Bioelectric networks: the cognitive glue enabling evolutionary scaling from physiology to mind

Each of us made the remarkable journey from mere matter to mind: starting life as a quiescent oocyte (“just chemistry and physics”), and slowly, gradually, becoming an adult human with complex metacognitive processes, hopes, and dreams. In addition, even though we feel ourselves to be a unified, single Self, distinct from the emergent dynamics of termite mounds and other swarms, the reality is that all intelligence is collective intelligence: each of us consists of a huge number of cells working together to generate a coherent cognitive being with goals, preferences, and memories that belong to the whole and not to its parts. Basal cognition is the quest to understand how Mind scales—how large numbers of competent subunits can work together to become intelligences that expand the scale of their possible goals. Crucially, the remarkable trick of turning homeostatic, cell-level physiological competencies into large-scale behavioral intelligences is not limited to the electrical dynamics of the brain. Evolution was using bioelectric signaling long before neurons and muscles appeared, to solve the problem of creating and repairing complex bodies. In this Perspective, I review the deep symmetry between the intelligence of developmental morphogenesis and that of classical behavior. I describe the highly conserved mechanisms that enable the collective intelligence of cells to implement regulative embryogenesis, regeneration, and cancer suppression. I sketch the story of an evolutionary pivot that repurposed the algorithms and cellular machinery that enable navigation of morphospace into the behavioral navigation of the 3D world which we so readily recognize as intelligence.

Understanding the bioelectric dynamics that underlie construction of complex bodies and brains provides an essential path to understanding the natural evolution, and bioengineered design, of diverse intelligences within and beyond the phylogenetic history of Earth.

All animals with advanced conventional cognitive capacities are the products of a very lengthy evolutionary process of accumulating gradual modifications of early microbial life. Even more remarkably, each of us personally took a journey in which matter acquired mind: we start life as a quiescent oocyte, and slowly remodel into an adult modern human with metacognitive, self-aware capacities and the ability to reason about entire counterfactual universes. How did the information processing capabilities of a single cell (Lyon 2006, 2015), with metabolic and physiological competencies, inflate to that of a human, who can pursue goals of planetary scale whose endpoint may be far after their expected lifespan (Fig. 1)? Taking evolution, and especially developmental biology, seriously means coming to grips with the continuous process of transformation that enables configurations of matter to express increasingly greater degrees and kinds of cognitive capabilities: decision-making, generalization, perception, causality detection, innate responses, learning, and advanced problem-solving. The emerging field of basal cognition (Baluška and Levin 2016; Levin et al. 2021; Lyon et al. 2021) strives to understand the evolutionary and embryological origins of our current behavioral capacities, and uncover the physical and computational dynamics by which simple minds emerge from chemistry and scale over time.Fig. 1Cognitive scaling. A Rat learning to press a lever to get a reward illustrates the principle of cognitive scaling: the rat is a collective of cells, some of which interact with the lever (skin of the paws) and some of which get the nutrient reward (intestinal cells).

However, no single cell has both experiences, and it’s the causal structure of the tissue network that enables credit assignment for behaviors, so that an associative memory between an action and a reward can form that belongs to the whole animal and none of its parts alone. B Simple homeostatic cycle (in this case, for pH control) indicative of the kinds of low-level goals that single cells can pursue. C Anatomical homeostasis, such as reliably regenerating toward a complete, correct salamander limb regardless of where it is cut, and stopping when that goal is achieved, is an example of large-scale (organ-level) collective behavior that emerges from the activity of a network of cells. D By merging together into computational networks, cells (neural or other) can increase the spatial scale of sensing, temporal memory and predictive power (sensing backward and forward in time, respectively). E Every agent has a “cognitive light cone” which demarcates the spatio-temporal scale of the goals toward which it can expend energy. This cone changes during evolution and during the lifetime of an animal from that of a single cell, with single-cell scale goals, to that of a morphogenetic system with organ-level goals and eventually a behavioral system with large-scale goals in 3D space. F This scaling of behavioral competencies and goal states in a given problem space can be implemented by gap junction (GJ) connections, which enable a connection of homeostats into a network that scales up the setpoint, measurement, and action steps of evolutionarily ancient homeostatic loops. All images created by Jeremy Guay of Peregrine Creative, used with permission. B-F taken with permission from (Levin 2022)

