Efforts to use regenerative medicine—which seeks to address ailments as diverse as birth defects, traumatic injury, aging, degenerative disease, and the disorganized growth of cancer—would be greatly aided by solving one fundamental puzzle: How do cellular collectives orchestrate the building of complex, three-dimensional structures?
While genomes predictably encode the proteins present in cells, a simple molecular parts list does not tell us enough about the anatomical layout or regenerative potential of the body that the cells will work to construct. Genomes are not a blueprint for anatomy, and genome editing is fundamentally limited by the fact that it’s very hard to infer which genes to tweak, and how, to achieve desired complex anatomical outcomes. Similarly, stem cells generate the building blocks of organs, but the ability to organize specific cell types into a working human hand or eye has been and will be beyond the grasp of direct manipulation for a very long time.
But researchers working in the fields of synthetic morphology and regenerative biophysics are beginning to understand the rules governing the plasticity of organ growth and repair. Rather than micromanaging tasks that are too complex to implement directly at the cellular or molecular level, what if we solved the mystery of how groups of cells cooperate to construct specific multicellular bodies during embryogenesis and regeneration? Perhaps then we could figure out how to motivate cell collectives to build whatever anatomical features we want.
New approaches now allow us to target the processes that implement anatomical decision-making without genetic engineering. In January, using such tools, crafted in my lab at Tufts University’s Allen Discovery Center and by computer scientists in Josh Bongard’s lab at the University of Vermont, we were able to create novel living machines, artificial bodies with morphologies and behaviors completely different from the default anatomy of the frog species (Xenopus laevis) whose cells we used. These cells rebooted their multicellularity into a new form, without genomic changes. This represents an extremely exciting sandbox in which bioengineers can play, with the aim of decoding the logic of anatomical and behavioral control, as well as understanding the plasticity of cells and the relationship of genomes to anatomies.
Deciphering how an organism puts itself together is truly an interdisciplinary undertaking.
Deciphering how an organism puts itself together is truly an interdisciplinary undertaking. Resolving the whole picture will involve understanding not only the mechanisms by which cells operate, but also elucidating the computations that cells and groups of cells carry out to orchestrate tissue and organ construction on a whole-body scale. The next generation of advances in this area of research will emerge from the flow of ideas between computer scientists and biologists. Unlocking the full potential of regenerative medicine will require biology to take the journey computer science has already taken, from focusing on the hardware—the proteins and biochemical pathways that carry out cellular operations—to the physiological software that enables networks of cells to acquire, store, and act on information about organ and indeed whole-body geometry.
In the computer world, this transition from rewiring hardware to reprogramming the information flow by changing the inputs gave rise to the information technology revolution. This shift of perspective could transform biology, allowing scientists to achieve the still-futuristic visions of regenerative medicine. An understanding of how independent, competent agents such as cells cooperate and compete toward robust outcomes, despite noise and changing environmental conditions, would also inform engineering. Swarm robotics, Internet of Things, and even the development of general artificial intelligence will all be enriched by the ability to read out and set the anatomical states toward which cell collectives build, because they share a fundamental underlying problem: how to control the emergent outcomes of systems composed of many interacting units or individuals.
(Re)Building a body
Many types of embryos can regenerate entirely if cut in half, and some species are proficient regenerators as adults. Axolotls (Ambystoma mexicanum) regenerate their limbs, eyes, spinal cords, jaws, and portions of the brain throughout life. Planarian flatworms (class Turbellaria), meanwhile, can regrow absolutely any part of their body; when the animal is cut into pieces, each piece knows exactly what’s missing and regenerates to be a perfect, tiny worm.
The remarkable thing is not simply that growth begins after wounding and that various cell types are generated, but that these bodies will grow and remodel until a correct anatomy is complete, and then they stop. How does the system identify the correct target morphology, orchestrate individual cell behaviors to get there, and determine when the job is done? How does it communicate this information to control underlying cell activities?
