Resting state functional connectivity predicts subsequent motor sequence learning

• ¹ Université Libre de Bruxelles, Belgium
• ² Université Libre de Bruxelles, Belgium

Introduction Activation in striatal, cerebellar and motor cortical regions of the frontal lobe contribute to motor sequence learning (Doyon et al., 2003). This study investigates, using a magnetoencephalography (MEG), the extent to which resting state functional connectivity (rsFC) in young adults can predict their ability to learn a new motor sequence. Methods Five minutes of resting state data (eyes open, fixation cross) were acquired in 14 young (7 females and 7 males, age range: 19–30 years) right-handed healthy adults with a whole-scalp MEG (Elekta). This resting session was followed by the execution of a Simple Movement Task (SMT) and by the learning of a Finger-Tapping Task (FTT). During the SMT, MEG data were recorded while participants performed auditory-cued key presses using all four left fingers simultaneously (index to little finger, 100 trials). During the FTT, participants had to reproduce twice per trial a 5-element sequence as fast and accurately as possible with the left hand (70 trials, sequence: little finger (4), index (1), ring (3), middle (2), little finger (4)). Execution times on the FTT were computed considering the 8 three-element chunks that composed the two sequences executed per trial (i.e. 413, 132, 324, 244, 441, 413, 132, 324) (see Hotermans et al., 2006, 2008). Learning performance was quantified for each participant as the characteristic time constant T of an exponential fit ETi∝exp⁡(-i/T) of mean chunk execution times ETi to chunk index i (estimated via semi-logarithmic linear regression). Resting state data were cleaned from artefacts using signal space separation as well as independent component analysis and visual inspection, and sources were reconstructed in the beta-band (12-30Hz) using Minimum Norm Estimation. Individual primary sensorimotor networks were then derived as the seed-based sources slow envelope correlation map (Wens et al., 2013) using seeds in the right sensorimotor cortex (rSM1) selected for each participant on the basis of event-related mu-rhythm synchronization during SMT execution as follows: (i) SMT data were cleaned from artefacts as for resting state data; (ii) event-related time-frequency power analysis (Morlet wavelet decomposition with a standard time-frequency compromise (Tallon-Baudry et al., 1996)) was first performed on SMT data at the sensors level to determine post-movement time (2653.29 ± 672.85 ms, mean±std across subjects) and frequency (21.07 ± 2.47 Hz) of maximum power increase decrease with respect to baseline (-800 to -200 ms pre-movement); (iii) the selected power map was projected on source space using sLORETA; (iv) the seed location was selected as the local peak of maximum power increase in the rSM1 cortex contralateral to hand movement. Finally, the relation between rsFC from rSM1 and learning performance was investigated via regression analysis (as implemented in SPM8, www.fil.ion.ucl.ac.uk/spm) on rsFC maps (normalized on MNI template and spatially smoothed with an 8 mm isotropic Gaussian kernel) with performance index T as covariate of interest. Reported results are significant at pcorr<0.05 after correction for multiple comparisons with an extent of the spatial threshold of minimum 100 voxels. Results Regression analysis indicated that rsFC between the seed (rSM1) and the left anterior cingulate cortex (ACC), the left dorsolateral prefrontal cortex (DLPFC), the right dorso-medial prefrontal cortex (DMPFC) and the lobule V and VI of the cerebellum varied with the motor learning performance (see Fig.1). These significant correlations between learning performance and rsFC were all positive (ACC: correlation value r = 0.76, uncorrected p-value puncorr = 0.002, MNI coordinates [-24 34 13] mm; the DLPFC: r = 0.65, puncorr = 0.01, [-49 28 39] mm; the DMPFC: r = 0.77, puncorr = 0.001, [10 53 39] and the cerebellum: r = 0.62, puncorr = 0.02, [-32 -40 -30]), suggesting that higher functional connectivity in these regions prior to learning facilitate motor sequence learning performance. Conclusion The ACC, the DLPFC and the cerebellum appear to be involved in the early phase of motor learning, as their activity progressively decreases with the automation of the motor skill (Doyon et al., 2003; Wu et al., 2004). The ACC may help to develop a mental representation of the motor sequence (Wu et al., 2004). The DLPFC may play a prominent role in orienting attention toward relevant mental processing (Abe & Hanakawa, 2009) needed to achieve the sequential movement. The cerebellum has been hypothesized to participate to sensorimotor integration and error correction that facilitates the first phases of learning (Penhune & Steele, 2012). In accordance with this hypothesis, Lehericy et al. (2005) evidenced that activation in lobule VI of the cerebellum is related to motor sequence accuracy. Our results suggest that high functional connectivity at rest between rSM1 and these regions might facilitate the subsequent early phase of skill learning in young healthy adults.