Forty young and forty older healthy adults participated in this study (young adults (YA): mean ± SD age = 23.38 ± 3.81 years, range = 18–35 years, 26 females; older adults (OA): mean ± SD age = 69.52 ± 5.42 years, range = 60–85 years, 21 females). Participants were recruited through a list of participants of previous studies, through posters and flyers in the university, libraries and social centers and through posts in social media. The gender distribution in both samples was not significantly different from an expected equal distribution, as indicated by a chi-square goodness-of-fit test (Young: X² = 0.783, p = 0.37, Old: X² = 0.08, p = 0.87; final samples). All participants reported the absence of neurological or psychiatric disorders and were evaluated with the MoCA battery to determine cognitive integrity and to exclude participants with MCI (YA: mean ± SD score = 27.4 ± 3.86, range = 24–30; OA: mean ± SD score = 26.9 ± 2.14, range = 22–30). One older participant scored below the cutoff point (score = 22; cutoff = 24, Islam et al., 2023) and was excluded for that reason. The participants also filled in the Edinburgh Handedness questionnaire. The handedness proportion of participants was representative of that reported for the general population (Willems et al., 2014): 37 young adults and 37 older adults were right-handed; 3 younger adults and 2 older adults were left-handed, and one older adult was ambidextrous. The experiment was approved by the local ethical committee of KU Leuven (protocol number: s61577) and all participants gave written informed consent.

Participants took part in two experimental sessions (scheduled no more than two weeks apart). During the first one, participants filled in self-reports and were tested with the MoCA questionnaire and a battery of Executive Functions (cognitive integrity evaluation). During the second session, participants took part in an MRI protocol. In order to investigate whether GSH is modulated by cognitive effort and sustained cognitive effort, GSH levels were measured in the IFC and in the IPL by MRS at baseline and during task (Rule Switching) performance. These regions were selected according to our meta-analysis results for executive functions paradigms that showed a consistent activation in the IFC and the IPL for the Rule Switching task (Rodríguez-Nieto et al., 2022). Individuals also performed the Rule Switching task during fMRI acquisition prior to the collection of MRS data.

Besides the MoCA test, individuals performed the tests of a Battery of Executive Functions, comprised by: two tests of short-term memory (verbal: Digit Span Forward, spatial: Corsi cubes Forward), two tests of working memory (verbal: Digit Span Backward and spatial: N-back test); two tests of inhibition (Go/No-go task and Stroop) and two tests of cognitive flexibility (Wisconsin Card Sorting Test and Number Letter task).

Prior to the MRI session, participants performed 16 practice trials of the Rule Switching task used to assess cognitive flexibility. Once positioned in the MRI scanner, they performed the Rule Switching task during the fMRI acquisition session (~6 min). Afterwards, the MRS protocol consisted of: (a) a high-resolution anatomical sequence (~6 min; placed after the fMRI to avoid the heat induced by the fMRI sequence to affect the quality of the MRS acquisition), (b) two MRS scans at baseline (resting-state MRS; ~12 min each) with the voxels centered at the IFC and IPL, (c) a short T₁-weighted scan to minimize the effect of intermediate head movement on voxel location by using it for the co-registration of images collected during the MR session (~ 3 min), and (d) two MRS scans with the voxels centered at the IFC and IPL while participants performed the task (task-related functional MRS; ~12 min each) (Figure 1A). The order of the two MRS scans in (b) and (d) was counterbalanced across participants.

In the Rule switching flexibility task (adapted from Sekutowicz et al., 2016; launched in PsychoPy, Peirce, 2007), participants were asked to respond using a button box whether the number shown on the screen was lower or higher than five (left or right button) when the number was preceded (cue) and simultaneously surrounded by a circle and to indicate whether it was odd or even (left or right button) if the number was preceded (cue) and simultaneously surrounded by a square (Figure 1B). In half of the trials, the rule was the same as in the previous trial (Repeat trial), and in the other half, the rule switched (Switch trial) based on a pseudo-randomized sequence. During fMRI, the task consisted of 94 trials (cue alone = 300 ms; cue + stimulus = 1700 ms; ITI = 3,000 or 4,000 ms). During MRS, the task consisted of 132 trials in each of two blocks (cue alone = 400 ms; cue + stimulus = 2,200 ms; ITI = 2000, 3,000, or 4,000 ms).