It is essential to expand the scope of this inquiry beyond traditional metrics of intelligence and behavior (Fig. 2). As humans, we are very good at recognizing intelligence of medium-sized objects moving at medium speeds in three-dimensional space: our sense organs face outward, and our capacities for detecting agency in this action space are strong (Mar et al. 2007; Repp and Knoblich 2007). However, assigning an IQ estimate to any being or object is, in effect, taking an IQ test ourselves: it is easy to miss intelligent behavior if one cannot recognize it. Imagine if we had a well-developed internal sense of our own body chemistry states—if we had the equivalent of a tongue that faced inwards into the bloodstream, it would then be natural for us to recognize all the remarkably clever things our liver and kidneys were doing with respect to managing us in the high-dimensional space of body physiology. Could recognizing diverse intelligence (in unfamiliar substrates) be essential to understanding the mechanisms and origins of conventional cognition and behavior?Fig. 2Multiscale competency architecture operates across problem spaces. A Biological systems are nested dolls in which molecular networks give rise to subcellular components, such as cytoskeletal networks, which give rise to cells, tissues, organs, organisms, and swarms. This multiscale architecture is not only structural, but rather functional: each level solves problems (with some degree of competency) in its own action space. Each level deforms the energy landscape for the levels below and above, influencing those components to do things they would otherwise not do. For example, when traversing the morphospace of different planarian head shapes, the bioelectric circuit controlling head morphogenesis alters the space of gene expression that is necessary to implement the specific anatomical outcomes.

B Familiar (conventional) behavior is navigation of a three-dimensional space, in which animal positions move to optimize certain reward functions. Although the human visual and cognitive repertoire is most accustomed to recognizing competency of typical (mid-sized, animal) agents navigating three-dimensional spaces, this deep concept integrates ideas across such wide-ranging fields as autonomous robotics (AI), evolutionary fitness landscapes, morphospaces, and cognitive behavioral studies. C There are in fact many problem spaces in which biological systems operate. Evolution progressively pivoted existing and novel mechanisms across problem spaces, to enable adaptive navigation toward specific goal states in metabolic, physiological, transcriptional, morphological, and ultimately behavioral space. All images created by Jeremy Guay of Peregrine Creative, except for the planaria image of panel A, which was created by Alexis Pietak. Used with permission; A,B taken with permission from (Levin 2022)

William James anticipated a cybernetic approach to this problem by defining intelligence as a degree of “the ability to reach the same goal by different means” (James 1890). This definition is not limited to brains or specific kinds of behaviors—it is intentionally agnostic about the composition of the agent. It challenges us to define a problem space, for any arbitrary system (including unconventional embodiments), and then formulate and test specific hypotheses about the competencies that the system can deploy in navigating that space. Traditional concepts of reflex responses, drives, memory, different kinds of learning, goal-directed activity, and even higher level capacities such as planning and creativity can all be defined in other problem spaces beyond the familiar 3D space, including metabolic, physiological, transcriptional, and anatomical spaces.

It has previously been proposed that the evolutionary path to conventional cognition involves evolutionary pivots across these problem spaces (Fields and Levin 2022). This suggests a broadening of the traditional systems to which these cognitive terms apply, and asking what is essential about them that is deeper than the specific frozen accidents that the course of terrestrial evolution has provided in its N = 1 history of life on Earth (Clawson and Levin 2022). Here, through the lens of a gradualist, evolutionary perspective I explore the possible origins of cognitive capacities and the extension of these concepts to diverse embodiments.

