1A. To define (or not define) complexity

If we are to understand how complex brains and cognition evolved in only three lineages, we must have some way of recognizing complexity when we see it. To our knowledge, there are presently no universally recognized standards or characteristics for doing so, although there are plenty of notions as to what complexity entails.

Perhaps the most common and enduring notion of cognitive complexity is that associated with intelligence (Morgan, Reference Morgan1882; Thorndike, Reference Thorndike1898; Jerison, Reference Jerison1973; Emery and Clayton, Reference Emery and Clayton2004; Reznikova, Reference Reznikova2007; Roth, Reference Roth2015; Seed and Mayer, Reference Seed, Mayer, Call, Burghardt, Pepperberg, Snowdon and Zentall2017; Zentall Reference Zentall and Sternberg2020), defined as “the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience” (Gottfredson’s Reference Gottfredson1997). Many other concepts, such as mental time travel (remembering the past and imagining the future), innovation and behavioral flexibility (e.g., inventing new behaviors to solve problems), theory of mind (knowledge of intent in other beings), consciousness (awareness of internal states and external surroundings), and tool use have also been associated with complex cognition in humans and other animals (Table 1; Emery and Clayton, Reference Emery and Clayton2004; Laland and Seed, Reference Laland and Seed2021).

Developing common standards is difficult for several reasons, not the least of which is that cognition, in general, is a very fuzzy concept, one that involves different defining characteristics in different disciplines (Bayne et al, Reference Bayne, Brainard, Byrne, Chittka, Clayton, Heyes, Mather, Ölveczky, Shadlen and Suddendorf2019), and one that, arguably, may not even be amenable to any strict categorical definition (Allen, Reference Allen2017). Similarly, finding objective standards for characterizing brain complexity across the diversity of vertebrate and invertebrate brains is challenging, made all the more difficult by fundamental differences in how protostome (invertebrates) and deuterostome (vertebrates and their allies) brains develop, and the varied and often confusing terminology used to describe invertebrate brain structures, sometimes even within the same phylum (see Appendix A).

In this paper, we attempt to circumvent this class of problems by identifying brain, cognitive and other traits that can be widely applied (and recognized) across lineages and that, by all appearances, vary from simple to complex. We then let the phyletic distributions of these traits tell us which, if any, are unique to CC lineages and which are not (see Sections1B and 6A for further detail).

1B. Working premises and general approach

We start with two premises: that cognition is embodied (Clark, Reference Clark1999) and that complex brains and cognition have evolved in only three lineages – vertebrates, cephalopod mollusks and euarthropods. We then selected a wide range of body, sensory, brain, motor, and behavioral/cognitive traits (5 - 8 in each category) that meet several key conditions for a good operational definition (see section 6A). In essence, we are trying to discover the defining characteristics of CC lineages, i.e., the embodied trait profiles that distinguish them from non-CC phyla. Thus, we determined the presence or absence of 35 traits in the three CC lineages, as well as in non-CC lineages, including three non-bilaterian phyla (Porifera, Ctenophora and Cnidaria) and eleven bilaterian phyla (Fig. 1). We started with a phylum-level analysis for non-CC lineages because we had no objective basis for selecting subphylum levels. We then let the distribution of all 35 traits across non-CC and CC lineages determine, which, if any, traits are unique to CC lineages. Finally, information from trait distributions were applied to a two-dimensional framework for understanding how different behavioral and cognitive outcomes might be understood in terms of a unique combination of body, sensory, brain and motor traits (first dimension), and how changes in the combination of traits led to increasing levels of cognitive complexity over time (second dimension).

2A. Multi-trait pyramids: Building brains and cognition from the ground up

Phyletic distributions of different traits are compiled into a single pyramid, with each building block of the pyramid representing a different trait, the length of which is proportional to the number of phyla (or lineages) that share that trait (Fig. 3A). Keeping in mind that each trait is common to all three CC lineages, the bottom of the pyramid represents very primitive (foundational) traits that are broadly shared with non-CC lineages. In contrast, the top of the pyramid (advanced traits) depicts more recently derived traits that are either unique to CC lineages or shared with only one other non-CC phylum. Filling the gap between foundational and advanced traits is a set of intermediate traits that are neither unique to CC lineages nor broadly shared with non-CC lineages.

Figure. 3.

(A) Pyramid of multiple traits that CC lineages share with non-CC lineages, (B) phylogenetic affinities of CC with non-CC lineages in each major bilaterian clade in terms of number of accumulated traits, and (C) the relationship between the number of intermediate and advanced traits in each lineage to show a clear separation of CC and non-CC lineages. The length of each horizontal bar in A represents the total number of lineages that share a given body, sensory, brain, motor, or behavioral/cognitive traits, the latter of which include central place foraging (CPF), and unlimited associative learning (UAL) abilities. For multiple lineages with the same number of intermediate or advanced traits in C, trait numbers have been displaced by a small fraction on either the x- or y-axis to separate individual data points. Within the non-CC cluster, Onychophora (On) has the highest number of intermediate traits, whereas Cnidaria (Cn) and Echinodermata (Ec) are the two phyla with one advanced trait. The dashed line in C is a curvilinear fit (a running average) to the data.

CC lineages clearly outstrip non-CC lineages in terms of total number of traits, as well as the number of traits in each category (Fig. 3B). As might be expected, non-CC lineages with close phylogenetic affinities with CC lineages have greater numbers of total shared traits than other non-CC lineages in the same clade (bracketed groups in Fig. 3B). These include sister groups to Chordata, Mollusca and Euarthropoda (Echinodermata, Annelida, and Onychophora, respectively)(Fig. 1).

When the number of intermediate traits is plotted as a function of the number of advanced traits for each lineage, three trends emerge (Fig. 3C). First, CC lineages can be easily distinguished from the vast majority of non-CC lineages. Although the number of both advanced and intermediate traits contribute to the separation, the number of intermediate traits vary widely among non-CC lineages, whereas the number of advanced traits provide the clearest separation between CC and non-CC lineages. Second, Annelida stands out as a phylum that is somewhere in between CC and non-CC lineages. Third, the relationship between the number of intermediate and advanced traits across lineages is curvilinear rather than linear. The curvilinear relationship is indicative of a cumulative process in which lineages tend to accumulate intermediate traits before acquiring advanced traits. Exceptions include cnidarians and echinoderms, each acquiring one advanced trait without having acquired all intermediate traits (see Section 5 for further discussion of these outliers). Finally, the appearance of more complex behavioral (e.g., central place foraging) and learning traits (unlimited associative learning, UAL) at the top of the pyramid suggest that a combination of several body, sensory, brain and motor traits, and not just one (as found in echinoderms and cnidarians), are pivotal to more complex behaviors and cognitive abilities.

A very interesting and somewhat unexpected finding is that Annelida stands out as sharing more advanced (4 out of 9) and intermediate (11 out of 11) traits with CC lineages than any other non-CC phylum (Fig. 3). As such, animals in this phylum can be thought of as being either on the evolutionary cusp of acquiring the full set of advanced traits, or on the tail end of a regressive loss of these traits, ideas that we explore in Sections 3C and 5. In any event, the ability of these traits to identify outliers like annelids gives us confidence that the distribution outcome was not entirely constrained by our trait selection criteria and working premises, and that the multi-trait analysis is capable of revealing surprises.

