Abstract
While the cerebellum is instrumental for motor control, it is not traditionally implicated in vision. Here, we report the existence of 5 ipsilateral visual field maps in the human cerebellum. These maps are located within the oculomotor vermis and cerebellar nodes of the dorsal attention and visual networks. These findings imply that the cerebellum is closely involved in visuospatial cognition, and that its contributions are anchored in sensory coordinates.
The purported role of the cerebellum has shifted from one that is exclusively sensorimotor-related to one that encompasses a wide range of cognitive and associative functions1. In fact, the majority of cerebellar cortex is functionally connected to a range of cognitive and associative cerebral networks2 and is coherently activated by cognitive and affective tasks3,4. Within sensorimotor areas of the cerebellum, a representation of the body (or ‘homunculus’) characterizes functional organization5. Yet, in the remaining cerebellar cognitive and associative networks, functional organization remains less well understood.
In cerebral cortex6-8 and subcortex9, orderly progressions of the visual field characterize functional organization in brain areas important for visual perception and cognition. Some of these areas are referred to as the dorsal attention network10, which was recently shown to have cerebellar counterparts11. One of these nodes was suggested to topographically encode visual field location12.
Here, we used the HCP retinotopy dataset13 to address the question of cerebellar visuospatial organization. The dataset contains population receptive field (pRF)14,15 parameters for the whole brain fitted on high-resolution 7T BOLD responses to visual retinotopic stimulation during fixation. The pRF model describes voxels’ visual field response preferences with a limited set of spatial parameters. The pRFs were fitted on data from individual subjects (N=181), and on an across-subject time course average (HCP ‘average subject’). As BOLD SNR is relatively low in the cerebellum this dataset provides an unprecedented opportunity to address cerebellar retinotopic organization, which we here use to identify 5 retinotopic maps in 3 separate cerebellar clusters. Fig. 1 shows two example pRFs in the cerebellum, as well as pRF polar angle in the cerebellar volume.
To better inspect the topographic structure of visual-spatial representations in cerebellum, we projected pRF parameters for each cerebellar voxel onto a flattened representation of the cerebellum16 (Fig. 2a-d; see Supplementary Figs. 1 and 2 for results in the volume, Supplementary Fig. 3 for individual subject results, and Methods). This revealed three clusters where the pRF model explained considerable variance (see Supplementary Fig. 4 for the voxel selection procedure). We refer to the clusters as OMV, VIIb and VIIIb (see Fig. 2 g-h upper panels). Fig 2c shows that the distribution of pRF centers within each cluster is characterized by representations of the ipsilateral visual field. This is in contrast to the contralateral visual field representations in subcortical and cortical retinotopic areas13 (see Supplementary Fig. 2). Yet, it matches the ipsilaterality of the cerebellar homunculi5, resulting from midline crossing of cerebellar connective fibers in the pons17. Quantifications of the progressions of polar angle (Fig. 2g) reveal a double representation of the lower visual field in OMV and VIIIb, as is common in cerebral visual cortex. Finally, smooth variations in preferred eccentricity take place in the direction roughly orthogonal to the direction of polar angle phase reversals, again mirroring the organization of cerebral visual cortex (Fig 2h). Fig 2. e-f provides a visual model summary of these retinotopic properties.
We next analyzed whether standard retinotopic properties (such as overrepresentation of the fovea and a strong correlation between pRF eccentricity and size14) were also present in the cerebellar visual field maps (Fig. 3). As Fig. 2g revealed double representations of the visual field in OMV and VIIIb, we split these clusters into a medial and lateral portion along their respective polar angle reversals (Fig 3a). Visualizing the eccentricity distributions (Fig 3b) quantifies the observation described above that eccentricity coverage is peri-foveal in OMV, extends somewhat into the periphery in VIIIb, and covers the full range of stimulated eccentricities in VIIIb. Importantly, Fig 3c reveals clear increases in pRF size with increasing eccentricity. Finally, Fig 3d highlights (1) the strong ipsilateral visual field representations, and (2) a strong overrepresentation of the lower visual hemifield in VIIIb. Together, this shows that the cerebellar visual field maps follow both known properties of retinotopic organization, albeit with unique idiosyncrasies (i.e. ipsilaterality, and strong overrepresentations of the fovea in OMV and of the lower visual field in OMVlat and VIIb).
