Eight people with paralysis from spinal cord injury, amyotrophic lateral sclerosis, or brainstem stroke attempt to move and speak through brain-computer interface electrodes implanted in their motor cortex, specifically the precentral gyrus, the strip of brain tissue running across the top and side of each hemisphere that drives voluntary movement. They are enrolled in the BrainGate2 consortium clinical trial (NCT00912041, with active sites at Massachusetts General Hospital, VA Providence, Stanford, Emory, UC Davis, and Baylor College of Medicine) and in the Pittsburgh-Chicago intracortical trial (NCT01894802), and their implants are 20 Blackrock Microsystems intracortical microelectrode arrays. In a Nature paper published online on 17 June 2026, recordings from those arrays across the eight participants produced the most detailed single-neuron map of the human precentral gyrus published to date. The headline finding is that body parts are highly intermixed throughout the precentral gyrus, with the entire body represented at every sampled location, and that two speech-preferential areas flank a broadly tuned orofacial-dominant zone in the inferior portion of the gyrus. The first author is Stanford neurosurgery researcher Darrel R. Deo. Co-senior authors are Jaimie M. Henderson and Francis R. Willett, both at Stanford’s Neural Prosthetics Translational Laboratory (NPTL). The paper appears in Nature under DOI 10.1038/s41586-026-10653-x. Darrel Deo began the work as a Wu Tsai Neurosciences Institute postdoctoral scholar in the lab of the late Krishna Shenoy and Jaimie Henderson, per the Stanford Report. Shenoy, the Stanford engineer who co-founded Neuralink and co-directed the NPTL with Henderson, died from pancreatic cancer on 21 January 2023.

What the eight-participant dataset shows

The participants performed an instructed-delay task in which a visual cue prompted one of 46 attempted movements spanning the whole body, with toe curl and hand close among the example movements the authors name in the abstract. The 20 analysed arrays (96-channel and 64-channel Blackrock Utah arrays, 1.5 mm electrode length) were placed in the left precentral gyrus and covered the dorsal hand-knob region, the middle portion of the gyrus, and the ventral portion. The authors report, in the abstract’s own framing, that “body parts were highly intermixed, such that the entire body was represented in all sampled locations of the PCG, although the relative strength of body parts was roughly consistent with the motor homunculus.” They also report “two speech-preferential areas with a broadly tuned, orofacial-dominant area in between them” and find that “homologous movements of different limbs (for example, toe curl and hand close)” had correlated representations. A superior ventral array in participant T12 (a left-handed woman with bulbar-onset ALS) reached 86 per cent decoding accuracy across movement types. Darrel Deo, speaking to the Stanford Report on the day of publication, called the superior ventral region “a jack-of-all-trades area” and said “it was quite exciting to see these whole-body representations appear.”

The 25-year arc of evidence against the strict homunculus

The classical motor homunculus most clinicians remember from medical school was popularised in Wilder Penfield and Theodore Rasmussen’s 1950 textbook The Cerebral Cortex of Man, redrawn from the original 1937 Brain paper by Penfield and Edwin Boldrey. The 1937 paper itself, based on intraoperative electrical stimulation during awake craniotomies in epilepsy patients, already noted substantial overlap and individual variability between body-part representations. The strict somatotopic version of the diagram, with a continuous and tidy body-map running from medial leg to lateral face, is a 1950 popularisation that does not appear in the underlying data.

Marc Schieber’s 2001 paper in the Journal of Neurophysiology argued, on the basis of accumulated single-unit recording data, that representations of single muscles and individual fingers are distributed and overlapping rather than spatially segregated. Tyson Aflalo’s 2015 Science paper from the Andersen lab at Caltech showed mixed body-part encoding at the single-unit level in the posterior parietal cortex of a tetraplegic human participant. Sergey Stavisky’s 2019 eLife paper found that the canonical “hand knob” modulates during speech and orofacial movements. Frank Willett’s own 2020 Cell paper, titled “Hand Knob Area of Premotor Cortex Represents the Whole Body in a Compositional Way,” demonstrated that the hand-knob already carries an interlinked, compositional representation of the entire body in BrainGate participants. Willett’s 2023 Nature paper on a speech neuroprosthesis reported that speech articulators (jaw, larynx, lips, tongue) are intermixed at the single-electrode level in area 6v. Evan Gordon and Nico Dosenbach’s 2023 Nature paper from Washington University in St Louis introduced the somato-cognitive action network (SCAN) and showed, using precision functional fMRI in individual humans, that the classical homunculus is interrupted by inter-effector regions that prefer multi-joint actions over single-effector control. Nature’s own editor summary at the time called the classical homunculus “discontinuous.” A Scientific American article by Dosenbach published in April 2023 was titled “How Our Team Overturned the 90-Year-Old Metaphor of a ‘Little Man’ in the Brain Who Controls Movement.”

The June 2026 Deo, Henderson and Willett paper enters this body of work as the highest-resolution and largest-cohort human single-neuron demonstration to date, not as the originator of the revisionist argument. The Stanford Report’s own framing on the day of publication, “Study upends decades-old map,” is institutional press shorthand for a more incremental scientific contribution.

Scale, spatial architecture, and the residual coarse map

Eight BrainGate2 and Pittsburgh-Chicago participants, 20 arrays, and the crown of the precentral gyrus systematically sampled at single-neuron resolution carry the paper’s scale claim. No prior paper that the authors cite had assembled a dataset of comparable breadth in living humans.

