Mental Visualization in the Cerebellum: Rapid Non-motor Learning at Sub-Lobular and Causal Network Levels

Abstract

It is generally understood that the main role of the cerebellum is in movement planning and coordination, but neuroimaging has led to striking findings of its involvement in many aspects of cognitive processing. Mental visualization is such a cognitive process, extensively involved in learning and memory, artistic and inventive creativity, etc. Here, our aim was to conduct a multidimensional study of cerebellar involvement in the non-motor cognitive tasks. First, we used fMRI to investigate whether the cognitive task of visualization from an immediate memory of complex spatial structures (line drawings) engages the cerebellum, and identified a cerebellar network of both strongly activated and suppressed regions. Second, the task-specificity of these regions was examined by comparative analysis with the task of perceptual exploration and memorization of the drawings to be later visualized from memory. BOLD response patterns over the iterations of each task differed significantly; unexpectedly, the suppression grew markedly stronger in visualization. Third, to gain insights in the organization of these regions into cerebellar networks, we determined the directed inter-regional causal influences using Granger Causal Connectivity analysis. Additionally, the causal interactions of the cerebellar networks with a large-scale cortical network, the Default Mode Network (DMN), were studied. Fourth, we investigated rapid cognitive learning in the cerebellum at the level of short-term BOLD response evolution within each region of interest, and at the higher level of network reorganization. Our paradigm of interleaved sequences of iteration between two tasks combined with some innovative analyses were instrumental in addressing these questions. In particular, rapid forms of non-motor learning that strongly drive cerebellar plasticity through mental visualization were uncovered and characterized at both sub-lobular and network levels. Collectively, these findings provide novel and expansive insights into high-order cognitive functions in the cerebellum, and its macroscale functional neuroanatomy. They represent a basis for a framework of rapid cerebellar reorganization driven by non-motor learning, with implications for the enhancement of cognitive abilities such as learning and memory.

Keywords: Granger Causal Connectivity; cerebellum; fMRI; learning; memory; mental visualization; plasticity.

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2021 Likova, Mineff and Nicholas.

Figures

FIGURE 1

Didactic summary of cerebellar functional anatomy based on human fMRI evidence. Data-driven analyses of cerebellar fMRI indicate three major fundamental poles of cerebellar functional neuroanatomy. (Left,Center) Motor processing (blue) is represented twice in each cerebellar cortical hemisphere (lobules I–VI; lobule VIII). Non-motor processes (red and yellow) are represented three times in each cerebellar cortical hemisphere (lobules VI-Crus I; lobules Crus II–VIIB; lobules IX–X). A specific anatomical order is conserved throughout the cerebellar cortex, and propagates from first motor toward first non-motor representation (i.e., from lobules I–VI to Crus I), from second motor toward second non-motor representation (i.e., from lobule VIII to Crus II), and from second motor toward third non-motor representation (i.e., from lobule VIII to lobule IX/X). (Right) The principal axis of macroscale functional organization in the cerebellar cortex progresses from motor, to attentional/executive, to default-mode processing. This progression is captured in the anatomical order of cerebellar functional territories as shown in the center and right panels, and also revealed by a data-driven analysis of functional gradients in the cerebellar cortex based on resting-state functional connectivity between cerebellar cortical areas (After Guell and Schmahmann, 2020; reproduced with permission).

FIGURE 2

A schematic preview that shows at glance the overall structure of the methodological design. Average cerebellar activation was mapped to an unfolded cerebellar flatmap (icon at left) where the pattern of cerebellar involvement in non-motor cognitive tasks was used to determine the activation levels in each local region of interest (ROI), and its rapid change over three repeats of each task (top row); Granger Causal connectivity was assessed within the respective networks of these ROIs, and the connectivity evolution over the three repeats plotted in terms of a Connectivity Density Index (bottom row). Two interleaved tasks: EM = Explore and Memorize; Vis = Visualization from memory (see Figure 3 in Experimental Design).

FIGURE 3

Functional Magnetic Resonance Imaging experimental design included a sequence of two tasks (i) perceptual exploration and memorization of complex graphic material (30 s) and (ii) visualization-from-memory of these images (30 s), interleaved with fixation periods (20 s); this sequence was repeated three times for each image.

FIGURE 4

(A) Three-slice volume view of the maximum probability atlas transformed by SUIT to the MNI152 T1 MRI scan used as the structural reference for registering multi-subject fMRI data in this study. (B) Probabilistic atlas of the cerebellar lobules projected to a flatmap of the cerebellum from the SUIT Matlab toolbox (Diedrichsen and Zotow, 2015; http://www.diedrichsenlab.org/imaging/suit_flatmap.htm).

FIGURE 5

Cerebellar flatmap representation of the average activation across the three Visualization repeats, thresholded at z-score of ±0.5. Positive BOLD activation coded in warm colors and negative BOLD activation in cool colors. The regions of activated voxels within each lobule were used to define the ROIs for further analysis.