The goal is not to review the steps of evolution of electrically excitable networks (Arendt et al. 2016; Brunet and Arendt 2016; Jekely 2019; Jekely et al. 2015; Keijzer and Arnellos 2017; Keijzer 2017; Levin et al. 2021; Lyon et al. 2021). Rather, I aim to provide a novel perspective on how to understand animal behavior in the broader context of “control of agential activity” in all its diverse guises. A deep unification is within reach, if we can find and exploit invariants—symmetries between a variety of processes and biological systems that shed light on generic principles for scaling of cognition and collective intelligence. In fact, all intelligence is collective intelligence—not just termite mounds and beehives. We too are emergent beings supervening on a collection of cells which all were once independent unicellular organisms. How do they work together to enable the creation of a novel being with goals and memories that belong to it and not to any of its parts? The key question is not only the scale-up of a unitary cognitive capacity, but the many-into-one transition: the emergence of minds from collectives (Levin 2019, 2021b), working in spaces beyond those of their component cells.

At stake is not only a better understanding of our evolutionary history, but also insight into the relationship between genome and functional forms, and strategies for modifying, improving, and building novel synthetic cognitive agents (with implications ranging from exobiology to regenerative medicine). The over-arching framework (Levin 2022) is fundamentally grounded in the continuity hypothesis, using biophysical and evolutionary approaches to identify invariants and symmetries across natural evolved forms, as well as hybrid and fully synthetic life forms (and perhaps, some day, truly alien beings) (Fields and Levin 2022; Levin 2022). Here, I illustrate one segment of this framework’s roadmap by focusing on one highly instructive example: developmental bioelectricity as a precursor of brain-like processes, which reveals not only evolutionary pivots between two different problem spaces, but also shows a path to solving the problem of collective intelligence across scales of organization. In this discussion, I intentionally avoid issues of first-person experiential consciousness, focusing instead on third-person observable capacities for behavior broadly. I describe how computation via voltage states enabled coordinate navigation of anatomical morphospace before brains and muscles enabled us to navigate 3D space, as an example of how familiar behavioral science concepts can be generalized to gain insight into the origin and mechanisms of cognition. Here, cognition is not restricted to advanced capabilities such as planning, self-aware metacognition, language, etc., but is considered broadly in accordance with its continuous developmental and evolutionary origins. It serves as an umbrella term for all degrees of adaptive information processing and problem-solving (Levin 2022), no matter how advanced or primitive.

The focus on bioelectricity below, as a kind of tractable cognitive glue, is one facet of a broader emerging research program designed to help detect, understand, and relate to a wide range of natural, artificial, and hybrid intelligences regardless of size scale or material implementation (Abramson and Levin 2021; Calvo et al. 2020; Fields et al. 2021; Fields and Levin 2022; Ginsburg and Jablonka 2021; Gokhale et al. 2021; Kuchling et al. 2020; Levin 2021b, 2022; Lyon 2019; Ramstead et al. 2022; Smith-Ferguson and Beekman 2020; Timsit and Grégoire 2021; Watson et al. 2022).

To begin to broaden the inquiry into the evolutionary origin and mechanisms of conventional cognition, it is important to consider several biological examples that stretch standard assumptions (Fig. 3). First, consider Physarum: a unicellular slime mold, which has been used as a popular model for basal cognition (Beekman and Latty 2015; Nakagaki and Guy 2007; Reid et al. 2016; Saigusa et al. 2008; Vallverdú et al. 2018). Recent work has shown that this organism can exhibit learning following repeated experience, eventually becoming willing to cross areas of noxious chemicals to receive a reward (Boisseau et al. 2016; Boussard et al. 2019; Vogel and Dussutour 2016). Moreover, when placed in an arena containing distant inert glass objects of different mass distributions, it uses a biomechanical mechanism to process information about its environment and then reliably grow towards the heavier object (Murugan et al. 2021). These examples of learning and decision-making occur in the absence of a brain, neurons, or cellularization. From the perspective of evolutionary change across behavioral domains, one interesting thing is that for Physarum, its behavior is its morphological change. In this system, changing body shape to exploit the environment is how it implements behavior. Distinctions between morphological problems and behavioral problems are blurred by the biology, and the classical (Cartesian) conceptual distinctions between mind and body, which lead to separate communities for developmental biology and behavioral science, are now increasingly seen as artificial, for example, in the field of morphological computation in robotics (Bongard and Levin 2021).Fig. 3Unconventional agents: plasticity and robustness to change. A Tadpole of the frog Xenopus laevis can be made to have no primary eyes (white arrow), but instead have an ectopic eye on its tail (red arrow).

