Part 1:Age‐related Decline in Cortical Inhibitory Tone Strengthens Motor Memory
Mar 20, 2022
Contact: Audrey Hu Whatsapp/hp: 0086 13880143964 Email: audrey.hu@wecistanche.com
Pierre Petiteta,b,1,∗, Gershon Spitza,c,1, Uzay E. Emird,e, Heidi Johansen-Berga, Jacinta O’Sheaa,f
a Wellcome Centre for Integrative Neuroimaging, FMRIB Centre, Nuffield Department of Clinical Neurosciences (NDCN), John Radcliffe Hospital, Headington, Oxford, United Kingdom
b Centre de Recherche en Neurosciences de Lyon, Equipe Trajectoires, Inserm UMR-S 1028, CNRS UMR 5292, Université Lyon 1, Bron, France
c Turner Institute for Brain and Mental Health, Monash University, Melbourne, Australia
d School of Health Sciences, Purdue University, West Lafayette, Indiana, USA
e Weldon School of Biomedical Engineering, Purdue University, West Lafayette, Indiana, USA
f Wellcome Centre for Integrative Neuroimaging, Oxford Centre for Human Brain Activity (OHBA), University of Oxford Department of Psychiatry, Warneford Hospital, Warneford Lane, Oxford, United Kingdom
a b s t r a c t:
Aging disrupts the finely tuned excitation/inhibition balance (E:I) across the cortex via a natural decline in inhibitory tone (-aminobutyric acid, GABA), causing functional decrements. However, in young adults, experimentally lowering GABA in the sensorimotor cortex enhances a specific domain of a sensorimotor function: adaptation memory. Here, we tested the hypothesis that as sensorimotor cortical GABA declines naturally with age, adaptation memory would increase, and the former would explain the latter. Results confirmed this prediction. To prove causality, we used brain stimulation to further lower sensorimotor cortical GABA during adaptation. Across individuals, how stimulation changed memory depended on sensorimotor cortical E:I. In those with low E:I, stimulation increased memory; in those with high E:I stimulation reduced memory. Thus, we identified a form of motor memory that is naturally strengthened by age, depends causally on sensorimotor cortex neurochemistry and may be a potent target for motor skill preservation strategies in healthy aging and neurorehabilitation.
1. Introduction
Motor capacities decline with age (Hunter et al., 2016; Krampe, 2002). As the brain and body become older, movements lose speed (Bedard et al., 2002; Jiménez-Jiménez et al., 2011), strength (Frontera et al., 2000), and coordination (Serrien et al., 2000). This natural loss of function is exacerbated by motor disorders which rise sharply with age (e.g. stroke, sarcopenia, Parkinsonism). As the elderly population increases (Leeson, 2018), there is a need for strategies to counteract and compensate for age-related motor decline.
During aging, the motor system must adapt continuously to ongoing neuro-musculoskeletal change. Brain plasticity enables this. Plasticity is essential to learn new motor skills, adapt and retain existing ones, and to rehabilitate functions impaired by disease (Dayan and Cohen, 2011; Sampaio-Baptista et al., 2018). Thus plasticity plays an important role in mitigating age-related motor decline (McNeil and Rice, 2018; Rozycka and Liguz-Lecznar, 2017).
Unfortunately, plasticity also declines with age (Burke and Barnes, 2006), especially in the motor domain (Bhandari et al., 2016; Freitas et al., 2013; Rogasch et al., 2009). A major cause is the dysregulation of the finely tuned balance between cortical excitation and inhibition (E:I) (Rozycka and Liguz-Lecznar, 2017). Across cortex, E:I is disrupted because -aminobutyric acid (GABA) – the major inhibitory neurotransmitter –has predominantly been reported to decline with age,2018), impaired ability to suppress automatic responses (Hermans et al., 2018a), and slower motor sequence learning (King et al., 2020).