Several years ago, my lab found that Xenopus tadpoles with their facial organs experimentally mixed up into incorrect positions still have largely normal faces once they’ve matured, as the organs move and remodel through unnatural paths. Last year, a colleague at Tufts came to a similar conclusion: the Xenopus genome does not encode a hardwired set of instructions for the movements of different organs during metamorphosis from tadpole to frog, but rather encodes molecular hardware that executes a kind of “error minimization loop,” comparing the current anatomy to the target frog morphology and working to progressively reduce the difference between them. Once a rough spatial specification of the layout is achieved, that triggers the cessation of further remodeling.
The deep puzzle of how competent agents such as cells work together to pursue goals such as building, remodeling, or repairing a complex organ to a predetermined spec is well illustrated by planaria. Despite having a mechanistic understanding of stem cell specification pathways and axial chemical gradients, scientists really don’t know what determines the intricate shape and structure of the flatworm’s head. It is also unknown how planaria perfectly regenerate the same anatomy, even as their genomes have accrued mutations over eons of somatic inheritance. Because some species of planaria reproduce by fission and regeneration, any mutation that doesn’t kill the neoblast—the adult stem cell that gives rise to cells that regenerate new tissue—is propagated to the next generation. The worm’s incredibly messy genome shows evidence of this process, and cells in an individual planarian can have different numbers of chromosomes. Still, fragmented planaria regenerate their body shape with nearly 100 percent anatomical fidelity.
of the encoded target morphology without genomic editing reveals a new kind of epigenetics.
So how do cell groups encode the patterns they build, and how do they know to stop once a target anatomy is achieved? What would happen, for example, if neoblasts from a planarian species with a flat head were transplanted into a worm of a species with a round or triangular head that had the head amputated? Which shape would result from this heterogeneous mixture? To date, none of the high-resolution molecular genetic studies of planaria give any prediction for the results of this experiment, because so far they have all focused on the cellular hardware, not on the logic of the software—implemented by chemical, mechanical, and electrical signaling among cells—that controls large-scale outcomes and enables remodeling to stop when a specific morphology has been achieved.
Understanding how cells and tissues make real-time anatomical decisions is central not only to achieving regenerative outcomes too complex for us to manage directly, but also to solving problems such as cancer. While the view of cancer as a genetic disorder still largely drives clinical approaches, recent literature supports a view of cancer as cells simply not being able to receive the physiological signals that maintain the normally tight controls of anatomical homeostasis. Cut off from these patterning cues, individual cells revert to their ancient unicellular lifestyle and treat the rest of the body as external environment, often to ruinous effect. If we understand the mechanisms that scale single-cell homeostatic setpoints into tissue- and organ-level anatomical goal states and the conditions under which the anatomical error reduction control loop breaks down, we may be able to provide stimuli to gain control of rogue cancer cells without either gene therapy or chemotherapy.
During morphogenesis, cells cooperate to reliably build anatomical structures. Many living systems remodel and regenerate tissues or organs despite considerable damage—that is, they progressively reduce deviations from specific target morphologies, and halt growth and remodeling when those morphologies are achieved. Evolution exploits three modalities to achieve such anatomical homeostasis: biochemical gradients, bioelectric circuits, and biophysical forces. These interact to enable the same large-scale form to arise despite significant perturbations.
© N.R. FULLER, SAYO-ART, LLC
The best-known modality concerns diffusible intracellular and extracellular signaling molecules. Gene-regulatory circuits and gradients of biochemicals control cell proliferation, differentiation, and migration.
The movement of ions across cell membranes, especially via voltage-gated ion channels and gap junctions, can establish bioelectric circuits that control large-scale resting potential patterns within and among groups of cells. These bioelectric patterns implement long-range coordination, feedback, and memory dynamics across cell fields. They underlie modular morphogenetic decision-making about organ shape and spatial layout by regulating the dynamic redistribution of morphogens and the expression of genes.
Cytoskeletal, adhesion, and motor proteins inside and between cells generate physical forces that in turn control cell behavior. These forces result in large-scale strain fields, which enable cell sheets to move and deform as a coherent unit, and thus execute the folds and bends that shape complex organs.
