Mutation as an event of information transmission and computation creates a network ofinformation flow through the generations, from [93]. Each box represents an individual, and in eachbox, the two sets of lines at the top represent that individual’s diploid genotype (genes A throughG), and the set of lines at the bottom represents a haploid genotype transmitted through the gamete.For the sake of demonstration, a small number of mutational events due to interactions betweengenes is shown in two parents and an offspring (large boxes), although many mutations occur inother genes and in other individuals at the same time. For example, C* represents a mutation inone of the alleles of gene C. Because the output of a mutational event in one generation—namelythe mutation itself (e.g., C*)—can serve as an input into mutational events at later generations(e.g., the event creating D*), non-accidental mutation creates a network of information flow andcomputation over the generations, from many genes into one and from one gene into many, as wellas from many individuals into one and from one individual to many.90 ew nest Start Go to first nest
Raising a family in diggerwasps
Do next provision phase for given nest New nest Only one nest built? Is the weather excellent? Is all provisioning completed? Sometimes may build another nest (up to 3-4 total) Build a nest Get caterpillar Lay egg Close nest entrance with special care Return Get caterpillar Fly away and hunt a caterpillar Return Carry paralyzed caterpillar back Reopen nest Put caterpillar in Close entrance temporarily with soil Do next provision phase for given nest Make an inspection visit Reopen nest Is all in order? Did egg hatch? May abandon nest May start a new nest (up to 3-4 total) Determine amount of food to bring: • Determine phase a or phase b of provisioning, probably based on age of larva. • Adjust amount to bring based on amount of food present. Phase a: Bring 1-3 caterpillars. Phase b: Bring 3-7 caterpillars. Get caterpillar Was the predetermined amount brought in? Return Close the nest entrance; extremely carefully if end of phase b Build a nest Multiple operations not shown Return … Yes No No Yes No Yes No No Yes Yes No Yes
Figure 3: A flowchart describing the algorithmic behavior of nest building, egg laying and provi-sioning in the digger wasp,