Evolutionary Connectionism: Algorithmic Principles Underlying the Evolution of Biological Organisation in Evo-Devo, Evo-Eco and Evolutionary Transitions

The mechanisms of variation, selection and inheritance, on which evolution by natural selection depends, are not fixed over evolutionary time. Current evolutionary biology is increasingly focussed on understanding how the evolution of developmental organisations modifies the distribution of phenotypic variation, the evolution of ecological relationships modifies the selective environment, and the evolution of reproductive relationships modifies the heritability of the evolutionary unit. The major transitions in evolution, in particular, involve radical changes in developmental, ecological and reproductive organisations that instantiate variation, selection and inheritance at a higher level of biological organisation. However, current evolutionary theory is poorly equipped to describe how these organisations change over evolutionary time and especially how that results in adaptive complexes at successive scales of organisation (the key problem is that evolution is self-referential, i.e. the products of evolution change the parameters of the evolutionary process). Here we first reinterpret the central open questions in these domains from a perspective that emphasises the common underlying themes. We then synthesise the findings from a developing body of work that is building a new theoretical approach to these questions by converting well-understood theory and results from models of cognitive learning. Specifically, connectionist models of memory and learning demonstrate how simple incremental mechanisms, adjusting the relationships between individually-simple components, can produce organisations that exhibit complex system-level behaviours and improve the adaptive capabilities of the system.

We use the term “evolutionary connectionism” to recognise that, by functionally equivalent processes, natural selection acting on the relationships within and between evolutionary entities can result in organisations that produce complex system-level behaviours in evolutionary systems and modify the adaptive capabilities of natural selection over time. We review the evidence supporting the functional equivalences between the domains of learning and of evolution, and discuss the potential for this to resolve conceptual problems in our understanding of the evolution of developmental, ecological and reproductive organisations and, in particular, the major evolutionary transitions.

Evolution by natural selection aims to explain biological adaptations—such as how the giraffe came to have a long neck. For some purposes it is sufficient to assume that there is, for example, variability in neck length, that long necks confer differential survival or reproductive benefit, and that neck length is heritable. However, research in the rapidly-expanding fields of evolutionary developmental biology (evo-devo), evolutionary ecology (evo-eco) and the evolutionary transitions in individuality (for which we propose the term “evo-ego”, Watson and Szathmáry 2015) recognises that for some purposes this is only part of the explanation. That is, evolutionary outcomes are significantly dependent on how developmental, ecological and reproductive organisations constrain or facilitate the variability of phenotypes, the organisation of their selective environment and the heritability of the relevant evolutionary units, respectively. For example, how is it that developmental constraints allow multiple phenotypic traits to change simultaneously whilst maintaining suitable functional integration between them and avoiding deleterious side-effects on other traits? What is it about the organisation of an ecological community that causes some ecological relationships to remain stable over long periods of selection and applies a strong selective pressure for changes in other ecological relationships (e.g. between a particular herbivore and a particular resource)? How is it that reproductive constraints (e.g.

reproduction through a single-celled population bottle-neck) come to define a multicellular organism like a giraffe as a Darwinian unit in the first place—suppressing fitness differences between the cells or genes within an individual (so they do not compete with each other for representation in offspring individuals) but enabling the inheritance of fitness differences between individuals (allowing them to compete for representation in the population)? Answering these questions requires a move to a different level of explanation—one that attempts to explain why variation, selection and inheritance have the forms that they do, rather than taking them as fixed axioms of the adaptive process.