2B. The relevance of pivotal traits to active visual sensing and visuomotor control

Among advanced traits is a subset of non-cognitive traits that align with advanced behavioral/ cognitive traits in CC lineages (Fig. 3A) and that we call ‘pivotal’ because of their potential for providing a neural and sensorimotor foundation for the emergence of complex cognition. Pivotal traits (flexible limbs, high resolution eyes, laminated brain structures, high-speed locomotion, eye mobility and 3D rotational senses) can be functionally united in terms of their involvement in active visual sensing. Active sensing can be defined as the use of behavior (motor outputs) to acquire or modulate sensory information (Stamper et al., Reference Stamper, Madhav, Cowan, Fortune, Carlson, Sisneros, Popper and Fay2019). It is fundamental to theories of how cognition emerges from the interactions of an animal with its own environment (Llinas, Reference Llinas and The2001; Varela et al., Reference Varela, Thompson and Rosch2018), and has long been recognized as an important strategy for acquiring information (Gibson, Reference Gibson1950, Reference Gibson1962; von Holst & Mittelstaedt, Reference von Holst and Mittelstaedt1950; Brembs, Reference Brembs2009; Schroeder et al. Reference Schroeder, Wilson, Radman, Scharfman and Lakatos2010; Egelhaaf et al., Reference Egelhaaf, Boeddeker, Kern, Kurtz and Lindemann2012).

Some animals, like echolocating bats and dolphins (Au and Simmons, Reference Au and Simmons2007), or electro-locating fish (Nelson and MacIver, Reference Nelson and MacIver2006), use specialized motor systems for producing a specific stimulus energy, in these examples, high frequency sounds and low-frequency electric fields, for the purpose of probing the environment. But for most animals and for visual senses, in particular, active sensing involves the movement of the head, body or eyes to modify sensory feedback. The swimming or flying motions of fish and insects, for example, produces a form of visual feedback (sensory reafference) known as optic flow, which can be used in closed-loop fashion to guide animals through cluttered environments or with respect to each other in a school of fish or a swarm of bees. Similarly, the independent movement of the head or eyes can be used in task-specific ways to e.g., track a moving prey, search for an item of interest, reach for, grasp or manipulate an object, or extract specific information, such as motion parallax cues for depth perception (Horridge, Reference Horridge1986). Head and eye movements can also be used to redirect attention (visual gaze) to novel stimuli or items of particular salience, such as predators or prey. Accordingly, the design of visual processing regions, like first order, laminated visual regions (a pivotal brain trait), is heavily influenced by the tight coupling between vision and behavior (Zeil et al, Reference Zeil, Boeddeker and Hemmi2008) (see also Appendix C on visuomotor pathways for different behaviors).

Perhaps somewhat surprisingly, the primary and likely first function of mobile eyes was not to track moving objects or direct attention to them, but rather to prevent object images from blurring when animals rapidly move their head or body (Walls, Reference Walls1962; Land, Reference Land1999, Reference Land2019). As Walls (Reference Walls1962) so aptly put it, “the original function of eye muscles was not really to move the eye but rather to hold it still with respect to the environment”. Since images produced by low-resolution eyes are already crude, image blur is primarily a problem for animals with high-resolution vision, and these animals use very similar strategies to prevent image blur (Land, Reference Land2019). They use body motion sensors, like the three-dimensional rotational accelerometers found in the semicircular canals of both vertebrates and cephalopod mollusks, to trigger compensatory eye movements for reflexively opposing the visual effects of turning the head or body (Land, Reference Land2015, Reference Land2019) (see Appendix B: Sensory traits).

As the above discussion reveals, active visual sensing relies on many elements, including high-resolution eyes capable of detecting objects, compensatory eye movements to prevent image blur, and motion sensors to trigger compensatory eye movements. In addition, first order, laminated visual structures in the brain extract the relevant spatiotemporal features that are necessary for object detection in the first place and that allow animals to detect relative movement and movement direction between self and an object of interest (see Section 2D). Finally, the capacity for high-speed locomotion frames the adaptive significance of compensatory eye movements in animals with high-resolution vision.

The importance of these combined pivotal traits to active visual sensing has profound implications, as active sensing also enhances learning abilities, fundamental to the evolution of cognitive complexity (Ginsburg and Jablonka, Reference Ginsburg and Jablonka2021). Animals in control of incoming sensory information learn more quickly than those that passively receive it (Brembs & Heisenberg, Reference Brembs and Heisenberg2000; Wolf & Heisenberg, Reference Wolf and Heisenberg1991). As but one example, active visual sensing is critical to the developmental learning abilities of human infants, who often manipulate the apparent size or position of an object within their central field of view, either by moving closer to the object or by repositioning the object closer to them (Suanda et al, Reference Suanda, Barnhart, Smith and Yu2019; Yu and Smith, Reference Yu and Smith2012).

With respect to learning, adaptive filter architectures (an intermediate brain trait, nominally shared with only two other non-CC phyla) are integral to active sensing (Fig. 3A). Adaptive filter networks use continuous sensory feedback generated by body movements to build knowledge and expectations (i.e., to learn) about the consequences of an animal’s own actions, making it easier for them to predict the consequences of future actions (Montgomery & Bodznick, Reference Montgomery and Bodznick2016, see also Appendix B: Brain traits).

2C. The significance of high-resolution eyes and first-order processing areas

Distal sensing and ‘true’ eyes are often linked together as key traits in the evolution of mind and complex cognition (Llinas, Reference Llinas and The2001; Godfrey-Smith, Reference Godfrey-Smith2020; Trestman, Reference Trestman2013; Feinberg and Mallatt, Reference Feinberg and Mallatt2018; Ginsburg and Jablonka, Reference Ginsburg and Jablonka2019, Reference Ginsburg and Jablonka2021). However, there are a number of reasons for thinking that distal sensing is not the only, or even most important advantage of ‘true’ (image-forming) eyes. One, several other senses – long-range olfactory and auditory senses, as well as short-range flow senses - provide animals with the ability to sense stimulus sources at a distance (see Appendix B: Sensory Traits). Two, since animals in many lineages possess multiple forms of distal sensing, distal sensing per se is not a characteristic that can distinguish CC from non-CC lineages. Three, image-forming eyes can be either low- or high-resolution. Although low-resolution eyes might enable animals to detect a suitable habitat at some distance away, high-resolution eyes can do considerably more, such as determine the identity and direction of a small moving object (Nilsson, Reference Nilsson2009). Four, depth perception based on stereopsis and binocular disparities between the two eyes is possible for animals with frontal vision like cats and primates, but for many others, even those with forward-looking eyes, binocular depth perception is not possible (Martin, Reference Martin2009; Land, Reference Land2019). In fact, the octopus, perhaps the most cognitively complex of all invertebrates, relies on monocular vision (Pungor and Niell, Reference Pungor and Niell2023). Thus, many species, even if they have high-resolution vision, cannot gauge distance unless motion cues are generated, either by the movements of the receiver or the stimulus source moving closer to or further away from the receiver (Kral, Reference Kral2003).

In short, distal sensing by image-forming eyes likely provided an adaptive advantage, but only with the aid of motion. Thus, we propose that the pivotal importance of ‘true’ (image-forming) eyes to cognition was not so much in terms of distal sensing abilities, but in terms of object vision and active sensing abilities, and the information processing demands that high- (but not low-) resolution eyes place on first-order visual processing areas in the brain. According to Nilsson (Reference Nilsson2013), the information processing demands of high-resolution eyes is approximately three orders of magnitude higher than those imposed by low-resolution eyes and four to eight orders higher than those imposed by non-imaging photoreceptive organs. Thus, first-order visual structures can be seen as complex brain structures that have evolved in parallel with the high processing demands of complex (high-resolution) eyes. Seen in this light, information-processing demands depend less on the complexity of information in the animal’s surrounding environment or the complexity of the problem to be solved, but more on the complexity of sense organs that deliver information to the brain and the complexity of body parts (e.g., flexible limbs) that have to be moved. This bottom-up perspective is at the core of embodied cognition (Pfeifer et al, Reference Pfeifer, Iida and Lungarella2014).

In this regard, the computational power of the vertebrate retina and the optic lobes of euarthropods and cephalopod mollusks, is nothing short of remarkable (Gollisch & Meister, Reference Gollisch and Meister2010). These laminated structures consist of several cell layers, beginning with a layer of photoreceptive cells and ending with a layer of efferent cells that send their ultimate computational results to mid- and high-level motor centers in the motor hierarchy (Fig. 4). Efferents form two major visuomotor pathways in the brains of CC lineages that, among other things, underlie fundamentally different types of orienting behaviors and cognitive abilities (Section 8; Appendix C.2). As the information travels to higher brain centers, the projections typically remain in retinotopic register, preserving the spatial structure of peripheral maps at each processing station along each of the two pathways.

