|Home | About | Journals | Submit | Contact Us | Français|
Convergent data from various scientific approaches strongly implicate cerebellar systems in non-motor functions. The functional anatomy of these systems has been pieced together from disparate sources such as animal studies, lesion studies in humans, and structural and functional imaging studies in humans. To better define this distinct functional anatomy, in the current study we delineate the role of the cerebellum in several non-motor systems simultaneously and in the same subjects using resting state functional connectivity MRI. Independent component analysis (ICA) was applied to resting state data from two independent datasets to identify common cerebellar contributions to several previously identified intrinsic connectivity networks (ICNs) involved in executive control, episodic memory/self-reflection, salience detection, and sensorimotor function. We found distinct cerebellar contributions to each of these ICNs. The neocerebellum participates in: 1. the right and left executive control networks (especially crus I and II), 2. the salience network (lobule VI), and 3. the default-mode network (lobule IX). Little to no overlap was detected between these cerebellar regions and the sensorimotor cerebellum (lobules V–VI). Clusters were also located in pontine and dentate nuclei, prominent points of convergence for cerebellar input and output respectively. The results suggest that the most phylogenetically recent part of the cerebellum, particularly crus I and II make contributions to parallel cortico-cerebellar loops involved in executive control, salience detection, and episodic memory/self-reflection. The largest portions of the neocerebellum take part in the executive control network implicated in higher cognitive functions such as working memory.
Resting-state functional connectivity studies represent a rapidly growing subfield of human brain mapping. Biswal et al. (1995) were the first to demonstrate the potential of this approach showing that low-frequency fluctuations of the blood-oxygen level dependent (BOLD) signal of the left motor cortex were temporally correlated with fluctuations in the right motor cortex and the bilateral somatosensory cortex. Subsequently, resting-state functional MRI (fMRI) studies have replicated this sensorimotor network (Xiong et al. 1999; Beckmann et al. 2005) and demonstrated additional intrinsic connectivity networks (ICNs) corresponding to basic functions such as vision, audition, language, episodic memory, executive function, and salience detection (Cordes et al. 2000; Fox et al. 2005; Fransson, 2005; Greicius et al. 2003; Kiviniemi et al. 2003; Seeley et al. 2007). Multimodal imaging has demonstrated that functional connectivity in these ICNs has electrophysiological correlates (Laufs et al. 2003) and reflects underlying structural connectivity in humans (Greicius et al. 2008) and non-human primates (Vincent et al. 2007). Functions have been attributed to the various ICNs based on their resemblance to networks activated by specific tasks, their relationship to specific cognitive disorders, and their correlation with cognitive and affective measures acquired outside the scanner (Greicius et al. 2004; Fox et al. 2005; Seeley et al. 2007). The focus in these studies has been on cortical contributions to ICNs. Some efforts have been made to examine subcortical contributions, but these have generally been limited to the basal ganglia, thalamus, and hypothalamus with scant attention to cerebellar contributions. Only two prior studies have explored resting-state cerebellar connectivity. He et al. (2004) and Allen et al. (2005) used a limited region-of-interest (ROI) approach to demonstrate cerebellar connectivity with the basal ganglia, thalamus, prefrontal and parietal cortices.
The general lack of attention to the cerebellum in ICNs is in keeping with a long-standing tendency to underplay cerebellar contributions to non-motor functions. This is despite the fact that a staggering amount of convergent, multimodal data has been gathered to build a strong case for the role of the cerebellum in various cognitive and affective functions. Several studies have substantiated prominent connections from associative and limbic cortices and hypothalamus to the neocerebellum (including the dentate nuclei) (Schmahmann, 1996; Haines et al. 1997; Middleton and Strick, 1997, 2001; Dum and Strick, 2003). These cerebral afferents quantitatively increase from apes to human (Ramnani et al. 2005), and mainly reach the most phylogenetically recent parts of the cerebellum: lobules VI–VII (Kelly and Strick, 2003). A recent meta-analysis of neuroimaging studies of the neocerebellum has documented its role in emotion, language, working memory and executive functions (Stoodley and Schmahmann, 2008). Furthermore, Schmahmann and Sherman (1998) followed by Levisohn et al. (2000) have described a variety of cognitive and affective impairments in patients suffering from focal cerebellar lesions. These impairments, broadly referred to as dysmetria of thought and emotion, were mainly observed in lesions of lobules VI–VII. Cognitive deficits in executive functions, memory and spatial cognition occurred with hemispheric lesions whereas affective disturbances occurred with vermian lesions. Therefore, converging data strongly suggest that the human neocerebellum contributes to parallel associative cerebrocerebellar networks involved in various aspects of cognition and emotion (Schmahmann, 2004). Despite this burgeoning evidence, cerebellar involvement in cognition is still a matter of some debate. For instance, some studies have failed to find significant cognitive impairment in cerebellar lesion patients and in those that have the deficts are often minor and potentially confounded by motor or oculomotor task demands (Helmuth et al. 1997; Thier et al. 1999; Haarmeier and Thier, 2007).
