2.1. Varieties of developmental channeling

Let’s suppose you are accomplished at the standing long jump and have a record of 3.3m. The sport was part of the Olympics from 1900 to 1912 and the record was then set by Ray Ewry at 3.47m. You might think to yourself, “If only I were a bit taller, I could easily defeat the Olympic record.” Now here is the interesting question: If we just scaled you up, made you bigger but exactly proportional to your previous you, what would this do to your standing jump? To answer this, let’s make a few reasonable assumptions: Your strength is proportional to the cross-sectional area of your muscles and thus increases proportional to the square of your height. Your weight, however, is proportional to the cube of your height.

Now we can see what happens if we increase your height. To make it as clear as possible, let’s say we double your height. How long a jump can you now make? (Let’s assume that doubling does not impair organs like your brain or heart, so the relevant factors are only how your strength and mass increase.) Your height is doubled, which means your mass increases eightfold, while your strength increases fourfold. Thus, the double-sized you is half as strong relative to your mass and could therefore jump only half as far relative to your height. But you are twice as tall, so jumping half as far relative to your (double) size means that double you jumps exactly as far as normal you. The same with a version of you half as tall. A grasshopper that jumps two meters no longer seems all that impressive. Shrink yourself to grasshopper size and you could do the same.Footnote ¹

What explain the constancy of jump length are basic physical truths, truths about natural laws and the geometry of spacetime. Such features of the world channel organismic forms: Phenotypes develop and change according to what is physically possible (i.e., they reflect physical constraints on development). Physical laws and properties prevent organisms from developing in (possibly adaptive) divergent ways, such as the scaling laws standing in the way of your Olympic record. This is the constraining side of this form of channeling, though the very same laws allow for incredible feats in very small organisms. A flea can typically jump more than fifty times its length (Krasnov et al. Reference Krasnov, Burdelov, Khokhlova and Burdelova2003), not unlike a two-meter-tall human jumping 100 meters, the length of a football field. This is the constructive side of channeling: The same properties that constrain the emergence of some phenotypes enable the appearance of others. Importantly, facts about the scaling possibilities of phenotypic changes are independent from the potential fitness value of the changes, such as jumping larger distances. The nature of scaling implies that insects can have the luxury of walking about on gossamer appendages, whereas elephants require blunt, stumpy limbs.

There are other kinds of physical features that channel organismic form in similar ways. The nature of chemical bonds, the rates at which molecules diffuse through liquids, and the oxygen level of air are all examples of physical features that channel organismic forms at the molecular level. These forms of channeling apply to all organisms and are what Maynard Smith et al. (Reference Maynard Smith, Burian, Kauffman, Alberch, Campbell, Goodwin, Lande, Raup and Wolpert1985) label universal constraints. This universal channeling can be based on physical laws, but it can also rest on universal features of life on Earth. Life on Earth is based on cells and thus the physical properties of cell growth and differentiation are mechanical properties of evolved systems that determine the available pathways leading to possible adult forms (Newman et al. Reference Newman, Forgacs and Müller2006).

Another feature of life on Earth is that each developing organism has a fixed set of genes and can draw only on these genes in its development.Footnote ² The availability of genes therefore channels the development of the organism: If it needs a particular protein, it cannot synthesize a segment of DNA for use in coding this protein. Furthermore, the roles that a gene can play in development are determined by elements such as its genomic and epigenetic context. The resulting phenotype is thus the effect of the generative rules of development guiding the interaction and expression of genes (Müller Reference Müller, Laubichler and Maienschein2007). These genes do not merely code for specific proteins, but they have a mechanistic function in particular developmental pathways (Gilbert Reference Gilbert, Beurton, Falk and Rheinberger2000). Such pathways channel phenotypic results using the interaction of coding and regulatory regions of DNA with other molecular, cellular, and tissue aspects of the embryo. The complexity of these interactions in turn channels the functions that particular genes can play in the development of specific traits.

