A total of 50 right-handed participants distributed across two age groups—younger adults (YA) aged 20 to 40 years (N = 25; mean±SD 25 ± 5 years) and older adults (OA) aged 60 to 80 years (N = 25; mean±SD 68 ± 5 years)—were included in this MRS study. All participants were right-handed [determined using the Edinburgh Handedness Inventory (EHI) (Oldfield, 1971); scores ranging from −100 to +100, where values above +50 indicated right-handedness; inclusion cut off ≥ 50]. All participants had normal or corrected-to-normal vision. Recruitment took place across Flanders, Belgium, through community and university channels, including social media, the research group’s website, advertisements for the experiment in lectures focused on learning after retirement, and posters. Additionally, older adults listed in the Movement Control & Neuroplasticity Research Groups’ database at KU Leuven were contacted via email or telephone.

Prior to participation, participants were extensively screened. Cognitive and executive functioning were evaluated using the Montreal Cognitive Assessment [MoCA; Nasreddine et al., 2005] using a cut-off score of ≤ 23/30 (Carson et al., 2018). To screen for self-reported depressive symptoms, participants completed the 21-item Beck Depression Inventory (BDI), with higher scores reflecting more severe depressive symptoms and using a cut-off score of >13 (Beck et al., 1961; Van der Does, 2002). The 90-item Symptom Checklist (SCL) was also administered to assess physical and psychological symptoms, with a cut-off score of >81 (Derogatis, 1975), ensuring participants were in good mental and physical health. Table 1 displays questionnaire scores and demographics per group. Finally, participants were excluded if they self-reported (a history of) central nervous system disorders, psychiatric conditions, use of medications that affect the central nervous system (e.g., sedatives, antidepressants), brain surgeries or injuries, health issues or treatments influencing the central nervous system, or a history of drug or alcohol abuse, or if the reported any contraindications to MRI (e.g., metal implants, claustrophobia).

The study protocol was approved by the local ethics committee (Ethics Committee Research UZ/KU Leuven; reference S66028) and participants gave full written informed consent prior to study participation, according to the latest amendment of the Declaration of Helsinki (World Medical Association, 2013).

Prior to inclusion, each participant underwent a screening session to ensure eligibility. This session encompassed an evaluation of the previously outlined inclusion and exclusion criteria (see Section 2.1 Participants) and a practical familiarization phase. For the latter, participants entered a mock scanner, a non-magnetic replica of the MRI machine, to simulate the actual experimental environment. Within this mock setup, participants performed 15 trials of the Bimanual Tracking Task (BTT), focusing only on the simple (1:1) condition (see Section 2.3 Bimanual Tracking Task). This procedure allowed assessment of participants’ comfort in the scanner environment and verified participants’ ability to understand and perform the required task. If participants were comfortable in the scanner and agreed to continue participation, they were scheduled for a second session at the university hospital on a different day. This session included a 105-min MRI scan with a 5 min break in between. The first part involved acquiring high-resolution T1 and T2-weighted anatomical images (~10 min) and MRS measures (~4 × 11 min), while the second part included diffusion-weighted imaging (DWI; ~15 min) and resting-state functional MRI (rs-fMRI; ~15 min). This paper will focus solely on the data from the first part of the multimodal acquisition in order to maintain coherence and flow. The experimental design is illustrated in Figure 1.

The bimanual tracking task (BTT) was used to assess bimanual coordination and short-term motor learning since it requires real-time control of complex movement patterns with both hands. Previous studies employing a similar task design have demonstrated its sensitivity to age-related changes in bimanual coordination performance (e.g., Pauwels et al., 2015; Santos Monteiro et al., 2017; King et al., 2018; Maes et al., 2021). Originally developed by our group (Sisti et al., 2011; Pauwels et al., 2015; Monteiro et al., 2019), the set-up has since been further optimized for the MRI procedures used in this study [also see Hehl et al. (2025)]. Specifically, a non-ferromagnetic version of the BTT setup was adapted to be placed bridge-like over the participants’ hips, allowing for effortless operation while they were in a supine position.