These parameters are not simply exogenous contextual details to the processes of evolution. Each of these organisations (developmental, ecological and reproductive interactions) is itself a product of evolution or is modified by the products of evolution (Wagner and Altenberg 1996; Odling-Smee et al. 2003, Okasha 2006; Laland et al. 2011, 2015). The phenotypic variation exhibited by a genetic lineage is modified by the evolution of developmental interactions (Brakefield 2006; Kirchner and Gerhart 1998; Wagner and Altenberg 1996; Toussaint and von Seelen 2007), the selection it experiences is modified by the evolution of ecological interactions (Post and Palkovacs 2009; Laland and Sterelny 2006; Laland et al. 1999; 2011; Odling-Smee et al. 2003, 2013), and even the identity of the evolutionary unit changes as a result of the evolution of new reproductive strategies and new mechanisms of inheritance (Jablonka and Szathmáry 1995; Szathmáry and Demeter 1987; Okasha 2006; Sigmund and Szathmáry 1998; Maynard Smith and Szathmáry 1995). Accordingly, each of the major components of the Darwinian machine (namely, variation, selection and inheritance) (Lewontin 1970) are themselves subject to evolutionary change (Watson and Szathmáry 2015). The ambitious aim of an extended evolutionary synthesis (Pigliucci and Muller 2010; Laland et al. 2015) can thus be seen as the aim to endogenise these organisations into evolutionary theory (Okasha 2006, p. 220)—explaining both how evolutionary processes shape these organisations (evo → devo/eco/ego) and, in the reverse direction, how these organisations affect evolutionary outcomes (evo ← devo/eco/ego). This is not easy to do (Pigliucci 2007; Lawton 1999; Laland et al. 2011, 2015; Okasha 2006).

In evolutionary developmental biology (evo-devo) (Brakefield 2006; Carroll 2008; Wagner and Laubichler 2004; Wagner et al. 2007; Wagner 2013; Hoekstra and Coyne 2007) it is recognised that the organisation of developmental biases and constraints controls the distribution of phenotypic variation that is produced under genetic (or environmental) variation and can thereby control the possible paths of evolutionary trajectories through phenotype space (Arnold et al. 2001; Jones et al. 2007; Schluter 1996; Toussaint and von Seelen 2007; Gerhart and Kirschner 2007; Wagner 2014) (evo ← devo). For example, some physiological changes in limb morphology, wing patterns or gene-regulatory circuits may be readily produced and selected whereas others may not (Wagner 2014; Brakefield 2006). The structure of this developmental organisation is itself subject to change over evolutionary time, e.g. via the evolution of gene-regulatory interactions or morphological architectures (Riedl 1977; Draghi and Wagner 2009; Wagner and Altenberg 1996; Crombach and Hogeweg 2008; Amundson 2005; Pavličev and Cheverud 2015) (evo → devo). This bi-directional interaction means that evolution could modify developmental organisation in a way that facilitates or frustrates future evolution—hence the evolution of evolvability, i.e. evolved changes that affect the future ability of a population or lineage to evolve (Wagner and Altenberg 1996; Kirchner and Gerhart 1998; Chicurel 2001; Partridge and Barton 2000; Hendrikse et al. 2007). Work on this topic shows that the evolution of gene-regulatory interactions or phenotypic correlations can enhance phenotypic robustness (Wagner 2008, 2013), accelerate adaptation under directional selection (Pavlicev et al. 2011), or evolve to mimic the structure of the selective environment (Watson et al. 2014; Riedl 1977), such as modularity (Lipson et al.

2002; Watson et al. 2014; Clune et al. 2013; Kashtan et al. 2007, 2009; Parter et al. 2008). But the general relationship between the evolution of individual developmental interactions and developmental organisations, and in particular the evolution of structures that facilitate evolvability, remains poorly understood and conceptually problematic (Pigliucci 2007; Chicurel 2001; Partridge and Barton 2000; Sniegowski and Murphy 2006; Hendrikse et al. 2007). In particular, the idea that natural selection might change the variability on which it depends introduces a self-referential element that is particularly difficult to characterise with conventional theory.

