Brain connectome correlates of short-term motor learning in healthy older subjects

in


Introduction
Motor learning refers broadly to a change in the capacity to execute a motor task as a result of practice.It can occur across different timescales, leading from temporary gains in motor performance, often termed motor adaptation, to permanent acquisition of motor skills (Weaver, 2015).Since motor learning relies on the integrative contribution of brain cortical and subcortical systems to different aspects of the process (Graydon, Friston, Thomas, Brooks, & Menon, 2005), it entails changes across multiple brain regions (Dayan & Cohen, 2011), which may be limited to functional changes or extended to structural changes depending on the timescale of motor learning (Landi, Baguear, & Della-Maggiore, 2011;Scholz, Klein, Behrens, & Johansen-Berg, 2009).While brain functional changes were often observed for local activation during a motor task (Orban et al., 2010(Orban et al., , 2011)), it has been increasingly understood that motor learning-induced functional changes could be manifested in terms of taskrelated or resting state functional connectivity (Coynel et al., 2010;Wu, Chan, & Hallett, 2008) and collectively the connection matrix of the brain, referred to as the brain connectome.Usually based on diffusion-weighted MRI (dMRI) and functional MRI (fMRI), the brain connectome has been suggested as structural (here referred to as the brain structural connectome) and functional (here referred to as the brain functional connectome) descriptions, respectively, of the brain.The support of the brain functional connectome for motor learning has been assessed in terms of network topology, specifically efficiency (Heitger et al., 2012;Sami & Miall, 2013;Zang et al., 2018) and modularity (Bassett et al., 2011), but behaviourally relevant changes in network topology that underlie motor learning ability have yet to be further clarified according to different timescales of motor learning.
The pattern of the brain functional connectome tends to be promoted or constrained by the architecture of the brain structural connectome, as can be simulated by generative models (Mess e, Rudrauf, Giron, & Marrelec, 2015), so that there are relationships between the brain structural and functional connectomes.Given dynamic changes in the brain functional connectome in motor learning (Bassett et al., 2011), it is likely that the correspondence between the brain structural and functional connectomes would evolve as well.Especially, in short-term motor learning during which the brain structural connectome could be supposed to remain static, adaptive changes in the brain functional connectome would directly shape alterations in brain structureefunction correspondence.In that regard, motor learning could be characterized by whether brain functional changes would lead to a coupling or uncoupling of the brain functional connectome from the brain structural connectome.
In the current study, for short-term motor learning with a visually guided sequential hand grip learning task, we sought to examine brain functional changes in terms of network topology of the brain functional connectome and, moreover, parallel changes in the correspondence between the brain structural and functional connectomes.Considering regional differences in brain structureefunction correspondence (V azquez-Rodrı ´guez et al., 2019), we assessed the motor learning-induced changes across distinguished cerebral networks as well as the whole brain network.In particular, we wondered whether the inter-individual behavioural variability of a cohort of older adults who may be accompanied by progressively impaired motor learning ability (Mattay et al., 2002;Sailer, Dichgans, & Gerloff, 2000) would relate to the variability of resting-state networks changes associated with older age (Jockwitz & Caspers, 2021).Considering the involvement of prefrontal (Halsband & Lange, 2006) and associative (Hardwick, Rottschy, Miall, & Eickhoff, 2013) areas in initial phases of motor learning, we hypothesized that behaviourally-relevant brain connectome changes would be manifested across different cerebral networks apart from the somatomotor network.

Methods
We report how we determined our sample size, all data exclusions, all inclusion/exclusion criteria, whether inclusion/ exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study.

Subjects
Forty-three healthy older subjects initially participated in the study, with exclusion criteria of psychoactive medication use, drug or alcohol abuse, pregnancy, inability to follow study procedures, or contraindications to MRI.Among those, 39 subjects (age: mean (SD) ¼ 69.7 (4.7) years, men:women ¼ 15:24) were finally included in the analysis, whereas the other four subjects were excluded due to missing or abnormal MRI data.Handedness of the subjects was confirmed to be right-handed according to the Edinburgh handedness inventory questionnaire (mean laterality quotient (SD) ¼ 83.6 (20.5)) (Oldfield, 1971).The study was approved by the cantonal ethics committee Geneva (project number: 2017e00224), and the written informed consent was obtained from all the subjects.The study conformed to the standards according to the Declaration of Helsinki.