A’ These ectopic eyes (white arrow) can connect to the spinal cord (red arrow). B Using an automated behavioral training and testing apparatus, these animals can be shown to be able to see out of those eyes in a color vision training assay (B’) despite a novel visual system architecture that had no evolutionary adaptation—a remarkable example of functional plasticity despite wild-type genetics. C Planarian flatworms can be trained to associate laser-etched circular regions (bumps) in a petri dish surface with food. When their heads are amputated (C’), their behavior shows recall of the original information (place conditioning), showing the ability of memory to be stored outside the head and imprinted on newly produced brain tissue (showing how functional, behavioral memories are dynamically integrated with the tissue-level patterning processes that create specific shapes in anatomical morphospace). D Caterpillars (and other insect larvae) metamorphose into very different forms, which requires extensive disassembly and rebuilding of the brain. Despite this, their memories persist, showing that individual agents change during their lifetime not only due to experiences and learning, but also can radically change with respect to anatomical structure. Panel D created by Jeremy Guay of Peregrine Creative. Panels A–C used by permission from (Blackiston et al. 2010; Blackiston and Levin 2013; Levin 2022; Shomrat and Levin 2013); C’,D used by permission from (Levin 2022). (color figure online)

It has been proposed that the mechanisms of memory establishment are the same as those which sculpt brain tissue developmentally (Galván 2010; Kandel and O'Dell 1992). Planaria offer another example, where anatomical and behavioral information are tightly linked (Saló et al. 2009). These free-living flatworms, with a true centralized, bilaterian brain (Pagán 2014; Sarnat and Netsky 2002), have the ability to regenerate their bodies—every piece of a planarian gives rise to a properly patterned new worm (Sheiman and Kreshchenko 2015). However, this process not only implements an anatomical memory of body structure. Tails of worms trained on specific tasks regenerate into animals that show recall of the original information: the behavioral memories are apparently also imprinted onto the new brain as it forms (McConnell et al. 1959; Shomrat and Levin 2013). Thus, the information processing required to restore a specific body shape and that required to propagate the results of past experience despite cellular turnover and maintenance are tightly linked.

Metamorphosis tells a similar story. Caterpillars must become moths or butterflies, requiring turning a controller that operates a soft body in a two-dimensional lifestyle into one that operates a hard body in a three-dimensional world. The brain is largely dismantled and rebuilt in a new configuration, but learned information persists (Alloway 1972; Blackiston et al. 2008; Sheiman and Tiras 1996). Such dynamic plasticity is not just for invertebrates. Tadpoles of the frog Xenopus laevis can be produced with no primary eyes, but one eye on their tails; these animals can see out of an ectopic eye, which can connect to the spinal cord (Blackiston and Levin 2013). In one generation, requiring no evolutionary adaptation, the cellular hardware of a frog embryo can adapt to this radical reconfiguration of its visual sensory system.

Thus, radical changes of behavioral repertoires occur not only on evolutionary time-scales, but also at the level of an individual being—paralleling embryogenesis in raising profound questions about the transformation of agents by rearrangements of their parts—a key aspect of recognizing even traditional animals as fundamentally collective intelligences. A key aspect of understanding both basal and traditional cognition is formulating paradigms for predicting the properties and capabilities of emergent Selves from those of the components that comprise them.

This is fundamentally a story of the multi-scale competency architecture that biology employs. Animals are nested dolls made up of cells and tissues but this arrangement is not merely structural; cells were once unicellular organisms and have many competencies in their own problem spaces [an agential material which evolution molds by behavior-shaping as much as changes in their hardware (Davies and Levin 2023)]. When one “trains a rat” to obtain a treat by pressing a lever, the cells that interact with the lever (skin) are not the same cells that obtain the metabolic reward (intestine). No single cell had both experiences, and the owner of the associative memory linking those two events is an emergent collective intelligence (Fig. 1A). Likewise, even gene regulatory networks can learn relationships between their experiences (such as Pavlovian conditioning) as a collective, by virtue of the activity of very simple components (transcriptional elements) (Biswas et al. 2021; Watson et al. 2010). Conversely, the voluntary act of raising one’s arm reveals the functional connection between the highest levels of executive function to the depolarization of muscle cells—information crossing levels from that of human thoughts to the physiological status of individual cells. What is the coordination mechanism that enables this cross-level integrated information processing? It is developmental bioelectricity.