In evolutionary ecology (evo-eco) (Matthews et al. 2011; Post and Palkovacs 2009; Schoener 2011; Müller 2007), the details of the relationships in an ecological community have a significant effect on the selective pressures acting on component species (evo ← eco). For example, the prevalence of a particular resource or competitor may be strongly influenced by the existing ecological relationships in the community, and may also dominate the selective pressures on an evolving population within that community. The structure of these ecological organisations is also itself modified by the changing nature of ecological relationships, due to the evolution and coevolution of the component species in interaction with one another (e.g. changes in individual characters that modify the overlap of resource utilisation profiles, or by traits that modify the energy, time or resources invested in exploiting one ecological relationship rather than another) (evo → eco)—see niche construction and environment engineering (Odling-Smee et al. 2003; Laland et al. 1999; Wright and Jones 2006; Post and Palkovacs 2009). The organisation of ecological relationships (both trophic and non-trophic) can affect the stability, resilience and homeostatic dynamics of ecological communities and complex ecological functions (Jax 2010; Lenton 2004; Holling 1973; Folke 2006; Gallopín 2006). Some argue that the reciprocal causation involved in evo-ecological interactions (i.e. evolution occurs in an ecological niche and the ecological niche is itself a product of evolutionary processes; Watson and Ebner 2014; Post and Palkovacs 2009) constitutes a significant departure from conventional evolutionary models (Laland et al. 2011; Mesoudi et al. 2013).

But the general relationship between the evolution of individual ecological relationships and community organisation, and in particular the evolution of ecological feedbacks that facilitate self-regulation or homeostasis, remains poorly understood and conceptually problematic (Lawton 1999; Cropp and Gabric 2002; Okasha 2005; Lenton and van Oijen 2002). In particular, the idea that natural selection might change the selection pressures that act on itself introduces a self-referential element that is difficult to characterise with conventional theory.

In the major evolutionary transitions (evo-ego) (Maynard Smith and Szathmáry 1995, Godfrey-Smith 2009; Michod 1999, 2007; Bourke 2011; Buss 1987; Okasha 2006; Bouchard and Huneman 2013), evolution has repeatedly reinvented itself—creating new evolutionary units at successive scales of biological organisation, e.g. from self-replicating molecules, to chromosomes, to simple cells, to multi-organelle eukaryote cells, to multicellular organisms, to eusocial groups. These are not just changes in the characteristics of an existing evolutionary entity, but the result of changes to the reproductive relationships between evolutionary entities such that “entities that were capable of independent replication before the transition can replicate only as part of a larger whole after the transition” (Maynard Smith and Szathmáry 1995). These changes, i.e. changes that prevent independent replication, act to suppress fitness differences between individuals at one level of organisation and may provide opportunities for natural selection to create heritable fitness differences at a higher-level of organisation (Godfrey-Smith 2009; Okasha 2006; Michod and Roze 2001; Michod and Herron 2006; Ryan et al. 2015). This may involve reproductive organisations that synchronise the transmission of information across generations (e.g. vertical transmission of symbionts, Margulis and Fester 1991; compartmentalisation of replicators, Sigmund and Szathmáry 1998; Szathmáry and Demeter 1987; or linkage of replicating molecules into chromosomes, Maynard Smith and Szathmáry 1993) or restrict the channels of communication (e.g., bottle-necked life-cycle, germ-soma separation; Godfrey-Smith 2009; Buss 1987).

Such features change the scale at which heritable variation in reproductive success is manifest, thus resulting in the evolution of new levels of Darwinian individuality (Maynard Smith and Szathmáry 1995; Jablonka 1994; Clarke 2010; Okasha 2006; Godfrey-Smith 2009; Wilson 1989; Bouchard and Huneman 2013) (hence, “evo-ego”). Note that the evolutionary unit can be defined by the level (or levels) of organisation where variation in reproductive success is heritable (evo ← ego) (Clarke 2010; Godfrey-Smith 2009), and the level of organisation that exhibits this can be modified by the products of the evolutionary process (evo → ego). Work to understand these evolutionary transitions in individuality is motivating a new research programme within evolutionary biology (Calcott and Sterelny 2011; Okasha 2006, Bouchard and Huneman 2013) that encompasses concepts such as the evolution of individuality, social group transformation, de-Darwinisation (of individuals) and Darwinisation (of groups) and export of fitness (from lower to higher units) (Buss 1987; Bourke 2011; Godfrey-Smith 2009; Okasha 2006; Michod 1999). But at present the organising principles involved in the evolution of reproductive dependencies, and in particular the evolution of new levels of evolutionary individuality, remain poorly understood and conceptually challenging (Okasha 2006; Godfrey-Smith 2009). In particular, the idea that natural selection might redefine the evolutionary unit, i.e. the reproductive heritability on which it depends, introduces a self-referential element that is difficult to characterise with conventional theory.