In the current study we sought explicitly to explore the role of the cerebellum in several previously defined ICNs. We hypothesized that the sensorimotor ICN would incorporate clusters in the motor regions of the cerebellum whereas the several ICNs linked to cognitive or affective processing would incorporate distinct neocerebellar clusters.
To test our hypothesis we first applied independent component analysis (ICA) to the resting-state fMRI data of 15 healthy control subjects. Using an unbiased template-matching procedure (Greicius et al. 2004) we identified the following ICNs: the default-mode network (DMN) (Greicius et al. 2003, 2004), the executive control network (Seeley et al. 2007) (ECN, divided by ICA into left and right hemisphere ECNs), the salience network (Seeley et al. 2007), and the sensorimotor network (Biswal et al. 1995; Xiong et al. 1999). The same 5 ICNs were then identified in a second dataset in order to examine, in detail, the replicable cerebellar connectivity in each ICN. Subjects in both datasets also underwent structural scans and various task-activation fMRI scans which were not used in this study.
Fifteen healthy subjects (ages: 19–40; mean age: 26.5; nine females, all right-handed) were scanned after giving written informed consent.
Functional images were acquired on a whole-body 3T scanner (Signa Horizon; General Electric Healthcare, Milwaukee, Wis.), using an eight-channel head coil. In each scanning sequence, thirty-two contiguous axial T2*-weighted gradient echo-planar images (echo time 40 ms, repetition time 2500 ms, field of view 30×30, matrix 128×128 mm zero filled to 256×256 mm, slice thickness 4 mm, interslice gap 0 mm), were obtained to encompass the entire brain, brainstem and cerebellum. Two hundred-sixteen volumes were acquired for the resting-state functional scan with four “dummy” volumes acquired at the start of the session to allow for steady-state magnetization.
For the resting-state scan subjects were instructed to keep their eyes closed and try to hold still. The scan lasted 9 minutes and 10 seconds.
Functional MRI data were format converted from dicom to analyze format using MRIcro (http://www.sph.sc.edu/comd/rorden/mricro.html) and then analyzed using SPM5 analysis software (http://www.fil.ion.ucl.ac.uk/spm). Images were realigned to correct for motion, corrected for errors in slice-timing, spatially transformed to standard stereotaxic space (based on the Montreal Neurologic Institute (MNI) coordinate system), resampled to 2 mm isotropic voxel size using sinc interpolation and smoothed with a 4mm full-width half-maximum Gaussian kernel to decrease spatial noise prior to statistical analysis. Translational movement in millimeters (x, y, z) and rotational motion in degrees (pitch, roll, yaw) was calculated based on the SPM5 parameters for motion correction of the functional images in each subject. No participants had a range of movement greater than 3mm translation or 3 degrees of rotation.
For each analysis, an additional four frames of the functional data were discarded to allow for magnet stabilization.
The data were temporally filtered using a high pass filter of 100 seconds and a low-pass filter of 2.8 seconds. These subject-specific timeseries data were then concatenated to form a single, group-level four-dimensional dataset from all 15 subjects. This group dataset was then decomposed into independent component maps using a spatial ICA implemented in the MELODIC software (Beckmann and Smith, 2004), part of FSL (http://www.fmrib.ox.ac.uk/fsl). For computational reasons, the number of components was fixed at thirty for the group-level ICA. This initial group ICA decomposition allowed for the generation of group-level templates for the ICNs which were used to select subject-specific ICA maps obtained from the single-subject analyses described below.