The functional properties of brains and nervous systems emerge from a network architecture in which neurons (and other cells like glia) communicate via changes in cellular resting potential (Vmem) and the resulting movement of neurotransmitter signals. Vmem, the bioelectric state of each nervous system component, is determined by an integrated balance of charges via the action of ion channels and pumps that enable the segregation of potassium, sodium, chloride, and protons (Fig. 4). Bioelectric states can propagate between cells via electrochemical synapses known as gap junctions (GJs) (Mathews and Levin 2017; Palacios-Prado and Bukauskas 2009). GJs form direct connections between the cellular internal milieus that enable the transfer of current and small chemical messengers. In addition, bioelectric states can also induce the movement of neurotransmitter molecules, such as serotonin, both through the extracellular space (via transporters and vesicles) and directly through gap junctions (Levin et al. 2006; Romero-Reyes et al. 2021). Crucially, every component in this network is functionally regulated by every other, and by its own action: ion channels and gap junctions are themselves often voltage-sensitive (Brink 2000; Palacios-Prado and Bukauskas 2009), while neurotransmitters move under electromotive force and regulation, and in turn regulate ion channel properties.Fig. 4Developmental and neural bioelectricity: a deep symmetry. A Familiar hardware of neurons consists of ion channels in the membrane which set voltage state, and electrical synapses (gap junctions) which enable those states to selectively propagate through the network. This enables a kind of software phenomenon—physiological events that process information and guide behavior.

The effort of neural decoding is the practical implication of the idea that all of the agent’s memories, plans, preferences, behavioral repertoires, etc. are instantiated in that bioelectrical layer of control and will be able to be read out (interpreted) once we understand the encoding. B Exact same architecture is used throughout the body, and forms the evolutionary precursor of the behavioral control system. All cells have ion channels, and most cells couple via regulated gap junctions to their neighbors, enabling the bioelectric physiology that guides growth and form during morphogenesis. Consistent with the evolutionary pivot model, these electrical networks also process information to enable navigation: prior to navigating 3D space by controlling muscle action (when brains appeared), this system was used to process information and make decisions, while bodies navigated anatomical morphospace during embryogenesis, regeneration, and cancer suppression. C This isomorphism between somatic and neural bioelectricity is what enables all of the tools of neuroscience to be used outside of the brain. Developmental bioelectricity is studied by voltage imaging dyes, and functional techniques such as genetic, chemical, and optical ways of regulating ion channels and gap junctions in vivo. The tools (and many concepts) are broadly compatible across tissue types, enabling the insights of behavioral neuroscience to be portable toward understanding outcomes in other problem spaces. All images created by Jeremy Guay of Peregrine Creative and used with permission; A,B taken with permission from (Levin 2022)

This extremely rich set of feedback loops establishes computational capacity; for example, ion channels and GJs, as voltage-gated current conductances, are in effect transistors and possess a fundamental property of historicity (memory in which past events impact current signaling state). These events do eventually impact other kinds of pathways (such as gene expression), but it is critical that the information processing in such networks is essentially physiological—the rapid propagation of signals via action potentials and slow waves across the network does not itself require transcriptional change. As a corollary, the information content of this network cannot be read out at the transcriptional or even proteomic level: channels open and close post-translationally, and the same channels can give rise to different voltage states depending on cells’ history, while diverse channels can give rise to the exact same voltage map.

