Functional and effective reorganization of the aging brain during unimanual and bimanual hand movements

2.1 Participants

MEG recordings as well as structural T1w images were collected from 12 healthy young adults (mean age = 23.7 years) and 11 healthy elderly individuals (mean age = 67.4 years). Details regarding data acquisition and preprocessing are described in the Supporting Information. Details regarding demographic information and behavioral performance are presented in Table 1. Both groups were matched with respect to gender and education. Inclusion criteria for all participants were as follows: (i) no present or previous history of a psychiatric conditions, (ii) aged between 18-30 years (young group) and 60–75 years (elderly group), and (iii) right-handed according to the Edinburgh Handedness Inventory (Oldfield, 1971). Exclusion criteria included: (i) contraindications for MRI, or other limitations that would interfere with MRI or MEG data acquisition (e.g., claustrophobia, metal implants) and (ii) a Mini-Mental State Examination (MMSE) score ≤ 24. Written informed consent was obtained from all participants. The study was approved by the Research Ethics Board of the Montreal Neurological Institute and Hospital, McGill University.

2.2 Experimental design

As detailed in Supporting Information, motor performance of both hands was assessed for each participant via measurements of (i) handgrip strength, (ii) fine manual dexterity (nine hole peg test; NHPT), and (iii) unilateral gross manual dexterity (box and block test; BBT). Motor performance scores for the dominant hand (right hand) and nondominant hand (left hand) were used for group comparison.

All participants underwent three separate 5 min resting-state sessions, interspersed with two isometric handgrip tasks (Figure 1). During each resting-state session, participants were instructed to keep their eyes open and fixate on a cross. The first motor task consisted of 50 unimanual, visually-paced, isometric right-handgrips, in which subjects had to apply force to track a ramp target. Prior to scanning, subjects were asked to grip the manipulandum with maximum force in order to assess their maximum voluntary contraction (MVC). These values were then used to set the subject-specific target forces of 15% and 30% of MVC. In each trial, participants had to maintain a steady force at 15% of MVC for 3 s, followed by a linear increase of 3 s to reach and maintain a steady force at 30% of MVC for 3 s. The second motor task consisted of 50 bimanual, visually-paced, isometric handgrips performed at 15% of MVC (6 s each).

2.3 Task-based functional connectivity and statistical analyses

The task-based functional connectivity analysis pipeline (Supporting Information Figure S1) was conducted in Brainstorm (Tadel, Baillet, Mosher, Pantazis, & Leahy, 2011) and MATLAB, and was similar to that used in previous research (Larivière et al., 2017; Larivière, Ward, & Boudrias, 2018; Whitman et al., 2016). Specifically, for every participant, the task-related data were down-sampled to 160 Hz and epoched offline with a poststimulus time window of 9,000 ms (unimanual task) and 6,000 ms (bimanual task) with the first time point (time = 0) corresponding to stimulus onset. A linearly-constrained minimum variance (LCMV) beamformer spatial filtering approach (Van Veen & Buckley, 1988) was used on the subject-specific task-averaged epoched data to reconstruct a single time series for each of the 148 cortical brain regions defined by the Destrieux sulcogyral-based atlas (Destrieux, Fischl, Dale, & Halgren, 2010). For each pre-defined source location (i.e., brain region), activity was estimated at each vertex and subsequently averaged to produce a single time series per brain region. Time-frequency decomposition of source time series was then performed using Morlet wavelets (Tallon-Baudry & Bertrand, 1999) for two frequency bands of interest: alpha (8–12 Hz) and beta (13–30 Hz; Crone et al., 1998; Willemse et al., 2010; Yuan et al., 2010). Trial averaging preceded time-frequency analysis to specifically capture task-locked modulations and minimize non-task-related signals (David, Kilner, & Friston, 2006). Frequency-specific source time series for every subject were combined to create two data matrices (one per frequency band), each with columns corresponding to brain regions and rows corresponding to poststimulus time points × subjects. Singular value decomposition (SVD), of which PCA is a special case, was performed on each of the four standardized data matrices. For every component extracted, the resulting decomposition yielded (i) a spatial pattern of task-related variance in brain activity (i.e., a network constrained to the dominant 15% of interconnected brain regions derived from component loadings), resulting in independent sources of variance reflecting task-specific brain networks and (ii) component scores (i.e., time series) providing an estimate of the network's engagement at each poststimulus time point during the length of the trial, with component scores at each time point reflecting the influence of the spatial pattern for that given network. In other words, component score time series represent the strength of each network's signal, with higher/lower component scores depicting a stronger/weaker influence from that spatial pattern, and a component score of zero corresponding to baseline. The network-level connectivity analyses of oscillatory power described above were performed separately for each motor task (i.e., unimanual and bimanual).