While some channeling is universal, other kinds differ greatly across taxa, due to their different forms and evolutionary histories. There is thus taxon- and trait-specific channeling. Such channeling is local instead of universal (and was labeled local constraints by Maynard Smith et al. Reference Maynard Smith, Burian, Kauffman, Alberch, Campbell, Goodwin, Lande, Raup and Wolpert1985). For example, Carrier’s constraint, proposed by Carrier (Reference Carrier1987), holds that two-lunged vertebrates that laterally flex their bodies while they locomote will have difficulty breathing while moving, due to the compression that such movement imposes on their lungs. Lizards are one such taxon. This is a constraint that has to do with the physics of breathing and of locomotion, and their combination in a particular quadrupedal form of movement. No vertebrate can have two lungs, laterally flex while walking or running, and have its breathing unaffected by this locomotory movement. Similarly, no vertebrate limbs have evolved into wheels, not because they would not be useful, but because—among other things—you can’t have blood vessels flowing into revolving structures.

Development is path-dependent, with later stages depending on earlier ones. One form of dependency leads to what Wimsatt (Reference Wimsatt and Bechtel1986, Reference Wimsatt and Love2015) labels generative entrenchment, which means that early stages of development tend to be more conserved because changes in them typically alter later ones as well. Development is thus canalized, in the sense that the process of development is like the flowing of a branching river (Waddington Reference Waddington1942)—once the system commits itself to one branch, this can have permanent consequences for subsequent developmental possibilities.

Some of the restrictions of generative entrenchment can be offset by modularity, in which development can operate in parallel modules. That is, the development of one body part can be a relatively autonomous process within the entire development of the organism (Schlosser and Wagner Reference Schlosser and Wagner2004). But modules are only partially isolated from other modules, and both integration and generative entrenchment typically remain within modules, even if modularity partially frees the development of some modules from that of others. In addition, some developmental processes are highly robust with respect to environmental or genetic perturbations, channeling the organism or the developmental module toward a particular target phenotype under a wide range of conditions (De Visser et al. Reference De Visser, Hermisson, Wagner, Meyers, Bagheri-Chaichian, Blanchard, Chao, Cheverud, Elena, Fontana, Gibson, Hansen, Krakauer, Lewontin, Ofria, Rice, von Dassow, Wagner and Whitlock2003). By contrast, other developmental processes show high levels of plasticity, meaning that the phenotypic outcomes are strongly environment specific (Pigliucci Reference Pigliucci2001).

Individual development extends into the production of gametes, and the channeling of this production is based on limitations on available developmental resources. A human female, for instance, is born with all the eggs she will ever use to make offspring. Each egg is unique, but they all result from the same meiosis process in which the formation of gametes involves the reduction of genes from diploid (two copies of each gene) to haploid (one copy). These haploid combinations are restricted to the possibilities created by the diploid cell and are thus channeled by both its genetic material and the meiosis process.

These various forms of channeling can have evolutionary implications, which we will explore in section 3. But before we do so, we first need to jump from the organism-level phenomenon of developmental channeling to the ensemble-level outcome of evolutionary dappling.

2.2. The dappling of organismic forms

All vertebrates have four or fewer limbs. Terrestrial vertebrates can be traced back to the evolution of tetrapods from bony fish in the Devonian, around four hundred million years ago (Clack Reference Clack2012). An early tetrapod was Tiktaalik, which bore features of fish (scales and gills) but had a flattened head like that of a reptile or amphibian (Daeschler et al. Reference Daeschler, Shubin and Jenkins2006). Tiktaalik had limbs capable of carrying it out of its submerged state and into shallow water or even dry land. From the first tetrapods descended all reptiles, amphibians, and mammals. There are tens of thousands of tetrapod species and the overwhelming majority retain four limbs. Some have two limbs: Whales evolved from terrestrial tetrapods, and in so doing lost their rear limbs and had their forelimbs transformed into flippers. And there are tetrapods that lost all four limbs, such as snakes, caecilians (limbless amphibians), and pygopodids (legless lizards).

Thus, while limb number for vertebrates could in principle be 0, 1, 2, 3, 4, 5, and so on, we find only 0, 2, and 4 represented by extant or ancestral taxa. Instead of the possibility space being uniformly filled, it is only partially filled—it is what we, here, label dappled. Some limb numbers would be unlikely to be adaptive were they to evolve. For example, it is difficult to imagine forms with an odd number of limbs having fitness advantages. But we should be careful not to attempt to explain the lack of particular limb numbers based on the imagined fitness cost of the limbs. Before fitness costs play a role, the traits must arise, and the development of limbs plays a key role in which possibilities are present.

