GABA and Glx Distinctively Predict Motor Learning and Retention in Young and Older Adults

Abstract

Gamma-aminobutyric acid (GABA) and glutamate are fundamental in neural plasticity. Motor learning is predicted by baseline levels of these metabolites and their modulation in the sensorimotor cortex (SM1), but less is known about the metabolic activity in other areas that support learning, such as the dorsolateral prefrontal cortex (DLPFC), as well as the practice-induced metabolic modulation and age-associated differences. We investigated whether: (1) motor learning induces a differential degree of metabolic modulation in the SM1 and DLPFC, (2) learning tasks with higher difficulty levels enhance metabolic modulation as compared with those with lower difficulty levels, (3) metabolic modulation during motor learning is age dependent, and (4) training-induced metabolic modulation may have a differential effect on motor learning and retention. Young (n = 25, 12 females) and older (n = 21, 10 females) human adults completed a 6 d motor learning protocol with magnetic resonance spectroscopy scans being administered before, during, and after a low and high task complexity training condition. We observed a training-induced reduction of SM1 GABA+, regardless of age and task difficulty level, but no significant changes in DLPFC. Neither region showed a significant Glx (combined glutamate and glutamine) modulation. In addition, baseline GABA+ levels predicted learning, but this effect was region and task difficulty dependent. Age-related differences emerged in the prediction of retention, with older adults showing a beneficiary role of task-induced increase in the SM1 inhibitory tone. These results highlight the complexity of metabolic dynamics in learning and retention, showing their dependency on age, brain region, and task difficulty.

Significance Statement

Excitatory and inhibitory metabolites are essential for learning. Although motor learning is linked to reduced GABA levels in the motor cortex, metabolic activity in other learning-related regions remains unclear. It is also unknown whether neuromodulations depend on task difficulty or age. We demonstrate that inhibitory metabolite levels decrease in the motor cortex, but not in the prefrontal cortex, independent of age and task difficulty. Notably, age-related differences emerged in the prediction of retention: in older adults, task-induced increases in SM1 inhibitory tone were associated with better retention performance, whereas no association was observed in young adults. These findings underscore the complex nature of metabolic dynamics underlying motor learning and retention, emphasizing their dependence on age, brain region, and task difficulty.

Introduction

Synaptic plasticity constitutes the underlying neurophysiological basis of learning (Carcea and Froemke, 2013). Gamma aminobutyric acid (GABA) and glutamate (Glu) are the primary inhibitory and excitatory neurotransmitters in the human brain and play key roles in shaping cortical plasticity and learning (Buzsáki et al., 2007). Magnetic resonance spectroscopy (MRS) enables accurate, in vivo quantification of neurometabolite concentrations within specific regions of the human brain. Previous studies have shown that MRS-assessed GABA levels in the sensorimotor cortex (SM1) decrease during motor training but remain stable during random movements or rest (Floyer-Lea et al., 2006; Kolasinski et al., 2019), while other work has linked individual changes in GABA or Glx to motor learning performance rather than consistent group-level effects (Maruyama et al., 2021; Bell et al., 2023; Rasooli et al., 2024). These findings seem to point to a favorable role of decreases of inhibitory tone in motor learning.

However, motor learning involves not only the execution of movements, controlled by motor cortical regions [including the primary motor cortex (M1)], but also the integration of sensory feedback and the progressive adjustment of motor plans across trials, requiring the recruitment of sensory processing and prefrontal regions. Previous work has shown that performance of more demanding bimanual coordination tasks is predicted by working memory capacity (Seer et al., 2021). These findings suggest that brain regions supporting working memory may play an important role in motor learning. Accordingly, the dorsolateral prefrontal cortex (DLPFC) is of particular interest, as it plays a central role in working memory and the cognitive control of action during complex motor skill acquisition (Pochon et al., 2001; Santos Monteiro et al., 2017; Rodríguez-Nieto et al., 2022). However, despite its well-established functional involvement, previous studies have only reported that increased GABA levels in the DLPFC are associated with motor performance (Rodríguez-Nieto et al., 2025), while direct evidence of neurometabolite modulation in this region during motor learning is still lacking.

Although both SM1 and DLPFC regions are involved in motor learning, older adults often rely more on compensatory activation of the prefrontal cortex for successful motor performance (Heuninckx et al., 2008; Santos Monteiro et al., 2017). Consequently, the relative involvement of SM1 and DLPFC in motor learning may shift with age. Moreover, as aging is associated with lower levels of GABA and Glx (Gao et al., 2013; Porges et al., 2021; Rodríguez-Nieto et al., 2023), such reductions may alter the regulation of these metabolites during motor learning in older adults (King et al., 2020). Specifically, although a transient, task-related decrease in GABA+ levels was observed across participants irrespective of age, only in older adults was the reduction in SM1 GABA associated with motor performance (Maes et al., 2021). Therefore, this study aimed to investigate age-related differences in the modulation of neurometabolites within these two brain regions during motor learning, as well as their association with motor performance.

Modulation of GABA within motor-related brain regions has not been consistently observed across studies. Such inconsistency may stem from differences in the cognitive and motor demands of the tasks employed (Li et al., 2024b). Specifically, greater GABA modulation has been observed under more demanding motor learning conditions compared with less challenging ones (Chalavi et al., 2018). Extrapolating these observations, the present study incorporated two levels of task complexity to examine its influence on neurometabolite modulation during motor learning.

Finally, although several studies have examined neurometabolite dynamics during motor learning, their role in motor task retention, a critical phase reflecting the consolidation of newly acquired motor skills, remains largely unexplored. Preliminary results have suggested that plasticity and stability may be expressed through distinct patterns of neurometabolite modulation (Yamada et al., 2024). Understanding whether such modulation supports memory consolidation is key to elucidating the mechanisms of lasting motor plasticity.

Therefore, we examined neurometabolite modulation during motor learning in the SM1 and DLPFC across age groups and task complexity levels and determined whether such modulation is associated with motor learning and retention performance.

Methods and Materials

Participants

Twenty-five young adults (YA; mean ± SD = 22.4 ± 3.29 years; range 18–31 years, 12 females) and 21 old adults (OA; mean ± SD = 71.71 ± 4.41 years; range 65–83 years, 10 females) participated in this study. They had normal or corrected-to-normal vision and reported no history of neurological disease or psychiatric disorders. Handedness was assessed using the Oldfield Handedness scale (Oldfield, 1971). In addition, the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005) test with a cutoff score of 26 was used to screen for cognitive impairment. Age groups did not differ with respect to gender (χ² = 0, df = 1, p = 1), handedness [laterality quotient (LQ) YA: mean ± SD = 89.21 ± 12.87; OA: mean ± SD = 94.16 ± 13.50; independent samples t test: t = 0.80, p = 0.21], and MoCA score (YA: mean ± SD = 28.52 ± 0.92; OA: mean ± SD = 28.14 ± 1.35; independent samples t test: t = 0.62, p = 0.26). Prior to testing, written informed consent was obtained from each participant. The protocol was approved by the local ethical committee of KU Leuven, Belgium, and was in accordance with the Declaration of Helsinki (2009).

