Bridging the Gap Between Functional andAnatomical Features of Cortico-Cerebellar Circuits
Using Meta-Analytic Connectivity Modeling
Joshua H. Balsters,1,2 Angela R. Laird,3 Peter T. Fox,4,5 andSimon B. Eickhoff6,7*
1Neural Control of Movement Lab, Department of Health Sciences and Technology,ETH Zurich, Switzerland
2Trinity College Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland3Department of Physics, Florida International University, Miami, Florida
4Research Imaging Center, University of Texas Health Science Center San Antonio,San Antonio, Texas
5South Texas Veterans Administration Medical Center, San Antonio, Texas6Institute of Neuroscience and Medicine (INM-1), Research Center Julich, Germany
7Institute of Clinical Neuroscience and Medical Psychology, Heinrich-HeineUniversity Dusseldorf, Germany
Abstract: Theories positing that the cerebellum contributes to cognitive as well as motor control aredriven by two sources of information: (1) studies highlighting connections between the cerebellum andboth prefrontal and motor territories, (2) functional neuroimaging studies demonstrating cerebellar activa-tions evoked during the performance of both cognitive and motor tasks. However, almost no studies todate have combined these two sources of information and investigated cortico-cerebellar connectivityduring task performance. Through the use of a novel neuroimaging tool (Meta-Analytic ConnectivityModelling) we demonstrate for the first time that cortico-cerebellar connectivity patterns seen in anatomi-cal studies and resting fMRI are also present during task performance. Consistent with human and non-human primate anatomical studies cerebellar lobules Crus I and II were significantly coactivated withprefrontal and parietal cortices during task performance, whilst lobules HV, HVI, HVIIb, and HVIIIwere significantly coactivated with the pre- and postcentral gyrus. An analysis of the behavioral domainsshowed that these circuits were driven by distinct tasks. Prefrontal-parietal-cerebellar circuits were moreactive during cognitive and emotion tasks whilst motor-cerebellar circuits were more active duringaction execution tasks. These results highlight the separation of prefrontal and motor cortico-cerebellarloops during task performance, and further demonstrate that activity within these circuits relates to dis-tinct functions. Hum Brain Mapp 00:000000, 2013. VC 2013 Wiley Periodicals, Inc.
Key words: cerebellum; meta-analytic connectivity modeling; cognition
Additional Supporting Information may be found in the onlineversion of this article.
*Correspondence to: Simon B. Eickhoff, Institut fur Neurowissen-schaften und Medizin (INM-1), Forschungszentrum Julich GmbH,D-52425 Julich, Germany. E-mail: S.Eickhoff@fz-juelich.de
Received for publication 8 February 2013; Revised 29 July 2013;Accepted 31 July 2013.
DOI 10.1002/hbm.22392Published online 00 Month 2013 in Wiley Online Library(wileyonlinelibrary.com).
r Human Brain Mapping 00:0000 (2013) r
VC 2013 Wiley Periodicals, Inc.
A number of authors have suggested that in order tounderstand the functional properties of a brain region onemust understand its anatomical features and connections[Crick and Koch, 2005; Eickhoff and Grefkes, 2011; Pas-singham et al., 2002]. A great deal is known about theintrinsic microstructure of the cerebellum [Eccles et al.,1967], and a large number of studies have mapped cortico-pontine and cortico-cerebellar connections in humans andnon-human primates [see Ramnani, 2011; Strick et al., 2009for review]. Theories of cortico-cerebellar informationprocessing have been in a large part driven by our under-standing of cortico-cerebellar connectivity. Studies in bothhumans [Buckner et al., 2011; Habas et al., 2009; Krienenand Buckner, 2009; OReilly et al., 2010; Ramnani et al.,2006] and non-human primates [Kelly and Strick, 2003;Middleton and Strick, 2000, 2001; Schmahmann and Pan-dya, 1997] have repeatedly demonstrated that the cerebel-lum receives inputs from a wide range of corticalterritories including (but not restricted to) the premotorand primary motor cortices, medial and dorsal prefrontalcortex, and parietal cortex. Studies in nonhuman primateshave also suggested that prefrontal and motor cortico-cerebellar circuits are completely independent of oneanother and do not exchange information at any pointwithin the loop except for within the frontal lobe. Kellyand Strick  showed in non-human primates that thearm area of the primary motor cortex projected to cerebel-lar lobules HV, HVI, HVIIb, and HVIII, whilst tracer labelinjected into the dorsal bank of the sulcus principalis(putatively Walkers Area 46) terminated in cerebellarlobules Crus I and Crus II. These same connections havebeen shown in humans using resting state fMRI [Buckneret al., 2011; Habas et al., 2009; Krienen and Buckner, 2009;OReilly et al., 2010]. Given that the cerebellum receivesinputs from prefrontal and parietal regions that are knownto process abstract information [Badre and DEsposito,2009], and that this information does not integrate withmotor cortico-cerebellar circuits, it would suggest that thecerebellum is not solely processing motor information.However, in order to further develop theories of cortico-cerebellar connectivity it is necessary to corroborate thesefindings with task-based information.