Each of these areas shares the common underlying conceptual problem caused by bi-directional interactions or reciprocal causation (Levins and Lewontin 1987; Post and Palkovacs 2009; Laland et al. 2011, 2015), i.e. evolution modifies organisations (often naturally described as networks) and these organisations modify the process of evolution. The notion that evolutionary processes can thereby change their own parameters is the root cause of theoretical and conceptual roadblocks in each field (Watson and Szathmáry 2015).

In particular, although it is clear that evolution can modify the parameters of variation, selection and inheritance, and this might facilitate evolvability, it is not at all clear that these organisations can be adaptations for increased evolvability. In the absence of a higher-level selective process that might favour evolutionary processes that are successful in the long term, it seems equally likely that such organisations might constrain or frustrate subsequent adaptation. For example, researchers in evo-devo struggle to agree on whether the long term evolvability of a population can systematically improve as a consequence of natural selection acting on developmental organisations (Pigliucci 2008; Sniegowski and Murphy 2006; Chicurel 2001; Partridge and Barton 2000; Pavlicev et al. 2011; Earl and Deem 2004). Evo-eco recognises the pressing need to predict how ecological resilience and ecosystem functions change over time (Holling 1973; Gallopín 2006; Jax 2010). Although evolution might change community organisation in a way that increases the resilience or self-regulation of the ecosystem (Cropp and Gabric 2002), when there is no selection at the ecosystem level it seems equally likely to become more susceptible to stresses and perturbations over evolutionary time, possibly resulting in catastrophic collapse in the long term (Holling and Gunderson 2002; Montoya et al. 2006; Schoener 2011). Work in evo-ego seeks to understand whether evolutionary transitions in individuality result from systematic adaptive pressures toward the creation of higher-level biological organisations or whether they are merely a collection of independent accidents (Maynard Smith and Szathmáry 1995; Okasha 2006).

It might be the case that the reproductive dependencies evolved at one level of organisation create higher-level organisations that are better-able to respond to selective pressures relevant at the higher level of organisation, but alternatively it might be the case that they encapsulate maladaptive relationships that oppose an effective response to higher-level selection.

Such problems motivate the growing recognition that the Darwinian machine needs an overhaul: That self-referential evolutionary mechanisms (where the products of evolution alter the processes of evolution) create serious problems for existing theory (Laland et al. 2011), and that a new and expanded theoretical framework is needed (Pigliucci and Muller 2010; Laland et al. 2015) that integrates “eco-evo-devo” processes (Blute 2008) (and we would add “ego” to this list also).

The same underlying problem of reciprocal causation is manifested differently in each domain. Whilst it is clear that the products of the Darwinian machine can modify the parameters of its own operation, it is not clear in what way it changes itself and, in particular, whether it is possible that the Darwinian machine changes systematically ‘for the better’, i.e. in a way that facilitates rather than frustrates subsequent adaptation. This problem arises in domain-specific versions:Evo-devo—implications for modifying variability, and the evolution of long-term evolvability: Can development be organised to facilitate future adaptation? That is, can evolution ‘predict’ or ‘anticipate’ what developmental organisations will enable adaptive variation in new environments that it has not yet been exposed to? Conversely, if all it does is find organisations that are fit over the set of past environments, then that seems to be conventional evolution, not the evolution of evolvability.Evo-eco—implications for modifying the selective context, and the evolution of ecosystem organisation: Can an ecosystem be organised ‘for’ anything if it is not an evolutionary unit? That is, how can natural selection at the level of individuals within multiple species result in ecological organisations that are self-supporting at the system level (homeostasis), or indeed, result in any kind of ‘ecosystem evolution’ that is more than the sum of the evolution of the parts? (Levin 2011; Leigh and Vermeij 2002; Lenton 2004). It is clear that by evolving its ecological relationships a species may modify the ecological dynamics of the community and hence it’s ecological context and hence the selection it experiences over subsequent generations (i.e. niche construction, Odling-Smee et al. 2003, 2013).