Experimental procedure

This study was part of a larger research protocol which consisted of six sessions, spread across a time window of 18.53 ± 5.87 (mean ± SD) days (for an overview of the experimental sessions, see Fig. 1A). During the first session (Day 1), participants were first screened for handedness, cognitive performance, and contraindication for MRI and/or transcranial magnetic stimulation (TMS). This was followed by a baseline test of the PC-based N-back task and a 5 min baseline test of the bimanual tracking task (BTT) in a dummy scanner. On Day 2 participants performed the N-back task during fMRI acquisition, which served to localize the DLPFC voxel for the MRS acquisitions used for this study. To investigate the underlying cortical processes of long-interval intracortical inhibition (LICI) on Day 3, TMS pulses were administered on the left DLPFC and left M1 of the participant, and electroencephalography (EEG) data were concurrently recorded during rest. Please note that these TMS–EEG data are not reported in the present manuscript because they do not fit with the research questions addressed in the present manuscript. As LICI is a paired-pulse TMS paradigm, it is considered to only induce transient inhibitory effects and is not expected to result in long-lasting cortical modulation. Therefore, the TMS protocol is unlikely to have influenced subsequent neurochemical dynamics and motor learning performance. On Days 4 and 5, participants were tested in the MRI unit, starting with one experimental condition (either low or high task difficulty condition) on Day 4 and returning on Day 5 to perform the other experimental condition. The MRI session consisted of MRS acquisition on the SM1 and DLPFC during (1) pretraining (Time Point 1, TP1), (2) twice during training (TP2 and TP3), and (3) posttraining (TP4; Fig. 1B). The experimental conditions consisted of the following: (1) a high difficulty level training condition (complex task), in which the participants were trained on a complex variation of the BTT, and (2) a low difficulty level condition, in which the participants were trained on an easier variation of the BTT (easy task). The order of the experimental conditions was counter-balanced across participants within each age group. Finally, on Day 6, a retention test (comprising the exact same task, as applied on Day 1) was administered in the dummy scanner. The data used for the current manuscript were extracted from Day 1 (Baseline condition), Days 4 and 5 (MRI: low/high difficulty level), and Day 6 (Retention).

Figure 1.

A, Overview of the experimental sessions. During Day 4 and Day 5, participants were trained on either the high challenging or low challenging variations of the bimanual tracking task. The order of the challenge condition was counter-balanced across the participants in each age group. B, Timeline of the experiment on Day 4 and Day 5. MRS data from the left DLPFC and SM1 voxels were collected at four different timepoints. The voxel order was counter-balanced across the participants in each age group. There was a pause after the MRS data acquisition at Timepoint 1 and Timepoint 3, where participants had the option to leave the MRI room for a short break. “+” indicates that participants were instructed to keep their eyes open and look at a fixation cross during the MRI data acquisition. C, Experimental setup. D, All possible bimanual coordination directions and frequency ratios. CCW, counterclockwise; CW, clockwise; DLPFC, dorsolateral prefrontal cortex; IN, inward; MRS, magnetic resonance spectroscopy; OUT, outward; SM1, sensorimotor cortex; rs-fMRI, resting-state functional MRI; sT1, short T1-weighted image; TP, timepoint.

With this study design, we aim to test the following hypotheses. Hypothesis 1: SM1 GABA levels will decrease during motor learning, whereas Glx will remain relatively stable. A comparable pattern was expected in the DLPFC, particularly among older adults; Hypothesis 2: the magnitude of the individual GABA decrease will be positively associated with motor learning performance; Hypothesis 3: task complexity influences the modulation of neurometabolites during motor learning, such that higher task complexity induces more pronounced neurometabolic changes. Furthermore, we expected age-related differences in neurometabolite modulation between the two task conditions, as older adults are typically more sensitive to increases in task complexity (Van Ruitenbeek et al., 2023). Hypothesis 4: the association between neurometabolite modulation and behavioral performance observed during learning (as proposed in Hypothesis 2) may differ from that observed during retention because the former process refers to the building and the latter to the maintenance of motor representations.

N-back task description and data analysis

This study aims to investigate brain regions supporting working memory and movement refinement during motor learning, as performance in complex bimanual coordination tasks has been linked to working memory capacity (Seer et al., 2021). The DLPFC, a key hub for working memory and cognitive control of movement, was therefore selected as the primary region of interest. To functionally localize this region, we employed the N-back task, a well-established paradigm for probing working memory, which reliably activates frontal and parietal cortices, especially the DLPFC (Ragland et al., 2002). This approach allowed us to identify an individualized and functionally relevant DLPFC location for each participant. More specifically, the participant was presented with a series of alphabetical letters at the center of the screen and was instructed to indicate when the current stimulus matches the one from “N” trials earlier in the series. In the present study, three blocks of 0-back (press if it is “X”, as a control condition) and three blocks of 2-back were presented in a pseudorandomized order. We used the 2-back condition as it has shown to require continuous working memory updating of the stored information. Each block consisted of 24 letters, 5 of which were target letters. At the beginning of each block, the instruction of that block was presented for 2 s, followed by a 1 s interval. Then the letters were presented for 1 s with an ITI of 1 s. There were 5 s of rest in between the blocks. Participants responded to the target letter by pressing a button on a keyboard (i.e., on Day 1) or on a key-box (on Day 2). The N-back data were recorded with LabVIEW software (National Instruments). The exact start time and duration of each block as well as total accuracy were obtained using MATLAB 2022.

Bimanual tracking task (BTT)

BTT description

Participants laid supine on the extendible table of the MRI or dummy scanner with forearms resting on pillows. To prevent head movements during task performance, foam cushions were placed around the participant's head. Using an LCD projector (Barco 6300, 1,280 × 1,024 pixels), visual stimuli were projected onto a double mirror placed in front of the participant's eyes. The task device was a nonferromagnetic apparatus with two dials (diameter, 5 cm), which was placed over the participant's thighs (Fig. 1C). The dials could be adjusted to the participant's anthropometry and had an angle of ∼45° for comfortable handling. The two dials controlled the direction and speed of a cursor on a computer screen: the left dial controlled the displacement along the y-axis and the right dial along the x-axis.

On the screen, a blue target line indicated the main coordination directions: the two hands could rotate clockwise (CW), counterclockwise (CCW), inward (IN), and outward (OUT; Fig. 1D). Each coordination direction could be performed at three different frequency ratios, that is, 1:1, 1:2, and 2:3 (left hand: right hand), which was visualized by the slope of the target line. For example, a target line with a 45° angle indicated the 1:1 frequency ratio, requiring both hands to rotate at equal speeds. A blue target line with a 26.56° angle indicated the 1:2 frequency ratio requiring the right hand to move twice as fast as the left hand. Each trial consisted of a preparation phase (2 s), an execution phase (8 s), and an intertrial interval (ITI) of 2 s. During the preparation phase, the target slope template was visualized but no movement was required, and the participants were instructed to plan their movement. The start of the execution phase was marked by the appearance of a white target dot that started moving along the target line at a constant speed starting from the center of the display toward the periphery. Participants were instructed to closely track the target dot as accurately as possible (both spatially and temporally) by rotating both hand dials simultaneously. Angular displacements were registered by means of nonferromagnetic high precision optical shaft encoders (HP, 2,048 pulses per revolution, sampling frequency 100 Hz).