During the task-related MRS acquisition of GABA, participants performed the BTT inside the MR scanner (± 52 trials in total over 11 min). The task was displayed via a projection screen mirrored at the participant’s eye level (projection [LCD projector: NEC NP-PA500U, 1920 × 1,200 pixels] at the cranial end of the MR scanner and viewed through a mirror sized ~14 × 9 cm, positioned approximately 13 cm from the participant’s eyes) and required participants to track a white target dot as it moved at a constant speed along a blue target line. Using coordinated movements of both index fingers, participants had to match the dot’s trajectory as accurately as possible. Each hand was placed on a handle, with the index fingers extended and positioned in the circular grooves of rotatable dials, which controlled movement along the x-axis (right finger) and y-axis (left finger). While maintaining contact with the handles, only the index fingers could move.

Each trial began with a blue target line on a black background that remained visible until the end of each trial. A large yellow dot initially covered the smaller white target dot, marking a 2000 ms anticipatory period. Participants were instructed to start moving as soon as the yellow dot disappeared, rotating the dials to control vertical (left dial) and horizontal (right dial) cursor movements, while a red line visualized the real-time trajectory of the participant’s cursor (see Supplementary material 1). The movement phase lasted 10,000 ms and ended when the white dot reached the opposite end of the blue target line. Each trial lasted 12,000 ms in total, with a 2000 ms rest between movement phases. The task comprised three conditions, each with distinct inter-hand movement frequency ratios: 1:1 (iso-frequency), 1:3, and 3:1 (non-iso-frequency). The 1:1 condition, in which both fingers moved at identical speeds, represented the least complex coordination task (e.g., Swinnen, 1998; Sisti et al., 2011). In contrast, the non-iso-frequency conditions (1:3 and 3:1) required asymmetric coordination, with, respectively, the right or left index finger moving three times faster than the other. The target moved along one of four possible trajectories: toward the upper right, upper left, lower right, or lower left quadrant, requiring different rotation directions (e.g., both hands rotating clockwise when moving to the upper right).

All MRS and MRI data were acquired on a 3 Tesla scanner (Philips Achieva dstream scanner; University hospital Leuven, Gasthuisberg), with a 32-channel receiver head coil (Philips, Best, The Netherlands). The protocol consisted of an anatomical T1-weighted scan, followed by four successive MRS acquisitions of the right cerebellar hemisphere, performed during approximately 44 min, including three GABA-edited acquisitions—one at rest (GABA_pre and Glx_pre), one during task performance (GABA_BTT and Glx_BTT), and another at rest (GABA_post and Glx_post)—and one GSH-edited acquisition (see Figure 1).

The T1-weighted high-resolution anatomical image was collected using a three-dimensional turbo field echo (3DTFE) sequence [echo time (TE) = 4.6 ms, repetition time (TR) = 9.7 ms, flip angle = 8°, field of view = 256 × 242 × 182 mm, 182 sagittal slices, voxel size = 1 × 1 × 1 mm3; acquisition time ~6 min] and was used for MRS voxel placement and correction of MRS quantification based on anatomical information.

For MRS, acquisition voxels were positioned in the right cerebellar hemisphere [30 (AP) x 25 (LR) x 25 (HF) mm³ = 18.75 mL], see Figure 2. Specifically, the VOI was positioned at the top of the cerebellum, aligned with the cerebellar border along the cerebrum, and placed as medially as possible to include the dentate nucleus, while avoiding contact with the fourth ventricle in both the sagittal and coronal planes. In exceptional cases—such as participant discomfort or technical issues—excessive head movement or briefly removing the participant from the scanner for an additional break required the T1-weighted anatomical scan to be repeated and the MRS voxel to be repositioned. This occurred for one older and one younger participant, where the voxel was repositioned after both the GABA_pre and GABA_post scans, and for one additional younger participant, where repositioning was necessary only after the GABA_pre scan.

