Fork the consciousness, or download the project and create your own.

Hugues Duffau, Slow-Growing Gliomas and the Plastic Adult Brain

Can an adult brain move a function out of the tissue that carries it, while the person stays awake and the same? The clinical work of neurosurgeon Hugues Duffau at Montpellier University Medical Center says yes, under one condition. The loss must be slow. Patients with slow-growing low-grade gliomas keep language and movement after surgeons remove large parts of cortex that a stroke in the same place would leave impaired. Duffau’s group also found where this brain plasticity stops. Cortex reorganizes well. The white matter tracts that connect it do not.

For a project that asks whether a person could move from one substrate to another, this is the strongest human evidence available. It shows a function leaving its tissue inside one living brain, with the person conscious and reporting throughout, and it shows the conditions under which that works and fails.

Slow lesions and acute lesions leave different brains

The starting point is a review in Brain by Michel Desmurget, François Bonnetblanc and Hugues Duffau, published in April 2007 (doi:10.1093/brain/awl300). Most of what neurology knew about plasticity came from acute strokes. Stroke data built the picture of a brain with specialized “eloquent” areas, where a lesion causes a permanent deficit. Low-grade glioma patients did not fit that picture. They underwent large resections of eloquent areas with no detectable functional consequence.

The review argues that brain plasticity cannot be understood without the time course of the injury. It collects human and animal data showing that functional recovery is much better after slow-growing lesions than after acute ones. It also reports a difference in how the brain reorganizes. With slow lesions, the recruitment of remote areas, in the same hemisphere and in the opposite one, is much more efficient.

A later clinical study gave the time effect a number. Jeffrey Wefel and colleagues at MD Anderson tested 119 patients with malignant glioma before surgery (Neuro-Oncology 2016, doi:10.1093/neuonc/now165). Patients with the faster-growing IDH1 wild-type tumors had worse memory, processing speed, language and executive function than patients with the slower IDH1-mutant tumors, although lesion volume did not differ. Lesion volume predicted the deficit only in the fast group. The authors call the rate of growth “lesion momentum”.

Watching a function leave a site

Awake surgery gives a direct view of this process. During the operation the surgeon stimulates small sites of cortex while the patient names pictures, counts or moves a hand. A site where stimulation disrupts the task is “positive” and is spared. When a tumor recurs years later, some patients are mapped a second time.

Derek Southwell, Shawn Hervey-Jumper, David Perry and Mitchel Berger at the University of California, San Francisco reviewed 561 mapped glioma cases (J Neurosurg 2016, doi:10.3171/2015.5.JNS142833). Eighteen patients had a repeat operation in which stimulation sites overlapped with the first, a mean of 4.1 years later.

Site status at first surgery Number of sites Status at repeat surgery
Positive (language, motor or sensory) 22 13 still positive, 9 negative
Negative 95 94 still negative, 1 positive

Nine sites that once carried language, movement or sensation no longer did, and none of those patients had a neurological deficit at the repeat operation. Function left these sites. Only one silent site became positive, so the data show where function left far more clearly than where it went. Only the exposed cortex is tested, and a function can move to tissue outside the surgical field or spread so thinly that no single site is required.

Duffau’s own serial mappings in Montpellier show the same remapping in the same patients over time, which allows a wider resection at the second operation (Cortex 2014, doi:10.1016/j.cortex.2013.08.005).

Cortex is plastic, white matter is the limit

The most useful result for any theory of substrate change is a map of where plasticity works. Guillaume Herbet, Maxime Maheu, Emanuele Costi, Gilles Lafargue and Hugues Duffau built a probabilistic atlas of functional plasticity from the MRI and intraoperative mapping data of 231 patients operated for diffuse low-grade glioma (Brain 2016, doi:10.1093/brain/awv394).