Group differences in the activation level of each functional brain network at every poststimulus time point were statistically compared using repeated measures analysis of variance (ANOVA). Notably, repeated measures ANOVAs are suitable to test for group differences between time series as they do not assume non-independence among the repeated observations and allow for correlation between repeated measures within subjects. As such, the amplitude envelope of each network's associated time series was extracted using the Hilbert transform and then submitted to repeated measures ANOVA to test for group and time differences. For every unimanual network identified, the Hilbert transform values at each time point and for each subject were submitted to a 1,441 × 2 repeated-measures ANOVA, with the within-subjects factor of Poststimulus Time (1,441 time points were estimated after stimulus onset), and between-subjects factor of Group (elderly individuals and young adults). Similarly, for the identified bimanual networks, these Hilbert transform values for every subject were submitted to a 962 × 2 repeated-measures ANOVA, with the within-subjects factor of Poststimulus Time (962 time points), and between-subjects factor of Group. Moreover, we carried out a power analysis using G*Power (Franz Faul, Universität Kiel, Germany) and calculated the required effect sizes for the current study, with a sample size of 23 (n₁ = 12, n₂ = 11), power = 0.8, and alpha = 0.05. Minimum effect sizes based on our sample size are Cohen's f = 0.046 (unimanual task) and Cohen's f = 0.051 (bimanual task).

2.4 Granger causality analysis of task-specific brain networks

Granger causality has been increasingly used to identify the presence of directional interactions (or causal relations) in physiological systems (Schiatti, Nollo, Rossato, & Faes, 2015). This approach relies on the concept that a causal influence from a source region to a target region can be assumed if past information about the source region (i.e., its corresponding time series) improves the prediction of future values of the target region. In other words, Granger causality can provide insights as to how information propagates from one brain region to another. We conducted Granger causality connectivity analysis using the narrow band time series from each region within the task-specific, PCA-derived functional brain networks. Specifically, Granger causality was performed on each subject individually, and binary outcomes were coded 0 for nonsignificant causal relations (p > 0.05) and 1 for significant causal relations (p < 0.05) among all pairs of brain regions. The model order parameter of our multivariate autoregressive model was optimized using the minimum description length criterion and statistical significance of every pairwise causal relation was detected using an F-test. Findings were corrected for multiple comparisons, controlling at a false discovery rate of p < 0.05. Significant group-level causality maps (constrained to the significant, subject-level causal relations) were then detected using binomial p value computation for testing proportions. Specifically, for a given causal link, the binomial test uses the mean of all coded binary outcomes within a group (i.e., 0 s and 1 s) to compute the number of subjects presenting this significant causal link that is required for this link to be significant at the group-level (Siegal, 1956). As such, the resulting task-based Granger causality maps display the dominant patterns of cortical information flow that were significant both at the subject- and group-levels for every task-specific brain network.

2.5 Resting-state functional connectivity and statistical analyses

For each participant and each of the three resting-state sessions, the MEG data were down-sampled to 300 Hz and epoched offline in 10 s windows to obtain meaningful activation from both low- and high-frequency modulations (Supporting Information Figure S2). The choice of time window was further determined from previous work showing fairly robust test–retest reliability when using epochs of 10 s resting-state data (Colclough et al., 2016; Jin, Seol, Kim, & Chung, 2011). Epochs in which significant signal artifacts were observed were rejected and the remaining “clean” 10 s windows were concatenated across time. The LCMV beamformer spatial filtering approach (Van Veen & Buckley, 1988) was subsequently used on the subject-specific, concatenated data to reconstruct a single time series for all of the 148 cortical brain regions defined by the Destrieux atlas (Destrieux et al., 2010). Each time series was corrected for signal leakage effects (i.e., spurious correlations between the inferred cortical sources) using a symmetric, multivariate correction method intended for all-to-all functional connectivity analysis (Colclough, Brookes, Smith, & Woolrich, 2015). The Hilbert transform was subsequently used to extract the instantaneous power and phase within six frequency bands of interest: delta (1–4 Hz), theta (5–7 Hz), alpha (8–12 Hz), beta (13–30 Hz), gamma “low” (31–80 Hz), and gamma “high” (81–150 Hz). Resting-state functional connectivity analysis was performed for each of the six frequency bands by systematically computing pairwise envelope correlations between all 148 source-reconstructed brain regions. The resulting all-to-all connectivity matrices (one per frequency band) were sorted in functional networks according to the recently proposed seven-network brain cortical parcellation estimated by intrinsic functional connectivity using resting-state fMRI data from 1,000 healthy adults (Yeo et al., 2011). This network parcellation provided spatial consistency across all subjects as well as between resting-state runs, thereby making direct comparison of functional network connectivity possible. Here, functional connectivity was defined as the mean connectivity strength (i.e., the mean of all pairwise correlations) within each of the pre-defined seven resting-state networks. Differences between groups (young vs. elderly), runs (resting-state 1, 2, and 3), and mean connectivity strength for each of the seven resting-state networks were identified by carrying out six 7 × 3 × 2 mixed-model ANOVAs (one per frequency band). Tests of sphericity were carried out for all ANOVAs and Greenhouse–Geisser adjusted degrees of freedom were checked. Original degrees of freedom are reported as any violations of sphericity did not affect the results.