For this task, we calculated the cognitive flexibility index by subtracting the number of hits in Repeat trials from the number of hits in the Switch trials. This estimation was done with the data from the fMRI block to avoid the effects of learning and fatigue on cognitive flexibility. A higher score indicated a better cognitive flexibility performance. One old participant was excluded due to a very low flexibility score (< 3 SD).

MRI scans were acquired using a Philips Achieva 3 T scanner with a 32-channel head coil. For structural imaging, a high-resolution three-dimensional T₁-weighted image was collected (MPRAGE, TR/TE/TI = 9.6/4.6/900 ms; 0.98 × 0.98 × 1.2 mm³ voxel size; FOV = 192 × 250 × 250 mm³; 160 coronal slices, scan time ∼7 min). Before the task-related MRS, a short T₁-weighted anatomical image was acquired to verify correct alignment for the MRS voxels.

Data were preprocessed and quantified using Gannet v3.3.0 (Edden et al., 2014). Preprocessing steps included line-broadening (3 Hz), eddy-current correction, robust spectral registration (Mikkelsen et al., 2020), signal averaging, residual water filtering and reconstruction of subspectra to generate a GSH-edited spectrum. All preprocessed data were visually inspected to check for spectral artifacts and to assess the spectral data quality. Nine GSH measurements (three OA from the IFC baseline, two OA from the IPL baseline, three OA from the IFC task condition, and one YA for the IFC task) were excluded for a fitting error (defined as the standard deviation of the model residuals divided by the amplitude of the fitted GSH signal) above 3 SD. Two datapoints from GABA+ (GABA plus macromolecules) in the IFC task condition were excluded for the same reason (presenting a fitting error above 3 SD).

The voxel masks for each analysis were co-registered to their respective anatomical image to determine gray matter, white matter and cerebrospinal fluid fractions using SPM12 (Friston et al., 1994). The metabolite levels were quantified in institutional units (i.u.) using tissue water as an internal concentration reference. GABA+ and Glx were tissue-corrected according to the alpha-correction method of Harris et al. (2015), where the WM: GM concentration ratio was assumed to be 1:2. For GSH, the ratio was assumed to be 1:1, as there are currently no studies demonstrating a bulk tissue dependence for MRS-measured GSH. Figure 2 shows the overlap of the voxels from all participants positioned for IFC and IPL, as well as the group-average spectrum for each condition (baseline and condition) and by age group, highlighting consistent voxel placement and excellent spectral quality. A group-average spectrum of GABA+ and Glx from this sample has been displayed in our previous report (Rodríguez-Nieto et al., 2024).

The mixed GLM model revealed an Age effect on the GSH levels from IFC with GSH being higher in older individuals regardless of the condition [F_(1,74) = 13.01, p < 0.001; YA: mean = 1.28, SD = 0.33; OA: mean = 1.73, SD = 0.86]. The same model applied for the IPL, revealed no age differences [Group: F_(1,76) = 0.75, p = 0.39; YA: mean = 1.97, SD = 1.03; OA: mean = 2.14, SD = 0.87].

There were no significant differences on GSH levels as an effect of cognitive performance (Rule Switching task), nor its interaction with age in IFC [Condition: F_(1,74) = 0.02, p = 0.89; Interaction: F_(1,74) = 0.07 p = 0.78] or in IPL [Condition: F_(1,76) = 0.16, p = 0.69; Interaction: F_(1,76) = 0.07, p = 0.78; Figure 3].

The correlations between baseline GSH levels and the MoCA and the tests from the Battery of Executive Function are presented in Table 1. Only two correlations survived multiple comparisons correction. The first one showed that lower baseline levels of GSH in the IPL were related to a better performance in the N-back task in young adults (Figure 4A). The second correlation showed that higher baseline levels of GSH in the IPL were related to a better performance in the Digit Span task in older adults (Figure 4B).