BTT training schedule

On Day 1, the target line consisted of a straight line with an angle of 45°, requiring a rotation of both dials at equal speed. Participants completed a familiarization block of 20 trials including five in each movement orientation (CW, CCW, IN, and OUT) while they received online visual feedback on their performance. Subsequently, in order to assess baseline performance prior to motor training, participants performed 24 trials, i.e., 2 trials per frequency ratio in each movement orientation, without providing any augmented feedback.

On the day of performance of the low difficulty condition, trials with a 1:1 hand frequency ratio in four movement orientations were presented to the participants (low task difficulty), whereas, on the day of the high difficulty condition, trials with 1:2 and 2:3 hand frequency ratios in four movement orientations were presented to the participants (high task difficulty).

Each training day (Days 4 and 5) started with a pretest block of the task variants to be learned, followed by four blocks of training, and ended with a posttest block of the practiced task variants. The pretest and posttest blocks, each consisting of 16 randomized trials, were performed in the absence of any type of augmented visual feedback to assess task learning. Each training block consisted of 48 randomized trials which were performed in the presence of different forms of augmented visual feedback. That is, within each block, half of the trials were presented with concurrent (online) visual feedback and half of them were presented with after-trial visual feedback. During the trials with concurrent visual feedback, participant's actual performance was displayed online on the screen with a red cursor while the targeted performance was shown with a white dot, upon which movements could be corrected. For the trials with after-trial visual feedback, the participant's complete performance cursor trajectory was displayed during the 2 s of the ITI to reduce dependency on online visual feedback. Finally, on Day 6, a retention test consisting of a block of 48 trials (i.e., four trials for each frequency ratio in each movement orientation) without any feedback was administered in the dummy scanner.

BTT analysis

Behavioral data were recorded and analyzed with LabVIEW software (National Instruments). The x and y coordinates of the target dot and the subjects’ cursor were sampled at 100 Hz. Offline analysis was carried out using MATLAB R2022 and Microsoft Excel. Performance accuracy was measured by calculating the error rate based on the average deviation (AvgDev) from the target line and target dot. This means that, for each trial, the track deviation was measured as the Euclidian distance between the target line and participant's cursor position at each point in time and then averaged. Better performance is thus reflected by lower values of AvgDev. For the statistical analyses, online trials without concurrent visual feedback were analyzed, to better test the degree of task learning. The behavioral data from two young adults were excluded, one due to withdrawal from the study after Day 4 and one due to being identified as an outlier (exceeding the group mean by more than two standard deviations).

Magnetic resonance imaging (MRI)

Image acquisition

MRI data were acquired using a 3 Tesla Philips Achieva magnetic resonance scanner (Philips Healthcare), with a 32-channel head coil. On Day 2, a high-resolution T1-weighted anatomical image was acquired using a chemical shift three-dimensional turbo field echo [3DTFE; TR/TE, 9.7/4.6 ms; flip angle, 8°; voxel size, 1 × 1 × 1 mm³; field of view (FoV), 256 × 242 × 182 mm³; 182 sagittal slices; scan duration, ∼6 min]. The task-based fMRI data during the N-back task were acquired with an ascending gradient EPI pulse sequence for T2*-weighted images (TR/TE, 2,000/29.80 ms; multiband factor 2; flip angle, 90°; voxel size, 2.5 × 2.5 × 2.5 mm³; FoV, 210 × 210 × 145.6 mm³; 54 transverse slices; interslice gap, 0.2 mm; 171 dynamical scans; scan duration, ∼6 min). On Day 4 and Day 5, a short T1 anatomical image (TR/TE, 9.6/4.6 ms; flip angle, 8°; voxel size, 1.5 × 1.5 × 1.5 mm³; FoV, 256 × 256 × 182 mm³; 182 sagittal slices; scan duration, ∼2 min) was acquired which was used for MRS voxel planning. Subsequently, MRS data from the left SM1 and DLPFC were acquired using the GABA-edited MEGA-PRESS sequence (Mescher et al., 1998; Edden and Barker, 2007) with the following parameters: TR/TE, 2,000/68 ms; flip angle, 90°; 2 kHz spectral width. ON and OFF spectra were acquired in an interleaved fashion, corresponding to an editing pulse at 1.9 or 7.46 ppm, respectively. Prior to each MRS acquisition, an automatic shimming procedure was performed. For each MRS voxel, 16 unsuppressed water averages were acquired within the same region using identical acquisition parameters. Considering the shape and dimensions of each region of interest and based on previous studies, voxel dimensions were set to 30 × 30 × 30 mm³ for the SM1 region and 40 × 25 × 25 mm³ for the DLPFC region. For the left DLPFC voxel, 160 averages were acquired (scan time, ∼11 min 12 s). However, since the number of averages can be reduced without affecting the data quality for the SM1 voxel, 112 averages were acquired for the left SM1 voxel (scan duration, ∼8 min). Acquiring spectra from both VOIs at each timepoint required 19 min 12 s, while each training block lasted 9 min 36 s. Accordingly, the MRS acquisition at each timepoint was temporally aligned with the preceding two behavioral blocks. MRS voxels were placed on a subject-to-subject basis (Fig. 2). The SM1 voxel was centered on the hand knob area of the precentral gyrus in the axial plane (Yousry et al., 1997) and rotated in the coronal and sagittal planes to align with the external surface of the brain (Fig. 2B). The DLPFC voxel was placed over the DLPFC peak activation mask obtained following the fMRI data analysis of the N-Back task and then it was rotated in all the planes to ensure alignment with the external surface of the brain (Fig. 2A). Of note, although spectral editing allows the detection and quantification of GABA, the edited signal contains a significant contribution from the macromolecular signal. Therefore, the GABA measurements are referred to as GABA+.

Figure 2.

MRS voxels and spectra. Exemplar voxel positioning over the left (A) DLPFC and left (B) SM1 cortices. The (C) DLPFC and (D) SM1 MR spectra obtained from the sample of young and older adults. DLPFC, dorsolateral prefrontal cortex; SM1, sensorimotor cortex.

On Day 4 and Day 5, the MRS data from each region (i.e., left SM1 and DLPFC) were acquired at four different timepoints (TPs): once before the start of the training (TP1), two times during the training (TP2 and TP3), and once after the end of the training (TP4).

During the MRS data acquisition at TP1 and TP4 (resting-state measures), participants were instructed to have their eyes open and look at the fixation cross, whereas, during TP2 and TP3, they were executing the BTT task (task-based measures). Because of the long duration of the experiment, two short breaks of ∼10 min were planned into the timeline, one after the MRS acquisition at TP1 and one after the MRS acquisition at TP3. Moreover, three additional short T1 anatomical images were acquired before the MRS data acquisition at each timepoint, to either place the MRS voxel after a break or verify whether subsequent voxel positioning needed to be adjusted in case of head movement.