By contrast, here we tested the hypothesis that, as M1 GABA declines with age, a specific form of upper limb motor function – adaptation memory – would increase. Across the lifespan, adaptation is that property of the sensorimotor system that enables individuals to counteract perturbations by adjusting their movements and thus maintain successful motor performance (Franklin and Wolpert, 2011; Wolpert et al., 2011). After this form of learning has taken place and the perturbation is removed, adaptation memory is expressed as an after-effect (AE) – a movement bias in the direction opposite the perturbation. The strength of adaptation memory is indexed by the persistence over time of this AE. There is a wealth of evidence that while older adults often demonstrate deficits during exposure to a sensorimotor perturbation (i.e. slower error reduction; Anguera et al., 2011; Bock, 2005; Buch et al., 2003; Fernández-Ruiz et al., 2000; Huang and Ahmed, 2014; Panouillères et al., 2015; Vandevoorde and Orban de Xivry, 2019), following removal of the perturbation the AE is preserved (Bock, 2005; Buch et al., 2003; Hegele and Heuer, 2008; Panouillères et al., 2015; Roller et al., 2002; Vandevoorde and Orban de Xivry, 2019) or even increased (Fernández- Ruiz et al., 2000; Nemanich and Earhart, 2015; Wolpe et al., 2020) compared to younger adults (although see: Malone and Bastian, 2016). From and safety contraindications for the MRS and tDCS measurements. The screening was performed by one of the experimenters, and participants’ medical history was determined by self-report. Written informed consent was provided by all participants. The study was approved by the U.K. NHS Research Ethics Committee (Oxford A; REC reference number: 13/SC/0163). In Experiment 1, all participants ( = 32) performed prism adaptation (PA) and tests of short (10-minutes) and long-term (24-hours) retention. A sub-sample underwent an MRS scan to measure neurochemistry in left sensorimotor cortex ( = 22) and in an anatomical control volume in the occipital cortex ( = 20; Fig. S2). A sub-sample consented to also participate in Experiment 2 ( = 25), consisting of two weekly sessions of PA combined with anodal/sham tDCS to M1. Full de- tails of which measurements were obtained for each individual are in Table S1.
In Experiment 1, the sample size ( = 32) was determined based on a power analysis run in G∗Power (Faul et al., 2007) (Version 3.1.9.2), in-formed by previous investigations of the association between behavior and age-related GABA change within the motor domain (Heise et al., 2013; Hermans et al., 2018a). The average effect size across these studies was || = 0.52. To detect an effect of this size requires a minimum sample of = 19 with a probability of a Type I error = 0.05, and power (1 − ) = 0.80 (based on a priori one-tailed correlational analysis). Our a neurochemical perspective, previous work showed that experimentally lowering M1 inhibitory tone during adaptation via brain stimulation had no influence on the rate of adaptation but increased persistence of the AE in young adults (Galea et al., 2010; O’Shea et al., 2017). Here, we reasoned that if AE retention depends causally on M1 inhibitory tone, then this form of memory may increase naturally with age owing to an age-related M1 GABA decline.
This hypothesis was confirmed in a cross-sectional study of thirty- two healthy older adults (mean age: 67.46 years, s.d.: 8.07). Using magnetic resonance spectroscopy (MRS) to quantify neurochemistry, we showed that M1 GABA declines with age. Using prism adaptation (PA; von Helmholtz, 1867), we showed that retention increases with age. A
mediation analysis subsequently confirmed that as GABA declines with age, adaptation memory increases, and the former explains the latter. To demonstrate causality, we intervened experimentally with excitatory anodal transcranial direct current stimulation (a-tDCS) – to try and further lower M1 GABA (Antonenko et al., 2017; Kim et al., 2014; Stagg et al., 2009) and thus further increase adaptation memory. On average, stimulation did not increase memory in this age group. Rather, a moderation analysis showed that how stimulation changed memory depended on individuals’ motor cortical E:I. Stimulation increased retention in individuals with low E:I, but decreased retention in individuals with high E:I.
In summary, we identified a specific domain of motor functional plasticity that improves with age, as a natural consequence of motor cortical inhibitory decline. This memory function can be further enhanced by neurostimulation, but only in individuals least affected by age-related dysregulation of motor cortical E:I. These findings challenge the prevail- ing view of aging as an inevitable functional decline. Whereas learning of new motor skills may decline, the capacity to maintain adaptation of existing skills improves naturally with age. That adaptation memory is enhanced naturally with age indicates it may have untapped potential as a target for training strategies that aim to preserve, improve or restore motor function in healthy or pathological aging (e.g. prism therapy for visuospatial neglect rehabilitation; O’Shea et al., 2017; Rossetti et al., 1998).