Motor learning ability
Inside an MRI scanner, the subjects performed two subsequent sessions of a visually guided sequential hand grip learning task adapted from the previously developed one (Wessel et al., 2020), with each session composed of nine training blocks containing 15 hand grip trials (three repetitions of a sequence of five hand grip trials) each (Fig. 1A and C).
With each session lasting around 12 min dependent on individual subjects' reaction times, the total time spent for the task was within half an hour by including a short break between the two sessions.The task involved applying force on a gripper (Fig. 1B) that controlled the height of a cursor on a computer screen to match the height of a target bar.The absolute height of a target bar was adapted according to each subject's maximum hand grip force, such that 70 % of the maximum force corresponded to 85 % of the height of the computer screen.During the task, the subjects were instructed to move a cursor to target bars in sequence as swiftly and accurately as possible by pressing and releasing the gripper, and they were expected to learn to track a sequence of hand grip trials demanding variable isometric force contraction with the non-dominant (left) hand.
Accuracy was evaluated by the proportion of hand grip trials that successfully reached target bars within a block.The elapsed time per trial was measured from the onset of cursor movement at the baseline to the stop of cursor movement at a target bar in successful trials or to the continued pause of cursor movement at least for 200 ms outside a target bar in unsuccessful trials.Motor task performance for each block was computed by the ratio of accuracy to the average elapsed time per trial.While it appears not to be consistent in the literature what motor learning ability refers to (Krakauer, 2019), we defined motor learning ability as a summary measure representing a change in motor task performance based on the capability to respond appropriately in the process of motor learning.Across the two sessions, individual subjects' motor learning ability was calculated by the ratio of the difference between later motor task performance (measured for the last two training blocks of the second session) and earlier motor task performance (measured for the first two training blocks of the first session) to earlier motor task performance: motor learning ability ¼ (later motor task performance e earlier motor task performance)/(earlier motor task performance).
The code for the gripper task is accessible at https://github.com/mdurandruel/grippertask.matrix size ¼ 112 Â 112, in-plane resolution ¼ 2.0 mm Â 2.0 mm, and multiband factor ¼ 7. T1-weighted structural MRI (sMRI) data composed of one volume image in sagittal planes were acquired with a 3D magnetization prepared rapid gradient echo (MPRAGE) sequence: TR ¼ 2,300 ms, TE ¼ 2.96 ms, number of slices ¼ 192, slice thickness ¼ 1.0 mm, matrix size ¼ 240 Â 256, in-plane resolution ¼ 1.0 mm Â 1.0 mm, and GRAPPA acceleration factor ¼ 2. For each subject, dMRI and T1-weighted sMRI data were obtained once after the hand grip learning task, whereas resting state fMRI data were acquired twice, before and after the hand grip learning task each (Fig. 1A).