Spatial ICA was also performed at the single-subject level using the same temporal filtering parameters. At the single-subject level, MELODIC’s automated dimensionality estimate was used to select the optimum number of components for each subject. The number of components per subject ranged from 35 to 48.
Twenty-two healthy subjects (ages: 19–21; mean age 20.6; eleven females, all right-handed) were scanned after giving written informed consent.
Images were acquired on a 3T GE Signa scanner using a custom-built head coil. Head movement was minimized during scanning by a comfortable custom-built restraint. A total of 29 axial slices (4.0mm thickness) parallel to the AC-PC line and covering the whole brain were imaged using a T2* weighted gradient echo spiral pulse sequence (TR = 2000 msec, TE = 30 msec, flip angle = 80°, 1 interleave) (Glover and Lai, 1998). The field of view was 20 cm, and the matrix size was 64 × 64, providing an in-plane spatial resolution of 3.125 mm. To reduce blurring and signal loss arising from field inhomogeneities, an automated high-order shimming method based on spiral acquisitions was used before acquiring functional MRI scans (Kim et al. 2002).
For the resting-state scan subjects were instructed to keep their eyes closed and try to hold still. The scan lasted 8 minutes.
A linear shim correction was applied separately for each slice during reconstruction using a magnetic field map acquired automatically by the pulse sequence at the beginning of the scan. Functional MRI data were then analyzed using SPM5 analysis software (http://www.fil.ion.ucl.ac.uk/spm). Images were realigned to correct for motion, corrected for errors in slice-timing, spatially transformed to standard stereotaxic space (based on the Montreal Neurologic Institute (MNI) coordinate system), resampled to 2 mm isotropic voxels using sinc interpolation and smoothed with a 4mm full-width half-maximum Gaussian kernel to decrease spatial noise prior to statistical analysis. Translational movement in millimeters (x, y, z) and rotational motion in degrees (pitch, roll, yaw) was calculated based on the SPM5 parameters for motion correction of the functional images in each subject. No participants had a range of movement greater than 3mm translation or 3 degrees of rotation.
For the Stanford data, the first five frames of the functional data were discarded to account for magnet stabilization.
Spatial ICA was again performed at the single-subject level using the same methods described for the Paris dataset. The only difference is that the Stanford data were not temporally filtered before applying ICA. The need for temporal filtering in resting-state ICA studies remains unclear. Here we have applied it in the first dataset and not in the second. As such our intersection maps show common cerebellar clusters independent of whether or not temporal filtering was applied prior to ICA. The number of components in the Stanford data ranged from 25 to 128.
Identification of resting-state brain networks for each subject was done in three distinct steps: (1) visual identification of five ICNs from the group-level components in the Paris dataset; (2) creating binary masks of each ICN; (3) using the mask as a template to select each individual subject’s best-fit component using a template-matching procedure as we and others have done previously (Greicius et al. 2007).
Previously described ICNs corresponding to the sensorimotor network, the default-mode network (DMN) (Greicius et al. 2003), the executive control network (ECN) (Seeley et al. 2007), and the salience network were visually identified from the 30 group-level independent components (Greicius et al. 2003, Seeley et al 2007, Damoiseaux et al. 2006). As has been shown previously, the ECN was divided by ICA into homologous left and right networks referred to here as LECN and RECN respectively (Damoiseaux et al. 2006). The five ICNs here were selected based, visually, on their correspondence to five ICNs identified by Damoiseaux and colleagues (Damoiseaux et al. 2006). Specifically our DMN corresponds to their Figure 3B, our LECN to their Figure 3C, our RECN to their Figure 3D, our salience network to their Figure 3J, and our sensorimotor network to their Figure 3H. We confirmed this match by calculating a spatial correlation between each of our 5 visually-selected templates and the eight ICNs shown in Figure 3 of Damoiseaux et al. (these maps were provided to us by Dr. Damoiseaux). In each case, our visually-selected template was more strongly correlated with the Damoiseaux et al. Figure 3 ICN described above than with any of the other 7 ICNs (i.e. our visually selected DMN was more strongly correlated with their Figure 3B than with any of the 7 other ICNs shown in their Figure 3). These group-level ICA maps were then binarized to serve as templates for the template-matching algorithm, used to select the best-fit component for each ICN from each subject’s individual ICA data. The template-matching algorithm derives a goodness-of-fit score for each of a given subject’s several components. The goodness-of-fit is calculated as the mean z-score of voxels within the template minus the mean z-score of voxels outside the template. That component with the highest goodness-of-fit score to the template is selected as the best-fit component. After running the template-matching algorithm, separately for each of the 5 ICN templates, every subject had 5 best-fit components, one for each ICN.