2. Materials and methods
2.1. Participants
Thirty-two right-handed men aged between 49 and 81 (mean age: 67.5 years, s.d.: 8.1) participated in this study. All were screened to rule out any personal or family history of neurological or psychiatric disorder sample sizes ( = 32 for behavioral analyses; = 20 for neurochemistry analyses) therefore had adequate power. In Experiment 2, the sample size was determined based on a comparable power analysis informed by the stimulation effect size reported in our previous work (O’Shea et al., 2017). In that study, left M1 a-tDCS enhanced long-term retention up to four days after adaptation, with an effect size of = 0.73. The minimum sample size required to detect an effect of = 0.73 with a probability of a Type I error = 0.05, and power (1 − ) = 0.80 was = 14 (based on a one-tailed difference of two dependent means). To allow for potential dropouts, twenty-six participants were recruited. One participant was lost to retention follow-up and was therefore not included in the final sample of = 25.
2.2. Prism adaptation protocol
In both experiments, PA was performed using a purpose-built automated apparatus (Fig. S1a). Participants sat with their head fixed in a chinrest, viewing a 32-inch horizontal touchscreen through a Liquid Crystal Display (LCD) shutter (Dispersion film, Liquid Crystal Technologies, Ohio, USA). The touchscreen was used to present the visual targets and record reach endpoints, and the LCD shutter was used to control visual feedback of the screen and limb. A button was attached to the pole of the chinrest and served as a starting position for all pointing movements. Participants were instructed to keep the button pressed at all times, and to only release it when initiating a reaching movement towards a target. On after-effect (AE) trials only, the release of the button triggered the LCD shutter to turn opaque, thus occluding visual feedback of endpoint accuracy. In addition, a fixed shutter prevented participants from seeing their limb at the starting position and during the first third of their reaching trajectory. Participants were instructed to not slide their finger across the surface of the touchscreen but to instead touch the screen only at the end of their reaching movement. Pointing errors were calculated as the angle formed between a straight line joining the starting position and the target, and a straight line joining the starting position and the recorded landing position. By convention, errors in the direction of the prismatic shift (rightward/clockwise) were coded as positive, while errors in the opposite direction (leftward/counterclockwise) were coded as negative. The task was programmed in MATLAB version 2014b (MathWorks; https://uk.mathworks.com) using Psychtool- box (Kleiner et al., 2007) version 3, run on a MacBook Pro laptop. On each trial, an audio voice recording instructed participants to reach and point with their right index finger at the target presented on the touch-screen. The target could either be located at the center of the screen(open-loop trials) or 10 cm to the left or right (closed-loop trials). The distance between participants’ eyes and the central target was 57 cm.
During PA participants alternated between two types of task block: closed-loop pointing (CLP) and open-loop pointing (OLP). On closed-loop trials, participants wore 10◦ right-shifting prism goggles (glacier goggles: Julbo, Longchaumois, France; lenses: OptiquePeter, Lyon, France) and were instructed to make rapid reaching movements (mean movement duration: 452 ms, s.d.: 119 ms) to either the left or right target in a pseudo-randomized order. Participants were trained to keep their finger at the landing position and correct their movement on the next trial as needed. To limit strategic adjustments and “in-flight” error correction (Redding and Wallace, 1996; 2001) visual feedback of sham) during behavioral testing. This was achieved by using blinding codes (“study mode” of the stimulator) provided by a researcher who was not involved in behavioral testing. Unblinding occurred at the statistical analysis stage, once data collection was completed.
In Experiment 2, participants performed two PA+tDCS sessions (anodal/sham, order counter-balanced), each separated by a minimum of one week (average interval: 10 days, s.d.: 6 days). This interval was chosen to allow both the effect of DCS on cortical excitability (Nitsche et al., 2003; Nitsche and Paulus, 2000) and the AE to wash out (O’Shea et al., 2017), to ensure a return to baseline pointing behavior and cortical excitability by the start of the other experimental session. The rationale for stimulating during PA – as opposed to before or after – was to interact the first third of each reaching movement was occluded with the fixed shutter, as in previous work (Inoue et al., 2015; O’Shea et al., 2017; 2014). At the end of every trial, visual feedback of the landing position lasted for 500 ms after the touch was recorded. After this time, the LCD shutter turned opaque and participants had to return to the starting position (i.e. press and hold the button) without visual feedback of their hand. This procedure limited prism exposure to the reaching movement as opposed to the return movement. On open-loop trials, prisms were removed and participants were instructed to point at the central target. Accuracy was emphasized over speed (mean movement duration: 799 ms, s.d.: 135 ms). Visual feedback was presented on each trial by the LCD shutter turning opaque at reach onset, thus occluding vision of the target, reach and endpoint error, and return movement. This enabled the leftward AE to be measured without participants actively de-adapting in response to visual error feedback.