N-back fMRI data analysis

N-back fMRI data were analyzed using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12). The standard procedure for fMRI analysis was implemented. In brief, images were aligned to the first image of the scan, slice time corrected, coregistered to the high-resolution T1-weighted scan, normalized to the Montreal Neurological Institute (MNI)-152 space, and spatially smoothed with a Gaussian kernel of 8 mm full-width at half-maximum. Subsequently, for each participant, a general linear model (GLM) was designed modeling the task conditions by convolving a boxcar reference vector with the canonical hemodynamic response function. Then the data were high-pass filtered with a cutoff of 128 s to remove low-frequency drifts, and contrast images were calculated to identify brain regions with greater activity during working memory performance relative to the control task (i.e., 2-back > 0-back). Finally, the DLPFC mask was defined as the individual local activation maximum in the left DLPFC region observed in the latter contrast.

MRS data processing

MRS data were analyzed using the GABA analysis Toolkit “Gannet (v3.2.1)” (Edden et al., 2014). In the first step, data were frequency and phase corrected by applying spectral registration (Mikkelsen et al., 2020). The ON spectra were subtracted from the OFF spectra, and the resulting difference spectrum was fitted between 4.2 and 2.8 ppm using a three-Gaussian function. The water signal was fitted using a Lorentz-Gaussian model and it was used as the reference metabolite. The GannetCoRegister and GannetSegment modules were used for partial-volume correction, using the SPM12-VBM8 toolbox to segment the brain tissues inside the voxel into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) fractions. To ensure a more accurate segmentation of the anatomical data, the high-resolution anatomical image acquired on Day 2 was coregistered to the short anatomical images that were used to verify optimal voxel positioning on Day 4 and Day 5, and it was used in the MRS data analysis. Subsequently, GannetQuantify was used in which the tissue fractions in the previous step were utilized to correct the obtained neurometabolite levels, assuming that GABA+ is absent in the cerebrospinal fluid and is considered to have a concentration twice as high in gray as compared with white matter (Harris et al., 2015; Eq. 5). Finally, as the tissue composition of the voxels of interest (VOIs) differs across age groups, GABA+ levels were normalized to the average voxel composition of each group (Harris et al., 2015; Eq. 6). Data quality was assessed by visual inspection of the spectra for lipid contamination and poor water suppression and by examining the fit error and signal-to-noise ratio. Due to technical issues, MRS data were not collected for some participants at one or more timepoints. Ultimately, complete datasets of neurometabolite levels were obtained from 20 young and 20 older adults.

Statistical analysis

Statistical analyses were carried out in R (4.1.2). As the first step, distributions of the behavioral or MRS data were screened. When the data had a normal distribution, parametric statistical methods were used. Conversely, when the data deviated significantly from a normal distribution, nonparametric statistical methods were used.

Statistical analysis of behavioral data

To assess whether performance before the start of the training (i.e., Baseline AvgDev obtained on Day 1) and after the end of training (i.e., Retention AvgDev obtained on Day 6) were different between young and older adults, a 2 × 2 × 2 (Age group × Task difficulty level × Timepoint) nonparametric ANOVA (nparLD in R) was used in which Age group (YA vs OA) was entered as a between-subject factor, and Task difficulty level and Timepoint were entered as within-subject factors. Furthermore, we investigated whether behavioral performance on each training day improved with practice and whether there was an effect of age group and/or task difficulty level. To do so, two nonparametric ANOVAs were used for different timepoints as explained next (nparLD in R). In the first 2 × 2 × 2 (Age group × Task difficulty level × Timepoint) ANOVA, the Timepoint factor had two levels (i.e., Pretest and Posttest), since these tests were similarly performed under no visual feedback. However, in the other 2 × 2 × 4 (Age group × Task difficulty level × Timepoint) ANOVA, Timepoint had four levels (i.e., four training blocks).

Statistical analysis of the MRS data

Baseline levels of neurometabolites. First, we investigated whether the initial (baseline) levels of neurometabolites measured at the first timepoint on each training day (TP1; that is before the start of the training) differed between the training days (i.e., Day 4 and Day 5). To do so, we used a paired sample t test for GABA+ measures and Wilcoxon rank-sum test for Glx levels. Results of these analyses revealed no significant effect of training day/task difficulty level for GABA+ (t = 0.35, p = 0.72) and Glx (V = 1,937.5, p = 0.50) levels, indicating that the initial levels of the neurometabolites were not different between the training days. Therefore, to obtain a more robust baseline measure, the initial levels of neurometabolites measured at the first timepoint on the 2 training days were averaged, and this measure was used as the measure of initial baseline levels of the neurometabolites for the following test.

We investigated whether there were differences in the regional baseline levels of the neurometabolites and whether these levels varied between age groups. To do so, a 2 × 2 (Age group × VOI) repeated-measures ANOVA with Age group as a between-subject factor and VOI as a within-subject factor was used.

Dynamic modulation of neurometabolites. We aimed to investigate whether learning a motor task under two levels of task difficulty results in differential modulation in the neurometabolite levels (task-related) and whether this effect is affected by aging. As such, a linear mixed effect model was designed in which neurometabolite levels from the SM1 and DLPFC were modeled using Age group [YA, OA], Task difficulty level [Low, High], and Timepoint [i.e., TP1 (at resting-state, pretraining), TP2 (first mid-training), TP3 (second mid-training), and TP4 (at resting-state, posttraining)]. Participant was modeled as a random factor.

Associations between levels of neurometabolites and behavioral measures

We investigated whether baseline levels of neurometabolites or training-induced modulations in the neurometabolite levels could partially predict learning and retention. For this purpose, we estimated learning and retention indices. There was a learning index for each condition (low or high difficulty) estimated by subtracting the Avgdev at the Posttest from the AvgDev at the Pretest from the respective condition (Pretest AvgDev − Posttest AvgDev). A high and low difficulty retention index was calculated by subtracting the AvgDev on Day 6 from the AvgDev on Day 1 [from their respective difficulty condition; i.e., Baseline AvgDev (Day 1) − Retention AvgDev (Day 6)]. A higher learning or retention index would indicate a greater improvement in performance.

Linear regression models (backward method) were built, in which the abovementioned learning and retention indices were modeled using Age group, baseline levels of the neurometabolites in the SM1 and DLPFC voxels, and their interactions. These models were conducted separately for each task difficulty level (Low and High).

We investigated whether training-induced modulations in the levels of neurometabolites could predict: (1) learning and (2) retention. To do so, linear regression models were designed, in which measures of learning were modeled using Age group, Timepoint, and modulations in the neurometabolite levels in the SM1 and DLPFC voxels (Neurometabolite level at TP4 − Neurometabolite level at TP1) and their interactions. These models were conducted separately for each task difficulty level (Low and High) and using the corresponding variables of each day (i.e., metabolites measures and behavior from each respective condition).

In the regression models, all the continuous variables were mean-centered to avoid issues with multicollinearity.

