The superior colliculus sits in the midbrain and serves as a major target of the optic nerve. Decades of work in non-human animals have established that it is organized into layers with distinct functions. Neurons in the superficial layers respond to visual stimuli arranged in retinotopic maps, while neurons in the intermediate and deep layers respond to visual, auditory and somatosensory signals, often in combination, and help control movements of the eyes, head and body. Evidence for a similar organization in humans existed, but it remained incomplete, largely because the superior colliculus is too small to resolve reliably with standard 3-Tesla fMRI, whose typical voxel sizes of two to three millimeters blur the layers together and risk contamination from neighboring structures such as the periaqueductal gray and cerebrospinal fluid.

The research team, led by Danlei Chen of the Massachusetts Institute of Technology and Massachusetts General Hospital together with colleagues at Northeastern University and collaborating institutions, overcame this limitation by scanning participants at 7 Tesla with a voxel size of just 1.1 millimeters isotropic. They also developed a careful preprocessing pipeline: subject-specific masks excluded cerebrospinal fluid and the periaqueductal gray, and a secondary alignment procedure using DARTEL registration brought individual superior colliculus masks into a common space, raising overlap between subjects from roughly 89.5 percent to nearly 95 percent. This technical groundwork was essential, because the questions the team wanted to ask required separating signals arising from different depths within a structure only about six millimeters across in each hemisphere.

Participants performed a probabilistic avoidance learning task across five runs, totaling 120 trials. Half the participants received visual stimulation in the form of affective images from the International Affective Picture System; the other half received mechanical pressure applied to the nail bed of the left thumb. Each trial unfolded in three phases. First, two shapes appeared on screen. Then came a two-second pre-stimulus decision period, in which either the participant chose one shape with a right-handed button press or the computer selected a shape at random. Finally, after a jittered interval, the sensory stimulus was delivered. Crucially, during the pre-stimulus decision period the sensory input and motor demands were identical for both groups; only the modality of the upcoming stimulus differed.

During actual stimulation, the results matched the classical animal literature with striking fidelity. Visual stimulation produced stronger blood-oxygen-level-dependent, or BOLD, responses in the superficial layers of the superior colliculus, while somatosensory stimulation drove stronger responses in the deep layers, a depth-by-modality interaction that was highly significant. A second interaction served as a sanity check: pressure applied to the left thumb produced a stronger response in the right superior colliculus, consistent with contralateral sensory processing, whereas visual stimulation showed no left-right asymmetry. The somatosensory responses were also markedly larger in amplitude than the visual ones, underscoring how powerfully the deep layers register touch.

The surprise came in the period before any stimulus was delivered. During the pre-stimulus decision period, when the two groups saw the same shapes and performed the same task, the superior colliculus already displayed a layer-specific pattern corresponding to the modality of the upcoming stimulation. Participants about to receive somatosensory pressure showed significantly greater activity in deep than superficial layers, mirroring the reactive pattern. Participants about to view images showed a numerically greater superficial response, though the difference between layers did not reach statistical significance in that group. Because the immediate sensory and motor conditions were matched across groups, the only factor that could explain the difference was what each group expected to feel next.

The team took several steps to rule out alternative explanations. Carryover from previous trials was tested by modeling the shape presentation period that intervened between one trial’s stimulation and the next trial’s decision. That period showed a different pattern of depth-dependent activity, and classifiers trained on shape presentation maps could not distinguish visual from somatosensory participants at above-chance levels in any test period. Multivariate pattern analysis using linear support vector machines, with leave-one-subject-out cross-validation, confirmed the key result: classifiers trained on pre-stimulus decision maps could decode the upcoming sensory modality both within that period and, remarkably, in held-out sensory stimulation maps, and vice versa. The anticipatory and reactive signatures were statistically interchangeable, while the shape presentation period was distinct from both.

The predictive signals extended beyond the superior colliculus. In the lateral geniculate nucleus, the thalamic relay for vision, participants anticipating visual images showed greater BOLD responses than those anticipating touch, both during stimulation and during the pre-stimulus decision period. In the ventral posterolateral nucleus, the somatosensory relay, stimulation of the left thumb produced the expected right-lateralized response, and during anticipation the response became bilateral, consistent with a contralateral motor signal from the right-handed button press combined with an anticipatory signal for the upcoming left-thumb pressure. This convergence across multiple subcortical sensory structures suggests that anticipatory signals are distributed through a broader circuit, likely shaped by descending cortical projections that coordinate prediction across sensory pathways.

The authors interpret these findings through the lens of allostatic and predictive processing theories, which hold that the brain does not passively wait for sensory input but actively predicts and prepares for it based on past experience. On this view, a brain organized purely around reaction would be poorly suited to regulate behavior in a fast-changing environment. Prior work had shown anticipatory responses in the superior colliculus in the context of eye movements, where neurons predict the visual consequences of impending saccades, but the new study extends prediction into the somatosensory domain and demonstrates it in living humans at the level of laminar organization. The authors also note caveats: the anticipatory somatosensory signal could partly reflect preparation for an unpleasant stimulus rather than a specific prediction about the thumb, eye movements were not recorded during scanning, and partial volume effects from adjacent structures cannot be entirely excluded, though control analyses suggest they do not account for the results.

What emerges is a revised picture of a structure often caricatured as primitive. The superior colliculus, and the thalamic nuclei alongside it, appear to be active participants in the brain’s predictive machinery, carrying modality-specific signals about expected sensory events before those events reach sensory surfaces. The study confirms that the laminar organization long documented in animal research is a genuine feature of the human brain, and it reframes subcortical nuclei not as passive relay stations but as nodes in distributed networks that anticipate the world. As the authors conclude, these structures both react to and predict sensory input, a duality that may prove fundamental to how the brain efficiently guides attention and action in an uncertain environment.

Subject of Research: Predictive and modality-specific sensory processing in the human superior colliculus measured with 7-Tesla fMRI

Article Title: Modality-specific reactive and predictive responses in the human superior colliculus

Article References: Chen, D., Barton-Zuckerman, M., Kross, Z., Kragel, P. A., Savoca, P. W., Wald, L. L., Bianciardi, M., Wager, T. D., Choi, J.-K., Zhang, J., Quigley, K. S., Satpute, A. B., Barrett, L. F., & Theriault-Brown, J. E. (2026). Modality-specific reactive and predictive responses in the human superior colliculus. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02480-0

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02480-0

Keywords: superior colliculus, 7-Tesla fMRI, predictive processing, allostasis, somatosensory, visual processing, midbrain, thalamic nuclei, lateral geniculate nucleus, ventral posterolateral nucleus, BOLD signal, layer-specific activity