The atlas shows high plasticity across most of the cortex. The exceptions are the primary unimodal areas and a small set of neural hubs. Plasticity is low in the connecting tracts, especially the long associative and projection tracts. Duffau’s 2014 review in Cortex draws the same conclusion from serial mappings. White matter connectivity is the main limit of brain plasticity, which explains why patients with extensive subcortical damage do not recover.

Brain structure Plastic potential in the atlas Consequence for resection
Most association cortex High Large parts can be removed in slow glioma
Primary unimodal cortex Low Removal leaves a deficit
A small set of cortical hubs Low Removal leaves a deficit
Associative and projection white matter tracts Low The tract is the boundary of the resection

Herbet and Duffau later turned these observations into the meta-networking theory of cerebral function (Physiological Reviews 2020, doi:10.1152/physrev.00033.2019). In this model, complex behavior comes from the dynamic integration of distributed, relatively specialized networks. A function is a property of the connections between areas. Remove a cortical node slowly and the network can route around it. Cut the tract and the network has no route.

What this tells us about consciousness

Every patient in these studies stays the same person, by their own report and to their families, while the tissue that carried their language or movement changes. Awake surgery adds a detail that no thought experiment provides. The patient is conscious and talking at the moment a site is tested and removed.

This bears on two positions that this site covers often. Hans Moravec’s gradual neural replacement argument assumes that a person survives if their parts are replaced one at a time. The glioma data are the nearest natural version of that thought experiment, and they add two conditions the thought experiment leaves out. The replacement must be slow enough for the network to reorganize, and the connecting tracts must stay intact.

The data also support the view that a function belongs to a pattern of activity across a network, more than to one patch of cells. If one patch of cortex were the function, nine of 22 positive sites could not have gone silent with no loss. That is a view about functions, and functions are not consciousness. Still, theories that place consciousness in network dynamics, such as the global workspace or the dynamical repertoire described on our consciousness fingerprint page, predict exactly this kind of relocation. A theory that tied a person’s experience to specific cells would have to explain why these patients report no change.

The limits are as informative as the successes. “No deficit” in these studies means a normal neurological examination and the tasks that can be tested in seconds on the operating table. Fuller neuropsychological testing finds subtler problems, and Martin Klein notes that brief screens can miss small cognitive changes (Neuro-Oncology 2012, doi:10.1093/neuonc/nos161). Capacities in daily use, such as speech and movement, are what the studies measure. They say much less about rarely used memories. The broader questions about whole brain emulation and gradual transfer are collected on our brain emulation page.

Comparison to The Consciousness AI

The Neutral Core is this project’s design for a synthetic hemisphere. Its current protocol ends with a single switch, where one biological hemisphere is suppressed and the Core takes over. In light of the glioma evidence, we added a section on a slow handover instead of one switch. A one-step handover has the time profile of a stroke. The Core’s interface also sits at the corpus callosum, which is a white matter tract, the least plastic tissue in the Herbet atlas.

None of this has been tested with the Core, which exists only as a simulation. The Substrate Console lets a reader run the kind of leaky integrate-and-fire network the Core design builds on, at a much smaller scale. The glioma data come from living tissue taking over from living tissue. They do not show that a non-living substrate could take over a function inside a brain. They do describe what any such handover would need to respect, in speed and in connectivity.

Limits of the slow-glioma evidence

Hugues Duffau’s work and the studies around it show that the adult human brain can move language and movement out of tissue that is lost slowly, and that white matter tracts set the limit. The evidence has clear boundaries. It comes from patients with tumors, whose brains may differ from healthy ones. The repeat mapping study has 18 patients. The measured functions are the ones easy to test during surgery. No study gives a safe speed for planned tissue loss in humans, and no study shows a stored memory, as distinct from a skill, moving to new tissue. Within those limits, slow-glioma brain plasticity is the best human evidence that a person can continue while the tissue that carries their functions changes. It belongs next to the measurement questions in how scientists race to define and test AI consciousness in 2026, since any claim that a mind survived a change of substrate will need the same kind of before and after evidence.

Researchers covered here