The GLM revealed order effects in both regions [IFC: R² = 0.18, F_(1,77) = 4.59, p = 0.035; IPL: R² = 0.27, F_(1,79) = 23.96, p < 0.001]. This implied that the GSH levels were higher during the second task block in IFC but lower when measured in the second task block in IPL in the full sample (Figures 5A,B; Table 2). Furthermore, the order effect showed a significant interaction with age in the IPL but not in the IFC [IFC: F_(1,77) = 0.13, p = 0.71; IPL: F_(1,79) = 4.52, p < 0.04]. In particular, the sustained effort had a stronger influence (accompanied by a steeper decrease) on GSH levels in IPL in younger individuals (Figure 5B; Table 2).

To assess whether this effect would simply be due to a natural fluctuation in time, we performed similar analyses for the baseline condition, finding no significant order effects [IFC: R² = 0.11, F_(1,79) = 0.75, p = 0.39; IPL: R² = 0.04, F_(1,78) = 1.21, p = 0.28] nor an interaction with Age [IFC: F_(1,79) = 0.01, p = 0.96; IPL: R² = 0.05, F_(1,78) = 0.92, p = 0.34; Supplementary Table S1; Supplementary Figure S1]. In addition, for complementary purposes, and to explore how this pattern compares to other metabolites, we performed similar analyses for GABA+ and Glx. In sum, GABA+ levels in the IPL of the full sample were lower in the second block of baseline but higher during the second block of the task, whereas Glx levels in the IPL of young individuals were lower during the second block of baseline (Supplementary Tables S2–S4; Supplementary Figures S2, S3).

In young adults, the regression models showed that the modulation of GSH in both regions predicted cognitive performance, showing that an increase in GSH was accompanied by a better performance in the Rule Switching task (R² = 0.39, F_(31,2) = 8.21, p = 0.002; ΔGSH IFC β = 0.35, t = 2.31, p = 0.03; ΔGSH IPL β = 0.51, t = 3.32, p = 0.003; Figure 6). In contrast, none of these variables predicted cognitive performance in the older adults group (R² = 0.065, F_(35,2) = 0.83, p = 0.44; ΔGSH IFC β = −0.18, t = −1.07, p = 0.28; ΔGSH IPL β = 0.11 t = 0.58, p = 0.56).

These relationships remained significant after controlling for local levels of GABA+ modulation (see section on the relationship of GSH with GABA+ and Glx; IFC r = 0.35, p = 0.005; IPL r = 0.52, p = 0.004).

The relationship between task-induced GSH modulation and the BOLD response (within the respective voxel) during cognitive effort was not significant in either group (Table 3). The whole-brain BOLD response is displayed in Figure 7 for the full sample (no age differences were observed in the Switch>ITI contrast).

The graph theory analyses revealed that GSH in IPL correlated negatively with GABA+ during rest in both age groups. GSH showed a positive correlation with Glx in the IPL only in older adults during the task performance. In addition, the GABA+ modulation (change from baseline to task performance) correlated with the GSH modulation in IFC in young adults, implying a co-variation of both metabolites as a product of cognitive performance (Figure 8). The correlation matrices for these analyses and the comparison of significant correlations between young and older adults can be found in Supplementary Tables S7–S11.

GSH levels were higher in older adults in the IFC, but no age differences were observed in the IPL. This observation is consistent with the suggestion that the age-GSH relationship may be brain-region dependent (Detcheverry et al., 2023). This finding is also consistent with one MRS and one post-mortem study that showed higher GSH levels in frontal cortices in older adults (Hupfeld et al., 2021; Tong et al., 2016). As protein oxidative damage and decline in oxidative stress homeostasis occur in the frontal lobe during aging (Cabré et al., 2017) and the frontal lobes exhibit the greatest age effects in the brain (e.g., gray matter reduction and volume loss; Raz et al., 1997; Salat et al., 2004; Jernigan et al., 2001), it is possible that higher GSH frontal levels emerge as an adaptive response.

