The cerebellum, traditionally considered a motor structure, is increasingly understood to play a broader role by virtue of its interactions with association cortex such as the parietal and prefrontal lobes. This shift has been driven by functional imaging and patient work in humans, and by anatomical studies in monkeys. Since the earliest days of functional imaging, cerebellar activations have been observed, often unexpectedly, in experiments with minimal motor demands including sensory (Gao et al. 1996
; Blakemore et al. 1999
) and linguistic tasks (Roskies et al. 2001
; Noppeney and Price 2002
; Xiang et al. 2003
; Ravizza et al. 2006
) and in executive function (e.g., Desmond et al. 1997
; for review see Stoodley and Schmahmann 2009
). In parallel, patient studies indicate wide-ranging cognitive deficits following cerebellar damage, including altered social and emotional behavior and a slowing of mental performance (e.g., Schmahmann and Sherman 1998
; Schmahmann and Caplan 2006
Anatomical studies in monkeys suggest that different cerebral–cortical projections form discrete “channels” and should therefore map onto different regions of the cerebellar cortex. Afferent fibers destined for the pontine nuclei (the cerebellar input nuclei) are segregated within their white-matter bundles by cortical region of origin (Schmahmann and Pandya 1992
), and retain this segregation in the pontine nuclei (Brodal 1978
). Similarly, cerebellar efferents arising in the dentate nucleus are organized according to the functional topography of the cerebral cortex (Dum and Strick 2003
). In one particularly important study in monkeys, Kelly and Strick (2003)
identified the polysynaptic connections between 2 cerebral–cortical areas, the primary motor cortex and area 46 of the prefrontal cortex, and specific territories in the cerebellar cortex. Cerebellar lobules HIV, HV, HVI, and HVIII were found to be interconnected with the primary motor cortex (a later study by Lu et al. (2007)
indicated that M1 also receives projections from Crus I). (We have followed the nomenclature of Schmahmann et al. (2000)
, “MRI Atlas of the Human Cerebellum” in which the cerebellar lobules are labeled I-X from the anterior/superior border of the cerebellum, through posterior, to the anterior/inferior border. Schmahmann's nomenclature was partly based on that of Larsell [e.g., Larsell and Jansen 1972
], in which the cerebellar hemispheres were distinguished from the vermis with the prefix H. We have included the H prefix where we are referring specifically to activity in the hemispheres, and the prefix “vermal” where we are referring specifically to vermal activity. Where we are referring to the whole lobule, we use no prefix.) In contrast, parts of Lobule VIIa, especially Crus II, were interconnected with prefrontal area 46. Importantly, the cerebellar-cortical regions which received input from each cortical area were found to send output back to the same cerebral area, forming parallel connectivity loops. In view of these results, Strick and colleagues have proposed that cerebro-cerebellar connectivity is characterized by discrete “parallel circuits,” reciprocally linking different parts of the cerebellum with their corresponding cerebral–cortical functional areas (e.g., Dum et al. 2002
However, despite extensive work mapping cortical connections with the pontine nuclei and dentate, the connectional topography of the cerebellar cortex itself (in relation to the cerebral cortex) remains largely unmapped. Anatomical tracer studies of cerebro-cerebellar connectivity have almost exclusively focused on the input and output nuclei of the cerebellum, because of the difficulty in tracing the multisynaptic circuits which link cerebellar and cerebral cortices (Kelly and Strick 2000
). Projections from the cerebral cortex synapse in the pontine nuclei, then the cerebellar cortex; reciprocal connections synapse first in the cerebellar dentate, then the thalamus, before reaching the cerebral cortex. Very few studies have successfully traced connections trans-synaptically from cerebral to cerebellar cortex (Kelly and Strick 2003
; Lu et al. 2007
). Furthermore, diffusion tractography, an imaging method that can provide information on anatomical connectivity in the human brain, is at present problematic between the cerebellum and the cerebral cortex for 3 reasons. First, cerebellar afferents and efferents decussate in regions of dense crossing fibers in the brainstem. Second, cerebellar efferents pass through a “bottleneck” in the superior cerebellar peduncle which is so narrow that its subregions cannot clearly be distinguished with the spatial resolution of diffusion imaging. Third, cerebellar-efferents synapse in areas of gray matter (notably the thalamus) before reaching the cerebral cortex. Finally, all anatomical studies face the difficulty that each region of cerebellar cortex receives inputs via at least 2 routes, relayed by either the pontine nuclei or the inferior olive.
Given the difficulty of tracing cortico-cerebellar anatomical connections “cortex to cortex,” it is at present difficult to relate our knowledge of the connectivity of the cerebellar input and output nuclei to a functional
topography of the cerebellar cortex. In accordance with the anatomical data of Kelly and Strick (2003)
, sensorimotor representations of the body have been found in the superior/anterior-most part of the cerebellum (lobules III–V or VI) and in lobule HVIII (MacKay and Murphy 1973
; Ojakangas and Ebner 1994
; Gao et al. 1996
; Jueptner et al. 1997
; Thickbroom et al. 2003
). However, using functional methods it is much more difficult to be specific about which areas of association cortex (prefrontal and posterior-parietal cortex) are linked with a cerebellar subregion because the functional roles of the cerebral–cortical areas in question are less clearly defined—there is probably no single task we could give to a monkey which, if associated with activity in a cerebellar neuron, would allow us to conclude that we had found a “cerebellar prefrontal” or “posterior-parietal” cell. Indeed, functional imaging studies indicate that the prefrontal cortex is generally active as part of a broad network of association cortex. Based on patient work, Schmahmann and colleagues (Schmahmann 1996
; Schmahmann and Sherman 1998
) have described a schema in which the posterior lobe of the cerebellum is involved with cognitive or executive functions: “in patients with lesions involving the posterior lobe of the cerebellum…[they observed] impairment of executive functions such as planning, set-shifting, verbal fluency, abstract reasoning and working memory; difficulties with spatial cognition including visual–spatial organization and memory…” (Schmahmann and Sherman 1998
)—but again, the executive functions described are not exclusively associated with one region of the cerebral cortex.