MRI data processing
Using tools in MRtrix3 (https://www.mrtrix.org/)and FSL (https://fsl.fmrib.ox.ac.uk/fsl/), images of dMRI data were corrected for Gibbs ringing artefacts, field inhomogeneity, susceptibility-induced off-resonance field, and head motion and eddy currents.By estimating the fibre orientation distribution function within each voxel via multi-shell multi-tissue constrained spherical deconvolution (Jeurissen et al., 2014), whole-brain tractography was conducted based on the probabilistic algorithm of the second-order integration over fibre orientation distribution (Tournier et al., 2019).A total of 10 million streamlines were generated by initiating them at each voxel of the white matter.
Using tools in SPM12 (https://www.fil.ion.ucl.ac.kr/spm/), images of resting state fMRI data were corrected for different acquisition time across slices, field inhomogeneity, and head motion, and they were spatially smoothed with a 6 mm fullwidth at half-maximum (FWHM) Gaussian kernel.In addition, nuisance covariates regression was applied to model effects of low-frequency fluctuations, head movement, and non-neuronal fluctuations on resting state fMRI signals.Nonneuronal fluctuations were estimated from the average of signals extracted from each mask of the white matter and corticospinal fluid.
For slice timing correction, we have adopted the common method as implemented in SPM12 (Update Revision Number ¼ 7771; https://www.fil.ion.ucl.ac.uk/spm/software/ spm12/).We used the slice timing, instead of the slice order, as an input to slice timing correction, since a slice order cannot represent multiple slices acquired at the same time (https://en.wikibooks.org/wiki/SPM/Slice_Timing).While the benefit of slice timing correction on fMRI data with short TR and multiband acquisition has been suggested (D.B. Parker & Razlighi, 2019), slice time correction may be less necessary in the case since the temporal difference between slices is reduced.

Brain connectome analysis
For the non-linear registration between the dMRI or resting state fMRI data native space and the standard space, T1weighted sMRI data were used to estimate transformation parameters.For 246 brain regions as defined by the Brainnetome atlas (https://atlas.brainnetome.org/) in the standard space, a brain structural connectome was constructed by selecting fibre bundles that connected each pair of the 246 brain regions among those over the whole brain.For the same 246 brain regions, a brain functional connectome was constructed by computing the correlation of signals between each pair of the 246 brain regions.That is, nodes were commonly defined by the 246 brain regions, while edges between the nodes were estimated by fibre bundles for the brain structural connectome and by signal correlation for the brain functional connectome.
Given the brain structural connectome at one time point (after the hand grip learning task) and the brain functional connectome at two time points (before and after the hand grip learning task), we assumed that the brain structural connectome remained static on the short timescale of motor learning, such that the brain structural connectome could be regarded not only as a baseline, but also as being unchanged thereafter.Thus, in addition to functional network topology measured for the brain functional connectome, the correspondence between the brain structural and functional connectomes was measured at each time point, so that changes in structureefunction correspondence as well as functional network topology over the two time points could be assessed.
For a brain structural connectome, a sparse binary network was defined by considering that an edge is not existent between a pair of nodes when there was no fibre bundle tracked between the two nodes.For a brain functional connectome, a sparse binary network was defined by supposing that an edge is not existent between a pair of nodes when the correlation of signals between the two nodes failed to pass the false discovery rate correction for multiple comparisons at the significant level of p .05.Given a sparse binary network derived from a brain structural connectome or a brain functional connectome, network topology was evaluated in terms of efficiency and modularity.Efficiency, as a measure of how efficiently a network exchanges information (Latora & Marchiori, 2001), was computed by averaging inverse shortest path lengths between nodes.Modularity, as a measure qualifying community structure in a network (Newman, 2006;Newman & Girvan, 2004), was computed by comparing the number of edges included in nonoverlapping groups in a given network against an equivalent network with edges connected at random.
The correspondence between the brain structural and functional connectomes was assessed by the multilinear regression fit of an observed brain functional connectome to a predicted brain functional connectome, as proposed before (V azquez-Rodrı ´guez et al., 2019).The predicted brain functional connectome was generated by the linear combination of the Euclidean distance, path length, and communicability in the brain structural connectome.The goodness of fit in terms of the coefficient of determination (R 2 value) between the observed and predicted brain functional connectomes provided structureefunction correspondence at each node.
For the brain structural and functional connectomes, in order to address variable involvement of different functional systems in motor learning, seven cerebral networks that have been divided according to the similarity of signals between brain regions (Yeo et al., 2011) were considered, so that network topology and structureefunction correspondence were measured for the seven cerebral networks as well as the whole brain network.The seven cerebral networks included the visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default mode networks that consisted of 34,33,30,22,26,26, and 36 cortical regions, respectively.The code for the analyses is accessible https://gitlab.com/chang-hyun.park/brainconnectome.