In both experiments, each PA session measured pointing accuracy during baseline, adaptation, short-term (10-minutes), and long-term retention (24-hours; Fig. S1). Baseline closed- and open-loop pointing ac- curacy was measured in two blocks of 20 and 30 trials respectively. Adaptation comprised of alternating pairs of closed- and open-loop pointing blocks, six in Experiment 1 and seven in Experiment 2 (Fig. S1). Retention of the AE was measured 10-minutes and 24-hours after the end of PA, by means of a single block of 45 open-loop trials. In Experiment 2, 10-minute retention was followed by a washout phase in which participants pointed without wearing prisms, observed their leftward errors, and therefore de-adapted. Washout consisted of 40 closed-loop trials and 45 open-loop trials distributed across six interleaved blocks (Fig. S1b). The purpose of washout was twofold. First, it enabled us to investigate whether, in the sham condition, older age was associated with a failure to de-adapt which could explain stronger AE at a later time point (see Supplementary Results). Second, we reasoned that, if memory formation was strengthened by stimulation during PA, then washout was more likely to interfere with long-term retention in the sham condition than in the anodal condition, which might increase sensitivity to detect the effect of stimulation at 24-hours.
2.3. Transcranial direct current stimulation
In Experiment 2, tDCS was delivered by a battery-driven DC stimulator (Neuroconn GmbH, Ilmenau, Germany) connected to two 7 × 5 cm sponge electrodes soaked in a 0.9% saline solution. The anodal electrode was centered over C3 (5 cm lateral to Cz) corresponding to the left primary motor cortex according to the international 10–20 electrode System (Herwig et al., 2003). The cathode was placed over the right supraorbital ridge. During anodal tDCS, stimulation was applied at 1 mA for 20 min, throughout the entire adaptation phase, as in our previous work (O’Shea et al., 2017). Impedance was monitored online and kept under 10 kOhm at all times during stimulation. The current ramped up and down over a 10 s period at stimulation onset and offset. During sham tDCS, the procedure was identical except that no stimulation was delivered during the 20 min. Instead, small current pulses (110 A over 15 ms) occurred every 550 ms to simulate the transient tingling sensations associated with real stimulation. Both experimenters and participants were blinded to the stimulation condition (anodal or with memory formation processes occurring during exposure to the visual shift, which is known to relate to long-term retention (Inoue et al., 2015; Joiner and Smith, 2008; Kording et al., 2007; Smith et al., 2006). We showed previously that M1 a-tDCS applied before – as opposed to during – PA had no effect on adaptation memory, demonstrating the importance of the interaction between neurostimulation and concurrent cognitive state (O’Shea et al., 2017).
2.4. MRS acquisition protocol
MRS data were acquired at the Oxford Centre for Clinical Magnetic Resonance Research (OCMR, University of Oxford), on a Siemens Trio 3-Tesla whole-body MR scanner and using a 32-channel coil. High resolution T1-weighted structural MR images (MPRAGE; 224 × 1 mm axial slices; TR/TE = 3000/4.71 ms; flip angle = 8◦; FOV = 256; voxel size = 1 mm isotropic; scan time = 528 secs) were acquired for MRS voxel placement and registration purposes. MRS data were acquired from two volumes of interest (VOIs; voxel size = 2 ×2 ×2 cm3) in two consecutive acquisitions. The first VOI was centered on the left motor hand knob (Yousry et al., 1997) and included parts of the pre-and post-central gyrus (Fig. S2c). The second (anatomical control) VOI was centred bilat- erally on the calcarine sulcus in the occipital lobe (visual cortex) (Engel et al., 1997; Ip et al., 2017; Lunghi et al., 2015) (Fig. S2c). This control region was chosen because it has, to our knowledge, not been implicated in the development and/or retention of prism AEs (for review, see: Panico et al., 2020; Petitet et al., 2017). B0 shimming was performed using a GRESHAM (64 × 4.2 mm axial slices, TR = 862.56 ms, TE1/2 = 4.80/9.60 ms, flip angle = 12◦ , FOV = 400, scan duration = 63 secs). MR spectroscopy data (spectra) were acquired using a semi-adiabatic localization by adiabatic selective refocusing (semi-LASER) sequence (TR/TE = 4000/28 ms, 64 scan averages, scan time = 264 secs) with variable power radiofrequency pulses with optimized relaxation delays (VAPOR), water suppression, and outer volume saturation (Deelchand et al., 2015; Öz and Tkáč, 2011). In addition, unsuppressed water spectra were acquired from the same VOIs to remove residual eddy current effects, and to reconstruct the phased array spectra (Natt et al., 2005). Single-shot acquisitions were saved separately (single-shot acquisition mode), then the frequency and phase-corrected before averaging over 64 scans.