There is no one-to-one mapping between the molecular state and the bioelectrical state, making it essential to study such systems in the living condition (unlike genetic and protein-level information, which can be studied in fractionated or fixed material, bioelectrical information disappears at cellular death). One implication of this feature is a critical separation of hardware and software (Boone and Piccinini 2016). Of course, the ability to form stable behavioral repertoires based on specific kinds of past experience, and the way in which it coarse-grains and generalizes from a sensory stream, are shaped by the network’s structure and physical properties. However, structurally identical networks can have learned different things after physiological experiences—bioelectric networks’ historicity means that their information content is not hardwired by their genetic specification but is dependent on past experience. One cannot know the informational content of a brain merely from knowing its neural layout and genome: the exact same brain can contain numerous different memories, goals, etc. This decoupling of the material state (protein content) from the information content is the first step to the most amazing aspect of neural networks: they enable mind to arise from matter. Specifically, neural networks are the functional layer in which physiology transitions to meaning: electrophysiological events encode memories, plans, preferences, behavioral capacities, and a first-person perspective—the content and shape of a cognitive Self, at whatever level of sophistication.

Neuroscience [via the research program of neural decoding (Huth et al. 2016; Naselaris et al. 2009; Nishimoto et al. 2011)] is committed to the idea that all of the content of minds, from the most primitive to the most complex, can be read out from (and thus resides in) the electrophysiological state of the network. Thus, just as inorganic electric circuits and transistors enable the jump from the physics of Ohm’s law to the truth tables of logic gates (basic elements of formal thought), biological electric circuits and ion channels/GJs enable the jump from chemistry to embodied meaning. A crucial aspect of this is that it provides a scaling up of agency: the homeostatic competencies of single neurons are integrated into a network that supports an emergent, higher order Self with memories, preferences, and other features that belong to itself but not to any of its components individually. Bioelectric networks, with their inherent plasticity, multiscale historicity, and learning capacity, are an ideal kind of “cognitive glue” that binds the primitive goal-directedness (in the cybernetic sense) of single cells into a higher order system with a larger cognitive light cone (Fig. 1E). However, perhaps, the most fascinating and far-reaching aspect is that the key features that enable the supervenience of active information on a material substrate are not unique to brains and nervous systems at all (Fig. 4).

The use of bioelectric networks to enable coherent computation to occur via a spatially and temporally distributed living medium is ancient, both in its molecular components (which date back to our unicellular ancestors) and in the algorithms by which it provides adaptive function. Even bacterial biofilms use electrical networks to synchronize activity within the proto-body of the colony (Martinez-Corral et al. 2019; Prindle et al. 2015; Yang et al. 2020). Evolution discovered long ago that networks made up of ion channels, gap junctions, and neurotransmitters as transduction machinery for electric circuit function provide a remarkably powerful and flexible way to process information. The current use of these components to implement behaviors in 3D space by controlling muscle activity represents an evolutionary pivot: their original usage was to navigate anatomical morphospace (Fields et al. 2020) (that is, to control all cell behaviors toward specific morphogenetic outcomes). This conservation of both molecular mechanisms and algorithms is the reason that the workhorse tools of neuroscience work everywhere in the body—they do not distinguish between neuronal and non-neuronal uses.

Clear morphogenetic analogs (Pezzulo and Levin 2015) exist for optogenetics, ion channel mutants, neurotransmitter drugs—the tools of neuroscience—as well as for neuroscience concepts, such as memory, representation, navigation, perceptual bistability, and many others (Table 1).Table 1Conceptual mapping between behavioral cognition and anatomical regulationCognitionMorphogenesisAction potential movement within an axonDifferential patterns of Vmem across single cells’ surfaceLocal field potential (EEG)Vmem distribution of cell groupIntrinsic plasticityChange of ion channel expression based on Vmem levelsSynaptic plasticityChange of cell:cell connectivity via Vmem’s regulation of gap junctional connectivityActivity-dependent transcriptional changesBioelectric signals’ regulating gene expression during patterningNeuromodulationDevelopmental (pre-nervous) signaling via neurotransmitters such as serotonin moving under control of bioelectrical gradientsDirect transmissionCell:cell sharing of voltage via nanotubes or gap junctionsVolume transmissionCell:cell communication via ion levels outside the membrane or voltage-dependent neurotransmitter releaseSynaptic vesiclesExosomesSensitizationCells become sensitized to BMP antagonists to stabilize neurogenesisFunctional lateralizationLeft–right asymmetry of body organsTaste and olfactory perceptionMorphogenetic signaling by diffusible biochemical ligandsActivity-dependent modification of CNSControl of anatomy by bioelectric signaling within those same cellsCritical plasticity periodsCompetency windows for developmental induction eventsAutonomic reflexesWound healingVoluntary movementRemodeling, regeneration, metamorphosisMemoryShorter term: regeneration of specific body organs.