Brain connectome association with motor learning ability
Given changes in functional network topology and structureefunction correspondence between before and after the hand grip learning task, the correlation between them was assessed to test whether changes in functional network topology could be directly transferred to changes in structureefunction correspondence.In addition, brain connectome correlates of motor learning ability were evaluated from two perspectives.Firstly, the association of changes in functional network topology and structureefunction correspondence during the hand grip learning task with motor learning ability was assessed to check what changes in the brain connectome could support motor learning ability.Secondly, the association of network topology and structureefunction correspondence before the hand grip learning task (at baseline) with motor learning ability was assessed to check what substrates of the brain connectome could be predictive of motor learning ability.All statistical inferences were conducted by adopting permutation tests, in which the null distribution of a test statistic was obtained by repeatedly computing the test statistic through 1000 times of rearrangements of the subjects' labels, after adjusting for the effects of the subjects' age and sex.Statistical significance was determined at p .05, specifically corrected for multiple comparisons by a false discovery rate approach in the case of considering the seven cerebral networks.

Motor learning ability
Among the 39 subjects, later motor task performance (mean (SD) ¼ 59.0 (20.1)) was improved compared with earlier motor task performance (mean (SD) ¼ 39.6 (15.1)) in 35 subjects (Fig. 2).Therefore, the subjects' motor learning ability (mean (SD) ¼ .64(.68)) was generally shown as positive values, while interindividual variations in them were large.Four subjects showed a decrease in motor task performance.These decreases might be associated to a reduction of attention supported by the fact that the VAS rating of attention of these subjects was lower than of the rest of the cohort.

Brain connectome change
While changes in functional network topology between before and after the hand grip learning task were not significant in the whole brain network and neither in the seven cerebral networks (Supplementary Figure 1A), efficiency and modularity showed trends for opposite directions in their changes, as seen before for a different form of shortterm motor learning (Sami & Miall, 2013).Brain structureefunction correspondence exhibited different relationships between the brain structural and functional connectomes across the seven cerebral networks, as observed previously (V azquez-Rodrı ´guez et al., 2019), at both time points (Supplementary Figure 2).Although there were changes in structureefunction correspondence in some brain regions, significant changes were observed in neither the whole brain network nor the seven cerebral networks (Supplementary Figure 1B).Between functional network topology and structureefunction correspondence, efficiency positively correlated with structureefunction correspondence (r ¼ .72,p < .01),whereas modularity negatively correlated with structureefunction correspondence (r ¼ À.66, p < .01)for their changes in the visual network (Supplementary Figure 3).
Motor learning ability was not related to network topology of the brain structural and functional connectomes before the hand grip learning task (Supplementary Figure 4).In contrast, greater structureefunction correspondence in the dorsal attention (t ¼ 1.90, p ¼ .04),ventral attention (t ¼ 2.33, p < .01),and frontoparietal networks (t ¼ 2.24, p ¼ .01)as well as the whole brain network (t ¼ 1.96, p ¼ .03)before the hand grip learning task was related to motor learning ability (Fig. 4).