2.6 Associations between task-based findings and resting-state connectivity changes

To relate resting-state connectivity changes to task-based functional network activity and motor task performance, we calculated Pearson correlations between resting-state connectivity changes following both unimanual and the bimanual tasks (i.e., taken as the difference in connectivity between the second and first resting-state scan as well as between the third and second resting-state scan, respectively) and (i) the levels of coordinated activity in each task-related network and (ii) motor task performance scores.

3.1 Behavioral results

As displayed in Table 1, the behavioral scores for each hand were entered into two-sample t tests to compare motor performance between young and elderly individuals. For both hands, young adults performed significantly better than elderly individuals on the BBT and NHPT (p values < 0.01), whereas grip strength did not differ between groups (p values > 0.09).

As for task performance, we defined task accuracy as the difference between the grip force applied and the position of the ramp target; higher accuracy was achieved when participants closely matched the target force, defined by the middle of the target ramp. We computed the mean accuracy of all trials for every participant (within each task independently), and subsequently tested for significant differences between young and elderly individuals for each task by entering the accuracy values into two-sample t tests. No significant differences were observed between both groups with respect to task accuracy during the unimanual task (using the dominant right hand; p > 0.3) as well as during the bimanual task (using the dominant right hand and the non-dominant left hand; p values > 0.1).

3.2 Task-based functional connectivity results

Inspection of the scree plot of singular values was carried out for the two frequency bands of interest (alpha, beta) and for each motor task (unimanual, bimanual). Visual inspection of every component (i.e., network) extracted from our task-based analysis was performed. Brain networks which did not include motor regions were excluded from the analysis and are not discussed below. This led to the inclusion of three unimanual networks (all beta-related) and three bimanual networks (two beta-related, one alpha-related). The brain regions and estimated time series associated with each network are displayed in Figure 2a–c (unimanual networks) and Figure 3a–c (bimanual networks), and described below. Anatomical descriptions for each network are presented in Supporting Information Tables S1–S3 (unimanual networks) and Supporting Information Tables S4–S6 (bimanual networks).

3.3 Functional networks underlying unimanual handgrips

3.4 Functional networks underlying bimanual handgrips

3.5 Association between behavioral scores and brain activity

To examine the relationship between behavioral motor performance and levels of task-related brain activity, we carried out Pearson correlation analyses between task accuracy scores and the level of engagement for each functional network (i.e., the component scores). Despite the low variability in task accuracy scores between participants, older adults demonstrated a positive relationship between task-related brain activity and behavioral motor scores in four functional networks: default-mode and motor networks (unimanual task; Supporting Information Figure S4a), as well as left- and right-dominant motor networks (bimanual task; Supporting Information Figure S4b).

3.6 Granger causality

We investigated the direction of information flow from and to every brain region derived from all six task-based networks extracted from the functional connectivity analysis. Binomial statistics revealed that causal links were significant at the group-level if the links were significant at the individual-level in at least five subjects (i.e., 5/12 for the young group and 5/11 for the elderly group), that is: p value = P(X ≥ 5 | p = μ_group) < 0.05. We used a multivariate autoregressive model of order 3, meaning that the time lag between interacting neuronal ensembles corresponded to 18.75 ms (i.e., 3/160). Granger causality maps for the unimanual and bimanual networks as well as for each group are depicted in Figure 4a–f, respectively. Brain region abbreviations for the Granger causality maps are listed in Table 2.