A J-difference-edited Mescher-Garwood Point-Resolved Spectroscopy (MEGA-PRESS) protocol was used to quantify either GABA or GSH (TR = 2,000 ms, TE = 68 ms for GABA-edited / 120 ms for GSH-edited MRS, 2,000 Hz spectral width, 1,024 spectral points, ‘on’/‘off’ editing pulse frequency = 1.9/7.5 ppm for GABA-edited / 4.56/7.5 ppm for GSH-edited MRS, ‘on’/‘off’ spectra collected in an interleaved fashion, duration ~11 min per acquisition) (Mescher et al., 1998; Edden and Barker, 2007). The automated second-order “pencil-beam” (PB) shimming procedure (Philips) was used, and water suppression was applied using the Multiply Optimized Insensitive Suppression Train (MOIST; bandwidth 140 Hz). For each of the four acquisitions, an additional 16 unsuppressed water averages were acquired within the same VOI using the same acquisition parameters and later used as reference metabolite. Since macromolecular signals were co-edited during the GABA-edited MRS acquisition, the signal detected at 3 ppm represents both GABA and macromolecules, and will be referred to as GABA+.

To evaluate the accuracy and reliability of voxel placement, the overlap across participants is visualized in Figure 2. Additionally, the mean difference spectra for both age groups, along with standard deviation (SD) and confidence intervals (CI), are presented.

A summary of all experimental procedures, in accordance with the “minimum reporting standards in MRS” consensus paper (Lin et al., 2021), is provided in Supplementary material 2.

Data processing was performed using Gannet (version 3.4.0-dev; RRID: SCR_016049) (Edden et al., 2014) and included Eddy current correction for both water and metabolite data, exponential line-broadening of 3 Hz, zero-filling to 32 K data points, and weighted averaging of the transients (Mikkelsen et al., 2020). Frequency and phase corrections were applied using robust spectral registration for all GABA+ data (Mikkelsen et al., 2020) and using spectral registration to NAA for GSH data. Since spectral registration of GSH-edited data is challenging, other alignment algorithms were used if visual inspection of pre- and post-alignment data suggested that the NAA spectral alignment algorithm was introducing error and, hence worsening spectral quality. A detailed description of this procedure can be found in Supplementary material 3.

For quantification purposes, the Gannet model for GABA+-edited spectra applies a five-parameter Gaussian-Lorentzian model to fit a Gaussian peak between 2.79 and 3.55 ppm. Specifically, the model incorporates three Gaussian peaks to represent GABA+ and the Glx [glutamate (Glu) and glutamine (Gln)] doublet between 2.79 and 4.1 ppm, along with a four-parameter curved baseline function including linear and quadratic terms. The Gannet model for GSH-edited spectra utilizes a six-parameter Gaussian model to fit the data between 2.25 and 3.5 ppm. This includes a four-parameter curved baseline function with linear and quadratic terms, one Gaussian assigned to model the GSH signal, and additional Gaussians to represent the complex co-edited aspartyl multiplet at approximately 2.6 ppm. Unsuppressed water signals were modeled using a Gaussian-Lorentzian lineshape. Consistent with previous MRS research on aging (Chalavi et al., 2018; Hermans et al., 2018b; Maes et al., 2018; Maes et al., 2021; Verstraelen et al., 2021), metabolite concentration estimates are reported relative to this internal water reference (i.e., GABA+/H2O).

MRS voxels were coregistered to the T1-weighted image using the GannetCoRegister module (SPM12), and the T1 data was then segmented with the GannetSegment module to identify the tissue fractions of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) within the voxels (Ashburner and Friston, 2005). During quantification via the GannetQuantify module, tissue correction was applied according to the Gasparovic method (Gasparovic et al., 2006). This approach accounted for the relative MR visibility of water in different tissues (GM, WM and CSF) as well as the distinct T1 and T2 relaxation values for both water and metabolites. Modulation of GABA+ and Glx was defined as the difference between during-task (BTT) and pre-task levels, calculated by subtracting GABA_pre and Glx_pre from GABA_BTT and Glx_BTT, respectively.

All statistical analyses were performed using SAS JMP® software (version 17, SAS Institute Inc., Cary, NC; RRID: SCR_022199) and the significance level was set at α = 0.05. All values are presented as mean±SD unless stated otherwise.