Figure 4.

Laminated visual centers that preserve the spatial order of photoreceptive inputs as retinotopic maps and that use parallel processing in different layers to extract different spatiotemporal features of high-resolution visual inputs. Projections from efferent neurons follow two major pathways that culminate in mid- (yellow) and high-level (blue) motor regions in vertebrate, euarthropod and cephalopod mollusk brains. Layers include the photoreceptive layer (red), intermediate processing layers of interneurons with lateral interconnections (blue) and efferent layers (green). Efferent cell layers provide the main outputs to other regions of the brain. Inspired by the text and figures in Sanes & Zipursky (Reference Sanes and Zipursky2010).

Laminated structures support multiple parallel but interacting subsystems (local circuits) that extract different salient features for directing biologically relevant behaviors (Fig. 4) (Sanes and Zipursky, Reference Sanes and Zipursky2010). The mammalian retina, for example, has at least 15 distinct efferent neurons (retinal ganglion cells), each of which encodes a different feature (Gollisch & Meister, Reference Gollisch and Meister2010; Masland, Reference Masland2001; Wassle, Reference Wassle2004). Examples of features that are extracted by local circuits in the retina and optic lobes of CC animals include: (1) light/dark edges and object boundaries, (2) polarized light directions, (3) different directions of narrow-field (small object) motion, (4) different translational (fore/aft, left/right and up/down) and rotational (pitch, roll and yaw) directions of wide-field (optic flow) motion, (5) the motion of approaching objects, (6) trajectories of objects in motion, and (7) missing, but anticipated elements in a spatiotemporal series of visual events (the so-called omitted stimulus response) (Gollisch & Meister, Reference Gollisch and Meister2010; Krapp, Reference Krapp, Bleckmann, Mogdans and Coombs2014; Wiederman et al., Reference Wiederman, Fabian, Dunbier and O’Carroll2017; Cheong et al, Reference Cheong, Siwanowicz and Card2020).

Our interpretation is that local circuits underlying features 1, 3, 5 and 6 function as elements involved in the identification and classification of objects. Features 6 and 7 are crucial components of, and precursors to advanced cognitive abilities related to anticipation and future planning. Feature 4 additionally plays a role in cognitive constructs of self by providing visual reafferent (optic flow) information on self-directed motions. Finally, motion direction, whether it is the direction from which a small entity enters the visual field, the direction of global optic flow as animals swim or fly through the environment, or the predicted direction of an object already in motion, is a common theme to features 3 – 6. None of these features could be extracted if it were not for retinotopic (or more generally, topographic) maps that preserve the spatial order of photoreceptor inputs in all three CC lineages. These maps are created in the sensory periphery by the physics of how lenses refract and direct light onto photoreceptor arrays, but they are often preserved at higher brain levels of the motor hierarchy, especially mid-level centers in the visuomotor hierarchy that direct egocentric orienting behaviors (Fig. 4, see also Appendix C.2).

Feature extraction is fundamental to object detection and thus, to intentionality, the ability to perceive, learn about, or conceptualize compound objects (Ginsburg and Jablonka, Reference Ginsburg and Jablonka2019). Feature extraction also reduces redundancy and noise. By extracting spatiotemporal features that are essential to the animal’s survival, laminated structures whittle down incoming information into a more manageable subset of efferent outputs that can be more efficiently processed by higher brain centers (e.g., Atick and Redlich, Reference Atick and Redlich1990). Some efferent neurons even rely on a form of predictive coding, which arises from lateral inhibitory circuits, and that suppresses predictable features, while also enhancing novel and unexpected information (Hosoya et al., Reference Hosoya, Baccus and Meister2005; Johnston et al., Reference Johnston, Seibel, Darnet, Renninger, Orger and Lagnado2019; Srinivasan et al., Reference Srinivasan, Laughlin and Dubs1982).

It is often argued, for good reason, that advanced cognitive abilities rely on higher-order brain regions, such as the cerebral cortex in humans (Rakic, Reference Rakic2009; Hofman, Reference Hofman2019). However, here we see how low-order visual areas can have a huge impact on the cognitive capacity of higher order brain regions. Not only do they extract spatiotemporal features that enable object and motion vision, but they also provide rudimentary predictive abilities, a key theme of complex cognition, as well as a key operating principle of the brain (Clark, Reference Clark1999; Emery & Clayton, Reference Emery and Clayton2004; Hull, Reference Hull2020; Llinas, Reference Llinas and The2001; Trestman, Reference Trestman2023).

2D. Pivotal traits: Laying the groundwork for complexly active bodies

Trait distributions identified in this study both confirm and augment the suite of body, sensory and motor traits hypothesized by Trestman (Reference Trestman2013) to be uniquely important for complexly active bodies. Trestman’s suite included (1) appendages with several degrees of freedom, (2) distal senses (e.g., true eyes) and (3) the motor capacity for manipulating objects. The proposed thread binding all these traits together was a capacity for fine, sensorimotor (and brain) control of goal-directed movements of the body and various body parts (e.g., legs, fins, chelipeds, mouthparts etc.) These traits are very similar in principle, if not identical, to four of the current study’s traits - two distributed as advanced/pivotal traits (flexible limbs and high-resolution eyes) and two as intermediate traits (3D mobility and object manipulation) (Fig. 3A).

Augmenting the originally hypothesized traits are several additional pivotal traits (3D body motion sensors, laminated, retinotopically organized visual centers in the brain, and motor capacities for mobile eyes and high-speed locomotion). As discussed in Sections 2B and C, all of these pivotal traits interact in ways that are critical to active visual sensing and the visuomotor control of body actions. Thus, they produce the physical and cognitive capacity for object-oriented and spatially complex actions in three dimensions. As such, we regard pivotal body, sensory, brain and motor traits as the ‘tipping points’ on the pyramidal groundwork that enabled animals in CC lineages to transition from slow and largely reactive motion in two dimensions to rapid and more proactive motion in three dimensions.

3A. The trait-linkage hypothesis

The hypothetical framework of Fig. 5 suggests several testable hypotheses. One is that there is a general tendency for certain traits to be linked together in time to produce different levels or permutations of cognitive and behavioral complexity. A corollary of this hypothesis is that transitions in one trait tract (e.g., from low- to high-resolution eyes in the sensory tract) are loosely linked in time to transitions in parallel tracts (e.g., from unlaminated to laminated visual regions in the brain tract and from unjointed to jointed/flexible limbs in the body tract). If traits and trait-transitions are indeed linked, one would expect to find different levels of complexity, defined by a distinct combination of traits, in different groups of animals. Table 3 provides preliminary evidence of how body, visual and brain traits are linked at different complexity levels in several subphylum taxa (classes) of animals in both CC and non-CC lineages (see Sections 6D and 7C for further details). Table 3 also lends further support to our decision to treat cephalopod mollusks (complexity level 4) as a CC lineage that is separate from other non-CC mollusk classes like Gastropoda (level 3) or Polyplacaphora and Bivalvia (level 1), and likewise, our decision to lump all three major euarthropod clades into a single CC lineage (complexity level 4) at the phylum level.

Table 3.

