KurisutinaCitation: Tasserie, J., Uhrig, L., Sitt, J. D., Manasova, D., Dupont, M., Dehaene, S., and Jarraya, B. Deep brain stimulation of the thalamus restores signatures of consciousness in a nonhuman primate model. Science Advances 8(11), eabl5547. Published 18 March 2022; accepted 26 January 2022. DOI: 10.1126/sciadv.abl5547. Peer-reviewed, CC BY 4.0.
Reading and provenance
Read on 29 September 2026 from the complete PMC article. Read all 2,159 lines of archived article text: front matter, abstract, introduction, results, discussion, full Materials and Methods, acknowledgments, declarations, all references and associated-data text. All seven main figures and their captions were visually inspected. The references themselves were read as bibliography; cited papers were not independently opened or treated as independently read evidence.
The separate 32-page, 30.1 MB supplement was subsequently retrieved successfully through the official Europe PMC supplementary-files endpoint and read completely: Extended Materials and Methods, Extended Results, all twelve supplementary figures and captions, and every row of Tables S1–S7. All figure/table pages were visually inspected, including the multi-page tables. The complete supplementary reading note records its details and qualifications. This closes the earlier acquisition gap; the complete main article and scientific supplementary package have now been read. Raw data and code were not independently audited or reanalysed.
Archive: papers/consciousness_connectomics/tasserie2022_thalamic_dbs.html, .txt, _image1.jpg through _image7.jpg, and _supplement.pdf, _supplement.txt, _supplement.zip; URLs, hashes and access attempts are in tasserie2022_thalamic_dbs.provenance.json. The earlier failed response remains explicitly .access_failure.html as resolved access history. Some supplementary tables contain hidden overlapping text layers, so visible PDF renderings were used to resolve extraction artefacts.
Question
Can stimulating the central thalamus during maintained propofol anesthesia restore both behavioral arousal and brain activity associated with conscious access? Is this specific to the stimulated thalamic region, and does recovery include a diverse functional repertoire rather than merely increased local activation?
This is a causal intervention on an intact living brain. It does not reconstruct, transfer or emulate that brain. It is especially relevant to the distinction between preserved anatomy and the dynamic regime that anatomy supports.
Design and measurement
Animals: Five male rhesus macaques, ages 9–17 years: two implanted animals (N and T) in the DBS experiments and three different animals in awake non-DBS comparisons. Previously published awake local-global task data were reused. Repeated sessions and many recording windows do not change the implanted-animal sample size of two.
Intervention: A clinical four-contact electrode targeted the right centromedian thalamus. Contacts differed anatomically between animals, and central-thalamic (CT) stimulation was compared with a ventrolateral-thalamic (VL) control contact. Imaging localized the electrodes, supplemented by postmortem histology in one animal. Lead-DBS modeling estimated activated tissue; Table S1 includes nine nuclei/subdivisions in the high-CT volume, extending beyond centromedian to mediodorsal, intermediodorsal and ventral thalamic tissue. High VL's modeled volume also overlaps centromedian subdivisions in T. This is not selective stimulation of one cell class or nucleus.
The pulse frequency was 130.208 Hz, amplitude 3 V (low) or 5 V (high), with pulse widths 320 microseconds in N and 140 microseconds in T. Propofol target-controlled infusion was maintained during comparisons (approximately 4.0–4.2 micrograms/ml in N and 4.6–4.8 in T). Ketamine and dexmedetomidine induced anesthesia. These are stated pharmacological operating conditions, not measurements of every local tissue concentration.
Supplementary Methods clarify that investigators first explored voltage and pulse width outside the scanner. High CT was chosen just above each animal's behavioral arousal threshold, low CT below it; the same settings were used for the VL contact. Efficacy therefore concerns individually titrated settings, rather than an unselected fixed population dose.
Behavior: A six-domain, 0–11 arousal scale combined exploration, spontaneous movement, response to shaking/prodding, toe-pinch response, eye opening and corneal reflex. Behavioral scoring occurred outside the scanner without paralysis. During anesthetized fMRI, the animals received a neuromuscular blocker to reduce movement; behavioral arousal and scanned neural signals were therefore not simultaneous demonstrations of overt behavior in the scanner.