Along with studies of anatomical and functional con-nectivity, task-based functional neuroimaging studieshave provided a wealth of evidence suggesting that thecerebellum is involved in processing both motor and non-motor information [see Stoodley, 2012 for review]. Petac-chi et al., , Moulton et al. , and Stoodley andSchmahmann  have all conducted meta-analysesinvestigating task-dependent cerebellar processing. WhilstPetacchi et al.  and Moulton et al.,  focusedon auditory and pain processing respectively, Stoodleyand Schmahmann  investigated cerebellar process-ing during a variety of tasks ranging from cognitive tomotor to emotion. They found that cerebellar lobules
Crus I and II were active in studies investigating execu-tive function, working memory, and language tasks,whilst motor control tasks consistently activated cerebellarlobules HV, HVI, and HVIII. This work thus providesfurther evidence that distinct regions of the cerebellumprocess distinct forms of information, both motor andassociative. Although these findings are in keeping withcortico-cerebellar anatomy (i.e., cerebellar lobules inter-connected with prefrontal cortex are active during asso-ciative tasks) it is essential to investigate cortico-cerebellarconnectivity during task performance in order to ascertainthe roles of cortico-cerebellar circuits in cognitive andmotor control.
This study uses a novel neuroimaging tool [Meta-Ana-lytic Connectivity Modelling (MACM)] to integrate con-nectivity information with behavioral information and assuch extend our understanding of cortico-cerebellar infor-mation processing. MACM assesses brain-wise co-activa-tion patterns of an anatomical region across a largenumber of databased neuroimaging results [Eickhoff et al.,2011; Laird et al., 2009a]. First, we identified for each voxelof the seed VOI those experiments in the BrainMap data-base that reported activation at that particular location. Byperforming an Activation Likelihood Estimation (ALE)meta-analysis over these experiments, we can generate awhole brain activation map showing all the brain regionsthat are active when voxels in the seed VOI are active. Dif-ferences in the coactivation patterns of the respective VOIscan be tested by directly contrasting the regional MACMpatterns. Finally, in order to confirm a functional separa-tion between the anatomical VOIs selected in this studywe can assess the behavioral domain and paradigm classmeta-data of experiments associated with the ensuing clus-ters. This manuscript describes the application of MACMto cortico-cerebellar connectivity and the ensuing behav-ioral differences.
Cerebellar lobules of interest were selected based onprevious studies of primate cortico-cerebellar connectivity,specifically Kelly and Strick . Kelly and Strick is the only study performed on non-human primates thattraced anatomical connections from regions of the frontallobe (dorsal bank of the sulcus principalis (Walkers Area46; Walker, 1940) and the hand/arm region of the primarymotor cortex) all the way to the cerebellar cortex. Wedecided to restrict our analyses to cerebellar lobules foundin Kelly and Strick , namely vermal and hemisphericlobules V, VI, Crus I, Crus II, VIIb, VIIIa and HVIIIb(accounting for 86.34% of the cerebellar cortex; Diedrich-sen et al., 2009). There are also additional practical reasonsto restrict our analyses to these lobules. For example,many fMRI studies do not include the very posterior
r Balsters et al. r
r 2 r
lobules of the cerebellum in the field of view, thus thereare fewer studies reporting activations within lobules IXand X. Anterior lobules I-IV can also be contaminated bynon-cerebellar signal originating from the occipital lobesdirectly above them, i.e., the ventral visual cortex [Die-drichsen, 2006]. Cerebellar lobular masks were extractedfrom the probabilistic cerebellar atlas of Diedrichsen et al. and combined to create masks of interest (see Fig.1). For example, the seed mask for the analysis of cerebel-lar motor lobules was created by combining masks ofcerebellar lobules V, VI, VIIb, and VIII (red in Fig. 1). Theatlas of Diedrichsen et al.  conforms to the anatomi-cal landmarks outlined by Larsell and Jansen . Usingthese cerebellar lobules as seeds we investigated differen-ces in task-based connectivity between motor-projectingcerebellar lobules (V, VI, VIIb, VIII) and prefrontal-projecting cerebellar lobules (Crus I and II). We will addi-tionally investigate differences in task-based connectivitybetween anterior motor-projecting cerebellar lobules (V,VI) and posterior motor-projecting cerebellar lobules (VIIb,VIII), given that posterior motor-projecting cerebellarlobules have selectively expanded in humans compared tononhuman primates [Balsters et al., 2010].