Animal classes that have body, visual and brain traits at the same level of complexity. Body traits - striated muscle (SM), nerve tissue/system (NS), bilateral symmetry (BS), paired appendages (PA), and flexible limbs (FL). See Table 2 for visual traits

3B. Evolutionary scenarios for the multi-trait framework

How well does the multi-trait framework of Fig. 5 fit with the current evidence and hypotheses of how various traits evolved in different animal groups? The evolution of the brain and nervous system is currently viewed through the lens of two competing hypotheses. One is that complex brains in CC lineages had a single (monophyletic) origin – that is, they were inherited from a single, common bilaterian ancestor (the bilaterian node in Fig. 1)(Hirth, Reference Hirth2010; Holland et al., Reference Holland, Carvalho, Escriva, Laudet, Schubert, Shimeld and Yu2013). The competing hypothesis is that complex brains in CC lineages had multiple (polyphyletic) origins (Moroz, Reference Moroz2009, Reference Moroz2015). Depending on the relative position of sponges and ctenophores (comb jellies) on the phylogenetic tree, neurons and the nervous system are also thought to have had either a monophyletic or polyphyletic origin (Moroz, Reference Moroz2009; Liebeskind et al, Reference Liebeskind, Hillis, Zakon and Hofmann2016).

The multi-trait framework cannot resolve the issue of monophyletic vs polyphyletic origins of brains or nervous systems, but it can be applied to both scenarios. To illustrate this point, we assume that sponges (Porifera) are the sister group to the common ancestor of all other phyla, as shown in Fig. 1, and that the nervous system evolved only once from this common ancestor. The mixed mono/polyphyletic scenario depicted in Fig. 6 thus illustrates a monophyletic origin of foundational traits, such as neurons (after transition 1) and bilateral symmetry (after transition 2), followed by a polyphyletic origin of intermediate and advanced traits, such as paired appendages after transition 3, and complex sensory (high-resolution eyes) and brain features (laminated visual structures) after transition 4. This mixed scenario may or may not turn out to be correct, but it is consistent with the current thinking on the polyphyletic origin of image-forming eyes (Nilsson, Reference Nilsson2009; Gehring, Reference Gehring2014).

Figure 6.

Mixed mono- and polyphyletic scenarios for how CC and non-CC lineages may have transitioned from one level of multi-trait complexity to the next, with each level of complexity being the outcome of one of four transitions (see Table 3 for levels of complexity exemplified by different lineages). Complexity levels ascribed to each lineage (in parentheses) are based on the information in Table 3. Scenario in A illustrates the parallel pathways followed by all three CC lineages, resulting in the progressive addition of traits that are accumulated over each of four transitions to produce increasing levels of complexity, culminating in complexity level 4, as illustrated in Fig. 5. Scenario B illustrates how the ancestors of some non-CC lineages may have reached CC status in the same way as CC lineages, but with subsequent loss of advanced traits later in time, resulting in a reverse transition from complexity level 4 to 3. Scenario C depicts progressive addition of traits, but with different non-CC lineages completing less than four transitions. Note that phyla at lower complexity levels, such as Tardigrada and Onychophora, could have theoretically attained their current level of complexity (in this case, level 3) by either failing to complete all transitions, as illustrated in C, or by regressive losses (transition reversals), as illustrated in B.

Figure 6 also illustrates the while all three CC lineages completed the last of several transitions to acquire advanced traits (Fig. 6A), many non-CC lineages did not (Fig. 6C). Some may have initially followed similar paths leading to the acquisition of advanced traits, only to lose all (e.g., Tardigrada, Fig. 6B) or only some of them later on (e.g., Annelida, see Section 3C). Others completed none (e.g., Porifera), one (e.g., Ctenophora), two (e.g., Rotifera) or three (e.g., Onychophora) transitions to acquire different numbers of foundational and intermediate traits, but no advanced traits (Fig. 6C).

As Fig. 6A illustrates, all three CC lineages appear to follow the same general progressive sequence of transitions to acquire advanced traits. However, this does not mean that they completed the sequences during the same time period or that transitions were always stepwise. For example, the path from less to more complex body traits (limbless to flexible limbs) in euarthropods appears to involve a discrete intermediate step involving the acquisition of unjointed limbs (Transition 3) before the transformation to jointed limbs (transition 4)(see Sections 4C and D). A similar, stepwise path to flexible limbs is less clear for vertebrates and cephalopod mollusks (see Section 4D). Moreover, it seems that euarthropods acquired their flexible limbs much earlier than either vertebrates or cephalopod mollusks (see Section 4D). These examples are a stark reminder that evolutionary histories are varied and determined by a complicated set of interacting factors that we don’t yet fully understand.

Nevertheless, we think the two-dimensional framework of Fig. 5 is a useful construct. It should not be interpreted as an orderly, one-way road map that describes the evolutionary history (path taken) of all lineages, but rather as an overall framework from within which the evolution of brains and cognitive abilities can be understood in terms of an interaction of multiple traits. This interaction produces increasing levels (or different permutations) of complexity that roughly unfold in the same sequence to produce cumulative effects in many, if not all lineages. The framework also suggests that multiple traits are linked and that certain combinations of traits are more frequently expressed than others (see Section 7C, Table 3).

3C. Selection factors and constraints in the evolution of pivotal traits

What selection factors might be operating in the evolution of multiple pivotal traits in the three CC lineages? Since most, if not all, pivotal traits appear in the fossil record by the end of the Cambrian (Fig. 7; Trestman Reference Trestman2013), the Cambrian explosion is highly relevant. The Cambrian explosion is perhaps best characterized as a rapid increase in the number of different body forms, which gave rise to all of the known phyla of today. But as Trestman (Reference Trestman2013) points out, it also describes a rapid increase in the complexity of body features, such as articulated appendages, multi-faceted eyes, and even differentiated brain structures, as evidenced from the fossil record (Fig. 7, Section 4D).

Figure 7.

Evolutionary timeline for some of the key traits and events discussed in this paper.

Suffice-it-to-say, many factors, interacting in complex ways, likely triggered or fueled the Cambrian explosion (reviewed by Smith and Harper, Reference Smith and Harper2013). These include abiotic factors, such as increasing levels of available oxygen and ambient light, a rising sea level and global warming, as well as biotic factors like the advent of bilateral symmetry, the emergence of complex food webs, and an escalating arms race involving the coevolution of predators and prey.

So, how does body complexity fit into this overall picture? Parker (Reference Parker2003) argued that complex, image forming eyes were key factors in the Cambrian explosion, rapidly increasing the visual abilities of early euarthropods, fueling or even triggering the arms race. As Shubin, Tabin and Carroll (Reference Pungor and Niell1997) so cleverly put it, however, the ‘arms’ race could also be regarded as a ‘limbs’ race to build better (more flexible and varied) appendages for multiple purposes. Needless-to-say, eyes and limbs function together as an integrated whole in visually guided behaviors. In this respect, Trestman (Reference Trestman2013, Reference Trestman, Andrews and Beck2018) proposed that the brain’s ability to extract spatiotemporal features of the visual world transformed the evolution of limbs and other active body parts in a distinctive way. Without sufficient sensory information and brain capacity to guide them, flexible limbs would have been of little functional use for complex, three-dimensional motion. Once these abilities were acquired, the adaptive value of flexible limbs (and variations in their morphology) for a wide variety of other, spatially complex tasks (e.g., grasping and manipulating objects) would have increased, leading to increased diversification of active body parts and speciation, a pattern clearly observed in Cambrian arthropods, and in the subsequent evolutionary radiations of vertebrates and cephalopods.

The evolution of pivotal traits is also constrained by the laws of physics and gene regulatory networks. As but one example of the former, physical principles constrain the ways in which light rays from different directions can be focused and redirected onto photoreceptor arrays to form visual images, with lenses being by far the most commonly adopted solution in the animal kingdom (Land, Reference Land2012). In contrast, gene regulatory networks constrain body plans by controlling the developmental processes, including the differentiation of the body along different body axes, as well as different internal tissue organizations like brains and exterior body parts, including heads, trunks and limbs (Carroll, Reference Carroll2008).

