Magdalena Götz and Mark Hübener Show Transplanted Neurons Learn Their Neighbors' Job
Can a neuron with no history of its own take over the job of a neuron that died, inside an adult brain that never makes new cortical neurons? A team led by Magdalena Götz and Mark Hübener in Munich answered yes for the mouse visual cortex. Embryonic neurons grafted into damaged adult cortex grew into normal pyramidal cells, received the same inputs as the neurons they replaced within four weeks, and after 10 to 15 weeks responded to visual stimuli like their host neighbors. The paper is Susanne Falkner, Sofia Grade, Leda Dimou, Karl-Klaus Conzelmann, Tobias Bonhoeffer, Magdalena Götz and Mark Hübener, Transplanted embryonic neurons integrate into adult neocortical circuits, Nature 539, 248-253, October 2016 (doi:10.1038/nature20113).
The result bears on a central question for any neutral substrate. A new element does not need a copy of what the old element knew. In this experiment the network taught it, through the inputs it received and the activity that ran through it.
Removing neurons and grafting new ones
The adult mouse neocortex does not add new neurons, so any replacement must come from outside. The team first removed a defined population. They injected beads coupled to a light-sensitive chemical, chlorin e6, into one hemisphere. Neurons in the upper layers of the opposite primary visual cortex (V1) that project across the corpus callosum took up the beads, and laser light on that area one to two weeks later killed them by apoptosis. Seven to ten days after the lesion, the team grafted cortical cells from mouse embryos at day 18.5 into the depleted area.
The details below come from the paper and from its accepted author manuscript, which the Helmholtz Munich repository hosts.
- By four weeks after transplantation, most grafted cells had the shape of mature layer 2/3 pyramidal neurons, and 77% of 200 counted cells carried the upper-layer marker Cux1.
- Grafted cells did not fuse with host cells. In ten mice the donor and host carried different fluorescent markers, and no cell carried both.
- Dendritic spines and axonal boutons reached adult-like densities within 4 to 8 weeks, following roughly the timing of normal cortical development.
The right inputs, with almost no wrong ones
To see who connects to the grafted neurons, the team used a modified rabies virus that crosses one synapse backwards from a starting cell. Grafted neurons received inputs from the areas that normally project to upper-layer V1 neurons, including topographically organized input from the visual thalamus (the dorsal lateral geniculate nucleus), along with other cortical areas, subcortical nuclei and the opposite hemisphere. Most of these regional connections were already present four weeks after transplantation and were still present eight weeks later. The authors report virtually no aberrant connections.
Tuning that starts broad and sharpens with activity
The functional test used two-photon calcium imaging while the mice viewed moving gratings in eight directions, from 4 to 15 weeks after transplantation.
| Weeks after transplantation | What the grafted neurons did |
|---|---|
| 4 | Synaptic inputs from almost all regions that normally project to V1 |
| 4 to 8 | Spine and bouton densities reach adult-like values |
| 5 | First signs of orientation or direction preference, broadly tuned |
| 5 to 9 | Preferred direction changes between sessions in 9 of 13 repeatedly imaged cells |
| 6 to 10 | Tuning sharpens and responses become more selective |
| 10 to 15 | Tuning stabilizes at the values of host layer 2/3 neurons in adult V1 |
Of 28 neurons with significant fluorescence changes, 27 responded to the visual stimuli, and 13 of those 27 had strong orientation or direction preference, in five mice. Some grafted neurons in the binocular part of V1 responded to both eyes. Stable tuning coincided with the end of high spine turnover after about eight weeks.
The authors compare this with normal development. In young mice, sharp tuning appears right after eye opening and reaches adult values within a week, which is thought to depend on related neurons sharing feature selectivity through early gap junctions. The grafted neurons skip that step. Their first connections are stochastic and broadly tuned, and the specific tuning emerges over weeks, which the authors suggest reflects activity-dependent refinement driven by visual input.
What this tells us about consciousness
This experiment is the clearest animal case of a blank element taking over a function from the network it joins. The grafted neuron arrives with no visual history. Its selectivity comes from where it sits, which inputs reach it, and the activity that passes through it over weeks. On the evidence of this study, the information needed to rebuild the function lives in the surrounding circuit and its traffic, and a replacement part can acquire it.
That supports the view this site takes on consciousness as an emergent property of organized activity, and it gives the word “neutral” in the Neutral Core a concrete meaning. A neutral template does not need a stored copy of a person. It needs the right connections and enough time for the person’s own activity to shape it.
The experiment also sets limits. The replaced neurons were a small population in one area, the host network around them stayed intact, and the new parts were living neurons of the same species. It does not show that a person’s experience continues through such a replacement, and it does not test a stored memory, only stimulus tuning that the visual input keeps training. Our article on Hugues Duffau’s slow-growing glioma work shows the human side of the same theme, where a function moves out of tissue that is lost slowly. The broader questions on gradual transfer sit on our brain emulation page.
Comparison to The Consciousness AI
The Neutral Core is this project’s design for a synthetic hemisphere, and its slow handover section cites this paper for the idea that new tissue takes its function from its inputs. The Core exists only as a simulation, and its neurons are leaky integrate-and-fire units of the kind a reader can run in the Substrate Console. Nothing in the project has shown that such units could do what these grafted neurons did. The paper gives the design a target to test against. A new element should reach the right inputs first, then sharpen its tuning through activity, and its final response properties should become indistinguishable from the tissue around it.
Limits of the transplanted neuron result
Magdalena Götz, Mark Hübener and colleagues showed that transplanted embryonic neurons integrate into damaged adult mouse visual cortex with the right inputs, almost no wrong ones, and host-like tuning after 10 to 15 weeks. The limits are clear. The work is in mice, with a selective lesion of one cell population and a graft of matched embryonic tissue. The functional sample is small, 28 cells in five mice. No behavioral recovery was tested. The authors themselves name stroke and traumatic injury as lesion types still to test, and other cell sources such as stem-cell-derived neurons. Within those limits the study is a standard for what a replacement part must achieve, and it belongs with the measurement debates in how scientists race to define and test AI consciousness in 2026, since any replaced part must be shown to work like the original.