Where linear mixed models (LMMs) or multiple linear regression models were employed, the final model was obtained using backward selection (Babyak, 2004). Specifically, starting from the initial model, this involved iteratively removing effects that lacked significance. To ensure robustness, model selection was cross-validated using the Akaike Information Criterion (AIC) (Burnham and Anderson, 2002). The normality and homoscedasticity of conditional residuals of all models were visually checked using the Q-Q plot and the residual-by-predicted plot, respectively. Where adequate, post-hoc pairwise comparisons correcting for multiple testing were performed using Tukey honestly significant difference (HSD) tests.

The GABA_post data of one young adult was excluded from the analysis because its residual represented a visual outlier on the Q-Q plot that could not be accounted for using data transformation. This was corroborated using JMP’s ‘Robust Fit Outliers’ method based on Huber’s M-estimation (K = 4).

There were no significant effects of timepoint (pre, BTT and post levels), suggesting that GABA+ and Glx levels did not modulate over time at group level. Furthermore, there was no age group effect and no age group*timepoint interaction (all, p > 0.05).

Bimanual motor performance was better in younger as compared to older adults. As expected, motor performance was better in the simpler (iso-frequency) as compared to the more complex (non-iso-frequency) task variants. Additionally, performance differences between age groups were greater in the iso-frequency compared to the non-isofrequency variant, favoring the younger group. These findings are consistent with previous research (Swinnen and Wenderoth, 2004; Boisgontier et al., 2018; Maes et al., 2022; Van Hoornweder et al., 2022; Van Ruitenbeek et al., 2023).

In addition, the short-term motor learning rate was significantly higher in younger as compared to older adults, consistent with previous literature (e.g., Swinnen, 1998; Shea et al., 2006; for a review see: Voelcker-Rehage, 2008). Nevertheless, although motor performance was generally lower in older than younger adults, older adults still retain the ability to acquire new motor skills, albeit at a slower rate than younger adults (Raz et al., 2000; Smith et al., 2005; Voelcker-Rehage, 2008; Gooijers et al., 2024). This ability is mediated by task complexity, among others, with younger adults showing significantly faster improvement as task difficulty increases (Voelcker-Rehage, 2008). Hence, the lower learning rate in older as compared to younger adults observed in this study might be attributable to the overall complexity of the bimanual tracking task, encompassing both the simple and complex non-iso-frequency conditions.

At the cortical level, previous studies demonstrated reduced GABA+ levels in older adults as compared to younger adults (e.g., Porges et al., 2017a; Chalavi et al., 2018; Cassady et al., 2019; Chamberlain et al., 2021; Dobri and Ross, 2021; Maes et al., 2021; Heise et al., 2022), but findings remain inconsistent, as other studies reported no significant differences between age groups (e.g., Chalavi et al., 2018; Hermans et al., 2018b; Maes et al., 2018; Cuypers et al., 2021). It is important to note that different brain regions were studied, i.e., mostly cortical regions and especially the SM1 and primary sensory cortices (visual, auditory, somatosensory) (Li et al., 2022). To the best of our knowledge, this study is the first to examine age-related differences in neurometabolite levels in the cerebellum using MRS, and we found no significant age-related differences in cerebellar GABA+ concentration between both age groups. Interestingly, despite an increased CSF fraction and a trend toward lower GM in the voxels of older compared to younger adults, uncorrected GABA+ levels did not differ between age groups either (see Supplementary material 5). This contrasts with some studies that suggest that age-related differences in GABA+ levels (e.g., in the frontal cortex and SM1 cortex) are driven by bulk tissue changes and thus depend on the tissue correction method used (Porges et al., 2017b; Maes et al., 2018). It has been suggested that age differences in GABA+ are region-specific, potentially emerging in certain cortical areas while being absent or emerging later in others (Hermans et al., 2018a; Maes et al., 2021; Li et al., 2022). The latter may also be the case in the cerebellum.

