In Vivo Brain Glutathione is Higher in Older Age and Correlates with Mobility

Abstract

Brain markers of oxidative damage increase with advancing age. In response, brain antioxidant levels may also increase with age, although this has not been well investigated. Here, we used edited magnetic resonance spectroscopy to quantify endogenous levels of glutathione (GSH, one of the most abundant brain antioxidants) in 37 young [mean: 21.8 (2.5) years; 19 female] and 23 older adults [mean: 72.8 (8.9) years; 19 female]. Accounting for age-related atrophy, we identified higher frontal and sensorimotor GSH levels for the older compared with the younger adults. For the older adults only, higher sensorimotor (but not frontal) GSH was correlated with poorer balance and gait. This suggests a regionally specific relationship between higher brain oxidative stress levels and motor performance declines with age. We suggest these findings reflect an upregulation of GSH in response to increasing brain oxidative stress with normal aging. Together, these results provide insight into age differences in brain antioxidant levels and implications for motor function.

aging, balance, edited magnetic resonance spectroscopy (MRS), gait, glutathione (GSH), Hadamard Encoding and Reconstruction of MEGA-Edited Spectroscopy (HERMES)

Introduction

The role of oxidative stress in brain aging has been studied since the emergence of the free radical theory of aging. This theory posits that the cumulative result of a lifetime of oxidative insult is diminished tissue functioning and the aging phenotype (Harman 1955). Although evidence exists both for and against the free radical theory of aging, the literature largely agrees that markers of brain oxidative damage increase with advancing age (for review, see Chakrabarti et al. 2011). The brain consumes 20% of the body’s total oxygen uptake, despite accounting for only 2% of the body’s total weight (Quastel and Wheatley 1932; Hyder et al. 2013). This high rate of oxygen consumption, along with high levels of oxidizable iron molecules and polyunsaturated fats, increases the propensity of the brain to form reactive oxygen species (ROS). ROS production is a natural phenomenon that contributes to cell signaling. Excessive ROS production can lead to oxidative damage and requires detoxification of ROS molecules by antioxidant sources to prevent oxidative stress. Therefore, it is important to understand whether antioxidant levels change in the brain with aging and whether these changes relate to declines in cognition and motor control.

Glutathione (GSH) is one of the most abundant antioxidant sources in the central nervous system and plays a key role in the maintenance of redox homeostasis (Rice et al. 2002). Within the brain, GSH abundance appears to vary by cell type (Raps et al. 1989; Huang and Philbert 1995; Langeveld et al. 1996; Rice and Russo-Menna 1997) and brain region (Perry et al. 1971; Calabrese et al. 2002; Srinivasan et al. 2010; Nezhad et al. 2017). For detailed reviews of GSH biochemical characteristics, functions, and locations, see Dringen (2000) Rae and Williams (2017), and Dwivedi et al. (2020). Although several studies have measured age differences in cortical GSH, current understanding of such changes remains equivocal. Rodent model studies have suggested that GSH decreases with age (Chen et al. 1989; Sasaki et al. 2001; Liu 2002), but others have found no changes (Hussain et al. 1995; Asuncion et al. 1996). Across different brain regions, postmortem human work has reported conflicting findings, including no age-related differences in brain GSH levels from infancy (1 day) to older adulthood (99 years old) (Venkateshappa et al. 2012; Tong et al. 2016), lower GSH levels across the lifespan (0 day–80 years) (Venkateshappa et al. 2012), and unchanged or increased GSH levels across adulthood (23–99 years) (Tong et al. 2016). Of note, these previous attempts to measure GSH levels with aging have been hampered by a lack of noninvasive procedures; furthermore, measurements in postmortem conditions are subject to GSH breakdown (Perry et al. 1981), complicating interpretations and comparison with in vivo GSH levels.

Advances in spectral editing make it possible to remove overlapping signals of more concentrated metabolites and selectively resolve GSH with magnetic resonance spectroscopy (MRS) (Terpstra et al. 2003). Only one study to date has used edited MRS to compare GSH levels between young and older adults (Emir et al. 2011). This study scanned the occipital cortex and reported that GSH levels were 30% lower for older compared with younger adults (Emir et al. 2011). In the present study, we used Hadamard Encoding and Reconstruction of MEGA-Edited Spectroscopy (HERMES) (Saleh et al. 2016, 2019) to examine age differences in GSH levels in the frontal and sensorimotor cortices, brain regions involved in cognitive function and mobility, respectively. One previous study has used HERMES to investigate age effects on brain GSH levels; this study focused on early childhood development and found no correlation between frontal GSH and age among children 5–14 years old (Saleh et al. 2020). In addition, of note, as a recent review (Cleeland et al. 2019) discusses, few studies have explored age-related changes in neurometabolite levels in the sensorimotor cortex, and no previous studies have characterized age differences in GSH levels within the sensorimotor cortex.

