Abstract
Introduction:
The deep layers of superior colliculus (SC) integrate sensory information from multiple modalities to create a coherent sensory representation of the world [1,2]. However, information about the human SC is limited due to the technical challenge of imaging small structures deep within the cranium [3]. Advances in ultra-high field MRI enable imaging with greater signal-to-noise ratio in smaller voxels, allowing us to probe functional responses within SC [4]. To understand how human SC integrates information across somatosensory and visual modalities, we utilized functional MRI (fMRI) at 9.4T during an integration task.
Methods:
We collected fMRI from 5 individuals at 9.4T using a 16-channel transmit/31 receive array [5]. Participants performed a somatovisual integration task in which air puffs delivered to their fingers cued them to attend (but not saccade) to a quadrant of the visual field. Participants were asked to count the number of "+" signs that appeared in dot patterns in the cued quadrant while ignoring "X" and other random patterns. Single air puffs were continuously alternately presented to the index and ring fingers; a random double air puff cued visual attention to the upper (via index finger stimulation) or lower (via ring finger stimulation) visual fields. Stimulation alternated between the left and right hands (cuing left and right visual fields) every 15 seconds, enabling sinoidal data analysis.
In one participant, a second session used a visually cued paradigm with no tactile stimulation, allowing us to compare visual-only to somatovisual collicular processing.
Functional images (point-spread function-corrected EPI) were collected with 1 mm isotropic voxels over 26 slices which include the colliculi and most of early visual cortex (TR = 1.25 s). T1-weighted anatomical images were acquired with an MP2RAGE sequence (0.6-mm isotropic voxels).
Brain regions were initially segmented from the T1-weighted images using FreeSurfer, followed by manual adjustment. Next, a level-set depth-mapping approach was used to compute unique associations-streamlines-from the collicular surface to the cerebral aqueduct, enabling quantification of BOLD responses as a function of collicular depth.
Functional data were processed using a variant of the MrVista package. We corrected data for slice timing and motion and then fit a sinusoid (with frequency matching the left-right stimulus alternation) to each voxel time series, extracting amplitude, phase, and coherence. Next, using the depth streamlines, we averaged responses at superficial (0.6–1.8 mm) and deep (3.5–5.5 mm) levels.
Results:
We found strong lateralization of BOLD responses in the SC in all participants, with the attended visual hemifield having increased contralateral collicular activity. Activation was widespread in rostral SC at multiple depths. In rostral SC at superficial depths, BOLD responses were strongly lateralized, contralateral to the attended visual stimulus. In caudal SC, where deep somatosensory processing is expected for fore-limb stimulation [1], we saw activation in at least one SC (Figure 1 bottom). Indeed, compared to the BOLD phase in a visual-only task, deep caudal SC was significantly stronger (p < 0.01) in the somatovisual integration task (Figure 2).
Conclusions:
Using high resolution fMRI, we identified regions in SC that respond to somatovisual integration. These correspond to deep layers of the SC, which are believed to represent multisensory information onto a visuotopic map. In confirmation, a visual-only version of the task resulted in much weaker responses in deep caudolateral SC, while maintaining strong responses in the rostral superficial SC, corresponding to predominantly visual layers that represent the visual stimulus. Overall, we found that ultra-high field fMRI is sensitive to somatosensory integration in deep layers of human superior colliculus, which to this point has only been accessible in animal models.