FIGURE 6

Activation patterns for the positive cerebellar ROIs identified in the present protocol. Visualization (Vis) and Explore and Memorize (EM) task periods are coded by the green and purple bars, respectively. Note that the activation strengths in both tasks was generally similar for left and right hemisphere in the most rostral and caudal ROIs (in the flatmap framework). Note that the responses are plotted as z-scores. For this sample, z-scores of >2.45 are significant at p < 0.05 (two-tailed) and z-scores >5.95 are significant at p < 0.001 (two-tailed), etc. All specified ROIs reach the higher criterion for significance in at least one repeat in each of the tasks.

FIGURE 7

Activation patterns for the task-negative cerebellar lobular ROIs, expressed as z-scores in the same format, and to the same statistical criteria, as for Figure 6. Note that, the responses were negative for both Visualization (Vis) and Explore and Memorize (EM) tasks, and tended to become increasingly negative over the three repeats in Crus 1 and Crus II bilaterally. To the significance criteria specified in Figure 6, responses for Vis in all specified negative ROIs reach the higher criterion for significance in at least one repeat, except for the VI-vermis-n ROI, where EM reached this criterion. Lobule IX showed a hemispherically asymmetric pattern in both tasks, as opposed to the symmetric pattern in Crus I/II, and to the predominant case for the task-positive Vis ROIs of Figure 4. The VI-vermis-n ROI produced weak negative Vis responses, but in contrast the rest of these ROIs, its EM responses inverted to positive activation; in general, both Vis and EM responses became significantly more negative over the repeats.

FIGURE 8

Slopes of the evolution of average response strengths (red dots) for the task-positive ROIs from Figure 5 as a function of task repeats for the Vis task (upper panel A) and EM task (lower panel B). Note that for the Vis task sequence (A), three ROIs in the left lobules VIIIa, VIIIb, and IX had significant positive slopes - increasing response strength as the learning progressed. In contrast, the EM task sequence (B), showed both significant decreases and increases in different ROIs during learning. Error bars are 95% confidence intervals for the difference of the slopes from a zero-slope; asterisks indicate significant slopes at p < 0.05.

FIGURE 9

Slopes of the evolution of response strengths (blue dots) for the task-negative ROIs from Figure 6, with the same format and significance criteria as for Figure 8. One ROI had positive activation (red dot) for the EM task (B). Note that most of the task-negative Crus ROIs had significant slopes for the Vis task sequence (A), with increasing negative response strength as the learning proceeded, thus deepening the suppression in these regions. A similar picture, with even greater suppression, is seen for the EM task sequence (B). Asterisks indicate significant slopes at p < 0.05.

FIGURE 10

Average Granger causal connectivity in Visualization (Vis) for the task-positive cerebellar ROIs with each other and with the cortical DMN. Connections are shown as arrowed lines if the directed connectivity is significant at p< 0.05, with line thickness coding connectivity strength, and red/blue color coding whether the influence is congruent or inverse relative to the BOLD signal (in separate upper and lower plots). (A–C): Granger causal connectivity for Vis 1, 2, 3 (the first through third visualization repeats). There is no congruent connectivity to the DMN in this task-positive ROIs, while the dominant inverse connectivity flows from almost all of these ROIs to the DMN initially, decreasing dramatically with task repeats.

FIGURE 11

Average Granger Causal connectivity for the perceptual exploration and Memorization(EM) task in the task-positive network; shown with the same format and significance criteria as for Figure 10. The connectivity patterns had some general similarities to that for the Vis task in Figure 10, but also major differences, in that the dominant inverse connectivity flowed from the DMN to these ROIs in the second repeat.

FIGURE 12

Average Granger causal connectivity for Visualization in the task-negative network (shown in the same format and significance criteria as for Figure 10). Upper row: Congruent causal influences within the network of cerebellar ROIs and the cortical DMN. Lower row: Inverse causal influences. The Granger Causal Connectivity demonstrates rapid network reorganization over the visualization task repeats. The DMN in strongly involved bilaterally, sending congruent GC influences to these task-negative ROIs during all three repeats, and receiving weak inverse influences from the Crus ROIs.

FIGURE 13

Average Granger causal connectivity maps in the EM condition for the task-negative network (shown in the same format and significance criteria as Figure 10). The congruent influences were dramatically weaker than for the visualization task, while inverse influences were reorganized. Influences flow inconsistently from and to the DM at different times, but strongly to the DMN from the Crus ROIs for inverse influences during the second repeat.

FIGURE 14

Overview of the density of connections under each circular plot of Figures 9–12. Error bars are 1 SEM of the counts. Sequential pairwise differences significant at p< 0.05 are indicated by asterisks.