What evolution can do—and has done—is to transform and repurpose limbs, or to eliminate them altogether. The forelimbs of tetrapods have evolved into adaptations for flight three different times (Rayner Reference Rayner1988). The pterosaurs evolved long, strong fifth fingers (pinkies), which formed the basis of their wings. (Pterodactylus means winged finger.) Birds fly with extended arms, whereas bats have elongated splayed fingers with connecting skin like a pitcher’s mitt. This shows some degree of underlying anatomical flexibility in supporting adaptive functions. But this flexibility—this multiple realizability of wings—when examined closely, shows strong conservatism. While evolution seems to be able to stretch and shrink and fuse bones, and sometimes even to multiply homologous structures, it is much more difficult for it to generate novel structures, that is, structures not homologous to those of other vertebrate species.

The irregularly occupied morphospace represented by vertebrate limb number is a widespread phenomenon. Extant phenotypic traits are in general irregularly distributed across what could be seen as an ideal space of all possible phenotypes (Oster and Alberch Reference Oster and Alberch1982). Evolutionary dappling refers to this irregular distribution, what Alberch (Reference Alberch and Bonner1982, 317) described as “the empty spaces and the ordered pattern in morphology-space.” This is distinct from morphologic disparity, which is defined as “the amount of difference between related phyla, classes, species, individuals, proteins, genes etc.” (Runnegar Reference Runnegar, Campbell and Day1987, 40; see also Gould Reference Gould1989). While disparity is about the degree to which taxa vary in their features, dappling has a broader scope and considers the entirety of morphospace and how it is populated.

While we may imagine that evolution can bring about any nomologically possible organismic form, the result of billions of years of evolution clearly shows the conservation of a relatively small number of forms around which most species are clustered. Explaining this clustering requires resources not found in traditional population-genetic approaches to evolution, which are mostly focused on genetic and fitness differences within inbreeding populations. Indeed, the higher-level biases encountered in morphospace have typically been approached from the comparative branches of biology, such as comparative morphology and embryology, and are currently under the scope of evo-devo (Hall Reference Hall2012; Love and Raff Reference Love and Raff2003).

Because the concept of morphospace is foundational to the concept of dappling, we should pause here for a moment to consider the former in more depth. Take the game of chess, with its 64 squares and 32 pieces. If we are allowed to freely place the pieces on the board, there is a vast number of arrangements we can create. Let’s call the full set of such arrangements the chess piece arrangement space. Now we can ask of this space which regions can be reached through the course of legal chess play. Because of the fixed starting arrangement of chess pieces combined with the strictures of the legal moves for each piece, only a subset of this space can be reached through legal chess play. The chess piece arrangement space is thus dappled by legal play.

Consider just the first move. There are twenty legal chess openings. How many illegal (though physically possible) openings are there? This is unanswerable without giving at least some constraints. If we begin with the normal setup and hold that white starts (as it does in legal chess), and that any white piece can move to any unoccupied space on the board, then the number of possible openings is 512, resulting in about 25 illegal moves for every legal move. The combination of the legal moves and a fixed starting position results in a vast space of impossible (as well as highly improbable) chess arrangements. Because of this, chess play is nonergodic. An ergodic system is one in which all regions of the state space will (in the long run) be occupied with equal frequency. This is clearly not a feature of chess play, where not all pieces can visit every point in the chess piece arrangement space, let alone with the same frequency.

When it comes to phenotypes, an evolving system would be ergodic if given enough time it would visit all regions of morphospace. However, evolving systems are nonergodic. Just as there is an initial arrangement and set of rules for chess play, so is there an initial arrangement of organisms and a set of developmental rules. These rules are not as perspicuous as are chess rules, but they have the same effect of dappling the possibility space. Further, just as the rules are clearer in chess than in biology, so is the space more easily defined. A chessboard is a closed space with discrete positions. There is either a white pawn at c2 or there is not. For organisms, many traits—height, for example—vary continuously and are unbounded, resulting in a much more complex, highly dimensional morphospace.