For each ICN and separately for each of the two datasets, group maps were calculated in SPM5 by means of a between-subjects random-effects analysis (i.e. a one-sample t-test on the best-fit independent components for each subject). We only report and display findings replicated in the two datasets. This was done by generating an intersection map for each ICN limited to voxels that were present in both groups at a p < 0.01 significance threshold (joint probability (Poline et al.1997), height and extent, corrected at the whole brain level, minimum cluster size 25 voxels). This step ensures that any reported clusters cannot be attributed to site-specific data acquistion or processing steps.
For display purposes, all statistical maps are overlayed on a T1-weighted MNI template using MRIcro. Cerebellar cluster localization was determined: by visual inspection using the MRI atlas of the human cerebellum provided by Schmahmann et al. (2000), and with the probabilistic atlas of the cerebellum (http://www.bangor.ac.uk/~pss412/imaging/propatlas.htm) using the FSL view option of FSL 4.1 (FMRIB Software Library http://www.fmrib.ox.ac.uk/fsl/).
In order to demonstrate both qualitiatively and quantitatively that the cerebellum makes relatively distinct contributions to the 5 ICNs studied here, we undertook two additional analyses. First, we produced a union map showing the cerebellar contributions from all 5 ICNs on a single set of cerebellar slices. Second, we generated intersection maps for each possible pairing of networks (DMN-LECN, DMN-RECN, etc.) and then identified and quantified any cerebellar clusters in these intersection maps.
Figure 1 displays an overview of the general cerebellar subdivisions highlighted in the various ICNs below.
The sensorimotor network comprises the sensorimotor cortex (M1/S1), the premotor cortex (BA 6), the supplementary motor area, the anterior cingulate cortex (BA 24), the occipital cortex (BA 19/37), the temporal cortex (BA 21) and the insula (Figure 2A). This circuit also includes: the lentiform and caudate nuclei, the ventral thalami, the rostral part of the left red nucleus and the cerebellum. Cerebellar clusters were bilaterally located within the hemispheric portion of lobules V and VI (Figure 2B) and within what is likely the dorsal portion of the dentate nuclei (Figure 2C) (though the limitations of our spatial resolution preclude a definitive distinction between the dentate nucleus and the lateral aspect of the neighboring interposed nuclei).
The DMN shows functional connectivity within the following cortical areas shown in Figure 3A: the dorsomedial prefrontal cortex (BA 9/10), the medial prefrontal cortex (BA 32), the superior parietal cortex (BA 7), the angular gyrus (BA 39), the posterior cingulate cortex (BA 23/31), the retrosplenial cortex (BA 29/30), the medial temporal lobe, and the ventral temporal cortex (BA 20). Subcortical clusters were also found in the thalamus, the left red nucleus and the midbrain. This diencephalo-telencephalic circuit also encompasses the cerebellum. Cerebellar clusters were bilaterally situated in the caudo-dorsal hemisphere of lobule IX which may include part of the second homunculus (Figure 3B). A small cluster was also noted in the right hemisphere of lobule VIIB. Clusters within the dorsomedian pontine nuclei were also observed (Figure 3C).