2.5. MRS data analysis
Metabolites were quantified using LCModel (Provencher, 2012; 1993; 2001) performed on all spectra within the chemical shift range 0.5 to 4.2 ppm. The model spectra were generated based on previously reported chemical shifts and coupling constants by Vespa Project (Versatile Stimulation, Pulses, and Analysis). The unsuppressed water signal acquired from the volume of interest was used to remove eddy current effects and to reconstruct the phased array spectra (Natt et al., 2005). Single scan spectra were corrected for frequency and phase variations induced by subject motion before summation. Glutamic (Glx) was used in the current study due to the inability to distinguish between glutamate and glutamine using a 3T MRI scanner. To avoid biasing the sample towards high concentration estimates, and expected relative Cramér-Rao Lower Bound (CRLB) was computed for each individual dataset given the concentration estimate and assuming a constant level of noise across all measurements (see Supplementary Information for detailed methods). Datasets for which the Pearson residual between the expected and observed relative CRLB exceeded 2 were excluded from subsequent analysis. Using this quality filtering criterion for -Aminobutyric acid (labeled GABA), Glutamix (Glutamine+Gutamate, labeled Glx) and total Creatine (Creatine + Phosphocreatine, labeled TCR), four V1 MRS datasets were discarded and no M1 MRS dataset was discarded.
Tissue correction is an important step in MRS data analysis, especially in older adults owing to brain atrophy, which has been proposed to correspond to the default option of the “tab_model” function of the sjPlot package in R (Lüdecke, 2021). We compared LMM model parameters directly to establish neuroanatomical and neurochemical specificity. Model parameters were compared using a general linear hypothesis test using the multi-comp package in R (Hothorn et al., 2008). For visualization purposes, Figs. 1b, 3, and 6 b show block-averaged data as measures of retention, but the statistical analyses were run on individual trial data with random intercepts and slopes. Measures of effect size are
reported for all substantial analyses, using the effectsize package (Ben- Shachar et al., 2020) in R. Cohen’s d was used to compute effect sizes for a one-sample t-test against zero for short- and long-term retention in Ex-to account, at least in part, for the frequently observed age-related de- cline in MRS-measured GABA levels (Maes et al., 2018; Porges et al., 2017b). LCmodel outputs metabolite concentrations for an entire volume of interest. So if the fraction of neural tissue within a volume of interest is low, owing to age-related atrophy (Good et al., 2001), metabolite concentration estimates will also necessarily be below. Several tissues correction techniques have been proposed to account for this potential confound, with currently no consensus in the literature (Harris et al., 2015; Maes et al., 2018; Porges et al., 2017b). Most of these techniques make assumptions about the distribution of the metabolite of interest within the different tissue compartments. However, such assumptions may not hold across the lifespan, as the normal aging process may affect some compartments more than others. Hence, all analyses reported in this paper used non-tissue corrected concentration estimates and instead included the percentage of grey matter (GM) and white matter (WM) in the MRS voxel as confounding variables of no interest (as in Scholl et al., 2017). Since this partial volume correction approach makes no assumption about the distribution of GABA and Glx within the different tissue types, it is particularly suitable for the present study (in which participants ranged in age from 49 to 81), and hence controls for atrophy while remaining agnostic about the differential impacts of aging on tissue types. The percentages of grey matter, white matter, and cerebrospinal fluid present in the VOIs were calculated using FMRIB’s automated segmentation tool (Zhang et al., 2001). They are reported together with MRS data quality metrics in Table S2.