Discussion
In short-term motor learning, correspondence between the brain structural and functional connectomes changes most likely due to learning-related changes of the functional connectome, while the brain structural connectome remains unchanged.In this study, we sought to track changes in the brain functional connectome and its effects on changes in brain structureefunction correspondence after a short period of motor learning.Motor learning ability was attributable to decreased efficiency and increased modularity of the brain functional connectome and correspondingly decreased correspondence between the brain structural and functional connectomes over the visual and cognitive networks (Fig. 5).
In addition, motor learning ability could be predicted by the brain connectome determined before starting the motor learning task.Interestingly, only structureefunction correspondence over cognitive networks, but not structural or functional network topology alone, allowed to predict motor learning ability.Indeed, higher baseline structureefunction correspondence was related to superior motor learning ability.
The rationale behind increasing attention to the notion of the brain connectome is that the brain can be seen as a network machine (Sporns, 2013) and there is a strong need of knowledge about the different processes occurring within this network.In this context, network interconnections are key elements in understanding brain functioning (Bargmann & Marder, 2013), specifically with respect to connectivity patterns, for instance, the interplay between segregation and integration (Tononi, Sporns, & Edelman, 1994).New insights have been offered by brain connectomics for plastic changes in the brain, such as during normal development (Tymofiyeva, Hess, Xu, & Barkovich, 2014), after brain diseases (Griffa, Baumann, Thiran, & Hagmann, 2013), especially in recovery after stroke (Guggisberg et al., 2019;Koch et al., 2021;Egger et al., 2021), and during training and learning (Taya, Sun, Babiloni, Thakor, & Bezerianos, 2015).Here, we focused on plastic changes in the brain during a short period of motor learning from the perspective of brain connectome changes.While average changes in the brain connectome across the subjects were not clearly seen on the short timescale, changes in the brain connectome in association with individual differences in motor learning ability were revealed.We suppose that lack of significant brain connectome changes at the group level may be due to inter-individual variations in brain connectome changes in line with the diversity of brain  reorganization in ageing (Stumme et al., 2020), rather than due to generally trivial brain connectome changes across the subjects.
While higher efficiency at baseline has been suggested as a predictor of motor learning ability (Zang et al., 2018), intelligence (Langer et al., 2012), and robustness to cognitive impairment (Tuladhar et al., 2016), we show here that changes in motor learning performance are related to changes towards a decrease in efficiency, representing the reduction of integration of information transfer within the visual, somatomotor, and frontoparietal networks.The modulation of integrity within the functional systems may reflect less demand for information exchange in consequence of more practice, while a need for integration between different functional systems may be arisen as motor learning progresses (Coynel et al., 2010).
In contrast, motor learning ability-related changes in modularity were in the opposite direction, indicating enhancement of the quality of modular structure over the same networks.The contribution of increased modularity to motor learning ability appears to represent selective adaptability or flexibility required for motor learning that could be furnished by modular structure (Bassett et al., 2011).A shift of functional network topology towards a modular organization may not be limited to motor learning, but could be a more general mechanism that underlies, for example, working memory (Stevens, Tappon, Garg, & Fair, 2012).
Although the brain functional connectome could be shaped by the brain structural connectome, brain structuree function correspondence is not fixed due to dynamic changes in the functional part of the connectome that can occur even on a short timescale.Indeed, here we revealed that changes in the brain functional connectome led to changes in brain structureefunction correspondence in short-term motor learning, such that decreased correspondence in the visual, ventral attention, and frontoparietal networks contributed to motor learning ability.According to the notion that relatively low structureefunction correspondence could promote functional flexibility (Baum et al., 2020), decreased correspondence may reflect enhanced flexibility in the functional systems that supported motor learning based on successful brain dynamics.Considering that flexibility could be a main attribute to drive desired motor learning (Bassett et al., 2011;Reddy et al., 2018), a demand for flexibility in motor learning appears to be represented in this study by detachment of the brain functional connectome from the brain structural connectome, along with a shift of functional network topology towards a modular organization.A demand for flexibility specifically in the visual and cognitive systems may be related to the establishment of new associations between environmental targets and motor actions for the development of automaticity in motor learning (Hardwick et al., 2013).
Assuming the static brain structural connectome during short-term motor learning, we suppose that increased segregation and decreased integration in the brain functional connectome generally led to its uncoupling from the brain structural connectome.However, in the somatomotor system, unlike the visual and cognitive systems, motor learning ability-related changes in functional network topology did not lead to decreases in structureefunction correspondence, reflecting a possibly reduced demand for flexibility.This may be related to the continued involvement of the somatomotor system in the process of motor learning, while the visual and cognitive systems tend to be more dynamically involved probably only in an early stage of motor learning (Hardwick et al., 2013;Berghuis et al., 2019).Besides, it would be notable that functional network topology is only a facet of the brain functional connectome, so that its changes may not comprehensively explain changes in structureefunction correspondence.
In the context of this study, the evaluation of brain connectomics did not only allow to reveal connectome changes that underlie inter-individual variability in motor learning ability, but also to identify basic connectome information for predicting the magnitude of motor learning.It is of interest that only structureefunction correspondence, but not structural and functional network topology of the brain connectome, allowed to predict motor learning ability.This suggests the value of relationships between structure and function in explaining individual differences in the potential ability of motor learning, not only in specific brain regions (Tomassini et al., 2011), but also over more widespread brain networks.
Brain structureefunction correspondence appears to contain information distinguished from the sourced brain structural and functional connectomes, providing signatures for the network organization of individual brains (Griffa, Amico, Li egeois, Van De Ville, & Preti, 2022).In particular, the frontoparietal network has been suggested to include information for individual fingerprinting and, moreover, individual differences in cognitive traits in terms of brain structureefunction correspondence (Petrovic, Liegeois, Bolton, & Van De Ville, 2020).In a similar vein, it may be proposed that brain structureefunction correspondence specifically over the cognitive networks including the frontoparietal network could serve as substrate of individual subjects' motor learning ability.
The current study was performed for healthy older subjects; thus, we cannot exclude that changes in the brain functional connectome and their relationships with motor learning might be confounded by aging effects.For instance, age-related alterations in functional network topology and brain structureefunction correspondence, featured by decreases in efficiency (Sun, Tong, & Yang, 2012) and brain structureefunction correspondence (Esfahlani et al., 2022), may be noted.Furthermore, a possibility of overexpressed involvement of the cognitive networks due to a need for greater brain resources in older subjects (Wu & Hallett, 2005) may be taken into account.In future investigations, it needs to be checked whether the current findings regarding enhanced flexibility in motor learning in the view of brain connectome changes apply across the lifespan or they may be rather specific to older age.
In addition, there are limitations of this study in that reproducibility of the current findings may be affected by the choice of methodological approaches.In resting state fMRI data processing, for instance, an advanced slice timing correction method (Parker, Liu, & Razlighi, 2017) could be employed for data with short TR and multiband acquisition like ours, and an appropriate approach to spatial smoothing needs to be decided by considering effects of spatial smoothing on network topology (Alak€ orkk€ o, Saarim€ aki, Glerean, Saram€ aki, & Korhonen, 2017).

