A puff of air crossed a rat’s whiskers, and human neurons responded.

Those neurons had begun as reprogrammed human cells, grown into small clusters of cortical tissue in a laboratory and transplanted into a newborn rat’s brain. Now, information from the outside world was reaching them. In another experiment, activating the transplanted neurons with light became a signal the rat learned to associate with water. Tissue grown in a dish had become part of a circuit that could influence behavior.

This finding, from a 2022 Stanford study led by Sergiu Pașca, is the kind of result that makes me pause over what medicine might eventually become. Growing human neural tissue is remarkable on its own. Seeing that tissue participate in a living nervous system raises a much larger possibility: perhaps we could eventually grow tissue that helps restore functions lost to brain injury.

Getting there will require solving problems that this experiment only begins to approach. But I think the path toward that future starts with something more immediate and equally consequential: using a person’s own cells to understand their disease.

Scientists can take adult cells, reprogram them into induced pluripotent stem cells, and guide their development toward particular neural identities. In three-dimensional cultures, those cells can organize into structures called organoids, which reproduce selected features of developing brain regions. They offer a way to study living human neural tissue while retaining the genetic background of the person it came from.

The possibility is surprisingly intimate. A small sample of someone’s cells could help reveal why their neurons develop differently, how their electrical activity becomes disrupted, or which interventions might correct a particular defect.

That changes what a model of disease can be. A diagnosis groups people according to shared features. A model derived from an individual could help investigate the biology beneath those features. I imagine a future in which clinicians can draw on both: the accumulated knowledge of a condition and experimental evidence from tissue carrying their patient’s genetic background.

Assembloids make this prospect more powerful by allowing researchers to study interactions between different tissues. Separate organoids can be brought together so that cells migrate between them and establish functional connections. Researchers have already assembled cortical tissue, spinal cord-like tissue, and skeletal muscle into a system in which stimulating the cortical component produces muscle contraction.

For me, this is where the science becomes especially interesting. A neuron’s behavior depends on the system it belongs to. Some problems emerge as cells find their positions, receive inputs, or begin coordinating their activity. Assembloids give us experimental access to parts of that process. They allow us to ask how a disruption travels through a developing circuit and whether a treatment restores useful function across it.

There is already evidence that this approach can guide therapeutic development. In a later study of Timothy syndrome, researchers used patient-derived organoids, assembloids, and transplanted neurons to test an intervention that altered RNA splicing in a calcium-channel gene. The treatment corrected several disease-associated cellular abnormalities, including defects in neuronal migration and calcium signaling. These were preclinical findings, but they showed how human tissue models could connect a molecular mechanism to a potential treatment.

I can imagine this becoming a much broader process: grow relevant tissue, characterize its dysfunction, compare interventions, and use those results to help select what should advance toward clinical testing. For that to become reliable personalized medicine, responses in the model will have to predict outcomes in patients. A culture cannot reproduce someone’s entire brain or life history. Its value will depend on knowing which questions it can answer well.

The transplantation study adds another dimension. In a dish, neural tissue develops with limited access to the inputs and physiological conditions of a living brain. After transplantation into newborn rats, the human neurons matured further and formed more elaborate branches and connections than their counterparts kept in culture. The host environment helped reveal capabilities—and disease-associated abnormalities—that were harder to observe outside it.

I keep coming back to the implications for repair. If newly grown human tissue can receive information and influence an existing nervous system, could we eventually guide that integration toward a therapeutic purpose?

The future I see begins with carefully defined injuries and specific functions. A person loses part of a neural circuit after a stroke or trauma. We grow tissue with an appropriate regional identity and cellular composition, transplant it, and help it establish useful relationships with surviving networks. Over time, rehabilitation could give those networks repeated opportunities to recruit the new tissue.

There is some experimental support for pursuing this direction beyond newborn animals. Researchers have transplanted human forebrain organoids into injured adult rat visual cortex and found that graft neurons could respond to visual stimulation, with some showing preferences for particular stimulus orientations. That demonstrated functional integration; it did not establish restoration of normal vision.