The right and left ECNs, RECN and LECN, respectively, likely represent two homologous ICNs that together constitute the unitary ECN we have described previously with an ROI-based analysis (Seeley et al. 2007). LECN. Cortical clusters were found in (Figure 4A): the dorsolateral, mid-dorsolateral, and dorsomedial prefrontal cortex (BA 45/46, 9 and 8), the orbitofrontal cortex (BA 47), the superior parietal cortex_(BA 7) and the angular gyrus (BA 39). Subcortical clusters were also located within the left caudate nucleus. Functional connectivity was also detected in several widespread neocerebellar regions including (Figure 4B): the right crus I and crus II, with a crus I predominance, and limited extensions into lobules VI and VIIB, in the right rostral hemisphere of lobule IX and in the left medial crus I and crus II. Clusters were observed in the left dorsal basis pontis (Figure 4C). RECN. Cortical clusters were found in (Figure 5A): the dorsolateral prefrontal cortex (BA ventral 44/45/46), the orbitofrontal cortex (BA 47), the caudal cingulate cortex (BA 23 bilaterally), the superior parietal cortex (BA 7) and the angular and supramarginal gyri (BA 39/40). Subcortical clusters were also located within the right caudate nucleus and the left red nucleus. Functional connectivity was also detected in several widespread neocerebellar regions comprising (Figure 5B) clusters located on the left side, in crus I and crus II with an extension into lobules VI and VIIB. Clusters were also present in the right dorsal basis pontis (Figure 5C).
The salience network demonstrates functional connectivity between (Figure 6A): the medial frontal cortex (BA 32), the dorsal anterior cingulate cortex (BA 24), the dorsolateral prefrontal cortex (BA 46), the frontoinsular cortex (BA 47/12), the thalamus and the red nuclei with a left predominance. Within the cerebellum (Figure 6B), clusters are located bilaterally in the lateral and ventral part of the hemisphere of lobule VI and the adjacent crus I near the posterosuperior fissure with a narrow extension in crus II and in the hemisphere of lobule VIIB. Within lobule VI, the salience network clusters are located more laterally and closer to the posterosuperior fissure than the more paramedian clusters found in the sensorimotor network. A small region of overlap between these salience and sensorimotor clusters was present posteriorly (between y = −62 and y= −67) on the left side. Clusters were also located in the dentate nuclei (figure 6C). Within the pons, clusters are situated in the region of the dorsomedian pontine nuclei.
The cerebellar contributions to the 5 different ICNs were largely non-overlapping. Across all 5 networks shown in Figures 2–6, 7779 voxels were indentified in the cerbellum. Of these, only 210 voxels (< 3 %) appeared in more than one ICN. The cerebellar contributions to all 5 ICNs are shown on the same overlay in Figure 7.
Consistent with recent task-activation studies showing replicable cerebellar responses to a variety of cognitive demands (Stoodley and Schmahmann, 2008), the current results support an expanded role of the cerebellum beyond motor control. Unlike previous ROI analyses of cerebellar connectivity (He et al. 2004; Allen et al. 2005), the current approach allows for a functional anatomic parcellation of the neocerebellum across several distinct ICNs. Our data clearly demonstrate functional coherence between the neocerebellum, particularly crus I–II, and the distinct cognitive ICNs examined here, but not with the sensorimotor network. These neocerebellar networks may represent cortico-cerebellar loops as the DMN, the LECN, the RECN and the salience network all included basis pontis clusters, presumably corresponding to the pontine nuclei which constitute the last relay of the corticopontine fibers prior to their targets in the cerebellum (Schmahmann and Pandya, 1989, 1997). For the sensorimotor and salience networks, clusters were found in the dentate nuclei, which represent, with the other deep cerebellar nuclei and the lateral vestibular nucleus, the sole cerebellar output channels.
The DMN, comprised mainly of the posterior cingulate cortex/precuneus, medial prefrontal/pregenual cingulate cortices, temporoparietal regions, and medial temporal lobes, is implicated in episodic memory retrieval, self-reflection, mental imagery and stream-of-consciousness processing (Raichle et al. 2001, Greicius et al 2003, 2004; Buckner et al. 2005). We found that the DMN includes lobule IX and a small cluster in the right hemisphere of lobule VIIB. However, the role of lobule IX remains unclear as in monkeys it is mainly connected to the vestibular nuclei and its resection produces no obvious deficit (Dow and Moruzzi, 1958). The cingulate cortex projects to the pontine nuclei (Vilensky and Van Hoesen, 1981), which project to the ventral paraflocculus (equivalent to lobule IX) in the cat (Brodal et al. 1991). Lobule IX has been implicated in various functional tasks including, thirst satiation (Parsons et al. 2000), sensation (Hui et al. 2005), motor synchronization (Jantzen et al. 2004), working memory (Desmond et al.1997), and perception of change in stimulus timing (Liu et al. 2008). Lobule IX, confirmed by referring to the cerebellar MRI atlas (Schmahmann et al. 2000), is involved in past and future event elaboration in conjunction with retrosplenial and precuneus cortices (Addis et al. 2007). As the most inferior portion of the cerebellum, lobule IX is often not covered by standard fMRI protocols and therefore more prone than other regions to false negatives. Nonetheless, our findings linking this region to the DMN are supported by a recent resting-state study that has also identified a DMN cluster in lobule IX (Filippini et al. 2009). The DMN, with its putative role in episodic memory retrieval and self-reflection, may be the most phylogenetically recent of the ICNs considered here and so the least likely to have obvious cerebellar homologues in non-human primates. The functional role of caudal lobule IX in the DMN remains unresolved. Given its replicable presence in the DMN of both datasets and in the study by Filippini et al, future studies of episodic memory, self-reflection, and other putative DMN functions should provide full coverage of the cerebellum to include this poorly understood region.