Longer term: morphological homeostasis over decades as individual cells senesce; altering basic body anatomy in planaria by direct manipulation of bioelectric circuitGeneralization via multi-layer neural networksGeneralization by bow-tie architectures of signaling pathwaysPattern completion ability of neural networks (e.g., attractor nets)Regeneration of missing parts in partial fragments (e.g., planaria)ForgettingCancer, loss of regenerative abilityAddictionLimb becomes unable to regenerate without nerve once exposed to nerveEncodingRepresentation of patterning goal states by bioelectric properties of tissuePerceptual bi-stabilityStochastic flipping between two target morphologies by planarian fragments in the cryptic stateMcGurk effectModification of interpretation of biochemical signals by bioelectric state of cellsVisual system feature detectionOrgan-level decision-making during morphogenesisMirror neurons and mirror focus in epilepsyContralateral bioelectric signals mirroring sites of amputation damage in frog legsHolographic (distributed) storageAny small piece of a planarian remembers the correct pattern (even if it has been re-written)InstinctHardwired patterning programs (mosaic development)Behavioral plasticityRegulative developmental programs and regenerative capacitySelf-modelingSurveillance of anatomical state by brainGoal-seekingEmbryogenesis and regeneration work towards a specific target configuration despite perturbationsSub-goaling in problem solving tasksDevelopmental modularityAdaptivity and intelligenceMorphological rearrangements carrying out novel, not hardwired, movements to reach the same anatomical configuration despite unpredictable initial starting stateTabula rasaCells could be a (semi) universal constructor, able to build any shape that can be specified via the pattern memory codeAttention and context salienceTissues responding to specifi

c bioelectric states only when relevant (e.g., when in damage state)Age-dependent cognitive declineAge-dependent loss of regenerative abilityOptogenetic insertion of false memoriesOptogenetic induction of regeneration or ectopic organsReading of semantic content from brain scansDetecting differences in target morphology from fluorescent voltage dye dataExecutive control (free action) filtering down to regulate muscle motionLarge-scale circuit decisions that dictate whole bodyplan axial patterning transduced to individual cell gene expression changes as needed to implement large-scale phenotypeConceptual mapping of ideas and phenomena between neuro-behavioral sciences to morphogenesis (listed in rough order of ascending levels of organization). These are discussed in more detail elsewhere (Pezzulo and Levin 2015, 2016).

The bioelectric system is so versatile that it was readily exapted for behavior when nerve and muscle evolved (Keijzer et al. 2013), with two major changes: a significant speed-up (milliseconds, instead of hours, as the primary time scale) and a focus on temporal signaling (spiking patterns) for behavior instead of development’s reliance on spatial bioelectric patterns across tissues. Despite these differences, profound symmetries between the problems of morphogenesis and the problems of cognition exist; remarkably, while only recently explored in detail, this idea was already obvious to Alan Turing as early as 1952 (Turing 1952) and to others in the following decades (Grossberg 1978). Much as electrophysiological neural events are the fundamental currency that underlies the emergence of a coherent behavioral Self with some degree of cognitive activity, their slower, ancient ancestor mechanisms enable individual cells to cooperate toward coherent anatomical goals and deploy problem-solving capabilities in morphospace that belong to the “embryo”—an emergent collective.

