Building novel biological formsTraditionally, 2D neuronal cell culture models are used to understand neural development and disease mechanisms.
This led to the development of 3D brain organoids that can self-organize into neural circuits with the capacity for basic learning.
“We still do not understand how these neurons organize themselves or what cues guide their development,” Fotowat says.
The team also found that no two neurobots were the same in terms of neuronal growth and architecture, probably because of inconsistencies in the manual implantation of neural cells.
Reference: H. Fotowat et al., Engineered Living Systems With Self-Organizing Neural Networks: From Anatomy to Behavior and Gene Expression.
The notion of “adapt or die” is an underlying principle of Darwinian evolution theory. This adaptation, known as plasticity, refers to the ability of organisms to alter their physical appearance or phenotype in response to environmental change. More specifically, neuroplasticity refers to the ability of nervous systems to change structurally and functionally in response to changes, including sensory input or physical injury.
This type of adaptation or evolution normally happens over years or millions of years — but what are the limits of neuroplasticity during a short developmental stage in a non-standard body? For example, engineering eyes into the tail of a Xenopus tadpole gives the tadpole a sense of light perception, even though these “eyes” were not developed through evolution. If this can be achieved, what else is possible?
To answer this, a novel biological system is needed.
Researchers at Tufts University and Harvard University created tiny living “neurobots” that grow their own brain cells and use them to move, offering insights into how nervous systems function in new environments.
Building novel biological forms
Traditionally, 2D neuronal cell culture models are used to understand neural development and disease mechanisms. However, such models do not reflect the complexity of neural circuits and the diversity of cell types, making neurodevelopment studies challenging. This led to the development of 3D brain organoids that can self-organize into neural circuits with the capacity for basic learning.
While brain organoids can aid the study of neuroplasticity and neuropsychiatric disorders, they are non-motile and unable to perform simple tasks, limiting their representation of human brains.
A more recent advancement is biohybrid robots — a combination of biological material such as muscles and neurons with synthetic materials. These robots have applications in biotechnology, but are not entirely biological and do not self-assemble.
To address these limitations, Michael Levin, a professor of developmental and synthetic biology at Tufts University, and his team developed a cost-effective biological model to understand the early formation of neural circuits, including their structure and function. They used micro-tweezers to remove a section of ectodermal tissue called the “animal cap” (embryonic tissue responsible for giving rise to the brain, skin, and tissues) from a Xenopus frog embryo.
Left alone, this tissue develops into a spherical cluster of motile, skin-like cells, that can propel itself through fluid using tiny hair-like structures called cilia. These self-powered organoids are known as biobots.
Self-organizing neural networks
To create neurobots, the researchers implanted neuronal precursor cells into biobots within the first few minutes of extracting them, and supplied the raw materials needed for growth.
The precursor cells, also derived from frog embryos, can be induced to become functional neurons under specific conditions. In these experiments, they matured into neurons that self-organize within the biobot, interconnecting with other neurons and extending their neural processes towards the neurobot surface. In addition to structural growth, the neurons are functional and exhibit neural activity.
“When these neural precursors are introduced into an intact animal cap, they mature into neurons within a body composed primarily of skin cells,” explains Haleh Fotowat, Senior Scientist at the Wyss Institute and first author of the research study, published in Advanced Science.
Once inside, the neurons begin organizing themselves — but exactly how remains unclear. “We still do not understand how these neurons organize themselves or what cues guide their development,” Fotowat says.
The team also found that no two neurobots were the same in terms of neuronal growth and architecture, probably because of inconsistencies in the manual implantation of neural cells.
What the researchers do know is that the cells form complex networks — extending axons and dendrites, forming synapses, and showing spontaneous activity detected through calcium imaging, which acts as a proxy for electrical signaling in the brain. Interestingly, some of these neural projections extend toward the outer surface of the neurobot, suggesting that the neurons may influence the cells responsible for movement.
More active, complex behavior
Compared with biobots and “sham” neurobots—bots implanted with neural precursor cells that were not given the chance to develop into functional neurons—the neurobots were not only more elongated in shape and larger in size, but also behaved differently. While many biobots remained still for long periods, neurobots were more likely to keep moving.
“Neurobots were generally more active than biobots and tended to exhibit more complex trajectories,” Fotowat says, a phenomenon the researchers believe is due to their neurons. If their findings are confirmed, it raises the question about how much influence a neurobot, a more advanced version of a biobot, has over movement.
To probe the role of neural activity, the researchers exposed both types of bots to a seizure-inducing drug. They expected only neurobots to respond, but the results were more complex. “To our surprise, we found that biobots responded more dramatically, with most reducing their movement,” Fotowat says. Neurobots, meanwhile, showed mixed responses — some became more active, while others slowed down.
The findings suggest that the drug affects not only neurons but also non-neural cells involved in movement, and that neural activity in neurobots may partially counteract these effects.
Unexpected genetic signals
At the molecular level, the differences were also striking. Genetic analysis revealed that, compared with biobots, the transcriptomics (RNA readouts in cells) of neurobots express more of the genes needed for nervous system development and visual perception.
“The most surprising finding was the coordinated overexpression of genes encoding proteins involved in multiple stages of visual processing, including those expressed in the lens, photoreceptors, and various layers of the retina,” Fotowat says. The team are particularly excited about this, and plan to investigate whether these genes translate into visual perception proteins and whether neurobots could, in some sense, respond to light.
Despite the many open questions here, these findings already indicate that unexpected capabilities may emerge in these systems.
Interestingly, the genes of neurobots seem to be more ancient and reflective of gene profiles of the past compared with biobots. The team thinks that this is because the neurobot is so early on in its evolutionary history that it is like starting from the beginning.
Rethinking the limits of the brain
Unlike all other organisms on this planet, neurobots have no natural history of selection for their traits, so studying them provides insight into the different forms and functions that a genome can give rise to when under no natural selection pressure. The current results already suggest that nervous systems may be far more flexible than previously thought.
“By pursuing this line of research,” Fotowat says, “we will define these limits, which will be instrumental for the development of fully biological robots.”
The researchers caution that it is still early days. “The field is extremely new, and there are so many unanswered questions,” Fotowat says. “We are only beginning to explore the range of possibilities for nervous systems in novel embodiments.”
Fotowat is particularly interested in identifying what drives their behavior: “I am most excited about diving deeper into the neural network composition of neurobots and discovering which sensory stimuli they respond to.”
The study goes beyond fundamental insights. “It has clear applications in biological engineering, including the development of fully biological microrobots capable of navigating hard-to-reach environments, self-repair, and biodegradation,” says Fotowat. “In parallel, these studies can inform regenerative medicine by revealing how cells and tissues reorganize, integrate, and regain function in non-native settings, offering new strategies for repair and reconstruction.”
Moving forward, automated methods would help standardize neurobot structure and morphology. Automation would also speed up neurobot production, enabling experiments to explore the effects of light and pharmaceuticals.
Michael Levin sees even broader implications. Because neurobots are not shaped by evolution like most animals, he thinks they may offer a unique opportunity to probe how minds arise, perhaps helping to visualize the “home worlds” of cyborgs and synthetic beings in the future.
“What non-existent world is their cognitive architecture tuned to?” he asked. “I think we can find out.”
Article written by Kerry Day and Jenna Flogeras.
Reference: H. Fotowat et al., Engineered Living Systems With Self-Organizing Neural Networks: From Anatomy to Behavior and Gene Expression. Advanced Science (2026), DOI: 10.1002/advs.202508967