The ECN, encompassing the dorsolateral prefrontal and lateral parietal neocortices, is required for the selection and maintenance in working memory of relevant information necessary for action preparation (Seeley et al. 2007). We have shown that the ECN includes pontine clusters and the major part of the neocerebellum (crus I–II). These results agree with anatomical studies in monkeys reporting possible reciprocal connections between crus I and prefrontal cortex and crus II and parietal cortex. Specifically, afferents from the prefrontal cortex via the rostral and medial pontine nuclei converge on crus I while afferents from parietal cortex converge on crus II via the lateral, dorsal and medial pontine nuclei (Brodal, 1979). Efferents from crus I and II complete the reciprocal connection via projections from the dentate nuclei to prefrontal and parietal cortices (Dum and Strick, 2003). In humans, functional imaging studies have highlighted the role of crus I in executive functions, such as: abstract reasoning (Monti et al. 2007), working memory (Chen and Desmond, 2005), information updating (Colette et al. 2007), and response selection (Desmond et al. 1997). Lesions of the neocerebellum can cause executive impairments in abstract reasoning, working memory, set-shifting, and planning (Schmahmann and Sheerman, 1998). A broad functional lateralization of the cerebellum has been demonstrated, corresponding to our strongly lateralized right and left ECNs. Functional imaging (Desmond and Fiez, 1998; Kim et al. 1999; Goel and Dolan, 2004; Monti et al. 2007) and clinical studies (Richter et al. 2007; Gottwald et al. 2007) preferentially implicate the right cerebellum in verbal processes and the left-cerebellum in spatial processes. Clinical and neuroimaging studies point to a consistent right neocerebellar involvement in verbal working memory and verbal fluency (Bellebaum and Daum, 2007). Impairment in verbal fluency is reported with lesions of the right crus II (Richter et al. 2007). Therefore, the neocerebellum may constitute a crucial node for verbal and non-verbal executive functions in the RECN and LECN, respectively. Finally, the right rostral lobule IX also contributes to the LECN, but as mentioned above, the function of lobule IX remains unresolved.
