henhao
Sign in
NeuroscienceeLife
Peer-reviewed journal article

Consolidation alters motor sequence-specific distributed representations

Basile PinsardiD, Arnaud BoutiniD, Ella GabitoviD, Ovidiu LunguiD, Habib BenaliiD, Julien DoyoniD

Abstract

FMRI studies investigating the acquisition of sequential motor skills in humans have revealed learning-related functional reorganizations of the cortico-striatal and cortico-cerebellar motor systems in link with the hippocampus. Yet, the functional significance of these activity level changes is not fully understood as they convey the evolution of both sequence-specific knowledge and unspecific task expertise. Moreover, these changes do not specifically assess the occurrence of learning-related plasticity. To address these issues, we investigated local circuits tuning to sequence-specific information using multivariate distances between patterns evoked by consolidated or newly acquired motor sequences production. Results reveal that representations in dorsolateral striatum, prefrontal and secondary motor cortices are greater when executing consolidated sequences than untrained ones. By contrast, sequence representations in the hippocampus and dorsomedial striatum are less engaged. Our findings show, for the first time in humans, that complementary sequence-specific motor representations evolve distinctively during critical phases of skill acquisition and consolidation.

PDFDOI
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
1,065
FWCI
86.03
Field percentile
top 1%
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (25 Jul 2018) · CC BY 4.0

In human fMRI, consolidated finger sequences are more discriminable than new ones in sensorimotor putamen, premotor and prefrontal cortices, but less so in hippocampus and caudate.

General linear model (GLM) activation contrasts in motor sequence learning (MSL) conflate sequence-specific plasticity with general task expertise, and prior multivariate pattern analysis (MVPA) work mainly applied classifiers to extensively trained sequences. Comparing consolidated and new sequences at equal task expertise with an unbiased multivariate distance isolates sequence-specific representational change across cortex and subcortex after brief training.

Key results

  1. ShownConsolidated sequences were more discriminable than new ones in bilateral sensorimotor putamen and thalamus, and in contralateral lateral and dorsal premotor, supplementary motor and dorsolateral prefrontal cortices.

    Non-parametric paired t-test with TFCE (p <.05, 10000 permutations, minimum cluster 10 features) in n=18, conjoined with within-stage maps. The same direction also held in bilateral anterior insula, ipsilateral cerebellar lobule IX, posterior cingulate and parietal cortex, and contralateral frontopolar cortex and Crus I. · Reorganization of the distributed sequence…; Statistical testing; Results

  2. ShownNew sequences were more discriminable than consolidated ones in bilateral hippocampi, the caudate body, subthalamic nuclei and ipsilateral cerebellar Crus II. The authors read this as a dorsomedial-to-dorsolateral striatal shift and a transient hippocampal role.

    Comes from the negative side of the same TFCE-corrected paired contrast (p <.05), conjoined with the new-sequence group map. Contralateral putamen also discriminated new sequences, but bilateral sensorimotor putamen discriminated consolidated sequences significantly more. · Reorganization of the distributed sequence…; Cortico-subcortical representational…

  3. ShownA distributed network discriminated sequences at both learning stages. It spanned primary visual and ventro-temporal cortex, premotor, supplementary and primary motor areas, posterior parietal cortex and dorsolateral prefrontal cortex.

    Based on a minimum-statistic conjunction of the new and consolidated group maps (TFCE, p <.05). The contralateral M1/S1 cluster sat around the hand knob and was smaller than in Wiestler and Diedrichsen (2013). Both consolidated sequences were separable even though they begin with the same finger. · A common distributed network for sequence…; A distributed network for the representation of…; Statistical testing

  4. SuggestedContralateral primary somatosensory cortex showed no significant consolidation-related change in sequence discriminability, although M1/S1 around the hand knob discriminated sequences at both stages. Cerebellar lobule V also showed no difference between stages.

    No significant change in contralateral primary somatosensory cortex appeared in the TFCE-corrected (p <.05) consolidated-vs-new contrast. The authors relate this to M1 geometry being shaped by natural finger co-activation and note that motor cortex plasticity during consolidation remains debated. Lobule V showed no significant difference. · A distributed network for the representation of…; Cortico-subcortical representational…

  5. ShownConsolidated sequences were executed faster and more accurately than new ones and stayed at asymptote, while new sequences sped up across blocks. Neither pair differed internally in speed or accuracy.