Results

Behavioral data

Table 1 lists results of the nonparametric repeated-measures tests on AvgDev. Results of the nonparametric ANOVA test, comparing AvgDev measured at Baseline and Retention days (Day 1 and Day 6, respectively) between age groups and task difficulty levels, revealed significant main effects of Age group, Task difficulty level, and Timepoint (ps < 0.001), indicating that the overall AvgDev errors were higher in older adults as compared with young adults, were higher in the high task difficulty as compared with the low task difficulty condition, and were higher at Baseline as compared with Retention, suggesting improvement in performance (Fig. 3A). Furthermore, the interaction effects of Task difficulty level × Timepoint (p = 0.007) and Age group × Task difficulty level × Timepoint (p < 0.001) were significant, which was driven by the fact that from baseline to retention, AvgDev error declined more in the high as compared with low task difficulty level in young adults (Fig. 3A).

Table 1.

Results of the behavioral data analysis (nonparametric mixed model ANOVA equivalent tests)

Factor	df	Statistic	p value
Baseline and retention model
Age group	1	48.67	<0.001 ***
Task difficulty level	1	58.66	<0.001 ***
Timepoint (Baseline vs Retention)	1	67.26	<0.001 ***
Age group × Task difficulty level	1	0.44	0.51
Age group × Timepoint	1	2.57	0.11
Task difficulty level × Timepoint	1	7.340	0.007 **
Age group × Task difficulty level × Timepoint	1	19.15	<0.001 ***
Pretest and posttest model
Age group	1	21.51	<0.001 ***
Task difficulty level	1	21.00	<0.001 ***
Timepoint (Pretest vs Posttest)	1	140.78	<0.001 ***
Age group × Task difficulty level	1	0.50	0.48
Age group × Timepoint	1	13.13	<0.001 ***
Task difficulty level × Timepoint	1	0.08	0.78
Age group × Task difficulty level × Timepoint	1	0.50	0.48
Training blocks model
Age group	1	49.72	<0.001 ***
Task difficulty level	1	83.64	<0.001 ***
Timepoint (Training blocks)	2.25	44.92	<0.001 ***
Age group × Task difficulty level	1	1.28	0.26
Age group × Timepoint	2.25	0.64	0.55
Task difficulty level × Timepoint	2.57	2.38	0.08
Age group × Task difficulty level × Timepoint	2.57	4.21	0.008 **

Values shown in bold indicate statistically significant results.

^**p ≤ 0.01.

^***p ≤ 0.001.

Figure 3.

Behavioral data. AvgDev error at (A) baseline and retention; (B) pretest and posttest in the high and low task difficulty conditions; (C) each training block in the high and low task difficulty conditions. AvgDev, average deviation; OA, older adults; YA, young adults.

Results of the nonparametric repeated-measures test (Table 1) on the AvgDev errors measured at Pretest and Posttest (from Day 4 and Day 5) revealed the main effects of the following: (1) Age group, reflecting that the overall performance was poorer in older adults as compared with their younger counterparts (Fig. 3B); (2) Task difficulty level, indicating that the overall performance was better in the low task difficulty as compared with the high task difficulty condition; and (3) Timepoint, indicating that performance improved from Pretest to Posttest. Moreover, the interaction effect of Age group × Timepoint was significant, as the difference between young and older adults was more pronounced during the posttest (Fig. 3B).

Similarly, results of the nonparametric 2 × 2 × 4 (Age group × Task difficulty level × Timepoint) ANOVA, comparing AvgDev errors measured during the four training blocks between age groups and task difficulty levels (Table 1, Fig. 3C) revealed significant main effects of the following: (1) Age group, indicating that the overall performance was poorer in older adults as compared with young adults; (2) Task difficulty level, indicating better performance in the low as compared with the high task difficulty level; and (3) Timepoint, indicating that performance improved over the course of training. Furthermore, the interaction effect of Age group × Task difficulty level × Timepoint was significant (p = 0.008). While performance of young adults almost reached a plateau level after the second and third training block in the low and high task difficulty conditions, respectively, performance of older adults continued to improve until the last training block. Therefore, at the end of training, the difference in performance between low and high task difficulty conditions became smaller in young as compared with older adults (Fig. 3C).

MRS data

Figure 2, C and D, presents the average spectra from the SM1 and the DLPFC in young and older adults.

Baseline levels of neurometabolites

Results of the 2 × 2 (Age group × VOI) repeated-measures ANOVA on the average baseline levels of neurometabolites are summarized in Table 2.

Table 2.

Results of repeated-measures ANOVA on the baseline levels of neurometabolites

	df	Statistic	p value
Baseline GABA+
Age group	1,44	20.938	<0.001 ***
VOI	1,44	75.325	<0.001 ***
Age group × VOI	1,44	1.5	0.23
Baseline Glxa
Age group	1	0.02	0.90
VOI	1	115.41	<0.001 ***
Age group × VOI	1	0.21	0.64

Values shown in bold indicate statistically significant results.

^aAs the MRS-assessed Glx levels were not normally distributed, nonparametric test (Wald statistic) was used. VOI, volume of interest.

^***p ≤ 0.001.

GABA+ levels. Results of the repeated-measures ANOVA on the average baseline GABA+ levels (Fig. 4A) revealed significant main effects of Age group and VOI, indicating that GABA+ levels were higher in young adults as compared with older adults and were higher in the SM1 voxel as compared with the DLPFC voxel. Post hoc t tests showed that young adults, as compared with older adults, showed higher average baseline GABA+ levels in both the DLPFC (p < 0.001) and SM1 (p = 0.005) regions. However, the interaction effect of Age group × VOI was not significant, indicating that the difference in average baseline GABA+ levels between voxels was not different between age groups.

Figure 4.

Baseline and dynamic levels of neurometabolites in the DLPFC and SM1 voxels. A, Baseline GABA+ levels were higher in YA, as compared with OA, and were higher in the SM1, as compared with the DLPFC voxel. B, Baseline Glx levels were higher in the DLPFC as compared with the SM1 voxel. However, no effect of age group was found on the baseline Glx levels. C, SM1 GABA+ levels decreased significantly during motor learning. D, Glx levels did not significantly change during motor learning. DLPFC, dorsolateral prefrontal cortex; OA, older adults; i.u., institutional units; n.s., non-significant; SM1, sensorimotor cortex; YA, young adults.

Glx levels. Results of the 2 × 2 nonparametric repeated-measures ANOVA on the average baseline Glx levels revealed a significant main effect of VOI (Fig. 4B), indicating that Glx levels were lower in the SM1 voxel as compared with the DLPFC voxel. However, the main effect of Age group and its interaction with VOI were not significant.

Dynamic modulations in the levels of neurometabolites

Table 3 reports results of the linear mixed effect models, in which levels of neurometabolites in SM1 and DLPFC were modeled using Age group, Task difficulty level and Timepoint.

Table 3.