We did not observe a relationship between GSH levels in either region and general cognitive integrity as measured by the MoCA. However, higher GSH levels in the IPL were related to worse working memory (n-back) in young individuals but to higher short-term memory (Digit Span) in older individuals. The latter finding mildly resembles the study from Hu et al. (2024), who reported a positive correlation between GSH levels in the posterior cingulate cortex (but not from the anterior cingulate cortex) and performance in a visuospatial memory task in a sample of individuals 20–70 years old; although the difference in age range of the sample and the brain region studied makes difficult a direct comparison. The IPL processes numerical and visuospatial information (Liu et al., 2021), skills being tapped by the N-back and Digit Span tasks. The GSH-cognition relationship with opposite patterns in both age groups suggests that GSH levels (in IPL) may be indicative of different underlying processes in young and older adults. Whereas higher levels of GSH in the IPL in younger adults may be indicative of excessive oxidative stress (impairing cognitive processing), in older adults higher GSH levels in the IPL may be indicative of a healthy adaptive antioxidant response. The reason why the relationship between GSH and cognitive performance was specific to the IPL remains to be elucidated.

The cognitive effort (or task performance) per se did not have any significant effect on GSH levels in either region or either age group. Nonetheless, sustained cognitive effort affected GSH levels, and, once more, this effect was region dependent. In the full sample, GSH in the IFC was higher during sustained cognitive effort. However, in the IPL, GSH levels were lower during the same condition (sustained cognitive effort), with this effect being stronger in young individuals. Two key yet distinct phenomena may play a role during sustained cognitive effort: automatization and/or fatigue processes. Whereas automatization implies that a particular region requires fewer neurometabolic (but more finely tuned) resources, fatigue implies a continuous effort accompanied by a sustained use of resources or even resource depletion. Such continuous cognitive effort, accompanied by a sustained metabolic activity, may eventually induce endogenous oxidative stress (through a sustained generation of ROS during energy production; see Grimm and Eckert, 2017) and evoke a consequent GSH modulation. We had previously hypothesized that the IFC may be more active in the initial phases of rule switching, whereas the IPL may be continuously engaged (Rodríguez-Nieto et al., 2024). Under this reasoning, lower levels of GSH in the IPL during the second block could be interpreted as a depletion of available GSH, whereas increased levels of GSH in the IFC could be a rebuttal effect after GSH consumption. These interpretations are speculative, and whether the differences that we observed are due to differential resource demands of the IFC and IPL at different points or to different default metabolic properties (perhaps derived from a different cytoarchitecture), remains to be studied.

In young adults, an increase of GSH in IFC and IPL (from resting state MRS to fMRS) predicted a better cognitive performance. The capacity to upregulate GSH levels seems, therefore, important in the physiology sustaining cognitive control. Whereas the literature typically reports the role of GSH as an antioxidant, the results of the present work appear to tentatively show that GSH modulation is a consequence of sustained cognitive effort (previous section) and that it is associated with the quality of cognitive performance, at least in young adults (present section). We cannot make inferences about the direction or mechanisms underlying the relationship between GSH and quality of performance, but we may hypothesize that an increase in GSH could potentially relate to a reduction of oxidative stress produced by a higher neuronal/physiological metabolism that is required for successful cognitive performance. Alternatively, the balance produced by the GSH oxidative action may be crucial to preserve an (oxidant-free) proper environment for better metabolism and better cognitive performance. Furthermore, these findings reveal that baseline GSH levels and the modulatory capacity of GSH show different properties of the system, as higher baseline GSH in IPL in young adults was related to worse cognitive performance (working memory). Therefore, in contrast to baseline GSH in IPL, a higher regulatory capacity of GSH (i.e. GSH modulation as result of task performance) in young adults showed to be favorable to the organism adaptation.