In line with Ding et al. (2016) and Maghsudi et al. (2020), both lifespan studies using whole brain magnetic resonance spectroscopy imaging (MRSI), we found no age-related differences in cerebellar Glx levels. This aligns with some studies looking at cortical Glx (e.g., Mooney et al., 2017; Simmonite et al., 2019) but contradicts others that did reported age-related declines in cortical regions (e.g., Levin et al., 2019). Similar to GABA+, it has been suggested that age-related differences in Glx may be region-specific (Kaiser et al., 2005; Marjańska et al., 2017), and hence might evolve differently in the cerebellum as compared to cortical areas.

The finding that GABA+ and Glx levels did not differ between younger and older adults aligns with a recent transcranial magnetic stimulation (TMS) study (Van Malderen et al., 2025) examining CBI. CBI is thought to reflect the GABA-mediated inhibitory influence exerted from the cerebellar cortex to the dentate nucleus and motor cortex via the dentate-thalamo-cortical pathway (Ugawa et al., 1995; Fernandez et al., 2018). Similar to the current results, no age-related differences in CBI were demonstrated. While the exact relationship between MRS and TMS measures remains unclear—TMS primarily captures receptor-level functional aspects of GABAergic and glutamatergic neurotransmission (i.e., fast-acting phasic inhibition), whereas MRS predominantly reflects broader neurometabolic pools related to tonic inhibition (Stagg et al., 2011b; Rae, 2014; Cuypers and Marsman, 2021)—the two methods are considered complementary in assessing glutamergic and GABAergic neurotransmitter function (Stagg et al., 2011a; Tremblay et al., 2013; Cuypers and Marsman, 2021). These complementary findings support the idea that there may indeed be no significant age-related differences in cerebellar GABA+ and Glx functioning at the group level.

Our findings indicated no age-related differences in GSH levels in the right cerebellar hemisphere, supporting the notion that cerebellar GSH levels may remain stable across aging. This aligns with a recent systematic review suggesting that age-related GSH alterations are region-specific, with the cerebellum, among other regions, showing relative stability (Detcheverry et al., 2023).

Given that GSH plays a critical role in buffering reactive oxygen species (ROS)—associated with oxidative stress in high levels— its stability in the cerebellum may reflect this region’s relative resistance to aging (Schieber and Chandel, 2014). Supporting this idea, post-mortem studies have shown that, among others, the cerebellum exhibits fewer mitochondrial DNA deletions (Corral-Debrinski et al., 1992), reduced oxidative damage (Mecocci et al., 1993; Venkateshappa et al., 2012), less age-related DNA methylation compared to cortical regions (Horvath et al., 2015). This resilience may be linked to the cerebellum’s inherently low regional glucose metabolism, fairly low aerobic glycolysis, and minimal age-related change in glucose metabolism—factors that may lower susceptibility to oxidative stress (Bentourkia et al., 2000; Vaishnavi et al., 2010; Goyal et al., 2014).

Oxidative stress, driven by increased ROS production, an inevitable byproduct of oxygen metabolism in aerobic cells—is a key factor in aging (Ames et al., 1993). GSH plays a crucial role in buffering these ROS (Dringen et al., 2000). Evidence from rodent studies suggests that GSH homeostasis becomes increasingly disrupted with age, with several reports indicating lower GSH concentrations in older animals (e.g., Chen et al., 1989; Maher, 2005; Zhu et al., 2006; Rebrin et al., 2007).

Findings in humans, however, are more mixed and vary by region. Specifically, peripheral studies investigating blood GSH levels report declining GSH levels with age (Yang et al., 1995; Jones et al., 2002; Giustarini et al., 2006), while postmortem analyses yield contrasting results, showing either no age-related differences in GSH levels in the frontal cortex and cerebellum (Venkateshappa et al., 2012), or even increases in the caudate, frontal cortex, and cerebellum with advancing age (Tong et al., 2016). In vivo, studies are similarly inconsistent, with reports of stable (Gong et al., 2022) or lower GSH levels in older adults, for example, in the occipital and parietal lobes (Emir et al., 2011), precuneus, as well as in the posterior cingulate cortex (Suri et al., 2017). Conversely, other studies found higher GSH levels in the medial frontal and sensorimotor cortices in older adults after accounting for age-related atrophy (Hupfeld et al., 2021). Taken together, previous studies remain inconclusive, however, they emphasize the region-specific nature of age-related GSH changes. Our results contribute to this body of literature by reinforcing the idea that the cerebellum may be particularly resistant to GSH depletion with aging.