Conclusions
In this study, we showed that ability in short-term motor learning was attributable to higher brain structureefunction correspondence in the cognitive networks at baseline and reduced brain structureefunction correspondence in the visual and cognitive networks, which have been induced by topological reorganization of the brain functional connectome, during motor learning.These findings underscore brain connectome correlates of motor learning, in terms of a demand for flexibility in the visual and cognitive system, as supported by increased segregation and decreased integration over the systems.While we are motivated to examine brain connectome correlates of motor leaning on a longer timescale in future studies, the value of brain structureefunction correspondence on top of the sourced brain structural and functional connectomes stresses the importance of multimodal views on brain functioning.
Fig. 1 e Experimental design of a hand grip learning task.(A) Two subsequent sessions of a hand grip learning task administered between acquisitions of resting state functional MRI data.(B) An MRI-compatible fibre-optic grip force sensor used for the task.(C) A session of the task composed of nine training blocks containing 15 hand grip trials each.

Fig. 3 e
Fig. 3 e Brain connectome changes in association with motor learning ability.(A) Changes in network topology of the brain functional connectome.(B) Changes in brain structureefunction correspondence between before and after motor learning.Here the statistic represents t values.Functional network topology and brain structureefunction correspondence were measured for the whole brain network and seven cerebral networks.*, statistical significance.

Fig. 2 e
Fig. 2 e Motor learning during the hand grip learning task.(A) Motor task performance.Earlier and later motor task performance was assessed by averaging the two first blocks of the first training session and the two last blocks of the second training session, respectively.(B) Motor learning ability evaluated based on the ratio of the difference between earlier and later motor task performance.

Fig. 4 e
Fig. 4 e Brain connectome bases of motor learning ability.Here the statistic represents t values.Brain structureefunction correspondence before motor learning was measured for the whole brain network and seven cerebral networks.*, statistical significance.