The deep conceptual and mechanistic parallels between behavior and morphogenesis (Grossberg 1978) suggest a research program that can catalyze novel progress in the life sciences: accessing multiple levels of anatomical control (Figs. 5,6). To truly understand the origin of cognition, and conversely, to deploy the insights of neuroscience for regenerative medicine and bioengineering, it is important to investigate the proto-cognitive capacities of morphogenesis, specifically, the range of behaviors and capabilities that the collective intelligence of body cells can deploy toward adaptive behaviors in anatomical morphospace (Figs. 7, 8). These range from purely emergent, hardwired morphogenetic cascades (corresponding to fixed, inborn instinctual behaviors) and complex, flexible ability to reach the correct target morphology despite novel circumstances (corresponding to a degree of intelligent problem-solving behavior). It is critical to recognize and study the ways in which morphogenesis is not simply an emergent result of local rules but has complex, multi-scale feedback mechanisms that implement a degree of goal-directed activity and problem-solving in a way that was not directly encoded by genome (a hallmark of cognitive mechanisms).Fig. 5Examples of non-neural bioelectricity guiding behavior in morphospace. A Much as brain imaging allows the reading of bioelectrical states in tissues to decode the properties of control circuits and behavioral competencies, voltage reporter dyes enable in vivo tracking of the information processing in morphogenetic decision-making. Here is shown one frame from a timelapse video of a frog embryo prior to formation of the face, showing a prepattern of resting potential states that demarcates the position of the future gene expression domains and craniofacial organs, such as the eyes, mouth, and lateral structures.

In contrast to this endogenous pattern, pathological patterns (such as those leading to tumors in A’) can be induced via, for example, oncogene injection. The location of tumors (A”, red arrowhead) can be predicted by the bioelectric dye signal which shows an aberrant electrical signature of cells that are disconnecting from the tissue-level network and reverting back to unicellular-scale behavior (i.e., metastasis and over-proliferation). B These bioelectric prepatterns are known to be instructive, because reproducing them elsewhere by misexpression of specific ion-channel mRNA, such as in this frog embryo, results in induction of whole organs, such as eyes (red arrow), which have the necessary internal tissue structure (immunohistochemistry in B’). This demonstrates a key aspect of behavior—binding complex downstream actions to a simple (low information-content) trigger. Moreover, this phenomenon exhibits the competency of recruitment (B”): when only a few cells are injected with the channel (cyan b-galactosidase marker label), they autonomously recruit normal neighbors to complete their morphological goal: building a normal-sized ectopic lens (brown tissue). C Method of inducing whole organs (control of large-scale movements in anatomical morphospace) by modulating bioelectric states of the tissue circuit (i.e., incepting false memories into the network) can also, for example, produce ectopic forebrain (red arrow), or ectopic limbs (C’, C”, red arrows). Panels reused with permission from (Chernet and Levin 2013a; Levin 2009; Levin et al. 2017; Pai et al. 2012; Vandenberg et al. 2011). (color figure online)Fig. 6Isomorphism between neuroscience, morphogenesis, and artificial intelligence. A Multiscale nature of information-processing architectures are shown, with the left column (neuroscience), middle column (morphogenesis), and right column (computer engineering).

In each case (going down from top row), high-level phenomena such as behavior and morphological competencies are mediated by real-time physiological dynamics that enable decision-making and learning, which operate on electrical networks that perform computations, which in turn are enabled by molecular machinery that implements voltage-gated current conductors (ion channels and gap junctions in the case of cells, and transistors in the case of digital computers). B In behavioral science, the full panoply of phenomena requires understanding of molecular synaptic and ion channel machinery, circuit functions, network capabilities, behavioral repertoires, and eventually executive-level goal-directed activity. Similarly, in developmental bioelectricity, the field aims for a “full stack” integration of transcriptional networks that drive ion channel expression, to tissue-level voltage dynamics, to organ-level decisions about size and shape, ultimately to high-level algorithms that control the axial bodyplan and organ layout. Images by Alexis Pietak, used by permissionFig. 7Morphogenesis as competent behavioral navigation. A Animals begin life as a single cell and then a ball of embryonic blastomeres, which eventually gives rise to the incredible complexity of the body (cross section of the human torso is shown here). B Viewing anatomical structure as a morphospace of parameters describing possible configurations (here simplified to 2 dimensions, principal components (PC) 1 and 2), one can imagine this process as a hardwired transition from a starting state (S1) to an ensemble of states corresponding to viable adult organisms (goal states G). C However, when probed by perturbation (as any novel animal’s behavior is studied), traversals of morphospace are revealed to exhibit considerable flexibility. Here is shown an example of metamorphosis in the frog Xenopus laevis.