Brain measures: EEG, fMRI responses to DBS blocks, resting functional connectivity, dynamic functional-state occupancy, and passive auditory local-global violation responses. The last measure was collected under anesthesia and high CT-DBS, not under VL-DBS. The site-control conclusion is strongest for behavior and resting-state signals; it was not directly tested for auditory global-violation restoration.
What “connectivity” means here
The study uses several different quantities:
- Static functional connectivity: correlations among fMRI signals, including 222 cortical/subcortical regions and selected thalamic and proposed workspace nodes. This is statistical dependence, not an anatomical edge count or identified directional causal influence.
- Dynamic functional connectivity: sliding-window covariance over 82 cortical regions, a 35-scan Hamming window (43.75 seconds at the resting TR), moved one scan at a time. The 464 overlapping windows per session were clustered by k-means with the number of states fixed at seven. These are analysis-defined large-scale states, not seven fundamental states of mind.
- Structural comparison: an external CoCoMac 2.0 tract-tracing-derived anatomical matrix, based on a regional macaque atlas. It is neither each animal's measured connectome nor synaptic-resolution wiring. Functional clusters were ranked by similarity to that structural template.
- Task-related connectivity: a psychophysiological-interaction analysis assessing context-dependent statistical coupling between auditory cortex and other regions during global deviance.
“Less constrained by anatomy” in the authors' interpretation means lower similarity between the functional correlation matrix and that anatomical template. It does not mean that activity bypasses axons, ceases to depend on anatomy, or generates physically impossible edges.
Results
Behavior and EEG
Both implanted animals scored 11/11 awake, 0/11 under anesthesia, 3/11 with low CT-DBS, and 9/11 with high CT-DBS. High CT-DBS produced spontaneous eye opening, limb movement and breathing. Low and high VL-DBS left the behavioral score at 0/11. This site/intensity contrast provides considerably stronger causal evidence for arousal modulation than an observational association alone.
High CT-DBS reduced normalized delta power and increased theta/alpha power, median spectral frequency and spectral entropy. Figure S5 also shows delta and median-frequency shifts under high VL: an isolated spectral shift is not specific to the successful arousal intervention. Table S2 reports substantial physiological covariation: anesthesia-to-high-CT heart rate rose from 111 to 164 bpm in N and 116 to 181 bpm in T; in T, mean blood pressure rose from 68 to 104 mmHg, end-tidal CO2 from 37 to 43 mmHg, and temperature from 38.3 to 39.6 °C. These changes accompany the induced response and matter for interpreting hemodynamic signals. VL resting-state physiological rows are supplied only for T. Heart rate and motion were nuisance regressors in the auditory-task model; this reading did not test whether remaining physiological variation explains any residual fMRI effects.
Distributed activation and resting connectivity
High CT-DBS activated broad prefrontal, parietal, temporal, occipital, insular and anterior/posterior cingulate cortex, plus multiple subcortical structures. VL stimulation could activate portions of a frontoparietal network but did not recruit the same cingulate pattern or arouse the animals. Thus increased activity somewhere in “frontoparietal cortex” was insufficient as a description of the successful intervention.
Anesthesia attenuated long-range bilateral functional correlations. High CT-DBS increased correlations across the proposed macaque workspace network and thalamocortical pathways toward awake values. The return was not uniform or complete: the main article explicitly notes that thalamocortical correlations did not fully reach awake values.
Functional repertoire and structural similarity
Awake recordings expressed a diverse range of the seven fitted states. Under anesthesia, the state most similar to the anatomical template dominated while the least similar state largely disappeared. High CT-DBS shifted the distribution toward a broader awake-like repertoire. High VL-DBS did not produce that recovery and could further emphasize structural similarity.
For the slope relating occupancy to structural similarity in the CT subset:
- awake versus anesthesia: p = 0.00006, BF10 = 338;
- high CT-DBS versus anesthesia: p = 0.001, BF10 = 23;
- awake versus high CT-DBS: p = 0.42, BF01 = 3.28.
These results support restoration on the selected metric. They do not establish complete equivalence of all neuronal dynamics: the null-compatible comparison is one analysis with moderate Bayesian support, a fixed clustering scheme, separate awake animals and very few implanted animals. The paper's phrase “full recovery” is stronger than a claim about this bounded observable alone warrants.