But it is not clear that it can do this in a way which is systematically beneficial to itself, except in the case where such benefits fall differentially on the individual bearing the niche constructing trait (which implies it can be treated as an extended phenotype of the individual). For example, an individual character that reduces competition for resources with another species offers no differential advantage to the trait bearer if all members of its species benefit from such reduced competition (Wilson 1980). Thus we may expect that the network of ecological relationships that evolve may alter ecological dynamics and attractors, but not necessarily in a manner that creates adaptive benefits to the species, let alone the community as a whole.To clarify, we use the term ‘ecosystem evolution’ to refer to the ecosystem-level consequences of natural selection acting on individuals within component species (Levin 2011, 2014), e.g., evolutionary change in the inter-specific relationships of an ecological community arising from individual-level selection. We do not suggest that ecosystems or ecological communities are units of selection (which would require a population of multiple ecosystems with heritable characteristics). Nonetheless, we ask whether there exist conditions where individual selection has emergent yet predictable consequences for the organisation and efficiency of the system as a whole (Levin 2011). Such questions are related to questions regarding collective welfare in economic systems under the assumption of individual utility-maximising agents (Levin 2014), and the reciprocal interaction of social behaviour with co-constructed social structures, e.g. institutions, a.k.a.

the agency-structure debate, (Bator 1957, Ritzer and Goodman 2004).Evo-ego—implications for modifying heritability, and the evolution of new evolutionary units: Can evolution at one level of organisation favour the creation of heritable evolutionary units that are adaptive at a higher-level of organisation? That is, can the evolution of reproductive organisations find new heritable units that are suitable for responding to selection at the higher level of organisation before that level of organisation exists? (Trestman 2013). Like the previous question, this is related to the formation of social structures that change individual incentives. But the outcome in this case is even more radical—not merely the evolution of interaction structures that incentivise cooperation in individuals, but the evolution of reproductive dependencies that create a new evolutionary unit, subsuming the Darwinian individuality of the original evolutionary units. Moreover, not only do we ask whether individual selection can create new evolutionary units, but whether the new units it creates are effective at facilitating adaptation at a higher level of organisation—or conversely, whether they frustrate further adaptation.

In each case, the conventional answer seems to be—it cannot. It is not possible for evolution by natural selection to produce adaptations for an environment it has not yet encountered, to produce organisation at the system level without selection at the system level, or to create new units that are adaptive for a level of selection that does not yet exist.

We present the case that all of these behaviours that seem impossible for evolutionary systems are possible, and that necessary and sufficient conditions can be characterised. We make this argument by recognising that analogous behaviours are possible, and are well-understood, in another domain—and because the underlying principles are mathematically equivalent, specific results and insights from one domain can be transferred to the other (Watson and Szathmáry 2015). This is a domain where the idea of a system that changes itself over time is not controversial—namely, learning systems.

A learning system is a system that improves its performance at some task with experience (Mitchell 1997). A simple kind of learning (often likened to natural selection) is reinforcement learning. This utilises a reward function to reinforce good behaviour or good outputs (or punish bad outputs) when they occur. An analogy between this type of learning and evolution by natural selection is common and intuitive (Maynard Smith 1986; Frank 1996; Skinner 1953; Bateson 1979) and mathematical isomorphisms exists between formal models of selection and formal models of learning (Harper 2009; Shalizi 2009; Frank 2009; Valiant 2013; Chastain et al. 2014).