The salience network, centered on the dorsal anterior cingulate (BA 24/32) and frontoinsular cortices (BA 47/12) connected with subcortical limbic structures, is involved in detecting, integrating and filtering relevant interoceptive, autonomic and emotional information (Seeley et al. 2007). The current study extends this network to two main, distinct cerebellar regions: the lateral portion of the right and left lobules VI and the adjacent crus I. Lobules VI–VII (crus I) are connected, through the pontine and dentate nuclei, with posterior and lateral hypothalamus and with the mammillary nuclei (Haines et al. 1984, 1997). As the lateral neocerebellum is mainly connected to associative cortices, we postulate that the frontoinsular and prefrontal clusters detected here are preferentially linked with lobules VI-crus I. Dimitrova et al. (2003) demonstrated the role of vermal lobule VI and the hemisphere of lobules VI-crus I in pain-related processes like grimacing, fear and startle reactions. Cerebellar strokes (hemispheres of lobules VII–VIII) cause impairment in the subjective experience of pleasant feelings (Turner et al. 2007). Frontoinsular and anterior cingulate activations have been detected during interoceptive awareness and physiological mismatch (Critchley et al. 2004; Gray et al. 2007, respectively); notably Gray et al. (2007) found activation in the vermis of lobule VI. We also found a small cluster overlapping the border of the vermal and paravermal lobule VI. Lastly, Stoodley and Schmahmann (2008) reported cerebellar activation, especially in lobule VI and crus I, during emotional processes like evaluating facial expression and empathizing. Therefore, lobule VI-crus I may contribute to estimating the valence of salient, emotional cues and selecting appropriate behavioral responses. To our surprise, we found no clusters in the posterior vermis where lesions cause affective impairments such as blunting of emotion (Schmahmann and Sherman, 1998). A small medial/paramedial cluster was present in left lobule VI. The vermis of lobule VI receives afferents from the hypothalamus (Azizi et al. 1981) and projects back to the hypothalamus through the fastigial nuclei (Haines et al. 1997). Fastigial and interposed nucleus lesions can induce autonomic impairments (Haines et al. 1997). Stimulation of the vermis has reportedly produced improvement in some psychiatric disorders (Heath, 1977) and increased theta-activity related to emotion and memory (Schutter and van Honk, 2006). Thus, vermis and paravermis of lobule VI, as detected here, could play a modulatory role upon the subcortical nodes of the salience network and may represent a phylogenetically older cerebellar emotional processor in conjunction with the posterior vermis and the hemispheres of the posterior lobe. While the lobule VI clusters in the salience and sensorimotor networks were largely distinct, a small lobule VI cluster was found in both networks. This overlap suggests an intracerebellar connection between these two networks that merits additional study and that might relate to limbic control of the motor system.
This study has several limitations that merit consideration. Chief among them is that the networks described here are detected in the absence of specific functional activity. We are inferring functional roles for the cerebellar clusters here based on their belonging to ICNs whose functional relevance is reasonably well-established in the literature. The fact that our subjects were resting with their eyes closed does, however, avoid the critique—levelled at standard cognitive studies of the cerebellum—that subtle motor or oculomotor requirements of the task account for the cerebellar activity (Haarmeier and Thie,r 2007). The exact functional relevance of ICNs—what benefits are accrued by maintaining temporal correlations in very low-frequency neural fluctuations—remains unclear, but it seems that brain regions that typically activate together during particular tasks remain tethered at some basal level even in the absence of their preferred task. We speculate that this basal, task-independent, intrinsic connectivity is important for avoiding disuse-related pruning of critical synapses (Luo and O’Leary, 2005). Others have suggested that this basal connectivity maintains networks in a primed state to improve response efficiency (Fox and Raichle, 2007). We were encouraged by the sensitivity of the approach used here to detect, in several ICNs, clusters in the pontine and dentate nuclei, the major convergence points for afferent and efferent connections respectively. It should be noted, however, that both the pontine and dentate nuclei should be present in each of the five ICNs studied here (Schmahmann, 2002). Our insistence on replicating clusters across both datasets may have been too stringent to allow for the detection of clusters in these smaller nuclei in all 5 ICNs. The same can be said for the unilateral contribution of the (left) red nucleus to the sensorimotor cortex where we would expect bilateral involvement of the red nuclei in this ICN. The spatial resolution of the current study does not allow us to definitively distinguish the medial aspect of the dentate nucleus and the lateral aspect (emboliform nucleus) of the adjacent interposed nucleus. Finally, while the convergence of our results with more precise tract-tracing studies in animals is reassuring it is important to note that the BOLD signal used in fMRI is only an indirect measure of neural activity.
In conclusion, the present study provides further support for the view that distinct neocerebellar regions are involved in distinct cognitive functions. The human neocerebellum, particularly lobules VI and VII (crus I–II), selectively contributes to parallel cortico-cerebellar loops involved in executive control, salience detection, and episodic memory/self-reflection. The greatest portions of the neocerebellum contributed to the ECN, a network involved with selection and maintenance in working memory of relevant multimodal information. Interestingly, lobule IX, whose functional significance remains unresolved, was a major component of the DMN and, to a lesser degree, the LECN. These findings should help guide subsequent investigations designed to specify the precise functional role of distinct cerebellar regions in higher order cognitive and affective processing.
This work was supported by the following NIH grants: NS048302, HD047520 and NS058899, and NSF grant: BCS/DRL 0750340.