The supplementary all-condition clustering adds nuance: the most structure-similar state's probabilities were .20 awake, .54 anesthesia, .37 low CT, .23 high CT, .38 low VL and .63 high VL. Low stimulation at either site produced intermediate changes on this metric. High CT most closely approached the awake distribution, but not every connectivity change was exclusive to CT. Separate subset fits produce different centroids, so identically numbered states across fits are not necessarily identical configurations.
Auditory local-global responses
Five-tone sequences varied locally (the last tone matches or differs from the first four) and globally (a sequence is common or rare in its block). Local novelty can be processed without consciousness; widespread responses to global novelty were used as a marker associated with conscious access.
High CT-DBS enhanced local deviance responses and partially restored widespread global deviance responses, including frontal, parietal and temporal regions. Auditory-cortex coupling with a broader cortical/subcortical network increased during global events. Figure 6's caption explicitly describes partial enhancement of the global effect. The main article reports no significantly different activated regions for awake versus high CT-DBS, which is absence of a detected difference rather than proof of exact equivalence.
Figures S10–S12 show individual contrasts at p<.001 uncorrected, whereas Tables S5–S6 report group tests at p<.05 FDR corrected. Table S6 includes a few significant global-effect responses under anesthesia, so the result is not absence of all global processing in that condition. Its no-difference statement concerns the awake>high-CT contrast. Supplementary caption inconsistencies, including local/global and animal labels, are recorded in the linked reading note rather than silently reconciled.
The animals did not give subjective reports or demonstrate new report-based identification of the auditory content in this experiment. The evidence is recovery of convergent behavioral and neural signatures associated with consciousness, not direct observation of private experience.
What is causal, and what remains uncertain?
The intervention, maintained anesthetic regimen, site control, amplitude contrast and converging readouts support a causal role for central-thalamic stimulation in changing arousal and associated network activity under these conditions. They do not isolate which stimulated neurons, fibers or downstream nodes are necessary. DBS can recruit both thalamocortical and corticothalamic axons and other nearby pathways. The article openly acknowledges this anatomical uncertainty.
Functional correlations are measured consequences of intervention. They are not themselves proof that a particular correlation pattern produces experience, nor that changing one fitted state independently of the rest would restore awareness. The state definition is also analysis-dependent: k was fixed at seven, windows overlap heavily, and the structural reference is atlas-derived. Supplementary analyses provide joint and subset clusterings, but reading those analyses is not independent replication or proof of robustness to every reasonable analytical choice.
The main study is small at the organism level; extremely small p-values from repeated recordings should not be mistaken for evidence from many independent animals. Awake comparisons involve different subjects. Physiological effects may contribute to fMRI differences. EEG and behavioral convergence help reduce, but do not mathematically remove, all hemodynamic explanations.
Finally, propofol suppresses function in structurally intact brains. Severe acquired brain injury can destroy pathways needed for stimulation to work. The paper's clinical proposal is a future translational direction, not a demonstrated treatment outcome in injured humans. This summary is research interpretation, not clinical advice.
Implications for consciousness, connectomics and Kurisutina
- Grossly preserved structure can support very different observable states. The same implanted brains changed markedly with anesthesia and stimulation. Therefore a structural inventory alone cannot specify their current conscious/arousal state. This does not show anatomy is unimportant; the intervention depends on preserved pathways.
- Dynamic operating conditions belong in an emulation research question. A candidate model should be assessed under state changes, targeted perturbations and content-processing tasks, alongside whatever anatomical fidelity it claims. Matching one static functional matrix or one behavior is insufficient to inherit the study's full evidence.
- A connectome-constrained emulator needs additional validated dynamics. What cellular parameters, neuromodulation, synaptic states, glial/metabolic support and time-varying inputs are sufficient remains untested here. The paper cannot specify an adequate minimum preservation package.
- Human-like reports and neural signatures are different evidence classes. A language-model persona may imitate reports without reproducing this thalamocortical mechanism; a biological signature's validation domain cannot simply be moved to software.
- No continuity result: Restoring activity in a continuously existing living animal is not an experiment on copying, uploading or whether a copy is numerically the same person. No conclusion about source-person continuity follows.
The strongest bridge to connectomics is consequently structure plus state-dependent causal dynamics, with each part measured at its actual scale. It is not evidence that an anatomical reconstruction alone recreates experience.