Across individuals, the total creatine (TCR) concentration estimate was negatively correlated with age in the M1 voxel ( (21) = −0.46, = 0.04) although not in the V1 voxel ( (17) = −0.06, = 0.81; Fig. S2b). Owing to this confound with age, TCR could not be used as a valid internal reference for metabolite estimates. Hence, throughout this work, we used absolute concentration estimates for GABA and Glx, rather than expressing the data as ratios of TCR.
2.6. Statistical analysis
Statistical analyses of behavior were performed in R (R Core Team, 2017). To control for inter-individual differences in pre-adaptation pointing accuracy, across all trials endpoint error data, were normalized by subtracting the average pointing error at baseline (across left/right targets for closed-loop blocks; middle target for open-loop blocks). Unless specified otherwise, all statistical tests were two-tailed. Analyses were performed using linear regression and included checks of the following assumptions: 1) linearity, 2) homogeneity of variance, and 3) normality of residuals. These assumptions were examined visually using plots of residuals vs. observed values (linearity), fitted values vs. residuals (homogeneity of variance), and distribution of residuals (normality of residuals). Linear mixed-effects models (LMMs) were used for analyses with a longitudinal/repeated-measures component (e.g. adaptation, retention) by including intercepts and slopes as participant random effects. This approach has two advantages compared to repeated measures analyses of variance (ANOVAs): it allowed us to 1) also consider within-block behavioral dynamics, as opposed to only block average errors, and 2) dissociate random sources of inter-individual variability from meaningful ones. All model specifications are reported in Supplementary Tables. P-values were estimated using the Wald test, which experiments 1, and for paired-samples t-tests of sham versus anodal stimulation on short- and long-term retention in Experiment 2. Approximate partial eta-squared () for linear mixed-effects regression analyses to summarise the proportion of variance associated with a particular fixed effect. Rules of thumb have been proposed for interpreting effect sizes. These norms for Cohen’s d are: small = [0.20; 0.49]; medium = [0.5; 0.79]; large ≥ 0.8. The norms for are: small = [0.01; 0.05]; medium = [0.06; 0.13]; large ≥ 0.14 (Cohen, 2013).
In Experiment 2, baseline OLP and CLP mean accuracy was analyzed in two ways. First, to check for the absence of an order effect (PA session 1 vs. PA session 2; using pairwise t-tests). Second, to check for the absence of a stimulation condition effect (anodal tDCS session vs. sham tDCS session; using pairwise t-tests on the same data reordered by neurostimulation condition). The former analysis ensured the one-week washout interval was effective (i.e. the behavioral effects of session 1 had dissipated by the onset of session 2), and the latter ensured that differences in performance between the anodal and sham tDCS conditions could be attributed to a neurostimulation effect as opposed to random systematic differences already present at baseline. To quantify the statistical evidence in favor of an absence of difference (i.e. what we aimed to achieve), a Bayes Factor (01) was computed for these quality con- trol analyses. A 01 > 3 was considered substantial evidence for the absence of difference, consistent with appropriate washout between the two experimental sessions.
Because GABA is synthesised from glutamate, the concentrations of these two neurotransmitters are typically correlated positively in the brain (Jocham et al. (2012); Stagg et al. (2011a); in our dataset, M1 GABA × M1 Glx: (20) = 0.34, = 0. 13; V1 GABA × V1 Glx: (14) = 0. 16, = 0.55). Therefore, when analyzing the relationship between the absolute concentration in GABA or Glx within a voxel and outcome, the concentration of the other neurotransmitter (GABA or Glx) was also included in the model. In addition, grey and white matter concentrations were also included as covariates of no interest in all models that included neurochemical data.
A mediation analysis was used to characterize the “mechanistic” links underlying the observed correlations between age, neurochemistry, and retention. This was performed using the R package mediation for causal mediation analysis (Imai et al., 2010). Mediation was conducted using regression with nonparametric bootstrapping (10,000 resamples) to ascertain whether M1 inhibitory tone accounted for the link between age and long-term retention. The model included: age as the independent variable (X); absolute concentrations of M1 GABA and Glx as mediators (M1, M2); block-averaged retention at 24-hours as the dependent variable (Y) (block mean error normalized by the baseline for each individual), and control for the fraction of GM and WM in the M1 voxel (C1, C2). The percentage mediation () was calculated as the fraction of total effect (c) accounted by indirect effects (ab1 or ab2).