It is common to think of learning systems as sophisticated machines (or intelligent organisms) with goal-directed intentions designed for the purpose of producing smart behaviours—which would make their abilities irrelevant to understanding evolutionary processes. But, in fact, all of the phenomena relevant to our evolutionary questions can be produced by intention-less algorithms with simple incremental improvement mechanisms. However, note that in a learning system the object of this incremental improvement process is different from a conventional optimisation process. Whereas a simple processes of optimisation (or incremental improvement) is usually applied to a solution or output directly, a learning process optimises a model of good solutions or outputs or an indirect representation of solutions. Evolutionarily, this is like the difference between adapting the parameters of a phenotype directly (e.g. the traits of a phenotype) vs adapting the parameters of a developmental process that produces fit phenotypes (this is a distinction which is lost when we assume a one-to-one mapping between genotype and phenotype). The learning process optimises the fit of the model to the observations (or minimises the discrepancy between the model and the observations) by incrementally adjusting the parameters of the model. The significance of this is that the model can then be used to recognise or generate new examples that have structural similarities with those that have been rewarded in the past but are not identical to them, and relatedly, examples that are far apart in ‘solution space’ can be placed next to each in ‘model space’ (Watson and Szathmáry 2015).

In advanced learning methods a model can be complicated and mechanisms for approximating the parameters of the model from observations can be quite sophisticated. But quite often the model can be simple; for example, a correlation model is a representation of how features in good solutions ‘go together’ or correlate. In practise, this often means making connections of some sort between different elements of a solution (causing their useage in solutions to become correlated). Correlation learning can be implemented via a very simple learning principle that adjusts the connections of such an organisation incrementally (Hebb 1949; Ackley et al. 1985). Mechanisms sufficient to learn a single correlation (between a pair of variables) by reinforcement can be trivial, i.e. just change a connection a small amount and see if it improves the output. This does not require sophisticated data structures, mathematical or statistical analysis, or complex computational machinery. Given variation in the connections of a network (that affects correlations between problem variables), it is simply an incremental improvement process applied to those connections. This is a crucial step-up from an incremental improvement process acting on the solution variables directly, however. Rather than merely finding good outputs, a learning process finds the structure underlying good outputs. It is this which enables new phenomenology compared to a simple optimisation process; For example, the ability to generate new patterns or behaviours that are different from the ones it was trained on but exhibiting the same underlying structure.

The consequences of correlation learning in larger systems can be surprisingly powerful. Under the right conditions, correlation learning is sufficient for the behaviours that seemed impossible for evolutionary systems (a, b and c above):Learning systems can perform well at novel tasks, i.e. tasks they have not previously been trained on.Learning systems can exhibit non-trivial collective behaviours without system-level feedback on performance.Learning systems can find new representations of a problem, that facilitate task learning at a higher level of organisation, before those higher levels of organisation exist.

These behaviours, described below, are uncontroversial and well-understood in learning systems, but knowledge about the conditions for and capabilities of such phenomena has not been previously transferred into evolutionary theory. Our long-term aim is to develop a unified predictive theory for these evolutionary questions by exploiting the existing concepts and extensive existing results from learning systems. It is the aim of the current paper to reinterpret the key open questions in these different biological domains from this unifying perspective, to describe how learning theory connects with these biological questions, and to review and synthesise our work thus far. In particular, we draw these works together to address their inter-related roles in major evolutionary transitions.

Developmental, ecological and reproductive organisations are structures that determine which things ‘go together’ and which things are independent. Specifically, the organisation of developmental interactions governs whether it is possible for multiple coordinated changes to occur in a way that preserves their functional dependencies without causing multiple unwanted side-effects on other aspects of the phenotype. Ecological interactions specify how a change in the density of one species modifies the selective pressures acting on other species and thus govern which species are mutually exclusive and which can coexist, for example. Reproductive organisations govern whether fitness differences among the components within evolutionary units are suppressed and whether fitness differences between different evolutionary units can be inherited. Understanding the evolution of developmental, ecological and reproductive organisations thus requires that we understand how evolution alters which things vary together, which things are selected together and which things are inherited together, respectively.