3.7 Resting-state functional connectivity results

A significant main effect of Run was observed in two frequency bands, specifically delta (F_{2, 42} = 11.1, p < 0.005) and beta (F_{2, 42} = 4.7, p < 0.05), whereas a significant main effect of Network was found in all frequency bands: delta (F_{6, 126} = 14.3, p < 0.001), theta (F_{6, 126} = 14.3, p < 0.001), alpha (F_{6, 126} = 56.3, p < 0.001), beta (F_{6, 126} = 14.3, p < 0.001), gamma “low” (F_{6, 126} = 8.8, p < 0.001), and gamma “high” (F_{6, 126} = 8.0, p < 0.001). Significant interactions involving Network, Run, or Group were solely observed in the delta and beta frequency bands and are described below.

3.8 Task-induced connectivity changes in the delta frequency band

Slow oscillatory connectivity (1–4 Hz) differed between young and elderly subjects across different resting-state sessions as evidenced by a significant Run × Group interaction (F_{2, 42} = 5.61, p < 0.05, η²_p = 0.21). Within-subjects contrasts yielded significant group differences from the first to second resting-state run (p < 0.005), and from the second to the third run (p < 0.05). As can be seen from Figure 5a, this interaction was caused by elderly subjects exhibiting a large increase in delta connectivity in the second resting-state run (i.e., after the unimanual task) relative to young adults. Correlations between unimanual task-based functional findings and resting-state connectivity changes (i.e., difference between second and first resting-state scans) revealed a significant relationship between ventral frontoparietal network activity and resting-state delta increases following the unimanual task in older adults (r = 0.53, p < 0.05). Moreover, a negative association between delta-related resting-state connectivity changes and task performance was observed in young adults (r = −0.50, p < 0.05) but was marginal in elderly individuals (r = −0.30, p > 0.18; Supporting Information Figure S5a). The analogous correlations involving resting-state connectivity changes following the bimanual task were all nonsignificant (Supporting Information Figure S6a).

3.9 Task-induced connectivity changes in the beta frequency band

The beta frequency (13–30 Hz) showed a significant Run × Network interaction (F_{12, 252} = 2.76, p < 0.05, η²_p = 0.12), indicating that beta-related resting-state network connectivity varies as a function of time (i.e., resting-state run). Figure 5b shows that this interaction can be interpreted by enhanced connectivity from the first to the second resting run (i.e., increased connectivity after the unimanual task), notably in the visual, dorsal attention, and sensorimotor networks. A significant Network × Group interaction was also observed (F_{6, 126} = 3.43, p < 0.05, η²_p = 0.14) suggesting that elderly subjects demonstrated slightly higher beta oscillatory connectivity than young adults in all resting-state networks (nonsignificant, p values > 0.36), with the exception of the visual network (p < 0.05; Figure 5c). Correlations between unimanual and bimanual task-based functional findings and resting-state connectivity changes did not yield any significant associations; however, older adults demonstrated a strong positive relationship between beta-related resting-state connectivity increases following unimanual handgrips and task performance (r = 0.64, p < 0.05), an association that was absent in young individuals (r = 0.20, p > 0.26; Supporting Information Figures S5b,S6b).

4.1 Age-related connectivity changes in unimanual movements

During movement and performance of visuomotor control tasks, primary and secondary sensorimotor networks are engaged to implement motor behavior, while networks comprising higher-order regions are recruited to ensure cognitive resources and goal-directed behavior (Serrien, Ivry, & Swinnen, 2007). Indeed, tasks that demand externalized attention to a visually presented stimulus have been shown to reliably activate regions within the frontoparietal network (Corbetta & Shulman, 2002; Spreng, Stevens, Chamberlain, Gilmore, & Schacter, 2010), with older adults showing higher activity increases in the prefrontal and parietal cortex (Cabeza et al., 2002; Madden et al., 2007). Our findings revealed a similar pattern of hyperconnectivity during the unimanual task, suggesting that attentional demands are significantly increased when elderly participants are required to perform handgrips. Accordingly, this greater cortical activation could further reflect a mechanism by which goal-directed attentional control is employed by elderly individuals in an attempt to dampen the processing of task-irrelevant stimuli (Milham et al., 2002). Unlike the attention network, in which increased activation was indicated by overall higher oscillatory amplitude, enhanced activity in the motor network was characterized by sharp beta peaks in elderly individuals. According to computational modeling studies on transient neocortical beta rhythms, these sharp beta oscillations likely reflect greater temporal synchrony in synaptic input on cortical pyramidal neurons (Sherman et al., 2016). In line with the latter finding, our effective connectivity analysis based on Granger causality revealed enhanced interhemispheric information flow from ipsilateral (right) primary and secondary sensorimotor regions as well as parietal regions to contralateral (left) M1 during unimanual handgrips in the elderly group. Combined, these findings are in agreement with an influential theory of brain reorganization suggesting a loss of asymmetry in connectivity patterns in older adults (Heuninckx et al., 2005, 2008; Ward & Frackowiak, 2003), possibly reflecting a mechanism whereby greater synchrony of excitatory input currents are recruited to counteract known age-related structural (Salat et al., 2004, 2005), myeloarchitectural (Bartzokis, 2004), and neurochemical changes (Mora et al., 2008). Future neuroimaging studies integrating network connectivity measures with microstructural features, such as myelination and tissue microstructure, may enrich our understanding of the intriguing associations existing between beta oscillatory rhythms and myelin in aging populations as well as in neurodegenerative disorders (Lariviere et al., 2018).