Two speech-preferential areas flanking a broadly tuned, orofacial-dominant zone in the inferior portion of the precentral gyrus is not, to the authors’ knowledge, present in the prior peer-reviewed literature. Earlier work, including Willett’s own 2023 speech neuroprosthesis paper, had reported intermixed articulator representation. The 2026 paper resolves a finer spatial structure than that earlier work could detect.

Whole-body representation at every sampled location across the precentral gyrus, while the relative strength of body parts remains roughly consistent with the classical homunculus’s coarse organisation, is the paper’s synthesis claim. The classical relative-strength gradient is preserved as a residual signal underneath the multi-effector distribution. The authors, in their own abstract phrasing, describe this as “an intermixed, interrelated and behaviour-centred organization of the motor cortex” with both a mosaic structure and a residual coarse map.

Implications for the implantable BCI cohort

Existing implantable arrays were not placed in the wrong location. Single-array recording in any region of the precentral gyrus carries decodable signals for a wider range of body parts and behaviours than the classical homunculus would predict, with three operational consequences for the commercial cohort.

Neuralink’s N1 device, with 1,024 electrode sites distributed across 64 threads placed in motor cortex, and Paradromics’ 421-electrode intracortical Connexus device, granted an FDA Investigational Device Exemption in November 2025 and implanted in its first Connect-One early feasibility study participant at the University of Michigan Health system in June 2026 under NCT07357428, both sit in the hand-knob-anchored category. The 2026 Deo paper implies these devices may, in principle, also decode signals from other body parts, including some speech and orofacial signals, from the same hardware.

Speech BCI placement strategies based on Willett’s 2023 speech neuroprosthesis paper used a single-region target. The 2026 paper identifies two speech-preferential areas with a broadly tuned orofacial-dominant zone between them, with direct implications for the next generation of speech neuroprostheses. Whether dual-array placement targeting the two speech-preferential areas can outperform single-array strategies is now a testable engineering question.

Synchron’s Stentrode, a 16-channel endovascular electrode placed in the superior sagittal sinus that records local field potentials through the sinus wall, and Precision Neuroscience’s Layer 7, a 1,024-microelectrode thin-film array placed on the cortical surface and cleared by the FDA under 510(k) (K242618) in April 2025 for up to 30-day implantation, each capture aggregated signals from the same multi-effector neural population that single-neuron intracortical arrays sample directly. Channel counts across modalities are not equivalent. Layer 7 records local field potentials at the cortical surface, Stentrode records local field potentials from inside a vessel, Neuralink and Paradromics record single-unit activity from inside the cortex. The 2026 paper does not adjudicate between these signal classes. It does suggest that the underlying neural population is multi-effector regardless of which signal class is being read.

What to watch

Replication across BrainGate2 sites (Massachusetts General Hospital, VA Providence Healthcare System, Stanford, Emory University, UC Davis, and Baylor College of Medicine), across the Pittsburgh-Chicago intracortical trial sites, and through Paradromics, Precision Neuroscience, and Synchron clinical-trial datasets adding further signal classes, would convert the bipartite-speech-area architecture from a single-paper observation into a consolidated mapping.

Array-placement strategy at the commercial implantable cohort would be visible through trial protocol amendments and through company technical disclosures. Absent a placement change, the paper functions as scientific validation for the existing single-array strategies the cohort has used since the early BrainGate2 trial arms.

Tier-1 general press coverage and the major specialist medtech trade press had not covered the paper eleven days after online publication. The visible coverage perimeter (Stanford Report, the Wu Tsai Neurosciences Institute release, and a small number of secondary outlets including MedicalXpress) leaves an open lane for institutional and commercial readings of the paper.

Sources

• Deo et al., “A mosaic of whole-body representations on the human precentral gyrus,” Nature, 17 June 2026 (DOI: 10.1038/s41586-026-10653-x)
• Stanford Report, “Study upends decades-old map of how the brain controls movement,” 17 June 2026
• Wu Tsai Neurosciences Institute, “It’s time to revamp the motor homunculus”
• Stanford Neural Prosthetics Translational Laboratory (NPTL)
• BrainGate consortium clinical trials page
• Stanford Engineering, “Krishna Shenoy, engineer who reimagined how the brain makes the body move, dies at 54,” 27 January 2023
• Penfield W, Boldrey E. “Somatic Motor And Sensory Representation In The Cerebral Cortex Of Man As Studied By Electrical Stimulation,” Brain 60(4):389-443, 1937
• Schieber MH. “Constraints on Somatotopic Organization in the Primary Motor Cortex,” J Neurophysiol 86(5):2125-2143, 2001
• Aflalo T et al. “Decoding motor imagery from the posterior parietal cortex of a tetraplegic human,” Science 348(6237):906-910, 2015
• Willett FR, Deo DR et al. “Hand Knob Area of Premotor Cortex Represents the Whole Body in a Compositional Way,” Cell 181(2):396-409.e26, 2020 (DOI: 10.1016/j.cell.2020.02.043)
• Willett FR et al. “A high-performance speech neuroprosthesis,” Nature 620:1031-1036, 2023
• Gordon EM, Dosenbach NUF et al. “A somato-cognitive action network alternates with effector regions in motor cortex,” Nature 617:351-359, 2023
• Inside BCI: UC Davis 3,800-hour at-home BCI, 16 June 2026 · Paradromics first Connexus implant, 17 June 2026 · Rockefeller ventral premotor cortex abstract thought decoding, 26 May 2026