The basal ganglia is a group of sub-cortical nuclei, which retrieves connections from almost the entire cerebral cortex, making the basal ganglia vital for integrating cortical brain processes and is involved in, e.g., action selection and learning, reward processing, cognition and emotion¹. The dorsal striatum (caudate and putamen) is the main input structure of the basal ganglia retrieving afferent topographically organized connections from cortex². The output of the basal ganglia goes back to cortex via thalamus, forming parallel cortico-basal ganglia-thalamo-cortico loops. To investigate the involvement of the striatum in sequential finger movements, we performed whole-brain functional MRI (fMRI) while healthy participants performed a discrete sequence production (DSP) task. We hypothesized that the distributed activity profile in the striatum will represent a specific finger sequence.

16 healthy right-handed participants (8 females, mean age 22.3±3.0) were scanned using a 3T Philips Achieva MR scanner using an Echo Planar Imaging (EPI) sequence (2.2mm isotropic voxels, TR/TE=2200/30ms, flip-angle=80˚, 42 axial slices). High-resolution T1-weighted (0.85mm isotropic voxels, TR/TE=6045/2.71ms) images were acquired and used for segmentation of putamen and caudate nucleus using FSL’s FIRST. Subjects performed six runs of a DSP task with the right hand (figure 1). Using a blocked design, four different sequences with increasing difficulty levels were performed. Preprocessing of the fMRI data included slice-time correction, re-alignment, de-spiking, and spatial normalization to MNI space using SPM. A first level fMRI model was constructed, which modeled each of the finger sequence and a contrast modeled the linear increase in BOLD signal due to increasing sequence difficulty levels. The mean effect across subjects of this linear increase was captured with a 1-sample t-test. We used Support Vector Machine (SVM) classification with a linear kernel to investigate whether voxels in cortex and striatum could discriminate between the different finger sequences. This was done in a leave-one-subject-out cross validation framework, where data from a single subject in turn was left out of model training and used for testing.

Sequence execution times and accuracies were stable after the two first runs (figure 3), so the following fMRI analysis only considers data from runs 3-6. At the group level, five cortical clusters showed significant linear increase in BOLD signal due to increasing difficulty level (figure 4) including left and right pre-motor cortex, left and right superior parietal lobule, and left inferior parietal lobule. No clusters in basal ganglia showed a linear increase of the BOLD signal with increasing difficulty. Multi Voxel Pattern Analysis (MVPA) classification based on SVM revealed a distributed set of voxels in left (contra-lateral) putamen, which significantly classified the four sequences (45.3% correct classification, p<0.001 permutation tests). Using voxels in left caudate, a classification accuracy of 34.4% correct was reached (p<0.05). In right putamen accuracy reached 34.4% (p<0.05) and right caudate 26.6% correct (not significant). Using all brain voxels, a classification accuracy of 54.7% was reached, and when using the voxels in the cortical clusters (figure 4), 59.4% correct classification was obtained. It should be noted, however, that the latter result is biased, since all data was used for voxel selection and is only included for comparison.

The distributed movement-related activity in the human striatum represents individual finger movement sequences. This finding points to a significant role of the basal ganglia in the control of finger sequences. The best classification was found in contra-lateral putamen, but also the distributed activity in contra-lateral caudate and ipsi-lateral putamen showed significant classification.