Article Figures & Data
Figures
- Figure 1.
Trial types and target sequences. a, Participants were trained to produce four sequences from memory via visual instruction in “go” trials. The target sequence for each trial was indicated by an abstract fractal cue which preceded a short preparation period. A black hand with a gray (instructed; a red cue indicated which finger to press with a set temporal structure) or green (from memory; participants had to produce the sequence without the visual cue) background appeared to indicate the go cue. A short fixation period followed, after which feedback was provided based on the accuracy and timing of the sequence. b, On 50% of trials, the go cue would be replaced by an extended fixation cross. Participants were subsequently rewarded for not making a press during these “no-go” trials. c, Participants performed the sequences using the right hand on a 10-finger force transducer. Finger press timing was defined as the time point when the channel force exceeded a fixed threshold relative to the baseline after the go cue. The color indicates which finger was pressed. Trained sequences consisted of permutations of two timings (Timing 1, Timing 2) and two orders (Order 1, Order 2), constituting S1 through S4.
- Figure 2.
Percentage signal change in subcortical regions during preparation and production. a, Volumetric slices of each subcortical region displaying percentage signal change for preparation and production, far left and right, respectively. The center shows the level of percentage signal change when averaged across each subcortical region. Error bars represent standard error of the sample mean. b, As above, for the CB. Explorative analysis across the whole CB revealed significant activity increase bilaterally in 4–6 including a significant cluster spanning cerebellar anterior motor regions bilaterally during production (black outline). Tha, thalamus; Cau, caudate nucleus; Put, putamen; Hip, hippocampus; **p < 0.01; *p < 0.05 (two-sided t tests against 0, Bonferroni-corrected for two trial phases within each ROI).
- Figure 3.
LDA results. a, Volumetric slices of each subcortical region are displaying classification accuracy for each classifier during preparation and production, far left and right, respectively. The center shows classification accuracy when averaged across each subcortical region. Error bars represent standard error of the sample mean. b, As above, for the CB. Explorative searchlights across the whole CB shows above-chance integrated decoding and significant cluster (black outline) in contralateral Crus 1. Tha, thalamus; Cau, caudate nucleus; Put, putamen; Hip, hippocampus; *p < 0.05 (one-sided t test against chance, Bonferroni-corrected for six comparisons within each ROI).
- Figure 4.
Simulated and empirical sequence and cross-phase distances using multidimensional scaling. a, RDMs (cross-validated Mahalanobis distance) and multidimensional scaling (MDS) plots of simulated fMRI data along the first three PCs, showing four sequence conditions across two phases. Red lines connect sequences with different timings but the same order, whereas blue lines indicate sequences with different orders but the same timing. Dotted gray lines are drawn between the same sequence during preparation and production. Preparation and production were either generated using the same distribution (left panel; no switch) or different distributions (right panel; switch). The highest dissimilarity and scaling of the distances was driven by phase (preparation to production) and most pronounced alongside PC1. b, RDM and MDS plots of empirical data from target ROIs showing all four sequences during preparation and production. c, Averaged Euclidean distance between preparation and production within sequences once PC1 is excluded, calculated relative to those same distances from the simulated no-switch data with matched signal-to-noise ratios. This represents differences for all sequences between preparation and production which cannot be explained by differences in overall activity. Orange horizontal lines represent the same distances in the simulations when a switch is induced. *p < 0.05; **p < 0.001; one-sided t test against simulated no distance with matched noise, 0 (Bonferroni-corrected for seven comparisons). Hip, hippocampus; Tha, thalamus; Cau, caudate nucleus; Put, putamen; Lob IV, lobule IV; Lob V, lobule V.
- Figure 5.
Schematic representation of sequence feature control during planning and execution. Higher-level sequence features, order (blue) and timing (red), are defined by the cortex and hippocampus during planning. During execution, the activity patterns in hippocampus, basal ganglia, and thalamus shift, while the cortex integrates order and timing into low-level sequence–specific trajectories (green) while maintaining higher-level independent movement timing. The green gradient fill in the thalamus, basal ganglia, and CB indicates regions that implement individual sequential elements. Tha, thalamus; BG, basal ganglia; CB, cerebellum; HC, hippocampus.