It now appears that much of the regulatory genome pre-dates the origin of bilaterian clades, appearing as early as almost a billion years ago (Erwin, Reference Erwin2020)(Fig. 7), suggesting that all phyla inherited very similar, if not identical genetic toolkits. Limb development, for example, appears to be controlled by the same gene regulatory networks in both vertebrates and invertebrates (Panganiban et al., Reference Panganiban, Irvine, Lowe, Roehl, Corley, Sherbon, Grenier, Fallon, Kimble, Walker, Wray, Swalla, Martindale and Carroll1997; Pueyo & Couso, Reference Pueyo and Couso2005; Shubin et al., Reference Pungor and Niell1997; Tarazona et al., Reference Tarazona, Lopez, Slota and Cohn2019), as does eye development (Halder et al., Reference Halder, Callaerts and Gehring1995; Gehring, Reference Gehring2004, Reference Gehring2011; Fernald, Reference Fernald2006; Vopalensky and Kozmik, Reference Vopalensky and Kozmik2009; Joly et al. Reference Joly, Recher, Brombin, Ngo and Hartenstein2016). Moreover, the eye toolkit not only regulates eye morphogenesis, but also that of nearby structures, including the head and target neuropils in the brain, ensuring that all these bits and pieces are developed in proper relationship to one another (Gehring, Reference Gehring2004, Carroll, Reference Carroll2008, Shubin et al, Reference Shubin, Tabin and Carroll2009; Joly et al., Reference Joly, Recher, Brombin, Ngo and Hartenstein2016). Finally, shared gene regulatory networks control the organization and development of different brain regions, structures and even corresponding circuitries, resulting in functional similarities between several vertebrate and invertebrate brain regions (Tomer et al., Reference Tomer, Denes, Tessmar-Raible and Arendt2010; Strausfeld and Hirth, Reference Strausfeld and Hirth2013; Wolff and Strausfeld, Reference Wolff and Strausfeld2015; Bridi et al., Reference Bridi, Ludlow, Kottler, Hartmann, Vanden Broeck, Dearlove, Göker, Strausfeld, Callaerts and Hirth2020)(see Appendix A).

Genetic mechanisms can also explain how some Annelids have come to have some, but not all advanced traits (Fig. 3). Fossil annelids (∼500 mya) appear to have had more complex neuroanatomical features than some of their more modern descendants, indicating that the nervous system was subsequently reduced several times over millions of years (Heuer et al., Reference Heuer, Müller, Todt and Loesel2010; Parry and Caron, Reference Parry and Caron2019). According to Isaeva and Rozhnov (Reference Isaeva and Rozhnov2021), regressive loss of traits via loss of individual genes, most notably Hox genes, is rather common among Metazoans. The simplification of non-CC tardigrade and nematode body plans (including the brain) is another putative example of trait loss due to the loss of Hox genes (Smith et al., Reference Smith, Boothby, Giovannini, Rebecchi, Jockusch and Goldstein2016; Minelli, Reference Minelli and Wanninger2015; Sommer, Reference Sommer and Wanninger2015; Isaeva and Rozhnov, Reference Isaeva and Rozhnov2021). Finally, secondary loss of advanced traits can even occur within CC lineages – the loss of limbs in snakes and many other vertebrate classes (Lande, Reference Lande1978), for example, or the loss of eyes/vision in both invertebrate and vertebrate species that inhabit dark environments like the deep sea, subterranean caves or underground burrows (Sumner-Rooney, Reference Sumner-Rooney2018).

4A. Transition 1: From undifferentiated tissues to differentiated nerve and muscle cells

The development of specialized nerve cells (neurons) for inter-cell communication and a nervous system of interconnected neurons is widely held to be a key innovation in metazoan evolution (Nielsen, Reference Nielsen2008) - one that enabled major advances in the way that information is encoded, transmitted, processed and used in the guidance of behavior. In essence, neurons function as analog-to-digital-to-analog converters, with input and output regions (the dendrites and axon terminal, respectively) having analog properties. Analog properties enable fine-scale integration of different inputs and likewise, outputs that can vary continuously in strength. Between input and output regions of the neuron is a single transmission line (axon) that supports all-or-none action potentials, a digital signal with temporal coding properties that propagates over long distances without degradation in strength. Regardless of whether the nervous system arose only once or multiple times (see Section 3B), the remarkable appearance of the polarized (one-way transmission of information) neuron with its combined analog and digital features had a major impact on the ability of animals to acquire and process information.

In parallel with the development of neurons was the development of striated muscle, although there is some controversy as to exactly when and how many times it arose (Burton, Reference Burton2008; Seipel and Schmid, Reference Seipel and Schmid2005; Steinmetz et al., Reference Steinmetz, Kraus, Larroux, Hammel, Amon-Hassenzhal, Houlistone, Wordheid, Nickel, Degan and Technau2012). Striated muscles have two important properties. One is that they are capable of rapid contraction and the second is that they depend on nerve stimulation for contraction, opening the door for rapid, but finely controlled, muscle-powered movements. Before the advent of muscle-based locomotion, animals relied on relatively slow and cumbersome ciliary or flagellar mechanisms, but as Biewener & Patek (Reference Biewener and Patek2018) put it, the capacity for animal motion “exploded” when actin and myosin were assembled into striated muscle cells.

Shortly after or concomitant with the advent of neurons and striated muscles, neurons became further differentiated into different neuron types, each for a dedicated function, consistent with the idea of a nervous system, rather than just one or two isolated nerve or sense cells. Neurons can be broadly classified into three types: (1) specialized sensory neurons (e.g., photoreceptors) for transducing stimulus energy (light, sound etc.) into the electrochemical currency of neuron language, (2) motor neurons for controlling the timing and amplitude of muscle cell contractions, and (3) interneurons for integrative handshaking between sensory and motor neurons.

Before specialized neurons, sensorimotor integration was strictly intracellular (Gehring, Reference Gehring2014; Randel & Jékely, Reference Randel and Jékely2016), as is still evidenced in the photoreceptive cells of extant sponge and cnidarian larvae (Gühmann et al., Reference Gühmann, Jia, Randel, Verasztó, Bezares-Calderón, Michiels, Yokoyama and Jékely2015; Kojima et al., Reference Kojima, Terakita, Ishikawa, Tsukahara, Maeda and Shichida1997). Interneuron populations in the distributed nervous systems of non-bilaterians were probably small, functioning as low-level controllers in the sensorimotor control and guidance of simple behaviors. The potential for mid- and high-level motor controllers (and thus multi-stage motor hierarchies) (Fig. 3; Appendices A and B) would not appear until brains were formed and further differentiated.

As neurons differentiated into different neuron types, individual photoreceptors were likely being organized into small clusters to form the first multicellular, light-sensitive ‘organs’ but without much directional sensitivity. Classified by Nilsson (Reference Nilsson2009, Reference Nilsson2013) as Class I detectors (Table 2), these rudimentary organs support a few simple abilities, such as the ability to respond to an overhead predator because of the shadow it casts (shadow response), or the ability to move from light to dark areas and vice versa during diel vertical migrations. As a modern-day example, comb jellies (Ctenophora) relying on nondirectional photoreceptors can move from a well-lit area to a darker area, but they do so by increasing overall motor activity (Baiandina et al., Reference Baiandina, Kirin and Krivenko2022), presumably to increase the probability of eventually landing in a darker place.

4B. Transition 2: From non-bilateral to bilateral symmetry

The advent of bilateral symmetry was arguably one of the most consequential steps in metazoan evolution in terms of its impact on animal lifestyle, behavior and other body traits, including the nervous system (Nielsen, Reference Nielsen2008; Erwin, Reference Erwin2020; Manuel, Reference Manuel2009)(see Body traits in Appendix B). Bilateral symmetry resulted in two orthogonal axes of polarity and differentiation – anterior/posterior (head/trunk) and dorsal/ventral (back/belly). These, in turn, enabled three other parallel transitions: (1) a nervous system transition from brainless, distributed nerve nets to a central nervous system (CNS) with a bilaterally-symmetric brain in the head and nerve cord/s down the body, (2) a sensory transition from widely-distributed sensors (or sense cells) around the body circumference to clustered, typically bilaterally-paired sensors on the head and (3) a locomotor transition from unpolarized directions of locomotion with respect to the main body axis to headfirst movements aligned with the longitudinal body axis. With little evidence for bilateral animals much before the Cambrian (Degan et al, Reference Degan, Isozaki, Xingliang, Jian and Maruyama2014), bilateral symmetry and associated traits likely played a critical role in the rapid diversification of bilaterian lineages during the Cambrian explosion (see Section 3C).