Despite these difficulties, an understanding of connectivity with the cerebral cortex is particularly important in the case of the cerebellum because the function of its subregions may be defined by their connectivity. The combination of the connectional diversity of the cerebellum with the extreme uniformity of cerebellar microcircuits fits a model in which the cerebellum applies a particular computational
function to information from a range of cortical areas, rather than having a functional specialization itself—motor, cognitive, or otherwise (Eccles et al. 1967
; Bloedel 1992
; Schmahmann and Sherman 1998
; Ramnani 2006
; Ito 2008
In the present study, we used resting-state functional magnetic resonance imaging (fMRI) to probe the topography of cerebral–cortical connectivity in the cerebellar cortex. This functional connectivity approach uses fMRI data acquired while the subject is at rest. The concept behind resting-state fMRI is that when the brain is “free-wheeling” (not involved in an externally cued task), correlations in slowly fluctuating spontaneous brain activity tend to reflect the intrinsic functional networks of the brain (Biswal et al 1995
; see Fox and Raichle 2007
for review). For example, cortical areas typically associated with motor function show a significant degree of covariation and are therefore thought to form a particular “resting state network” (RSN) while cortical areas associated with visual processing form a separate RSN (Beckmann et al. 2005
). Independent components analysis indicates that a large percentage of the fMRI signal in the resting human brain can be explained in terms of just a few (8–10) RSNs (Damoiseaux et al. 2006
). These primary RSNs are highly consistent across time and space and between individuals, suggesting they represent something fundamental about the functional organization of the brain. Further discussion of the resting-state approach and its advantages and caveats is given in the Discussion.
We mapped resting-state functional connectivity voxel-wise across the cerebellar cortex for a set of cortical regions or masks. The result of this analysis was a set of correlation maps across the cerebellar cortex, representing resting functional connectivity with each of a number of cortical regions. The approach used in this study is particularly useful for investigating whether subdivisions exist within a structure, because mapping within the structure of interest is voxel-wise. We defined a set of cerebral–cortical masks representing known functional systems, and tested the correlation of each cerebellar voxel with these cortical masks. In contrast, a typical resting-state approach would be to divide the cerebellum into regions a priori, and then use the cortical RSNs associated with each cerebellar region to infer its function or connectivity. The present approach is similar to that introduced by Zhang et al. (2008)
to map thalamo-cortical connectivity. We used the method to address a series of specific hypotheses about cerebro-cerebellar connectivity.
First, we asked whether separate motor and prefrontal zones could be defined in the cerebellar cortex, based on resting functional connectivity. The motivation for this analysis was to replicate the findings of Kelly and Strick (2003
, described above) with the resting state method; the division between motor and prefrontal also reflects the broad distinction between motor and executive zones proposed by Schmahmann (Schmahmann and Sherman 1998
Second, we further probed the validity of the resting-state approach through the lateralization of correlations. The white matter connections between the cerebellum and the cerebral cortex are crossed, so if resting-state correlations reflect neural connectivity, the correlation maps should show a contralateral organization, with voxels in the left cerebellum correlating more strongly with right cortical regions and vice versa. Such contralateral relationships are much more likely to be neural than vascular in origin. We might a priori expect artifactual (non-neurally generated) correlations to be stronger in the nearer, ipsilateral hemisphere. We therefore compared the correlation maps for left- and right-hemisphere cortical masks (using motor and prefrontal cortical masks as above).
Third, we asked how the broad distinction between motor and prefrontal zones related to other cortical areas. We extended the set of cortical masks to 6 large zones, representing different functional systems (prefrontal cortex, motor and premotor cortex, somatosensory cortex, posterior-parietal cortex, superior temporal cortex, and visual area middle temporal [MT]). Between them, the masks covered the regions of cortex reported to have significant cerebellar-afferent connections with neurons in the pontine nuclei or cerebellar-efferent connections with the dentate in nonhuman primates (see above).
Finally, we focused on the connectivity of the cerebellum with supramodal association cortex (prefrontal and posterior-parietal cortex), and asked which subregions within
the prefrontal and posterior-parietal masks contributed to the correlation patterns we observed in the cerebellum. The motivation for this analysis was that some researchers (Glickstein 2007
) have suggested that only the subregions of prefrontal and parietal cortex which are involved in motor control are linked to the cerebellum. If this were the case, we would expect to see the strong correlation with the cerebellar supramodal zone in the frontal eye fields and/or area 8 in the case of the prefrontal cortex mask, and for the parietal cortex, the anterior intraparietal cortex (aIP), which is involved in grasping, and superior parietal lobule which contains many regions involved in the planning of action, thought to be the homologs of monkey intraparietal sulcus (IPS) regions (see Culham and Valyear 2006
for review) including lateral intraparietal area (LIP; eye movements), medial intraparietal area (MIP; reaching), and ventral intraprietal area (VIP; movements in head centered space).
In the Supplementary Information, we present additional analyses mapping correlation with the Eigen time series of each mask across the cerebral cortex. These additional analyses give a picture of which regions within each mask contribute most strongly to the correlations described below, and indicate the strength of correlation between the cortical masks.