Meta-Analytic Connectivity Modeling
The BrainMap database [www.brainmap.org; Fox andLancaster, 2002; Laird et al., 2005, 2009a, 2011] wasemployed for the retrieval of relevant neuroimaging
experiments. At the time of assessment, the databasecontained coordinates of reported activation foci andassociated meta-data of over 11,000 neuroimagingexperiments. For our analysis, only whole brain studies ofhealthy subjects reporting activation in standard stereo-taxic space were considered, while all experiments thatinvestigated age, gender, handedness, training effects orinvolved a clinical population were excluded. As the firststep of the analysis we identified (separately for each seedregion) all experiments that featured at least one focus ofactivation within the respective seed (MNI space). In orderto facilitate such filtering, coordinates from Talairach spacewere converted into MNI coordinates by using Lancastertransformation [Lancaster et al., 2007]. Then, all experi-ments activating the currently considered seed were iden-tified. The retrieval was solely based on reportedactivation coordinates, not on any anatomical or functionallabel.
Functional connectivity of the different seeds was eval-uated using meta-analytic connectivity modelling [MACM;Robinson et al., 2012, 2010]. The key idea behind MACMis to assess which brain regions are coactivated abovechance with a particular seed region in functional neuroi-maging experiments [Eickhoff et al., 2010; Laird et al.,2009b]. MACM entails to first identify all experiments in adatabase that activate a particular brain region asdescribed above and then test for convergence across (all)foci reported in these experiments. Obviously, as experi-ments were selected by activation in the seed, highest
Cerebellar lobular masks. Red lobules are classified as motor lobules (V, VI, VIIb, and VIII),
blue lobules are classified as prefrontal lobules (Crus I and Crus II). Masks are overlayed on
the FSL standard template moving from anterior-> posterior.
r Meta-Analytic Connectivity Modeling of Cortico-Cerebellar Circuits r
r 3 r
convergence will be observed in the seed region. Signifi-cant convergence of the reported foci in other brainregions, however, indicates consistent coactivation, i.e.,functional connectivity with the seed. The whole brainpeak coordinates of the identified experiments were down-loaded from BrainMap database for each seed region.Coordinates were analysed with the modified activationlikelihood estimation (ALE) algorithm [Eickhoff et al.,2009, 2012] to detect areas of convergence. This approachmodels each focus as a Gaussian distribution reflectingempirical estimates of the uncertainty of different spatialnormalization techniques and intersubject variability as afunction of the number of subjects. Modeled activation(MA) maps are calculated for each experiment by combin-ing the Gaussian distributions of the reported foci [Turkel-taub et al., 2012]. Taking the union across these yieldedvoxel-wise ALE scores that describe the convergence ofresults at each particular location of the brain. To distin-guish true convergence between studies from randomconvergence, i.e., noise, in the proposed revision of theALE algorithm [Eickhoff et al., 2012], ALE scores are com-pared to an empirical null-distribution reflecting a randomspatial association between experiments [Eickhoff et al.,2012; Turkeltaub et al., 2012]. The p-value of an observedALE is then given by the proportion of this null-distribution (precisely, its cumulative density function)corresponding equal or higher ALE values. The ALEmaps reflecting the convergence of coactivations with anyparticular seed region were subsequently thresholded atP< 0.05 cluster-level corrected (cluster-forming threshold:P< 0.001 at voxel-level) and converted into Z-scores fordisplay.
For further investigation of commonalities and distinc-tions between the functional connectivity of differentseeds, conjunction and difference analyses were per-formed. For conjunction analysis the minimum statistic[Nichols et al., 2005] was used, yielding voxels thatshowed significant values in both coactivation maps. Theresult corresponds to the intersection of the (cluster-levelcorrected) MACM maps [Caspers et al., 2010]. Differencemaps were established by calculating the voxel-wise differ-ences of the Z-scores obtained from the ALE maps of thetwo MACM analyses. When calculating difference maps,activation foci common to both conditions were removed.The difference maps were then tested against an ALE dif-ference map assuming the null-distribution, which wasgenerated from a r...