The distinction matters because restoring function asks much more of a graft than surviving or becoming electrically active. The tissue needs adequate blood flow, controlled growth, appropriate inputs and outputs, and activity that contributes to the surrounding circuit. Existing studies show that graft maturity, vascularization, immune conditions, and connection patterns all influence the outcome.

I suspect that a successful treatment would therefore extend far beyond the operation. Growing the tissue and placing it would begin a longer process of integration. Rehabilitation, sensory experience, and perhaps carefully targeted stimulation could help shape what the new circuit becomes. This is still a research vision, and the newborn-rat study did not demonstrate repair of an injured human brain. It did, however, show a capacity for integration that makes the repair question worth pursuing.

There is something deeply compelling about the possibility of using the brain’s ability to learn as part of how we rebuild it. Newly grown tissue would not arrive carrying the memories or learned patterns that were lost. But perhaps surviving networks could teach it to contribute to movement, sensation, or other functions again.

That is the future I find myself imagining when I read about human neurons responding to a rat’s whiskers. First, we learn to grow tissue that helps us understand a patient. Then we learn to use those models to develop better treatments. Eventually, we may learn to grow tissue that can become part of the treatment itself.

For someone living with a brain injury, that could mean the possibility of recovering an ability they had been told was permanently lost. The scientific challenge is enormous. So is the reason to pursue it.

In The Matrix, you can hand someone an entire world through a cable. When I watched it, I kept wondering how much we would have to understand about the brain to make a convincing afternoon, let alone a lifetime.

The fly-brain maps coming out of Google and its collaborators bring back some of that feeling. In September, a Janelia-led team working with Google Research and researchers in Cambridge released a reconstruction of the male fruit fly’s central nervous system. Together with the FlyWire Consortium’s earlier female-brain map, it gives us an extraordinary view of the cells and connections inside an animal that senses, learns, and moves through the world.

Looking at those reconstructions, I find myself imagining them coming to life. Give each neuron a model, feed the system an input, and watch activity travel through it. Keep improving the map and the models until the simulated circuit responds the way the real one does.

It’s an appealing path toward understanding a brain. It also leaves me wondering what we need to carry from the living tissue into the computer.

From a map to a model

A synaptic connectome describes which neurons connect to which. At one level, it is a directed graph: neurons are nodes, synaptic connections are edges, and the number of synapses between two neurons can serve as a weight. The underlying reconstructions can also preserve detailed cell shapes and the locations of those synapses.

That is an enormous amount of information. It constrains which circuits are possible and gives experiments a concrete anatomical foundation.

To simulate those circuits, we also need to describe how they behave. A synapse count does not, by itself, tell us the strength or time course of a connection. A neuron’s response depends on its membrane conductances, its recent activity, and the inputs arriving at that moment. Researchers building models from connectomes already work to bring this physiology into the picture.

My interest is in another part of that translation. If we specify the neurons and their synapses, what have we assumed about everything around them?

A brain is densely packed living tissue. Cells share an extracellular environment, and their activity changes it. Some of those changes feed back into other cells. To reproduce that behavior, a model needs information about interactions that a synaptic graph alone does not describe.

Who touches whom

One way to recover more of that physical setting is through a contactome: a map of which cells are in physical contact, including contacts without a conventional synapse.

Salova and Kovács have constructed contactomes from fly, mouse, and human tissue datasets. More recently, a preprint by Matelsky and colleagues examined mouse cortical contact networks and found many more physical contacts than synaptic connections. These efforts let us ask how much of a cell’s anatomical neighborhood is missing from its list of synaptic partners.

A contact is a place to investigate, not proof that a signal passes there. Its function still has to be established. But the broader perspective matters: distance, membrane arrangement, and the spaces between cells can help determine how they interact.

Neurotransmitters, for example, can spread beyond the synapse where they were released and affect nearby cells. Recent cerebellar experiments by Santos-Valencia and colleagues showed how glutamate spillover recruits a pathway that enhances calcium signals in Purkinje cells. The chemical influence of a neuron can extend beyond its conventionally mapped connections.