What is important to see is that the space of possible chess arrangements is larger than the arrangements reachable through legal chess play. Analogously, the space of possible organismic forms is larger than is reachable through evolution. Our argument in this article is that just as chess rules make the exploration of chess piece arrangement space nonergodic, channeling makes the evolutionary exploration of morphospace nonergodic. Many piece arrangements are not reachable because of how each individual piece can move at any given time. Similarly, many regions of morphospace are unreachable (or unlikely to be reached) using the evolutionary process due to the developmental possibilities of organisms—that is, to developmental channeling.

While channeling concerns the individual-level developmental process, dappling is a higher-level product. Channeling concerns the constraints and drives controlling the course of development, whereas dappling concerns the way the space of possible forms is occupied. Dappling does not concern one particular level but can be observed at a variety of levels. For instance, we can inquire into the dappling within the scarabs, within the beetles, within the insects, or within the arthropods. In the following section, we will explore how developmental channeling can be a cause of evolutionary dappling.

4.1. Distinguishing evolvability from channeling and dappling

As discussed previously, one might think that evolvability could play the role of channeling or dappling, which would undermine our arguments for the usefulness of these concepts. Let’s now consider how evolvability relates to the channeling-dappling framework. One difficulty with juxtaposing our framework with evolvability is that evolvability is not univocal (Lynch Reference Lynch2007; Villegas et al. Reference Villegas, Love, de la Rosa, Brigandt, Wagner, Hansen, Houle, Pavličev and Pélabon2023) and has been the subject of discussion within different branches of evolutionary biology (e.g., Wagner and Altenberg Reference Wagner and Altenberg1996; Hansen and Houle Reference Hansen and Houle2008; Pavličev et al. Reference Pavličev, Cheverud and Wagner2011; Payne and Wagner 2019) as well as the philosophy of biology (Love Reference Love2003; Brown Reference Brown2014; Brigandt Reference Brigandt and Love2015; Nuño de la Rosa and Villegas Reference Nuño de la Rosa and Villegas2022). Some have argued that rather than being a single notion, evolvability refers to “a family of related concepts” (Pigliucci Reference Pigliucci2008, 75).

What is agreed upon is that evolvability refers to a capacity or disposition, but what this capacity is of and for is not always clear (Brigandt et al. Reference Brigandt, Villegas, Love, Nuño de la Rosa, Hansen, Houle, Pavličev and Pélabon2023). Bearers of evolvability can range from traits (Hansen and Houle Reference Hansen and Houle2008) to lineages (Kirschner and Gerhart Reference Kirschner and Gerhart1998), while what evolvability is the capacity for is often conceptualized differently depending on the explanatory context. For instance, it can be the capacity to respond to selection (Hansen and Houle Reference Hansen and Houle2008) or the ability to produce improved phenotypic variation (Wagner and Altenberg Reference Wagner and Altenberg1996). Under some conceptions, evolvability is linked with selection, while others hold that it is a capacity involving only internal features of the population, not their selective consequences. If one considers populations as bearers of evolvability, a central internal feature affecting this disposition is the developmental channeling of the organisms composing the population. But in addition, some views of evolvability may include other populational components affecting the ability to evolve. In a well-known account of evolvability, Brown (Reference Brown2014, 550) takes this latter approach:

Under Brown’s framework, channeling is certainly a part of evolvability, but evolvability involves more. To take just one example, population structure is one of the listed features of evolvability. Population structure is clearly a property of populations, not individuals. Nor is it a simple additive result of individual properties (in the way that the mass of a population is simply the additive result of individual masses). Thus, whether or not one takes fitness values to bear on evolvability, evolvability is clearly not identical with channeling or with dappling.

To get a clearer sense of evolvability’s relationship with dappling and channeling, let’s take a simplified species with only two traits: height and weight. We can place every member of the species on a simple two-dimensional plane in which one trait (height) is on one axis and the other (weight) is on the second axis. Now add fitness as the third dimension, forming valleys of low fitness and peaks of high fitness. Each organism will occupy a particular point on the landscape and a population will be a cloud of points.