There are many reasons for treating bilateral symmetry, along with associated traits, as foundational building blocks in the evolution of advanced cognitive abilities, as argued previously by others (e.g., Llinás, Reference Llinas and The2001; Godfrey-Smith, Reference Godfrey-Smith2020; Ginsburg and Jablonka, Reference Ginsburg and Jablonka2021; Barron et al., Reference Barron, Halina and Klein2023). First, a central brain becomes the ‘master controller’ (Barron et al., Reference Barron, Halina and Klein2023) of the body, enabling local control of individual body parts as well as coordinated control of multiple body parts during complex actions. Second, the clustering of sensors on the head (cephalization) sets the stage for animals being able to detect obstacles, habitats or other animals that are in line with their forward motion.

The coupling of forward-looking sensing abilities with polarized (headfirst) locomotion also sets the stage for the evolution of forecasting abilities – i.e., being able to detect things before they are encountered so that plans can be made, in advance, to e.g., engage, avoid or ignore what’s ahead. Bilateral symmetry also confers improved ability to change directions (Hollo and Novak, Reference Hollo and Novak2012). In essence, bilateral symmetry trumps asymmetry and other symmetries in terms of maneuverability, giving animals the enhanced capacity to execute pre-calculated motor plans to avoid or pursue distant targets.

4C.Transition 3: From limbless bodies to bodies with paired appendages

The evolution of muscle-powered, paired appendages gave animals at least three major advantages: (1) a faster mechanism of locomotion, (2) increased overall maneuverability during locomotion via braking and steering actions of appendages, and (3) a modifiable platform for the diversification of appendage functions (see Appendix B). Lobopods, the first paired appendages in euarthropods, date back to the earliest, now extinct Lobopodians (ca.540 mya) (Ortega-Hernández, Reference Ortega-Hernández2015), whereas pectoral and pelvic fins, the first paired appendages in vertebrates evolved with the first jawed vertebrates about a 100 million years later (Brazeau & Friedman, Reference Brazeau and Friedman2015).

At some point in time, multicellular animals must have transitioned from the two-dimensional world of the seafloor biomats during the Ediacaran to a three-dimensional world in the early Cambrian, when the first known pelagic eumatazoans emerged (Shixue et al., Reference Shixue, Steiner, Maoyan, Erdtmann, Huilin, Liangzhong and Weber2007; Vannier et al., Reference Vannier, Garcia-Bellido, Hu and Chen2009) (Fig. 7). A pelagic lifestyle would have placed increased demands on three-dimensional processing of sensory information from both interoceptors (e.g., body motion sensors) and exteroceptors (e.g., eyes). The ability to get off the benthos and maneuver mid-water was likely enabled or at least improved by the development of jet propulsion and appendicular forms of locomotion (see modes of locomotion under motor traits in Appendix B).

With increased locomotion speeds and maneuverability came upgrades in the visual abilities of animals to sense conditions at greater distances and in greater detail. The invagination of photoreceptors into lens-less cups on the skin surface and/or the addition of new structures, like lenses (albeit still unfocused), enabled animals to parse spatial differences in ambient light intensity and thus, to form crude, low-resolution images of large, nearby environmental features (Class III photoreceptors, Table 2). Elaborations of eyes and visual processing regions in the brain were likely accompanied by corresponding elaborations of descending visuomotor pathways and motor hierarchies in the brain (Butler, Reference Butler2000). These elaborations not only allowed animals to avoid obstacles when moving at moderate speeds (see also Section 4E), but also enabled more general forms of visually guided behaviors, including habitat selection (Nilsson Reference Nilsson2013, Reference Nilsson2022). Low-resolution vision also enabled a form of active visual sensing, unlike anything possible with non-imaging eyes, in which animals could control, by their own movements, the flow of incoming sensory information, thereby gaining additional time-dependent information about the features of the surrounding environment and the animal’s own relationship to them (Nilsson, Reference Nilsson2022).

Body movement also produces a form of visual reafference known as optic flow (Krapp, Reference Krapp, Bleckmann, Mogdans and Coombs2014), which greatly enhanced the available sensory information (Brembs, Reference Brembs2009; Egelhaaf et al., Reference Egelhaaf, Boeddeker, Kern, Kurtz and Lindemann2012; Heisenberg, Reference Heisenberg1994) and enabled animals to monitor self-motion (Nilsson, Reference Nilsson2009). Optic flow is the apparent motion of the visual surround when an animal moves relative to a stationary background. It is used as feedback control for closed-loop behaviors (Krapp, Reference Krapp, Bleckmann, Mogdans and Coombs2014; Zeil et al., Reference Zeil, Boeddeker and Hemmi2008) and plays an important role in many different behaviors, especially in flying and swimming organisms that move rapidly through a three-dimensional world. In flying insects, for example, optic flow aids in course correction after being blown off course, in collision avoidance when flying through a cluttered environment (Srinivasan, Reference Srinivasan1992; Srinivasan et al., Reference Srinivasan, Lehrer, Kirchner and Zhang1991, Reference Srinivasan, Zhang, Lehrer and Collett1996), and more generally, in the control of flight speed and various flight maneuvers (e.g., landings) (Linander et al., Reference Linander, Dacke and Baird2015; Mauss & Borst, Reference Mauss and Borst2020). Once sensory feedback is available for monitoring the consequences of an animal’s actions and memories of these consequences can be stored, the stage is arguably set for predicting the consequences of future actions.

Optic flow poses a problem, however, when flying or swimming animals are buffeted about or swept downwind/downstream by wind or water currents. Since the earliest bilaterians were exposed to shifting currents on the seafloor (Darroch et al., Reference Darroch, Rahman, Gibson, Racicot and Laflamme2017), they would have been confronted with the problem of distinguishing between optic flow caused by their own actions and that imposed by the water currents around them. The same is true for body motion detected by inertial sensors. Thus, the selective pressures for developing neural strategies, such as adaptive filters and motor efference copy (also known as motor corollary discharge), to distinguish between self- and exogenously imposed body movements would have evolved (see Appendix B: Brain traits). Neural solutions like these have been proposed as one of the first steps towards body self-awareness (Jékely et al., Reference Jékely, Godfrey-Smith and Keijzer2021; Montgomery & Bodznick, Reference Montgomery and Bodznick2016) and even consciousness (Merker, Reference Merker2005; Vallortigara, Reference Vallortigara2021).

Increased mobility also enhanced depth perception. In fact, many animals, even those with forward-looking eyes and binocular overlap, rely on motion cues for depth perception (see Section 2B). This would certainly have been the case for early euarthropod lineages during the Cambrian explosion, since the fossil evidence indicates that they had laterally placed, widely spaced eyes that were probably incapable of depth perception using binocular overlap. Depth perception cues include changes in the apparent size of an object as the distance between receiver and the object changes, and motion parallax cues in which the images of nearby objects translate faster across the field of view than those of far-away objects when an animal moves. Head bobbing to acquire depth perception cues, for example, is a behavior used by a variety of CC animals, including birds, praying mantises and octopuses (Kral, Reference Kral2003; Mather & Alupay, Reference Mather and Alupay2016; Nyakatura and Andrada, Reference Nyakatura and Andrada2014).