The preservation of GABA+, Glx, and GSH levels in the aged cerebellum may be attributed to its relatively slower aging process as compared to other (cortical) brain areas, potentially related to the so-called cerebellar reserve (Liang and Carlson, 2020; Arleo et al., 2024). While studies show age-related declines in cerebellar volume and corresponding reductions in motor performance (Jernigan et al., 2001; e.g., Bernard and Seidler, 2013; Bernard and Seidler, 2014), genetic and epigenetic evidence suggests that the cerebellum undergoes slower aging compared to other brain regions and may potentially be more resistant to the development of pathological markers of neurodegeneration (Liang and Carlson, 2020; Arleo et al., 2024). This resilience has been suggested to be due to the cerebellum’s greater epigenetic stability (Horvath et al., 2015), fewer age-related deletions in mitochondrial DNA compared to the cortex (Corral-Debrinski et al., 1992), and lower levels of oxidative damage to both mitochondrial and nuclear DNA (Mecocci et al., 1993). Therefore, while the average age of older adults in our sample was comparable to that of studies showing neurometabolic differences in cortical areas (e.g., Chalavi et al., 2018; Chamberlain et al., 2021; Maes et al., 2021), it is possible that they were not yet old enough to show detectable neurometabolic changes in the cerebellum. In addition, the older adults in our study were physically active, which may protect against metabolic decline, as physical activity is known to have a protective effect on the brain and may help preserve overall brain integrity (Boisgontier et al., 2018) and neurometabolite levels [GABA+ and Glx (Bütefisch et al., 2000; Maddock et al., 2016) as well as GSH (Rai et al., 2018)]. Future studies could consider including a less active, older group (e.g., >80 years) to further investigate the potential impact of age and physical activity on neurometabolite levels in very old age.

Alternatively, while age-related differences in neurometabolite levels may exist, they may have fallen below the detection threshold of our methodology. Specifically, just as different parts of the cerebellum undergo atrophic changes to varying degrees (Andersen et al., 2003; Paul et al., 2009), it is possible that specific metabolite changes occur on a smaller scale within functional cerebellar subregions. However, these subtle differences may be lost within the broader cerebellar voxel, which encompasses nearly the entire right cerebellar hemisphere and thus has limited spatial resolution. In addition, it could be that a longitudinal study is needed to detect such subtle differences and account for individual variability. Indeed, Zuppichini et al. (2024) demonstrated GABA differences in the right auditory, bilateral sensorimotor and ventrovisual regions between young and older adults in a longitudinal analysis, whereas no differences were observed in a cross-sectional comparison. Finally, it is important to consider the contribution of macromolecules to the GABA signal. Given that older exhibit higher macromolecule levels as compared to younger adults (Aufhaus et al., 2013; Marjańska et al., 2018), potential age-related differences in GABA itself (without macromolecules) may be underestimated.

GABA and Glx concentrations in a specific brain region may exhibit rapid modulation—either increasing or decreasing—as a function of task performance or motor learning, depending on the region’s precise role in the process being examined (e.g., Pasanta et al., 2023; Li et al., 2024b). Despite the increased interest in the modulation of GABA and Glx levels recently, evidence is still scarce, with only a few studies investigating visuomotor coordination learning tasks and just one study specifically focusing on the cerebellum (Nettekoven et al., 2022). Some studies reported GABA reductions in SM1 during motor performance and learning, which were linked to enhanced motor learning and improved visuomotor task performance (Floyer-Lea et al., 2006; Kolasinski et al., 2019; Maes et al., 2022) while others reported an increase in Glx which was correlated to motor actions but did not predict visual performance (Kurcyus et al., 2018; Volovyk and Tal, 2020). Nevertheless, findings are inconsistent, with considerable variability in effect sizes and directionality for GABA levels (Pasanta et al., 2023). Furthermore, prior research has shown that neurometabolite changes are highly region-specific, with some studies failing to detect modulation in certain areas while observing significant effects in others (Chalavi et al., 2018; Coxon et al., 2018; Frangou et al., 2019). Finally, specific features of the task also seem to influence temporal neurometabolite changes (Chalavi et al., 2018; Hehl et al., 2025).