Now we can ask what the capacity is for a population to evolve from the place it currently occupies to another place on the landscape. One reply to this question would be that it is both the fitness gradients across the landscape, as well as internal features of the population, that constitute this capacity. But one could instead follow Brown (Reference Brown2014) and others in separating the tendencies afforded by fitness from evolvability. If one does so and considers populations as bearers of evolvability, developmental channeling will play a role in it. But in addition, other higher-level components affecting the ability to evolve, such as population structure, will be included within evolvability. Evolvability is thus a higher-level disposition resting on the foundation of (organism-level) channeling as well as population-level properties.

Now take the simplified landscape mentioned in the preceding text and add more and more traits. And instead of a single population, or even a single species, let’s populate the landscape with all the species from a particular higher taxon. We can frame channeling, evolvability, and dappling with respect to this landscape. If we zoom into individual organisms, we can ask about their possible life history trajectories, how they are constrained and driven in specific ways. This is channeling.

Now we can zoom out to populations and ask about their capacity to evolve from one place in the landscape to another. Setting aside fitness and the challenges of crossing adaptive valleys, we are asking here only about the capacity that the population has to go from occupying one place in the landscape to another. This is evolvability.

Finally, we can ask about the phenotypic patterns exhibited over this landscape and observe how the space is filled by the various populations over the range of species. Instead of being evenly distributed, the landscape will be patchy at all taxonomic scales, with rich clumps of species, genera, families, and so on separated by broad deserts. The landscape is thus dappled.

Evolvability therefore plays different roles than channeling or dappling. The channeling-dappling framework links individual-level dispositions with evolutionary outcomes. Evolvability, however, points to a disposition not of organisms, but of evolvable units such as populations or species. The evolvability of a population can be changed, for instance, merely by changing the size of the population. It is an important question how population size has evolutionary consequences, but this is a distinct question from asking how individual development has evolutionary consequences. Both are important questions, highlighting the importance of both evolvability and channeling.

4.2. The extended evolutionary synthesis and reciprocal causation

There is an ongoing debate about the nature and explanatory resources of evolutionary theory. This debate is often framed in terms of “standard evolutionary theory” (SET) versus the EES (Pigliucci and Müller Reference Pigliucci and Müller2010). EES proponents view their account as needed to overcome the limitations of SET. The first thing to note is that it is misleading to hold that there is a univocal SET. Instead, evolutionary theory developed and changed over time and is not understood in the same way by all SET advocates (Stoltzfus Reference Stoltzfus2017). Setting these complications aside, what EES defenders argue is that the evolutionary causes acknowledged by SET are overly restrictive. One case they point to is the traditional assortment of causes into two categories, proximate and ultimate (Mayr Reference Mayr1961). To take the classic example of bird migratory behavior, proximate causes include things like a change in day length that triggers the behavior in particular birds (Pokrovsky et al. Reference Pokrovsky, Kölzsch, Sherub, Fiedler, Glazov, Kulikova, Wikelski and Flack2021), whereas ultimate causes include the fitness benefits gained from spending part of the year in one region, part in another (such as spending summers in temperate zones with a seasonal abundance of food and reduced predators, and winters in tropical regions with warmer temperatures), which explain why the trait evolved (and is maintained).

Within SET, proximate and ultimate causes are traditionally conceptualized as distinct and nonoverlapping. EES proponents, instead, argue that the causes are intertwined, that proximate causes can be evolutionary causes. One way this argument is made is by asserting that there can be “reciprocal causation,” in which the actions of organisms modify the selection pressures acting on those very organisms (Laland et al. Reference Laland, Uller, Feldman, Sterelny, Müller, Moczek, Jablonka and Odling-Smee2015), making some proximate causes also ultimate causes. Niche construction is a key example of this (Aaby and Ramsey Reference Aaby and Ramsey2022). Earthworms change the characteristics of the soil, which then exert selection pressures on the behavior of the worms (Ramsey and Aaby Reference Ramsey and Aaby2022). Does the channeling-dappling framework have implications for this debate as well?