4D. Transition 4: The emergence of flexible limbs

The unjointed appendages (lobopods) of eaurthropod ancestors are thought to have transitioned to jointed legs some 510 – 570 mya (Shubin et al., Reference Pungor and Niell1997). Fossil forms of nectocaridids with body features similar to those of cephalopods, suggest that flexible tentacles in mollusks may have evolved around the same time (∼530 mya) (Smith, Reference Smith2013, Reference Smith2020), potentially from a mobile foot (a creeping pedal sole) from benthic snail-like ancestors (Shigeno et al., Reference Shigeno, Sasaki, Moritaki, Kasugai, Vecchione and Agata2008). However, the affinity of nectocaridids with cephalopods is uncertain, and modern coleoids with internalized shells for increased mobility (squid, octopus, cuttlefish) did not appear until some 100 years later in the mid-Palaeozoic (Kröger et al., Reference Kröger, Vinther and Fuchs2011; Pohle et al., Reference Pohle, Kröger, Warnock, King, Evans, Aubrechtová, Cichowolski, Fang and Klug2022). Add yet another 100 million years or so for the final step in the evolution of flexible vertebrate limbs - the so-called fins-to-limbs conversion during the water-to-land transition.

The evolution of flexible limbs is perhaps the most complicated in vertebrates because it involved several parallel tracts of independent evolution of median unpaired and lateral paired fins, each characterized by increasing degrees of complexity and differentiation (Coates, 1994; Reference Coates2003; Larouche et al., Reference Larouche, Zelditch and Cloutier2017; Stewart et al., Reference Stewart, Bonilla, Ho and Hale2019; Zhikun et al., Reference Zhikun, Qiang, Ferron, Keating, Junqing, Donoghue and Min2022). Lateral paired, pectoral and pelvic fins eventually gave rise to forelimbs and hindlimbs, respectively, and this ‘fins-to-limbs’ conversion just prior to the water-to-land transition (∼360 mya) was the last major step in the evolution of flexible tetrapod limbs (Coates et al., Reference Coates, Jeffery and Ruta2002). Tetrapod forelimbs differ from pectoral fins of primitive aquatic vertebrates in having mobile wrists, ankles and digits. However, the pectoral fin joints in Devonian tetrapod-like fish fossils indicate several degrees of freedom, including rotation, flexion, extension, protraction and retraction (Shubin et al., Reference Shubin, Daeschler and Jenkins2006). Thus, limb flexibility may have been well underway in paired fins before the water-to-land transition that resulted in tetrapod limbs.

The advent of flexible limbs, however or whenever it occurred, was accompanied by the corresponding development of neuromuscular (and in cephalopod mollusks, also hydrostatic) control systems, which together were a transformative innovation that contributed to the incredible success of Euarthropods (Shubin et al., Reference Pungor and Niell1997, Reference Shubin, Daeschler and Jenkins2006; Boxshall, Reference Boxshall2004, Esteve-Altava et al., Reference Esteve-Altava, Pierce, Molnar, Johnston, Diogo and Hutchinson2019) and cephalopod mollusks (Tarazona et al., Reference Tarazona, Lopez, Slota and Cohn2019). In euarthropods, the structurally and functionally diverse antennae, mouth parts, walking legs, swimmerets and grasping appendages were all derived from a basic jointed limb structure that defines the phylum (Shubin et al, Reference Pungor and Niell1997). Although the flexible limbs of cephalopod mollusks and vertebrates may not have diversified into as many structurally different appendages as those found in euarthropods, they were nevertheless multifunctional, being used in locomotion, communication, capturing prey, defensive maneuvers, tactile (and in some cases, chemo) sensing, and grasping, pushing, pulling and otherwise manipulating objects (Kennedy et al., Reference Kennedy, Buresch, Boinapally and Hanlon2020).

The vertebrate jaw, which has been likened to an ‘opposable thumb for your face’ (Jane Sheldon, as quoted in Godfrey-Smith, Reference Godfrey-Smith2020), is a different type of flexible body part capable of complex actions, analogous to the mouth parts that evolved from flexible forelimbs in euarthropods. This important vertebrate innovation, appearing before the final fins-to-limbs conversion, may be particularly useful to vertebrates like fish and birds with forelimbs (fins and wings) that have limited manipulative abilities. New Caledonian crows, for example, use their beaks to manufacture and manipulate tools (Table 1), while some parrots use theirs to grasp tree limbs in a novel form of arboreal locomotion coined ‘beakiation’ (Dickinson et al., Reference Dickinson, Young and Granatosky2024).

With the emergence of flexible limbs came a transition from low- to high-resolution eyes for better visuomotor control of limbs, although it is unclear whether high-resolution eyes or limb flexibility came first. Anamalocaris, a large, free-swimming apex predator (now extinct) dating back to ∼515 mya, had large compound eyes with high-spatial resolution (Paterson et al., Reference Paterson, Garcia-Bellido, Lee, Brock, Jago and Edgecombe2011) and jointed, frontal limbs with high flexibility for grasping prey (Briggs, Reference Briggs1994; De Vivo et al., Reference De Vivo, Lautenschlager and Vinther2021).

In any event, high-resolution eyes, which greatly increased information processing demands (Table 2, Section 2C), would have been accompanied by parallel expansion of brain processing areas, which include the optic lobes of invertebrates and the vertebrate retina. Simple, two or three-layered structures expanded to multi-layered structures with the number of interconnections increasing exponentially, as evidenced in the developing visual system of larval lamprey (Rubinson, Reference Rubinson1990; Suzuki and Grillner, Reference Suzuki and Grillner2018). The expansion to multiple layers would have been accompanied by changes in visuomotor pathways and behaviors, perhaps like those that allow the transition from passive filter feeding in larval lamprey to visually guided, active predation by adults (Rubinson, Reference Rubinson1990). Such an expansion leads to a massive increase in the computational power of these laminated visual regions (Section 2C), perhaps the single most important brain feature to develop on the road to complex cognition.

Low-resolution eyes made visually guided orientation to the environment possible via crude optic flow images. High-resolution vision, especially when coupled with flexible limbs with greater degrees of freedom and range of motion, dramatically increased the diversity, precision and complexity of spatiotemporally controlled, visually guided behaviors. Equipped with local computational networks that encoded features such as object edges and motion direction, animals could now extract salient information from their visual surroundings with greater efficiency (sec. 4.3). They could also better determine the visual consequences of their own movements relative to discrete objects or entities (animate or inanimate). Much of this would not have been possible without the parallel evolution of mobile eyes to stabilize visual images and prevent image blur as animals moved at higher speeds through their environment (Land, Reference Land2015, Reference Land2019). High-resolution vision, aided by both (a) eye mobility and (b) translational and rotational body-motion sensors needed to trigger compensatory movements of the eye (Section 2C), gave animals the newly evolved potential for detecting, discriminating, locating, manipulating or pursuing small targets (objects) of interest. As Nilsson (Reference Nilsson2022) put it, ‘object vision’ gave them a whole ‘new world’.

4E. The interaction of motor and sensory traits in the evolution of proactive behaviors

The transition from colliding into something (no advance planning) to proactively avoiding it (advance planning) likely spans several of the trait transitions covered here and depends heavily on the interaction of motor and sensory factors, as illustrated for active sensing by weakly electric fish (Snyder et al. (Reference Snyder, Nelson, Burdick and Maciver2007). The sensorimotor ability of animals to react to environmental features depends on both the sensing range (how far away features can be detected) and distance range of motion during a single locomotor cycle (e.g., the active burst/passive coast cycle of many swimming animals) before motion is halted by drag or friction (the so-called stopping distance). Although these can be measured as three-dimensional sensory and motor volumes (MacIver & Finlay, Reference MacIver and Finlay2022; Snyder et al., Reference Snyder, Nelson, Burdick and Maciver2007), the ramifications of overlapping and different-sized sensory and motor volumes can best be illustrated for the simplified case of a 2-dimensional, unidirectional motion pathway and sensory range. If the longest distance at which an animal can detect an oncoming feature or obstacle (the sensory distance, SD) is less than the motor stopping distance (MD), the animal will collide with the obstacle (SD<MD, collision mode, prior to Transition 2 in the motor tract of Fig. 5). If the sensory distance is equal to the stopping distance, then the animal can react, but only barely (SD=MD, reactive mode, between Transition 2 and 3 in Fig. 5). But if the sensory distance is greater than the stopping distance, then the animal can act proactively to e.g., put on the brakes and/or turn to avoid collision or encounter (SD>MD, proactive mode after Transition 3 in Fig. 5). In terms of visually guided behaviors, animals with Class I photodetectors are thus stuck in collision mode (unable to avoid obstacles, since they cannot form visual images of either near or far objects). For small animals moving at slow speeds, collisions are not a problem. The same cannot be said for the larger, faster animals that would appear later. This is because the collision force (= mass x acceleration) is much greater for them than for their smaller, slower counterparts. The short-range flow-sensing or electroreceptive abilities of weakly electric fish to detect obstacle-created distortions in their own, self-generated flow or electric field (Windsor, Reference Windsor, Bleckmann, Mogdans and Coombs2013; Snyder et al, Reference Snyder, Nelson, Burdick and Maciver2007) might be good examples of the intermediate reactive mode or a very short-range proactive mode, but long-range, proactive abilities would not have been possible until animals acquired high-resolution, image-forming eyes. Because the transmission of light in air is far greater than that in water, the visual range of animals that made the water to land transition much later in time (see Fig. 7) would become even greater and especially critical to the evolution of vertebrate brains (MacIver & Finlay, Reference MacIver and Finlay2022).