In older adults, the absence of a significant statistical correlation between GSH modulation and cognitive performance could possibly be masked by certain neurodegenerative processes affecting cognitive performance.

Contrary to what we expected, we did not find a significant relationship between task-induced GSH modulation and the BOLD response in either region or group. It is possible that reactive stress may emerge during sustained cognitive effort, whereas the present protocol only measured the BOLD response during a short period of cognitive effort. Furthermore, GSH protects against not only reactive oxygen but also nitrogen and sulfur species (Lushchak, 2012). As such, these other GSH functions could possibly mask a relationship between oxygen consumption and GSH modulation.

GSH showed a relationship mainly with GABA+, and the significance and direction of this relationship were dependent on age, region, and cognitive state. In young individuals, GABA+ and GSH in the IFC were not only positively correlated during the task but also co-varied (showing a parallel modulation), suggesting that they may participate in the same process. In the same sample but in the IPL region, GABA+ levels co-varied with Glx levels. As GABA and glutamate hold opposite effects modulating the neural activity (either decreasing or increasing excitability), it may be possible that the relation of these neurometabolites with GSH derives from the energy supply needed to hold either inhibitory or excitatory connections and not from the energy of neural activity itself. Both, excitatory and inhibitory connectivity require energy and consequently may be considered to produce ROS, which may be underlying the parallel regulation of GSH.

These explanations are speculative in nature and future research may dig deeper in the origin of the relationship among these neurometabolites.

A negative GSH–GABA+ relationship in IPL was observed in both groups at rest. During the task, young individuals showed the same negative correlation, but older individuals showed a positive relation with Glx instead, showing a possible age-related adaptation such that, in IPL, higher levels of GSH are associated with a lower inhibitory tone (lower GABA+ or higher Glx). However, there was no GSH–GABA+ co-variation, possibly suggesting that these metabolites hold an indirect rather than a direct relationship in the IPL.

These opposite findings regarding the GSH–GABA+ relationship in the IFC and IPL suggest that the GSH function may differ according to the region, age and cognitive state of the individual.

Finally, it is important to consider that next to the oxidative role of GSH, this metabolite has also shown to be a signaling agent leading to the release of GABA in retinal neurons to endorse cell protection (Freitas et al., 2017). In addition, GSH is synthesized from glutamate, cysteine and glicine (Bottino et al., 2021). These relationships could partially account for our observations, but given the temporal and spatial disparity between cell biological and MRS observations, it remains difficult to draw firm conclusions.

One limitation of this study is the temporal disparity between the collection of fMRI and fMRS sequences. The metabolic activity during both time frames may be different due to learning, habituation and fatigue effects. This limitation could be addressed with a simultaneous MRS-fMRI acquisition (e.g., Craven et al., 2024); although to our knowledge no study has explored this approach with the measurement of GSH.

In our protocol there was a period of approximately nine minutes between the fMRI and the first resting state MRS block. Although we consider that this period could have been enough for the neurometabolic activity (occurring during the fMRI block) to return to baseline, there are no definitive parameters regarding the time that neurometabolites return to baseline levels after cognitive effort. As our study did not show neurometabolic changes in GSH (nor in GABA or Glx; Rodríguez-Nieto et al., 2024) after one block of cognitive effort, and, taking into account our counterbalanced design, it is unlikely that any possible carry-over effects impacted the present results.

It is also important to mention that in vivo ¹H MRS cannot detect the oxidized form of GSH (glutathione disulfide (GSSG)) or measure the GSH/GSSG ratio (Brix et al., 2019; Satoh and Yoshioka, 2006), which is considered a more sensitive indicator of oxidative stress than GSH alone. Furthermore, the GSH levels measured by MRS do not strongly correlate with peripheral GSH levels measured through serum (Mischley et al., 2016). Therefore, our conclusions are limited only to the relationships ascribed to the reduced form of glutathione (i.e., GSH) and to the brain regions that we have specified.

Finally, although one of our exclusion criteria was the presence of psychiatric and neurological diseases, we did not screen participants for their intake of medication. We advise performing this screening in future studies.