Results of the linear mixed effects models^a modeling training-induced modulations in the SM1 and DLPFC neurometabolite levels

Factor	df	F value	p valueb
GABA+ levels at 4 timepoints—SM1
Age group	1,39	21.16	<0.001 ***
Task difficulty level	1,273	0.003	0.95
Timepoint	3,273	3.75	0.01*
Age group × Task difficulty level	1,273	0.37	0.53
Age group × Timepoint	3,273	0.46	0.71
Task difficulty level × Timepoint	3,273	0.39	0.75
Age × Task difficulty level × Timepoint	3,273	1.18	0.31
GABA+ levels at 4 timepoints – DLPFC
Age group	1,39	8.56	0.005 **
Task difficulty level	1,273	0.06	0.81
Timepoint	3,273	2.25	0.08
Age group × Task difficulty level	1,273	0.34	0.56
Age group × Timepoint	3,273	0.11	0.96
Task difficulty level × Timepoint	3,273	0.54	0.65
Age × Task difficulty level × Timepoint	3,273	2.29	0.08
Glx levels at 4 timepoints—SM1
Age group	1,45.83	0.16	0.69
Task difficulty level	1,305.35	0.38	0.53
Timepoint	3,303.88	0.67	0.56
Age group × Task difficulty level	1,305.35	0.46	0.49
Age group × Timepoint	3,303.88	20.39	0.11
Task difficulty level × Timepoint	3,303.88	0.99	0.39
Age × Task difficulty level × Timepoint	3,303.88	0.73	0.53
Glx levels at 4 timepoints—DLPFC
Age group	1,45.36	0.04	0.83
Task difficulty level	1,304.67	0.08	0.77
Timepoint	3,302.45	11.23	0.33
Age group × Task difficulty level	1,304.67	24.01	0.12
Age group × Timepoint	3,302.45	0.71	0.54
Task difficulty level × Timepoint	3,302.45	0.11	0.95
Age × Task difficulty level × Timepoint	3,302.45	0.45	0.71

Values shown in bold indicate statistically significant results.

^aNeurometabolite level ∼ Age group × Task difficulty level × Timepoint.

^bp values obtained using Type III ANOVA with Satterthwaite's method.

^*p ≤ 0.05.

^**p ≤ 0.01.

^***p ≤ 0.001.

GABA+ levels. Results of the linear mixed effect model showed significant main effects of Age group in both SM1 and DLPFC voxels (p < 0.001 and p = 0.005, respectively). Additionally, a significant Timepoint effect (p = 0.01) was observed for GABA+ levels in the SM1. Post hoc analysis using Wilcoxon test revealed that the SM1 GABA+ levels declined over time and were lower in the following: (1) TP3 versus TP1 (V = 1,942.5, p = 0.025); (2) TP4 versus TP1 (V = 2,150, p = 0.002; Fig. 4C).

Glx levels. Results of the mixed effect model for Glx in SM1 and DLPFC did not show any significant effect for Age, Task difficulty level, Timepoint, or their interactions (Fig. 4D).

Regression analyses

Associations between behavioral measures and measures of neurometabolite levels

Predicting learning based on baseline levels of neurometabolites. Results of the stepwise (backward) regression identified that baseline GABA+ levels in DLPFC and in SM1 predicted learning (Pretest AvgDev − Posttest AvgDev) during the low task difficulty condition, with lower levels of GABA+ in SM1 and higher levels of GABA+ in DLPFC predicting a greater behavioral improvement (F_(2,40) = 3.13, R² = 0.13, p = 0.05; Table 4; DLPFC GABA+: β = 5.48, p = 0.04; Fig. 5A; SM1 GABA+: β = −4.56, p = 0.05; Fig. 5B).

Table 4.

Results of the linear regression models^a predicting learning based on baseline neurometabolite levels

Factor	Std. beta	t value	p value
Predicting learning (in low task difficulty level) based on baseline neurometabolite levels
R2 = 0.13, F(2,40) = 3.13, p = 0.05
Intercept	−0.09	−0.14	0.88
Baseline DLPFC GABA+	5.48	2.16	0.04 **
Baseline SM1 GABA+	−4.56	−2.01	0.05 *
Predicting learning (in high task difficulty level) based on baseline neurometabolite levels
R2 = 0.18, F(4,38) = 2.22, p = 0.08
Intercept	−0.08	−0.12	0.91
Baseline DLPFC GABA+	−6.88	−1.91	0.06
Baseline SM1 GABA+	0.06	0.02	0.98
Baseline DLPFC Glx	1.03	0.91	0.37
Baseline SM1 Glx	−1.21	−1.08	0.28

Values shown in bold indicate statistically significant results.

^aStepwise regression with backward elimination method. The predictor variables are exclusive to the condition (low or high).

^*p ≤ 0.05.

^**p ≤ 0.01.

Figure 5.

A, Baseline levels of GABA+ in the SM1 negatively predicted learning (low difficulty) in the full sample. B, Baseline levels of GABA+ in the DLPFC positively predicted learning (low difficulty) in the full sample. C, Baseline levels of GABA in the DLPFC negatively predicted learning (high difficulty) in the full sample. Blue dots represent participants in the YA group, and red dots represent participants in the OA group. DLPFC, dorsolateral prefrontal cortex; i.u., institutional units; SM1, sensorimotor cortex.

In contrast, there were no significant predictors of learning in the high task difficulty condition (F_(4,38) = 2.22, R² = 0.18, p = 0.08; Table 4), although baseline levels of DLPFC GABA+ showed a trend toward significance (β = −6.88; p = 0.06; Fig. 5C), with lower levels of DLPFC GABA+ being related to greater learning.

Predicting learning based on modulation of neurometabolites. The stepwise (backward) regression analyses showed that the interaction of Age with the training-induced neurometabolic modulation of SM1 Glx predicted learning during the low task difficulty condition (F_(3,39) = 2.73, R² = 0.17, p = 0.05, β = 4.71, p = 0.04; Table 5). While in younger adults a practice-induced increase in SM1 Glx was associated with a greater performance improvement, there was no such association in the older participants (Young r = 0.59, p = 0.003; Old r = −0.05, p = 0.83; Fig. 6).

Table 5.

Results of the linear regression models^a predicting learning based on neurometabolic modulation

Factor	Std. beta	t value	p value
Predicting learning (in low task difficulty level) based on neurometabolic modulation
R2 = 0.17, F(3,39) = 2.73, p = 0.05
Intercept	−0.31	−0.32	0.75
Age group	0.11	0.08	0.93
Modulation SM1 Glx	−0.41	−0.23	0.81
Age × Modulation SM1 Glx	4.71	2.06	0.04 *
Predicting learning (high task difficulty level) based on neurometabolic modulation
R2 = 0.32, F(4,37) = 4.35, p = 0.005
Intercept	−2.08	−2.21	0.03
Age group	4.11	3.05	0.004 **
Modulation SM1 GABA+	−6.49	−1.47	0.15
Modulation DLPFC Glx	−1.91	−1.63	0.11
Age × Modulation SM1 GABA+	10.25	5.91	0.09

Values shown in bold indicate statistically significant results.

^aStepwise regression with backward elimination method. The predictor variables are exclusive to the condition (low or high). df, degrees of freedom.

^*p ≤ 0.05.

^**p ≤ 0.01.

Figure 6.

Modulation of Glx in the SM1 predicted learning (low difficulty) in young adults (YA). Positive values in the modulation indicate an increase in Glx levels, whereas a negative value indicates a decrease. These findings indicate that in YA, a task-induced increase in SM1 Glx levels was associated to a higher learning gain (low difficulty). No association was observed in older adults (OA).