To get at this question, let’s revisit the tripartite structure introduced in Table 1. This structure highlights organism-level dispositions that play key roles in evolution. We identified three: fitness, driftability, and channeling. These dispositions become evolutionarily important only through the population-level bridge. This is where reciprocal causation—and thus the conflation of proximate and ultimate causes—can be identified. Let’s thus examine this bridge in more detail, beginning with fitness and driftabiliity before turning to channeling.

In the case of fitness, it is clear that organismic activities can have evolutionary consequences. Behaviors are not merely the passive result of selection, but they can actively shape selection pressures. The case of the earthworms just mentioned is an interesting example since, as Ramsey and Aaby (Reference Ramsey and Aaby2022) discuss, an important activity they perform is to lower the soil matric potential, which is the amount of energy required to extract moisture from soil. Without performing this activity, the worms wouldn’t just be less well off, but they wouldn’t even be able to live where they do. Importantly, these fitness-affecting activities have evolutionary effects in virtue of variation in fitness values within a population. Thus, the population-level bridge needed for organismic fitness to produce adaptation, as we saw in Table 1.

With driftability, organismic activities can also have effects on evolution. Driftability concerns variance in possible offspring number and organisms can change driftability values from one generation to the next by changing the variability of life histories (by, e.g., altering environmental heterogeneity). Organisms can also have an effect on population structure, and in so doing, can change effective population size. This can occur in the absence of driftability changes, but because it is driftability in conjunction with effective population size that determines drift probabilities (Ramsey Reference Ramsey2013b), these changes have evolutionary effects.

With channeling, we can examine the proximate causes that lead to particular organismic forms (and not others). But, as we have seen, channeling also has evolutionary effects. Think again of the set of possible life histories and the range of phenotypic forms they represent. Changes to this possibility space that change the phenotypic forms or their probabilities of occurring can take place through changes in the heritable material or the environmental arena. For example, sexual reproduction generates new organisms with new sets of possible life histories. The recombination of alleles involved in such a process will help determine the possibility space for the offspring.

In section 2.1, we outlined some of the developmental properties that channel organismic form. Take, for example, the modularity of forelimbs and hindlimbs in bats. This relative autonomy of limb development is similar to that of humans but unlike most other mammals (Young and Hallgrímsson Reference Young and Hallgrímsson2005). When bats reproduce, the modular development of their limbs allows new allelic combinations to produce small wing changes that don’t affect hindlimbs, and vice versa. Thus, the properties that channel individual forms (in this case forelimb and hindlimb modularity) help to determine evolutionary trajectories, implying that channeling is a proximate cause with ultimate effects. To see this, notice that the developmental properties that determine possible organismic forms, like modularity, are also considered to be crucial components of the evolvability of populations (Pavličev and Wagner Reference Pavličev and Wagner2012).

It is thus clear that proximate causes have evolutionary effects. Does this also imply that the reason for this is reciprocal causation? The answer to this question requires thinking carefully about the relata in reciprocal causation. As Hazelwood (Reference Hazelwood2023) points out, EES proponents often frame the reciprocal relationship as obtaining between processes (like the process of niche construction and the process of natural selection) instead of entities (such as organisms and their environment). If the reciprocity is between processes, there is a puzzle: How can a process like niche construction simultaneously be an evolutionary process yet undergo reciprocal causation with the very same evolutionary process? As Hazelwood argues, it is better to simply view niche construction as a part of the evolutionary processes without positing another process with which it is reciprocally interacting. In other words, niche construction is not a process interacting with evolution in a reciprocal way, but a part of evolution. Similarly, the channeling of individuals can be seen as part of the evolutionary process rather than as a separate process causally interacting with population-level ones.

Regardless of how one conceptualizes reciprocal causation, it is clear that proximate and ultimate causes are imbricated. However, this need not imply that the proximate-ultimate distinction is incoherent. Some have argued that one can understand the distinction in terms of overlapping causes (Ramsey and Aaby Reference Ramsey and Aaby2022) or in terms of explanatory abstractions (Scholl and Pigliucci Reference Scholl and Pigliucci2015). Our framework is compatible with these possibilities.