Electrical interactions take this question in a direction that has interested me for a long time. They can act through the shared environment without a direct contact between the interacting cells.

Connection without contact

When I was younger and first learning about membrane potential, I remember being unsure what to do with the outside in the definition.

A membrane potential is the voltage inside a cell relative to the voltage outside it. But the outside contains other active cells. They move ions, generate currents, and change the electrical conditions around them. If a neuron’s voltage depends on that environment, and its neighbors are changing the environment, shouldn’t they be coupled through it?

At the time, I had no way to tell whether the effect would be negligible or important. That was the part I wanted to understand.

The electrical interaction behind that question has an established name: ephaptic coupling. Currents generated by neural activity produce extracellular potential differences that can affect other neurons, without a synapse or a gap junction connecting them.

Working out what follows requires much more than noticing that membrane voltage has two sides. Intracellular and extracellular potentials evolve together. Current spreads through a medium whose conductivity and geometry help determine the resulting fields. Cell shape, orientation, and membrane state help determine the response.

A small change in the extracellular potential is therefore not a universal instruction to every nearby neuron. Its effect depends on where it occurs, how it varies across the cell, and what the cell is already doing.

Studying physics and working on neuromodulation have made this question more concrete for me. The tissue surrounding a neuron helps determine how electrical activity reaches it. And that surrounding tissue is itself active.

Timing is part of the signal

There are already experiments showing what this can mean.

In the fly’s olfactory system, neurons housed in the same sensory compartment can inhibit one another through ephaptic interactions. In the mammalian cerebellum, ephaptic coupling can help synchronize neighboring Purkinje cells. Other work has shown that a complex spike in one Purkinje cell can briefly suppress firing in its neighbors in awake mice.

These findings make it difficult for me to think of ephaptic coupling as a niche curiosity. It participates in sensory processing and ongoing neural activity. How much it contributes across different circuits is a question we still need to answer.

Part of what makes the problem interesting is that the size of a voltage change does not directly tell us the size of its effect on a circuit. Neurons are nonlinear systems. A perturbation arriving near spike threshold can have a different consequence from the same perturbation arriving a moment earlier. Anastassiou and colleagues showed that weak extracellular fields could influence cortical spike timing while producing small changes in membrane voltage.

Changing when a neuron fires can change whether its signal arrives alongside another input. Downstream, that can change how those inputs combine. Timing is part of what the circuit does.

What baffles me is how little room this possibility seems to get when neural communication is first introduced. I tried to check that impression by searching three open introductory neuroscience textbooks for “ephaptic.” None contained the term in their searchable text. That is a small, nonrepresentative sample, and two of the books share material; it cannot tell us what percentage of textbooks cover the subject. Specialist texts do discuss it.

Still, I wish this possibility appeared earlier in the story. It would have helped me understand that the space around a neuron belongs in our picture of how it works.

What a convincing model can miss

This is the question behind research I’ll be sharing soon.

My current computational work examines what happens when we model neural circuits with and without their spatial electrical interactions. The preliminary results point to a gap that interests me: a model can reproduce broad features of activity while missing important parts of the response to a particular input.

Averaging can make that gap harder to see. Two activity traces may look similar in aggregate while differing during the interval when a stimulus arrives or the circuit recovers. That raises a practical question about how we judge a model. Which measurements would reveal that we have left out an interaction that matters?

I’m also investigating what information helps close that gap. How much do we need to know about geometry, the extracellular environment, or the interactions themselves? What can we recover from existing reconstructions, and what needs an additional measurement?

I’m looking forward to sharing that work. It has made me think more carefully about what we mean when we say a brain model is complete. Complete enough to reproduce which behavior, under which conditions?

The challenge is to work out which interactions a model can safely leave out for the behavior we want to understand. Better anatomical maps give us a much stronger foundation for doing that.

The future I find exciting is one in which we can ask a simulated circuit a new question and trust that the real tissue would answer similarly. Getting there may require paying much closer attention to what happens between the neurons we’ve worked so hard to map.