4.2 Age-related connectivity changes in bimanual movements

Neurophysiological and neuroimaging studies of bimanual hand movements support the view that the dominant hemisphere, as opposed to each contralateral hemisphere independently, controls the organization of bimanual hand movements (Serrien, Cassidy, & Brown, 2003). In line with these earlier observations, our whole-brain multivariate analysis of the bimanual task data revealed three functional networks of interest: a left-dominant, a right-dominant, and a bilateral motor network. Akin to our unimanual findings, we found age-related connectivity increases in the left-dominant and bilateral motor networks during sustained isometric bimanual handgrips. Interestingly, while the unimanual task only elicited networks in the beta band, both alpha and beta motor networks emerged during the bimanual task. Based on previous electrophysiological and fMRI studies highlighting the importance of the dominant hemisphere during bimanual hand movements (Jancke et al., 1998; Serrien et al., 2003; Viviani, Perani, Grassi, Bettinardi, & Fazio, 1998), we speculate that task-related activity from the nondominant side (i.e., right hemisphere) is suppressed—or driven into a more “rest-like” state—by the dominant hemisphere, an interaction that may be accompanied by more pronounced alpha brain waves. Alternatively, it has been suggested that beta-related activity preferentially relates to motor aspects of coordination while alpha-related activity underlies somatosensory processing (Pfurtscheller, Stancak Jr., & Neuper, 1996; Salmelin & Hari, 1994). In line with our current findings, this alpha-beta dichotomy may reflect different systems whereby beta-related dominant (i.e., left) and bilateral motor network activity is involved in the initial production of bimanual handgrips, whereas alpha-related nondominant (i.e., right) motor network activity rather reflects a mechanism by which the nondominant hemisphere processes sensory information from the dominant hemisphere and adjust its motor output accordingly.

The effective connectivity maps for the bimanual task-based networks revealed that young adults rely more on prefrontal regions, such as the frontal pole, suborbital sulcus, orbitofrontal cortex, and medial prefrontal cortex. This pattern of strongly interconnected frontal regions was largely absent in the elderly group, which is in agreement with previous cognitive studies reporting age-related activity decreases in prefrontal regions (Cabeza et al., 2002; Damoiseaux et al., 2007). Yet another striking difference in connectivity patterns between young and older adults lies in the importance of the left temporal pole region in the bilateral motor network. According to our Granger causality analysis, this brain area appears highly integrated within the bilateral motor network in the young group as it receives cortical information from several parietal and frontal regions (including left PMd), but also has a causal influence on the posterior cingulate cortex, a connectivity pattern that is largely absent in elderly individuals. Our findings further demonstrate that bimanual movements in elderly individuals appear to be controlled from each contralateral hemisphere independently—with minimal interhemispheric connectivity—thus expanding upon previous research suggesting that the neural organization of the bilateral motor network in adults is driven by the dominant (left) hemisphere (Serrien et al., 2003). This lack of coordination between hemispheres may result from altered white matter integrity of the corpus callosum, which plays a key role in allowing both hemispheres to communicate during bimanual movements (Kennerley, Diedrichsen, Hazeltine, Semjen, & Ivry, 2002; Ota et al., 2006). Indeed, previous research investigating the interhemispheric connections of the temporal lobes in monkeys demonstrated that the corpus callosum receives extensive fibers from the temporal pole (Demeter, Rosene, & Van Hoesen, 1990). This reinforces the importance of the left temporal pole in coordinating movement in both hands during bimanual movements. Our results further suggest that this region is a central hub responsible for mediating information flow between the two hemispheres. Due to their long-distance connectivity and their central role in communication and integration of information in the brain, cortical hubs are known to be represent a high-cost feature of brain connectivity and as a consequence become highly vulnerable in aging (Crossley et al., 2014).