For the model to predict degree of learning during the high task difficulty condition, only age was the significant predictor (F_(3,39) = 4.35, R² = 0.32, p = 0.005; β = 4.11, p = 0.004; Table 5), with younger adults showing a greater improvement.

Predicting retention based on baseline levels of neurometabolites. The stepwise (backward) regression identified that higher baseline levels of SM1 GABA+ were related to a greater retention for the low task difficulty condition (F_(4,35) = 4.45, R² = 0.33, p = 0.005; Table 6; β = 4.96, p = 0.04; Fig. 7A). Furthermore, younger individuals showed greater retention in the low task difficulty condition than older adults (β = −3.01, p = 0.03), and there was an interaction of Age with baseline SM1 Glx levels, where higher SM1 Glx levels were related to a lower retention in young adults but to a greater retention in older adults (β = −2.96, p = 0.05; Young r = −0.49, p = 0.02, Old r = 0.42, p = 0.05; Fig. 7B).

Table 6.

Results of the linear regression models^a predicting retention based on baseline neurometabolite levels

Factor	Std. beta	t value	p value
Predicting retention (in low task difficulty level) based on baseline neurometabolite levels
R2 = 0.33, F(4,35) = 4.45, p = 0.005
Intercept	0.31	0.34	0.73
Age group	−3.01	−2.31	0.03 *
Baseline SM1 GABA+	4.96	2.18	0.04 *
Baseline SM1 Glx	2.31	1.94	0.06
Age group × Baseline SM1 Glx	−2.96	1.48	0.05 *
Predicting retention (high task difficulty level) based on baseline neurometabolite levels
R2 = 0.21, F(4,38) = 2.41, p = 0.06
Intercept	0.25	0.26	0.79
Age group	1.43	1.05	0.31
Baseline DLPFC Glx	−1.43	−1.49	0.14
Baseline SM1 Glx	3.27	2.28	0.03 *
Age × baseline SM1 Glx	−3.31	1.64	0.05 *

Values shown in bold indicate statistically significant results.

^aStepwise regression with backward elimination method.

^*p ≤ 0.05.

Figure 7.

A, Baseline levels of GABA+ in the SM1 positively predicted retention (low difficulty) in the full sample. B, Baseline levels of Glx in the SM1 positively predicted retention (low difficulty) in older adults (OA) but negatively predicted retention in young adults (YA). C, Baseline levels of Glx in the SM1 positively predicted retention (high difficulty) in the older adults, whereas no association was observed in young individuals.

For retention in the high task difficulty condition, higher baseline levels of SM1 Glx were related to greater retention gains (F_(4,38) = 2.41, R² = 0.21, p = 0.06; Table 6; β = 3.27; p = 0.03). In addition, this pattern was different between young and older adults, with older adults exhibiting a greater retention associated to higher baseline levels of SM1 Glx, while this was not the case in young adults (β = −3.31, p = 0.05; Young r = −0.34 p = 0.11, Old r = 0.46, p = 0.04; Fig. 7C).

Predicting retention based on modulation of neurometabolites. Results of the stepwise (backward) regression identified a model (F_(6,33) = 4.58, R² = 0.45, p < 0.001; Table 7), in which a decrease in SM1 Glx from TP1 to TP4 was related to a greater retention in the low task difficulty condition (Glx: β = −4.06, p = 0.004). In addition, the interaction of this factor with age was significant (β = 4.57, p = 0.02); whereas a decrease in SM1 Glx was related to a greater retention in older adults (Low difficulty), no association was observed in younger adults (Young r = −0.27, p = 0.22, Old r = −0.61, p = 0.004; Fig. 8A).

Table 7.

Results of the linear regression models^a predicting retention based on neurometabolic modulation

Factor	Std. beta	t value	p value
Predicting retention (in low task difficulty level) based on neurometabolic modulation
R2 = 0.45, F(6,33) = 4.58, p = 0.001
Intercept	0.05	0.06	0.95
Age group	−2.87	−2.49	0.01 *
Modulation DLPFC GABA+	−3.23	−1.57	0.12
Modulation SM1 GABA+	−1.82	−0.67	0.51
Modulation SM1 Glx	−4.06	−3.05	0.004 **
Age × Modulation SM1 GABA+	−6.44	−1.52	0.14
Age × Modulation SM1 Glx	4.57	2.49	0.02 *
Predicting retention (high task difficulty level) based on neurometabolite modulation
R2 = 0.23, F(3,39) = 3.92, p = 0.01
Intercept	0.98	1.05	0.29
Age group	0.97	0.75	0.46
Modulation SM1 GABA+	13.79	3.16	0.003 **
Age × Modulation SM1 GABA+	−11.19	−1.95	0.058

Values shown in bold indicate statistically significant results.

^aStepwise regression with backward elimination method. df, degrees of freedom.

^*p ≤ 0.05.

^**p ≤ 0.01.

Figure 8.

A, Modulation of Glx in SM1 negatively predicted retention in older adults (OA), showing that a task-induced decrease in this metabolite was associated to higher retention in the low difficulty condition. No association was observed in young adults (YA). B, Modulation of GABA+ in SM1 positively predicted retention in older adults; that is, a task-induced increase in this metabolite was associated to higher retention in the high difficulty condition. No association was observed in young adults.

Finally, the stepwise regression model for the high task difficulty condition showed that SM1 GABA+ modulation predicted retention in the high task difficulty condition (F_(3,39) = 3.92, R² = 0.23, p < 0.01, β = 13.79, p = 0.003; Table 7), indicating that an increase in SM1 GABA+ was associated to a greater retention. The interaction of SM1 GABA+ modulation with age showed a trend toward significance, where a more accentuated effect was observed in older adults (β = −11.19, p = 0.058; Young r = 0.22, p = 0.31, Old r = 0.59, p = 0.004; Fig. 8B).

Discussion

We observed that the baseline GABA+ levels in both SM1 and DLPFC regions were lower in older adults as compared with young adults. Furthermore, we observed a reduction of GABA+ in the SM1 during bimanual coordination practice regardless of age and difficulty level. Finally, we observed that the prediction of learning and retention performance based on the neurometabolic baseline levels and their training-induced modulation was dependent on the brain region, age, and task practice condition.

Age and brain regional effect on neurometabolites

GABA levels in both regions were lower in older as compared with young adults, replicating previous findings (Hermans et al., 2018; Cuypers et al., 2020; Rodríguez-Nieto et al., 2023). The age-related decline in GABAergic neurons (Hua et al., 2008; Stanley et al., 2012) and the decline in the efficacy of GABA production (Caspary et al., 1995; Leventhal et al., 2003) might be the biological processes underlying these observations. Age-related GABA+ reductions could indicate diminished inhibitory regulation, potentially contributing to less efficient motor learning in older adults. Although Glx was lower in older adults, this effect was not significant. Previous evidence suggests that age-related differences in Glx may be region dependent (Cleeland et al., 2019; Rodríguez-Nieto et al., 2023).

In both age groups, GABA levels were higher in the SM1, as compared with the DLPFC, and Glx levels were higher in the DLPFC than in the SM1 region, replicating previous findings (Maes et al., 2021; Li et al., 2024a). The lower GABA+ levels and higher Glx levels observed in the DLPFC compared with SM1 may reflect region-specific differences in inhibitory tone.