The oculomotor vermis (OMV) is implicated in the deployment of spatial attention and in the generation and adaptation of saccades18. Direction selectivity of OMV Purkinje cells was shown to arise as a function of saccade error direction, and are also organized along an anatomical gradient19. In addition, OMV neurons encode saccade amplitude by the duration of a population response rather than by amplitude tuning20. This could potentially explain the incomplete eccentricity coverages we find in the OMV cluster.
The anatomical location and extent of clusters VIIb and VIIIb overlap closely with cerebellar nodes of the dorsal attention network2. Visual field coverage was strongly biased to the lower visual field in VIIIb. Behavioral performance is known to be superior in the lower compared to the upper visual field for stimuli that are associated with visuomotor coordination21. In addition, activity in area VIIIb22 was shown to be related to (the observation of) reaching and grasping movements. This suggests that cluster VIIIb may be involved in the integration of visuospatial information for the guidance of effector movements.
Our results uncover 5 visual field maps in three retinotopically organized clusters in the cerebellum. This implies a much closer involvement of the cerebellum in visuospatial cognition than is classically assumed.
Methods
Data set
The pRF results presented in this manuscript are part of the 7T HCP retinotopy dataset. Please see the accompanying publication13 for details on data collection and model fitting procedures, and for links to the online repositories. Briefly, 181 subjects performed a discrimination task at the fixation mark while viewing expanding and contracting rings, rotating wedges and traversing bar stimuli filled with fast-changing, random visual stimuli, for a total of approximately 30 minutes scan-time. The maximum eccentricity for these stimuli was 8 degrees of visual angle. Visual selectivity for each ‘gray-ordinate’ is modeled as a population receptive field (pRF) model. This is a uniform Gaussian distribution with free parameters of center (x and y), size (standard deviation), amplitude (with fixed sub-additive normalization constant of 0.05) and a baseline parameter.
Voxel selection procedure
In order to examine voxels that respond robustly to retinotopic stimuli, we first dismissed voxels where the pRF model explained little variance (see Supplementary Figure 4, first column; thresholds determined in original paper13 at 9.8% for the average and 2.2% for the individual subjects). Second, the non-linear spatial transformations that were employed to align data across subjects resulted in activity from ventral visual cortex to be smoothed into the cerebellar cortex. We were able to identify these voxels as these voxels were located between the cerebrum and cerebellum, and as these voxels were characterized by stark deviations in pRF parameter values (polar angle, eccentricity, size and explained variance; see areas indicated by white ovals in Supplementary Figure 4). The resulting mask left many voxels with extremely low eccentricity and size, without clear polar angle progressions across voxels. We hypothesized the following as a generative mechanism for these voxels’ results. As subjects performed a task on the fixation mark, this task became periodically more difficult when the retinotopic mapping stimulus passed behind the fixation mark. This means that responses of voxels sensitive to cognitive effort expended to maintain fixation (in a space-invariant manner) are in fact well captured by an extremely small and foveal pRF model. We therefore included an additional and conservative ‘fixation mark’ mask of voxels that had both eccentricity and size smaller than 0.15 degrees of visual angle (see Supplementary Figure 4, third column).
Cerebellar flatmaps
We used the SUIT toolbox16 to project pRF results from three-dimensional volume space onto a flattened two-dimensional representation. Note that this flattened representation is compressed in the vertical dimension relative to a flattened representation that takes into account microscopic folding of individual cerebellum anatomy23.
Subject ranking
In order to provide an estimate of the stability of the retinotopic maps in individual subjects (Supplementary Fig. 3), we ranked subjects based on the median explained variance across voxels within the three retinotopic clusters determined in the average subject. In creating visualizations of polar angles in these subjects, we masked voxels that fell outside the three retinotopic clusters as identified in the average subject and that were below the individual subject explained variance threshold of 2.2% (as determined in the HCP retinotopy paper13).
Code availability
The analysis code for creating the figures presented in this manuscript will be published, upon acceptance under www.github.com/daanvanes/hcp_cerebellum_retinotopy.