    Mixed-effects model: new sequences were slower (β =.365, SE = 0.047, p <.001) and less accurate (β = −0.304, SE = 0.101, p <.001). Speed gain across blocks, new vs consolidated: β = −0.018, SE = 0.002, p <.001. No speed change for consolidated sequences (β = −0.006, p = 0.192). · Behavioral performance; Behavioral statistics

Methods, data and reproducibility

Participants trained two 5-element left-hand finger sequences on consecutive days. On D3 they performed them with two new sequences in 64 short fMRI blocks. Least-square separate beta maps fed a searchlight cross-validated Mahalanobis distance, tested at group level with TFCE permutations and conjunctions.

  • Right-handed healthy young adults (enrolled cohort: 19 females, 25 ± 6.2 yr), screened for sleep quality, mood, chronotype and musical or typing training · human · n = 18 analyzed (34 enrolled, 26 completed)
  • D3 MVPA-task fMRI: four sequences (two consolidated, two new) in 64 blocks across two runs, 16 blocks per sequence; 3T EPI, TR 2160 ms, 3.44 × 3.44 × 3.3 mm voxels · human · n = 18 participants
  • T1-weighted multi-echo MPRAGE anatomical scans (1mm isometric) used for cortical surface reconstruction and subcortical segmentation · human
  • Keypress data from the MVPA task (mean correct-sequence duration and number of correct sequences per block) used in mixed-effects models · human · n = 18 participants, 16 blocks per sequence
  • Screening-night EEG in a mock MRI scanner and wrist actigraphy (Actiwatch 2) used to check sleep ability and schedule compliance · human
  • Human Connectome Project 32k cortical surface sampling and subcortical atlas used as registration targets · human (Human Connectome Project templates)
Methods & tools
Freesurfer (surface reconstruction, subcortical segmentation), ANTs (non-linear registration), FSL Topup (EPI distortion correction), PyMVPA with custom extensions, Least-square separate (LS-S) single-event GLM regression, Cross-validated Mahalanobis distance with Ledoit-Wolf shrinkage, Searchlight (64 nearest gray-matter neighbors), Threshold-Free-Cluster-Enhancement (TFCE) permutation testing, Minimum-statistic conjunction analysis, Linear mixed-effects models (behavior), De-Bruijn cycle block ordering, Actiwatch 2 actigraphy
Reagents & environment
Custom preprocessing pipeline (slice-wise motion and intensity correction with gray-matter BOLD extraction) plus Freesurfer, ANTs, FSL Topup and PyMVPA. No software versions or containers are reported.
Reproducibility gaps
No code repository, data deposit or accession is given. The custom preprocessing and PyMVPA extensions are described but not shared, and the full mixed-model outputs are referred to supplementary material.
Limitations (6)
  • Only two sequences were used per learning stage, so each condition yields a single multivariate distance. The authors note this may limit statistical power. · stated by the authors
  • The D3 task added demands absent from single-sequence training: instruction processing, working memory, switching and inhibition. These could induce new learning of the consolidated sequences, and switching alone may engage dorsal striatum and prefrontal cortex. · stated by the authors
  • The representational analysis ignored behavioral performance, even though consolidated sequences were executed faster and more accurately than new ones. · stated by the authors
  • The design cannot test the information content (for example, extrinsic vs intrinsic coding) of hippocampal, striatal or cerebellar representations. After limited training, these representations may keep changing with more practice or sleep. · stated by the authors
  • Sequence identity is confounded with learning stage. As described, every participant trained the same two sequences, and both consolidated sequences start with the same finger while the new pair do not, so differences between the pairs could shape the distances. · noted in this summary
  • Consolidation is inferred from overnight intervals, with no wake control and no link to sleep measures in this report. Results from 18 participants are shown only as thresholded maps, without distance magnitudes or effect sizes. · noted in this summary

Who should read:Read this if you study motor sequence learning or memory consolidation and want a whole-brain representational (cross-validated Mahalanobis searchlight) contrast of consolidated versus newly learned sequences.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISYoon '23King '22Dolfen '25Maceira-Elv '22Yewbrey '24Berlot '20
82% semantic matchNeuroscience
Seojin Yoon, Sungshin Kim·bioRxiv·Aug 28, 2023