Modulation in the levels of neurometabolites

The SM1 GABA+ levels decreased during bimanual coordination training, consistent with previous findings using finger-sequencing tasks (Floyer-Lea et al., 2006; Kolasinski et al., 2019), although some studies have reported null effects (Maruyama et al., 2021; Bell et al., 2023). Among other things, inconsistency across studies may relate to training duration, as both our study and those reporting similar decreases measured neurometabolites over periods longer than 20 min, with significant GABA reductions emerging only toward the final measurement.

We had hypothesized that more challenging task conditions would elicit greater plasticity and thus larger GABA reductions. However, task difficulty level did not modulate metabolic changes in either region. This suggests that although SM1 GABA+ modulation may be necessary for task learning, it does not appear to be scaled up as a function of increasing task difficulty.

We had hypothesized a similar reduction of inhibition in the DLPFC, given its role in cognitive control during (initial) motor learning. However, no learning-related changes in GABA+ or Glx were observed in this region. Two possible reasons may account for this null result. First, the DLPFC may play a more supportive rather than primary role in this particular coordination task, which could lead to smaller and thus less detectable neurometabolic changes compared with SM1. Second, although the DLPFC contributes to learning, it may be less critical than SM1 for establishing the memory representation of the movement.

Prediction of learning

Baseline neurometabolite levels are often considered markers of the brain's default excitability. In the full sample, lower baseline SM1 GABA+ predicted greater performance improvement in the low difficulty condition, but no such relationship emerged in the high difficulty condition. A similar absence of correlation was also observed in a high challenge motor learning task (Hehl et al., 2025). This pattern suggests that the influence of baseline inhibitory tone on learning may depend on task demands. A tentative interpretation is that lower baseline SM1 GABA+ may support early performance gains when the task is relatively easy, as individual differences in inhibitory tone can directly facilitate motor cortical excitability. In contrast, in the high difficulty condition, performance is likely constrained by greater task complexity and strategy demands, which could engage additional neural processes and obscure any straightforward relationship between baseline GABA+ in SM1 and learning. However, a positive association between learning and M1 GABA has also been reported (Li et al., 2024a). This discrepancy may reflect differences in voxel localization, as the VOI in that study was positioned closer to frontal regions rather than the more typical SM1 location used here. Another possible reason is the difference in practice duration: ∼38 min in our study compared with 24 min in Li's study.

We had hypothesized that neurochemical properties in the DLPFC would also predict motor learning, given its established role in early learning stages. In the low-difficulty condition, lower baseline DLPFC GABA+ levels were associated with reduced learning, whereas in the high-difficulty condition, lower baseline GABA+ levels were associated with greater learning. This pattern suggests that a lower baseline inhibitory tone in the DLPFC, potentially reflecting higher neuronal activity, may facilitate performance improvements when motor coordination demands are high.

In young adults, greater training-induced increases in SM1 Glx were associated with better learning in the low-difficulty condition. Neurometabolic modulation is thought to reflect the brain's capacity to adjust neurochemical levels in response to task demands, and previous work has shown that the reduction of inhibition within SM1 is a key mechanism supporting early motor learning (Floyer-Lea et al., 2006; Stagg et al., 2011; Kolasinski et al., 2019). In this context, our findings suggest that an accompanying enhancement of excitatory tone in SM1 may further facilitate learning when motor coordination demands are relatively low.

Prediction of retention

Higher baseline SM1 GABA+ levels were associated with greater retention in the low difficulty task. This result aligns with previous findings reporting a positive relationship between baseline SM1 GABA+ and motor memory retention (Li et al., 2022; Hehl et al., 2025). Although this finding was not observed in our study in the high task difficulty condition, it may be possible that generally higher levels of baseline SM1 GABA+ facilitate the consolidation of motor learning, as measured during the retention test. A higher inhibitory tone can favor stability (in contrast to plasticity) during the posttraining epoch (Yamada et al., 2024), supporting retention by overcoming interference among the task variants in our study and promoting distinctiveness of these task representations.

Interestingly, only older adults appeared to benefit from higher baseline SM1 excitability (i.e., higher Glx) for successful retention across both task difficulty conditions. In contrast, in young adults, lower baseline SM1 Glx levels predicted better retention in the low difficulty condition. These opposing patterns suggest that an optimal excitability range may be necessary for effective long-term retention and that this optimal range may vary with age but further research is warranted.

In older adults, a task-induced decrease in SM1 Glx was associated with better retention in the low-difficulty condition. Furthermore, greater training-induced increases in SM1 GABA+ appeared to support retention in the high-difficulty condition. Together, these findings suggest that increases in inhibitory tone during training may benefit retention in this older age group. As noted earlier, stronger inhibitory processes may help minimize interference among subtask elements during consolidation.

It is important to acknowledge the current limitations of the MRS technique in terms of temporal as well as spatial resolution. Our measurements were obtained in blocks of almost 10 min, and these reflect the average neurochemical activity across multiple task elements within each block, including different feedback conditions and movement phases (planning and execution). These elements likely contribute jointly to the observed metabolic signal, which should therefore be interpreted as an integrated measure of the underlying processes occurring during the entire task period. Moreover, given the relatively large voxel size, it was impossible to determine the MRS signal from subregions of the DLPFC.

General discussion

From the group-level analyses, we found that SM1 GABA+ levels decreased during bimanual coordination training, consistent with previous findings. At the individual level, greater training-induced increases in SM1 Glx were associated with better learning in the low-difficulty condition. Together, these results suggest that a reduction in inhibitory tone, with/without an accompanying enhancement in excitatory tone, may support the plastic processes that underpin motor learning.

In contrast, the patterns associated with long-term retention pointed to a different neurochemical profile. Increased inhibitory tone, or reduced excitability tone, in SM1 was associated with better retention, manifested as higher baseline GABA+ in the full sample, lower baseline Glx in young adults, and training-related increases in GABA+ or decreases in Glx in older adults. These effects were most evident in the low-difficulty condition. This may reflect that higher inhibitory tone supports consolidation, potentially by enhancing the specificity of neural representations through selective activation of relevant neuronal populations and/or by reducing interference among the different subtasks. Although reduced SM1 excitability generally supported retention, with most effects observed in the low-difficulty condition, one notable exception became evident in the high-difficulty condition. In older adults, higher baseline SM1 Glx was associated with better retention, contrasting with the broader pattern linking reduced excitability to improved long-term performance. One possible explanation is that older adults may require greater baseline excitatory resources to support the consolidation of more complex coordination tasks.

Conclusion

Beyond the group-level effect showing a training-induced reduction in SM1 GABA+, we identified specific metabolic features that predicted individual differences in learning. We also characterized the much less studied metabolic features associated with motor task retention. The metabolic predictors of learning and retention varied with task difficulty, suggesting that increased coordination demands engage a more complex interaction between excitatory and inhibitory processes. Finally, clear age-related differences emerged, with metabolic features played a more prominent role in predicting retention in older than in younger adults.