7A. Trait distributions

Figure 8 shows that trait distributions across lineages in each category range from narrow to broad, illustrating that our criteria for selecting traits were relatively even-handed in identifying a mix of traits that were not only shared with CC lineages but also non-CC lineages. Several other points are worth making about these distributions. One is that the wide distribution of many body traits (e.g., nerve tissue, striated muscle, bilateral symmetry, and cephalization/brains), is completely expected, as these are the primitive, defining characteristics of bilaterian body plans (Nielsen, Reference Nielsen2008; see Appendix B: Body traits). Likewise, mobility and polarized locomotion are also primitive characteristics of bilaterians, and these depend on foundational body traits, such as neurons, striated muscle and bilateral symmetry. Distal sensing abilities - olfaction, vision, audition and flow sensing are also primitively shared traits, as are multimodal sensing capabilities. The presence of a brain (no matter how rudimentary, see Appendix A) is also a shared characteristic of most bilaterians (a trait linked to bilateral symmetry), but brains that are structured and divided into recognizable parts with distinct functions (Fig. 3) are more rarely shared.

Figure 8.

The number of lineages that share a given trait in different trait categories: Body (A), sensory (B), brain (C), motor (D), and behavior/cognition (E). Note that the cephalization/CNS/brain trait in the body category could easily double as a brain trait (i.e., presence or absence of a brain), thus extending the maximum number of lineages that share a trait in the brain category from a non-majority (8) to a majority (11).

7B. Dichotomous trees for distinguishing CC from non-CC lineages

Another way to look at different trait contributions is from the perspective of a dichotomous key in which different lineages are peeled off from the main road to CC status by asking if a given trait is present or absent (Fig. 9). Focusing on body traits first, different lineages can be seen to exit the road to CC status via various off-ramps (Fig. 9A). Sponges (Porifera) take the first off-ramp due to lack of both striated muscle and a nervous system, both foundational traits, whereas animals like bristleworms (Annelida), water bears (Tardigrada) and velvet worms (Onychopora) take the last off-ramp due to the lack of flexible limbs, a pivotal trait. The dichotomous key also accounts for the uneven distribution of CC attributes within chordate and mollusan phyla, with non-vertebrate chordates and non-cephalopod mollusks taking the off-ramps associated with lack of paired appendages, an intermediate trait, or flexible appendages, an advanced trait. Keeping the same foundational body traits (striated muscle/nervous system, bilateral symmetry, and cephalization) at the bottom of the dichotomous key, but switching in different sensory (Fig. 9b), motor (Fig. 9c) or brain (Fig. 9d) traits at intermediate or pivotal levels produces similar results, suggesting that traits from different categories may be linked in a parallel fashion. Regardless of how one analyzes the distribution of traits, lineages with animals that possess all of the pivotal traits reach CC status, whereas those without any or with some, but not all, do not. Furthermore, complex cognitive (unlimited associative learning) and behavioral (central place foraging and construction behaviors) traits are, with but one exception (construction behavior in Annelida), found only in the three CC lineages, lending further support to the idea that the ‘big three’ are among the few lineages to exhibit complex cognition.

Figure 9.

Dichotomous keys showing various off-ramps from the main road to CC status. More narrowly distributed traits are at the top of the tree to show pivotal body (A), sensory (B), brain (C) or motor (D) traits. More widely distributed, foundational traits are towards the bottom and in some instances, intermediate traits are in the middle Note that Echinodermata is technically classified as a bilaterian phylum owing to the bilateral symmetry of their larval forms.

7C. Frequency of linked traits at different stages of complexity

The results of the trait-linkage analysis revealed that a majority (∼58%) of classes in the 17 lineages studied here had animals with body, visual and brain traits all at the same level of complexity (Table 3; see also supplemental materials, Complexity level.xls, for the ‘raw data’ and assignment of traits). In contrast, only 3% of classes had animals with all three traits at different levels of complexity. The remaining 39% had two out of three traits matched in complexity, most often with the third category just one level higher or lower than the two matched categories.

As might be expected, examples of matched traits at the lowest levels of complexity (0 and 1) came from non-bilaterian phyla - sponges (Porifera) and comb jellies (Ctenophora), respectively, with animals from two out of six classes of cnidarians (hydra and corals/anemones) also exhibiting level 1 complexity across the board. In contrast, matched traits at the highest level came from CC lineages, with 64% of euarthropod classes and 90% of vertebrate classes having matched, level 4 traits. Among mollusks, only 1 (Cephalopoda) out of 5 classes (20%) had matched, level 4 traits (Table 3).

Evidence of matched level 4 traits can also be also found in the fossil record. The extinct Anamalocaris (∼515 mya) has large complex eyes with high spatial resolution and jointed frontal appendages (Briggs, Reference Briggs1994; Paterson et al., Reference Paterson, Garcia-Bellido, Lee, Brock, Jago and Edgecombe2011). Similarly, complex brain features in the form of nested optic neuropils (laminated visual structures) can be found in fossil arthropods with jointed appendages from the Cambrian period (Xiaoya et al., Reference Xiaoya, Xianguang, Edgecombe and Stausfeld2012).

Classes with matched level 2 or 3 traits were found in non-CC lineages only, including two classes of non-cephalopod mollusks (Table 3). Fossil lobopodians, believed to be the ancestor to panarthropods (euarthropods, tardigrades and onychophorans) (Smith and Ortega-Hernández, Reference Smith and Ortega-Hernández2014), also exhibit matched level 3 characteristics in the form of paired (but unjointed) appendages and small simple (not large faceted) eyes similar to those found in their modern-day tardigradian and onychophoran descendants.

Among classes that exhibited sequential combinations of traits, level 2/3 combinations were the most frequent (11 classes in 5 phyla), followed by level 3/4 combinations (5 classes in 2 phyla) and lastly, one class only for 1/2 (Staurazoa, phylum Mollusca) and 0/1 (Myzozoa, phylum Cnidaria) combinations. All in all, the vast majority of classes sampled (86%) had evidence for either the same level of complexity across all three trait categories (37 classes) or two out of three at the same level with the third at a sequential level (18 classes), while the remaining 14% had a combination of traits that were either all different (2 classes) or two at the same level with the third at a non-sequential level (7 classes).

Mollusks were particularly instructive for the diversity of combinations represented (Table 3) and the correspondence between complexity level and combined body and motor traits. Clams (Bivalvia), for example, are at level 2 on the complexity scale and are without limbs and mostly immobile, whereas octopuses (Cephalopoda) have highly flexible limbs and are extremely mobile (complexity level 4). Pelagic nudibranchs (Gastropoda), some of which use paired, parapodial flaps to propel themselves through the water (Farmer, Reference Farmer1970) are arguably somewhere in between bivalves and cephalopods at complexity level 3.