It is common to describe developmental and ecological interactions as networks (e.g. a gene-regulation network or food web/community matrix, respectively). These networks describe who interacts with whom, in what way and how much (i.e. how one gene-expression level affects changes in the expression of other genes, or how the density of one species affects the population growth of another species). Whilst it is not as common to describe reproductive dependencies as networks, these relationships also have the basic property of controlling how the inheritance of one evolutionary unit is, or is not, independent of the inheritance of another evolutionary unit. It is therefore useful to characterise all three (developmental, ecological and reproductive) organisations as networks. We can then ask how the structure of that network (topological changes including the strength and sign of connections, in some cases) change over evolutionary time, and how those structural changes then modify evolutionary processes.

When the structure of a network affects the dynamics that occur on it, and the dynamics that occur on the network affect changes to network structure, this is known as an adaptive network (Gross and Sayama 2009), e.g. where agents on a network can both choose behavioural strategies that are suitable for the current organisation, and can also choose to re-wire connections on the network to suit the current behaviours (Jackson and Watts 2002; Pacheco et al. 2006; Traulsen et al. 2008; Van Segbroeck et al. 2010). We argue that developmental, ecological and reproductive organisations (like other complex adaptive systems, Farmer 1990) exhibit this two-way property.

More specifically, we investigate the hypothesis that the evolution of developmental, ecological and reproductive organisations are all subject to the same underlying organisational principle, a simple principle of positive feedback between the topology of an interaction network and the behaviours that the network structure governs. That is, a connection between two components or nodes in this network causes them to exhibit correlated behaviour, and when nodes have correlated behaviours this causes natural selection to create or strengthen the connection between them. In short, entities that co-occur together ‘wire’ together (and entities that wire together co-occur together). For example, (as expanded below) genes that are selected together are wired together via the evolution of gene-regulatory interactions that cause them to co-vary (be co-expressed) in future, species that co-occur in high-density are wired together by the evolution of ecological relationships that cause them to be co-selected in future, and evolutionary units that reproduce together are wired together by changes to reproductive relationships that cause them to be co-inherited in future. Empirical observations are consistent with this simple positive feedback principle and in this paper we present theoretical support for each of these cases:Evo-devo: The more often that two genes are selected together (at the same time/in the same environment) the more selective advantage there is to strengthening developmental interactions between them. This organisational change causes their expression in the phenotype to be correlated in future (e.g. via an increase in the gene-regulatory connection between them; Wagner et al. 2007; Pavlicev et al. 2011; Watson et al. 2014; Kashtan et al. 2009).

See also, for example, the principle of ‘coevolution of coexpressed traits’ and conversely ‘the rule of independent selection’ (West-Eberhard 2003).Evo-eco: The more often that two species populations grow to high-density together (at the same time/in the same environment) the more selective advantage there is for individual traits that strengthen ecological interactions between them. These interactions change in a way that causes their population growth to be more correlated in future, e.g. via reductions to competitive interactions between them (Lewis 2009; Power et al. 2015). For example, this type of feedback is part of the backstory involved in ‘invasional meltdown’ (Gallardo and Aldridge 2015; Simberloff and Von-Holle 1999) where species that have been in prolonged contact with one another in one environment facilitate one-another’s invasion into another environment because they “have had a long evolutionary time to develop a cosy relationship with each other” (Gallardo and Aldridge 2015). This might involve character displacement that reduces niche overlap when species are driven into contact (Brown and Wilson 1956; Dayan and Simberloff 2005).Evo-ego: The more often two evolutionary units reproduce together (at the same time/in the same environment) the more selective advantage there is to individual traits that strengthen reproductive dependencies between them. Such reproductive interactions (controlled by individual selection in either participant) cause their reproduction to be more strongly correlated or centralised in future (e.g. via evolution of co-dispersal behaviours or vertical inheritance) (Watson et al. 2009b, 2011b, submitted).