In the present study, we did neither observe task-related modulation of GABA+ or Glx levels at the group level, nor did we find age-related differences in the modulation of these neurometabolites in the cerebellum. These findings align with previous studies that report mixed evidence for GABA and Glx modulation depending on the specific task and brain region, and might point to the absence of a modulation of GABA and Glx in the context of bimanual coordination learning in the cerebellum. Alternatively, it could be that transient changes in neurometabolite levels within a specific cerebellar region occur on a small scale within a specific functional sub-region. Indeed, prior work using magnetic resonance spectroscopic imaging (MRSI) has demonstrated GABA changes in the dentate nucleus but not in the cerebellar hemisphere (Nettekoven et al., 2022), suggesting that modulation might occur at a finer spatial scale than our voxel, which encompassed most of the right cerebellar hemisphere, could capture. It should be noted that the cerebellum is composed of a mosaic of functional subregions, each specialized for distinct roles ranging from fine motor control to social and linguistic processing (King et al., 2019; Nettekoven et al., 2024). As such, our cerebellar voxel likely captured a broad and heterogeneous area, in contrast to cortical MRS studies that often target more functionally discrete and homogeneous regions. Alternatively, modulation might take place on an even smaller scale, given the presence of multiple pathways underlying synaptic long-term potentiation (LTP) in the cerebellum (Binda et al., 2016) and other forms of cerebellar plasticity occurring at microscale (Schonewille et al., 2011). Due to the relatively large voxel in MRS, and hence, limited spatial resolution, averaging across multiple layers, subregions and functional domains—each serving distinct sensorimotor, cognitive, or affective functions—subtle, region-specific neurometabolic differences might go undetected. Future studies employing higher spatial resolution and region-specific targeted approaches may provide further insights into the subtle dynamics of neurometabolite regulation in the cerebellum.

Along the same lines, our use of 11-min task blocks may have limited our ability to detect modulation of metabolite levels. On one hand, dynamic fluctuations may have occurred outside the temporal sensitivity of our block-averaged approach, potentially masking transient changes (Bell et al., 2023). On the other hand, it is also possible that the duration of task engagement was insufficient for eliciting measurable neurometabolic changes, as longer or more intensive task practice might be required for such effects to emerge. The absence of detectable GABA+ or Glx modulation in our study may therefore reflect limitations in temporal resolution rather than a true absence of change. Future studies could benefit from exploring sliding-window analyses to capture transient modulations that are otherwise obscured by long block durations with a block-averaged analysis (Chen et al., 2017; Bell et al., 2023) as well as prolonged task paradigms to test for delayed or cumulative effects.

Since we found that neither baseline GABA+, Glx and GSH levels, nor GABA+ and Glx modulation predicted motor performance or short-term motor learning at a group-level, it may be that cerebellar neurometabolite concentrations remain relatively stable and not directly relate to age-related differences in behavioral outcomes. Instead, their effects could be mediated by other age-related factors influencing cerebellar function. For example, declines in norepinephrine levels may weaken GABAergic signaling (Watson and McElligott, 1983; Heron et al., 1996; Zhang et al., 2010), while age-related changes in Purkinje cell electrophysiology (Rogers et al., 1981; Chung et al., 2003) and synaptic signaling (Takahashi et al., 2009) could also play a role. A detailed investigation of these mechanisms is, however, beyond the scope of the current study.

It is also plausible that age-related differences in behavior and motor learning are not primarily driven by age-related changes in cerebellar neurometabolite levels but rather by broader alterations in cerebellar morphology, and both functional and structural connectivity. Indeed, Bernard and Seidler (2014) have argued that reductions in cerebellar volume, and particularly diminished connectivity with cortical and subcortical regions, may contribute significantly to age-related motor deterioration. To further clarify the complexity of cerebellar contributions to motor performance in aging, future studies may benefit from integrating morphometric analyses (e.g., voxel-based morphometry) with neurometabolic, connectivity (diffusion-weighted imaging and functional magnetic resonance imaging), and behavioral data. Such multimodal approaches could help disentangle whether structural, functional, and neurochemical differences interact to mediate motor outcomes in older adults.

