Cortico-cerebellar patterns differentiate maintenance and manipulation in working memory tasks

1116 

Abstract Submission 

Joshua Tan¹, Eli Müller¹, Claire O'Callaghan¹, James Shine¹

¹University of Sydney, Sydney, NSW

Joshua Tan  
University of Sydney
Sydney, NSW

Eli Müller  
University of Sydney
Sydney, NSW

Claire O'Callaghan  
University of Sydney
Sydney, NSW

Mac Shine  
University of Sydney
Sydney, NSW

Working memory involves the temporary storage and manipulation of information. It plays an important part of our day-to-day lives, such as when organising your daily schedule and engaging in conversations. However, the mechanisms of working memory are poorly understood, due in part to largely cortical explanations for the phenomenon. Intriguingly, working memory has previously been associated with regions of the dorsolateral prefrontal cortex that share extensive connections with the cerebellum, and rodent studies have implicated this circuit in working memory tasks. We hypothesised that working memory mechanisms are facilitated by cerebellar engagement, which may provide a parsimonious understanding for the neural mechanisms of working memory.

Twenty-four right-handed individuals (mean age = 23.8, SD = 2.6, 16 women) participated in the study and completed both a working memory spatial map and mental rotation task (King et al., 2019). Participants underwent three days of training before the first scanning session. Each participant completed 16 imaging runs (10 min per run) spread across two days. Each task consisted of an instruction period (5 s) and a response period (30 s). For the mental rotation task, there were 9 trials per run (duration = 3 s; ITI = 300 ms), with three difficulties with different degrees of rotation (easy = 0º, medium = 50º, hard = 150º). For the spatial map task, there were 6 trials per run (duration = 4.8 s; ITI = 200 ms) with three difficulties with differing number of digits to memorise (easy = 1, medium = 4, hard = 7). fMRI scans were pre-processed using fMRIprep, denoising and a high-pass filter (0.01) was applied using python scripts. Time-series signal was extracted and z-scored from a parcellation of 502 regions (400 cortical regions; Schaefer et al., 2018, 28 cerebellar regions; Diedrichsen, 2006, 74 subcortical regions; Tian et al., 2020). General linear models were fitted to the subject ROI time-series, comparing the effect of task difficulty on the BOLD response. Linear discriminant analysis (LDA) was used to find a low-dimensional manifold that separated both the spatial map and mental rotation tasks. Time-varying functional connectivity was estimated using the multiplication of temporal derivatives, a windowed approach (Shine et al., 2015). Edge connectivity across the whole brain was compared across tasks using permutation testing.

General linear models were constructed comparing the effect of task difficulty on regional BOLD response. From these models, increased difficulty in the mental rotation task was associated with increased BOLD in the dorsolateral prefrontal, parietal cortices, and cerebellar lobules V and VIII (p < 0.05). The spatial map task was associated with increased BOLD response in the dorsolateral prefrontal, premotor, parietal, cortices, as well as lobule VI, Crus I and Crus II of the cerebellum (p < 0.05). The beta coefficients from these models were then compared using LDA and the principal eigenvector was analysed. The principal eigenvector described a gradient separating the middle lobe from the anterior and flocculonodular lobes of the cerebellum. Where the middle cerebellar lobe was related to maintenance in the spatial map task, and the anterior and flocculonodular lobes were related to manipulation in the mental rotation task. At the cortical level, there were minimal differences between the two tasks. Time-varying analyses revealed increased connectivity over time between dorsolateral regions with Crus I and Crus II of the cerebellum during spatial mapping. During mental rotation there was a decoupling between dorsolateral regions and Crus I and Crus II of the cerebellum.

These results provide evidence that task demands of different working memory processes are differentiated in the cerebellum. By going beyond cortico-cortical interactions, we find distinct differences in neural patterns that differentiate processes of working memory.

Working Memory ¹

Activation (eg. BOLD task-fMRI) ²

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling

Subcortical Structures

Cerebellum

Cognition

fMRI CONTRAST MECHANISMS

Memory

Sub-Cortical

Other - working memory