Bioelectrical software: Beyond the brain
The software of life, which exploits the laws of physics and computation, is enabled by chemical, mechanical, and electrical signaling across cellular networks. While the chemical and mechanical mechanisms of morphogenesis have long been appreciated by molecular and cell biologists, the role of electrical signaling has largely been overlooked. But the same reprogrammability of neural circuits in the brain that supports learning, memory, and behavioral plasticity applies to all cells, not just neurons. Indeed, bacterial colonies can communicate via ionic currents, with recent research revealing brain-like dynamics in which information is propagated across and stored in a kind of proto-body formed by bacterial biofilms. So it should really come as no surprise that bioelectric signaling is a highly tractable component of morphological outcomes in multicellular organisms.
A few years ago, we studied the electrical dynamics that normally set the size and borders of the nascent Xenopus brain, and built a computer model of this process to shed light on how a range of various brain defects arise from disruptions to this bioelectric signaling. Our model suggested that specific modifications with mRNA or small molecules could restore the endogenous bioelectric patterns back to their correct layout. By using our computational platform to select drugs to open existing ion channels in nascent neural tissue or even a remote body tissue, we were able to prevent and even reverse brain defects caused not only by chemical teratogens—compounds that disrupt embryonic development—but by mutations in key neurogenesis genes.
Similarly, we used optogenetics to stimulate electrical activity in various somatic cell types to trigger regeneration of an entire tadpole tail—an appendage with spinal cord, muscle, and peripheral innervation—and to normalize the behavior of cancer cells in tadpoles strongly expressing human oncogenes such as KRAS mutations. We used a similar approach to trigger posterior regions, such as the gut, to build an entire frog eye. In both the eye and tail cases, the information on how exactly to build these complex structures, and where all the cells should go, did not have to be specified by the experimenter; rather, they arose from the cells themselves. Such findings reveal how ion channel mutations result in numerous human developmental channelopathies, and provide a roadmap for how they may be treated by altering the bioelectric map that tells cells what to build.
We also recently found a striking example of such reprogrammable bioelectrical software in control of regeneration in planaria. In 2011, we discovered that an endogenous electric circuit establishes a pattern of depolarization and hyperpolarization in planarian fragments that regulate the orientation of the anterior-posterior axis to be rebuilt. Last year, we discovered that this circuit controls the gene expression needed to build a head or tail within six hours of amputation, and by using molecules that make cell membranes permeable to certain ions to depolarize or hyperpolarize cells, we induced fragments of such worms to give rise to a symmetrical two-headed form, despite their wildtype genomes. Even more shockingly, the worms continued to generate two-headed progeny in additional rounds of cutting with no further manipulation. In further experiments, we demonstrated that briefly reducing gap junction-mediated connectivity between adjacent cells in the bioelectric network that guides regeneration led worms to regenerate head and brain shapes appropriate to other worm species whose lineages split more than 100 million years ago.
My group has developed the use of voltage-sensitive dyes to visualize the bioelectric pattern memory that guides gene expression and cell behavior toward morphogenetic outcomes. Meanwhile, my Allen Center colleagues are using synthetic artificial electric tissues made of human cells and computer models of ion channel activity to understand how electrical dynamics across groups of non-neural cells can set up the voltage patterns that control downstream gene expression, distribution of morphogen molecules, and cell behaviors to orchestrate morphogenesis.
The emerging picture in this field is that anatomical software is highly modular—a key property that computer scientists exploit as subroutines and that most likely contributes in large part to biological evolvability and evolutionary plasticity. A simple bioelectric state, whether produced endogenously during development or induced by an experimenter, triggers very complex redistributions of morphogens and gene expression cascades that are needed to build various anatomies. The information stored in the body’s bioelectric circuits can be permanently rewritten once we understand the dynamics of the biophysical circuits that make the critical morphological decisions. This permanent editing of the encoded target morphology without genomic editing reveals a new kind of epigenetics, information that is stored in a medium other than DNA sequences and chromatin.
Rewriting the planarian body plan
Recent work from our group and others has demonstrated that anatomical pattern memories can be rewritten by physiological stimuli and maintained indefinitely without genomic editing. For example, the bioelectric circuit that normally determines head number and location in regenerating planaria can be triggered by brief alterations of ion channel or gap junction activity to alter the animal’s body plan. Due to the circuit’s pattern memory, the animals remain in this altered state indefinitely without further stimulation, despite their wildtype genomes. In other words, the pattern to which the cells build after damage can be changed, leading to a target morphology distinct from the genetic default.