Moreover, as already discussed at length in the above sections, it might be that the neurometabolite levels in specific functional sub-regions rather than the cerebellar hemisphere as a whole are linked to behavioral outcome. Indeed, while Nettekoven et al. (2022) did not observe a group-level reduction in GABA during an adaptation task using MRSI, they reported that greater early adaptation-driven GABA decreases in the right dentate nucleus were associated with better performance. Due to the relatively large voxel size in our study, we were unable to differentiate between cerebellar subregions such as the hemisphere and the deep nuclei, which may have masked subtle but functionally relevant differences.

Finally, it is important to consider that bimanual coordination relies on a distributed network of regions beyond the cerebellum, including—but not limited to— primary motor (M1) and sensory (S1) cortices, supplementary motor area (SMA), premotor cortex, cingulate motor area, basal ganglia (Laplane et al., 1977; e.g., Brinkman, 1981; Brinkman, 1984; Tanji et al., 1987; Tracy et al., 2001; Debaere et al., 2004; Swinnen and Wenderoth, 2004; Jantzen et al., 2009). Indeed, age-related atrophy and alterations in functional and structural connectivity within this network have been shown to contribute to impaired bimanual coordination (e.g., Goble et al., 2010; Solesio-Jofre et al., 2014; Serbruyns et al., 2015; Fujiyama et al., 2016; Serbruyns et al., 2016). It might therefore be interesting if future work would use a network-level approach to better understand how aging affects the complex interplay between cerebellar, other subcortical and cortical regions involved in motor control and learning.

Most of the limitations of the study have been mentioned earlier, but for clarity, a summary is provided here.

First, it is of importance to acknowledge the contribution of macromolecules to the GABA+ signal. Specifically, since macromolecule levels are higher in older as compared to younger adults (Aufhaus et al., 2013; Marjańska et al., 2018; Genovese et al., 2024), potential age-related differences in GABA+ may be under-estimated. Moreover, the use of macromolecule-suppressed GABA measurements may enhance the relationship between GABA levels and motor performance (Mikkelsen et al., 2018), put still poses several methodological difficulties.

Second, Glx is a compound measure including both glutamate and glutamine. Its measurement as a single entity may lead to different age-related effects than assessing glutamate and glutamine independently (Hädel et al., 2013; Roalf et al., 2020). Specifically, in certain cortical brain regions, glutamate has been shown to decline with age, whereas glutamine demonstrated an increase (Hädel et al., 2013). Due to the considerable spectral overlap between their resonances, higher magnetic field strengths are recommended as they will improve separation, enabling more precise quantification of both metabolites (Hetherington et al., 1997). Furthermore, other MRS sequences—such as PRESS or STEAM—are better suited for quantifying Glx than the difference spectrum derived from the GABA-optimized MEGA-PRESS sequence used in this study (Maddock et al., 2018).

Third, neurometabolite levels were measured within a relatively large voxel to ensure an adequate signal-to-noise ratio. However, as previously discussed, age-related changes and modulation of neurometabolite concentrations might occur on a much smaller spatial scale, which is particularly relevant for the cerebellum, given its high neuronal density and multiple functionally distinct subregions (King et al., 2019; Nettekoven et al., 2024).

Finally, it is important to interpret comparisons between this study and previous research with caution, as substantial methodological differences exist across different MRS studies [e.g., edited vs. unedited MRS, voxel size, and TE–which may be suboptimal in HERMES and HERCULES scans when quantifying multiple metabolites simultaneously (Detcheverry et al., 2023)].

Future research using a higher resolution of MRS—using higher field strengths (7 T) —should address age-related differences in baseline neurometabolite levels and their potential modulation in the context of motor performance and learning on a smaller scale.