4.3 Task-induced modulation of resting-state network connectivity

Previous research studying the effects of intensive motor learning on subsequent resting-state brain activity in healthy adults have confirmed the presence of task-induced short- and long-term changes in cortical properties (Bellec et al., 2015; Vahdat, Darainy, Milner, & Ostry, 2011). In line with these reports, our analysis of the resting-state runs acquired before and after each motor task revealed significant connectivity increases following the unimanual task relative to baseline (i.e., the first resting-state session), but significant connectivity decreases following the bimanual task (relative to the second resting-state session). This enhanced connectivity pattern was observed in both young and elderly subjects, and was particularly noticeable in resting-state cortical networks that included regions previously engaged during the unimanual task (e.g., visual, dorsal attention, sensorimotor, default-mode networks). On the other hand, the lack of task-induced connectivity increases in the third resting-state scan (i.e., following the bimanual task) may relate to longer trial duration in the unimanual task (9 s unimanual handgrips vs. 6 s bimanual handgrips) which in turn yielded a stronger and/or longer-lasting postmovement rebound effect. Notably, our data suggest that unimanual hand movements can induce similar neuronal flexibility, as reflected by an increase in resting brain connectivity relative to baseline, even in the absence of external stimulation or intensive motor learning. However, future research protocols encompassing an aspect of motor learning, whereby motor performance changes across time, may be better suited to study the aging brain's ability to adapt to task demands. While the evidence reported herein indicates that healthy adults, irrespective of age, retain the capacity for task-induced connectivity changes at a systems level, elderly individuals also showed task-induced increases in slow wave oscillatory connectivity. Considered alongside cognitive studies in aging which have reported enhanced interhemispheric delta coherence (Maurits, Scheeringa, van der Hoeven, & de Jong, 2006; Vecchio, Miraglia, Bramanti, & Rossini, 2014), the large increase in delta connectivity following the unimanual task may play a role in modulating attentional or cognitive resources, and may therefore represent a marker of healthy neurocognitive aging.

4.4 Limitations, methodological considerations, and future directions

We investigated group differences in activity levels of commonly shared task-based brain networks which precluded us from determining whether, and how, spatial reorganization occurs during healthy aging. While these questions were outside the scope of this study, our Granger causal connectivity approach provided initial insights into age-related changes in the spatial pattern of information flow involved in motor processes. Further, the relatively small number of participants in each group may have contributed to low statistical power and an inability to detect significant group differences. While results reported here are consistent with previous fMRI and EEG studies with larger sample sizes (Maes et al., 2017; Sala-Llonch, Bartrés-Faz, & Junqué, 2015), reproducibility of our findings should be assessed in a similar, but larger, cohort of young and elderly individuals. Lastly, our choice of processing pipeline may have precluded the identification of modulatory effects that were not time-locked to the production of isometric handgrips. While our trial averaged connectivity analyses may have led to nonphase-locked task-related signal loss in alpha and beta frequency bands (Brookes et al., 2016), our analytical framework was primarily designed to target oscillations that were directly evoked by the production of isometric handgrips while also suppressing non-task-related signals as well as physiological and measurement noise (David et al., 2006).

4.5 Conclusions

The present work shows that despite matching levels of task accuracy, brain organization in elderly individuals is characterized by hyperactivity in MEG networks specifically underlying the performance of unimanual and bimanual hand movements. Furthermore, spatial patterns of information flow involved in motor processes significantly differed between young and older adults, with the latter group showing enhanced connectivity from ipsilateral frontal and parietal motor regions to contralateral M1 during unimanual movements. Interestingly, while frontal and temporal regions acted as integrative hubs for the coordination of interhemispheric information flow in young adults, elderly individuals demonstrated relatively little fronto-temporal information flow during the bimanual task. Finally, the combination of task-related and resting-state protocols described here represents a valuable tool for future research aiming at understanding brain organization during healthy aging. Our integrative experimental approach may further contribute to the refinement of rehabilitation strategies aiming at enhancing neural flexibility in movement-impaired populations by means of exercise programs targeting hand movement.