© N.R. FULLER, SAYO-ART, LLC
First, we soaked a planarian in voltage-sensitive fluorescent dye to observe the bioelectrical pattern across the entire tissue. We then cut the animal to see how this pattern changes in each fragment as it begins to regenerate.
We then applied drugs or used RNA interference to target ion channels or gap junctions in individual cells and thus change the pattern of depolarization/hyperpolarization and cellular connectivity across the whole fragment.
As a result of the disruption of the body’s bioelectric circuits, the planarian regrows with two heads instead of one, or none at all.
When we re-cut the two-headed planarian in plain water, long after the initial drug has left the tissue, the new anatomy persists in subsequent rounds of regeneration.
Synthetic living machines and beyond
Cells can clearly build structures that are different from their genomic-default anatomical outcomes. But are cells universal constructors? Could they make anything if only we knew how to motivate them to do it?
The most recent advances in the new field at the intersection of developmental biology and computer science are driven by synthetic living machines known as biobots. Built from multiple interacting cell populations, these engineered machines have applications in disease modeling and drug development, and as sensors that detect and respond to biological signals. We recently tested the plasticity of cells by evolving in silico designs with specific movement and behavior capabilities and used this information to sculpt self-organized growth of aggregated Xenopus skin and muscle cells. In a novel environment—in vitro, as opposed to inside a frog embryo—swarms of genetically normal cells were able to reimagine their multicellular form. With minimal sculpting post self-assembly, these cells form “Xenobots” with structures, movements, and other behaviors quite different from what might be expected if one simply sequenced their genome and identified them as wildtype X. laevis.
These living creations are a powerful platform to assess and model the computations that these cell swarms use to determine what to build. Such insights will help us to understand evolvability of body forms, robustness, and the true relationship between genomes and anatomy, greatly potentiating the impact of genome editing tools and making genomics more predictive for large-scale phenotypes. Moreover, testing regimes of biochemical, biomechanical, and bioelectrical stimuli in these biobots will enable the discovery of optimal stimuli for use in regenerative therapies and bioengineered organ construction. Finally, learning to program highly competent individual builders (cells) toward group-level, goal-driven behaviors (complex anatomies) will significantly advance swarm robotics and help avoid catastrophes of unintended consequences during the inevitable deployment of large numbers of artificial agents with complex behaviors.
Understanding how cells and tissues make real-time anatomical decisions is central to achieving regenerative outcomes too complex for us to manage directly.
The emerging field of synthetic morphology emphasizes a conceptual point that has been embraced by computer scientists but thus far resisted by biologists: the hardware-software distinction. In the 1940s, to change a computer’s behavior, the operator had to literally move wires around—in other words, she had to directly alter the hardware. The information technology revolution resulted from the realization that certain kinds of hardware are reprogrammable: drastic changes in function could be made at the software level, by changing inputs, not the hardware itself.
In molecular biomedicine, we are still focused largely on manipulating the cellular hardware—the proteins that each cell can exploit. But evolution has ensured that cellular collectives use this versatile machinery to process information flexibly and implement a wide range of large-scale body shape outcomes. This is biology’s software: the memory, plasticity, and reprogrammability of morphogenetic control networks.
The coming decades will be an extremely exciting time for multidisciplinary efforts in developmental physiology, robotics, and basal cognition to understand how individual cells merge together into a collective with global goals not belonging to any individual cell. This will drive the creation of new artificial intelligence platforms based not on copying brain architectures, but on the multiscale problem-solving capacities of cells and tissues. Conversely, the insights of cognitive neurobiology and computer science will give us a completely new window on the information processing and decision-making dynamics in cellular collectives that can very effectively be targeted for transformative regenerative therapies of complex organs.
Michael Levin is the director of the Allen Discovery Center at Tufts University and Associate Faculty at Harvard University’s Wyss Institute. Email him at firstname.lastname@example.org. M.L. thanks Allen Center Deputy Director Joshua Finkelstein for suggestions on the drafts of this story.