Although our current understanding of how aging affects brain GSH levels is limited, some evidence suggests that cortical GSH may be associated with cognitive and sensorimotor function. Normal aging results in cognitive decrement (Anstey and Low 2004), as well as widespread motor decline, including difficulties with fine motor control (Seidler and Stelmach 1995), balance (Downs et al. 2014), and walking (Rantakokko et al. 2013). Past work has found lower brain GSH levels in patients with mild cognitive impairment (MCI) and Alzheimer’s disease (AD) compared with healthy aging (Mandal et al. 2012, 2015). Lower levels of brain GSH in the frontal cortex (Mandal et al. 2012, 2015), parietal cortex (Oeltzschner et al. 2019), and hippocampus (Mandal et al. 2015) have been associated with a larger degree of cognitive impairment in MCI and AD. Despite this, the limited work in aging has not found relationships between MRS-measured brain GSH levels and cognitive status (Emir et al. 2011; Chiang et al. 2017). Moreover, relationships between motor function and brain GSH levels have not yet been tested for older adults, although there is some support for a relationship between GSH and motor function, given that GSH levels are altered in various movement disorders. For instance, MRS-measured GSH levels are decreased in multiple sclerosis (motor cortex, Srinivasan et al. 2010, frontal cortex, Choi et al. 2011), amyotrophic lateral sclerosis (motor cortex, Weiduschat et al. 2014; Weerasekera et al. 2019), and spinocerebellar ataxia (cerebellum, Doss et al. 2015). Although these studies did not report relationships between brain GSH levels and motor performance or disease severity, past rodent work has found that transient basal ganglia GSH depletion results in pronounced sensorimotor impairments (Díaz-Hung et al. 2014). Taken together, it is plausible that alterations in regional brain GSH levels may affect cognitive and sensorimotor function, although it is unclear whether this relationship would be evident in aging or only in pathological conditions. In the present work, we tested associations between brain GSH levels and performance. We predicted regionally specific relationships in which frontal GSH levels would be associated with cognitive performance, and sensorimotor GSH levels would be associated with motor performance.

Overall, it remains unclear how human brain GSH levels alter with aging and whether brain GSH is associated with cognitive or motor function. Based on the limited in vivo human work (Emir et al. 2011) and the larger body of animal and postmortem human studies, it is plausible that brain GSH levels would be lower in older adults. If GSH levels are lower in older age, this may indicate greater oxidative stress burden, thereby exhausting brain antioxidant capacity. However, if GSH levels are higher for older adults, this could suggest an upregulation of GSH in response to increased oxidative stress. For instance, there is evidence that mild stress increases brain GSH levels; this upregulation of GSH is thought to provide protection against more severe oxidative stress (for review, see Maher 2005). Thus, it is possible that aging could be associated with higher brain GSH levels as a response to generalized aging processes.

The aims of the present study included: (1) to determine whether there are age differences in in vivo MRS-measured brain GSH levels in the frontal and sensorimotor cortices; (2) to characterize regional differences in brain GSH levels; and (3) to assess the relationships between brain GSH levels and cognitive and motor function.

Materials and Methods

The University of Florida’s Institutional Review Board provided ethical approval for the study, and all participants provided their written informed consent at the first testing session.

Participants

We recruited 37 young and 23 older adults from the Gainesville, FL community. Due to the coronavirus disease 2019 global pandemic, data collection was terminated slightly early before collection of older adult participants was complete. Exclusion criteria included history of any neurologic condition (e.g., stroke, Parkinson’s disease, seizures, or a concussion in the last 6 months) or psychiatric condition (e.g., active depression or bipolar disorder). We also excluded those who self-reported smoking, consuming more than two alcoholic drinks per day on average, or a history of treatment for alcoholism. All participants were screened for magnetic resonance imaging (MRI) eligibility; we excluded those with any contraindications (e.g., implanted metal, claustrophobia, or pregnancy). All participants were right handed and self-reported an ability to walk unassisted for at least 10 min and to stand for at least 30 s with their eyes closed. Participants disclosed all current prescribed and over-the-counter medications.

Prior to enrollment, we screened participants for suspected cognitive impairment over the phone using the Telephone Interview for Cognitive Status (de Jager et al. 2003). We excluded those who scored <21 of 39 points; this is equivalent to scoring <25 points on the Mini-Mental State Exam and indicates probable cognitive impairment (de Jager et al. 2003). At the first testing session, participants were rescreened for cognitive impairment using the Montreal Cognitive Assessment (MoCA) (Nasreddine et al. 2005); we excluded those who scored <23 of 30 points (Carson et al. 2018).

Testing Sessions

Prior to the first session, we collected basic demographic information, including age, sex, years of education, and medical history, as well as information regarding self-reported exercise, handedness, and footedness. We also collected basic anthropometric information, such as height, weight, and leg length.

Participants then completed behavioral testing, followed by an MRI session ~1 week later (Fig. 1). For 24 h prior to each session, participants refrained from consuming alcohol, nicotine, or any drugs other than the medications they previously disclosed. At the start of each session, participants completed the Stanford Sleepiness Questionnaire, which asks for the number of hours slept the previous night and for a rating of current sleepiness (Hoddes et al. 1972).

Figure 1

Methods overview. Left: During Session 1, participants first completed the MoCA test of cognition. Participants were then instrumented with six IMUs (sensor placements are pictured) and completed a 30-s balance task in which they stood as still as possible with their eyes open, gazing at a blank white wall. Next, participants completed a 4-min walk at a self-selected speed across a 32-foot room. Right: During Session 2, participants completed an MRI protocol, which included a T₁-weighted anatomical scan and two edited MRS scans to quantify neurometabolites in a frontal and sensorimotor voxel.

Session 1: Behavioral Testing

MoCA Test

Participants first completed the MoCA (Nasreddine et al. 2005). We added one point to the scores of participants with ≤12 years of education (Nasreddine et al. 2005).

Balance Task

Participants completed the four-part Modified Clinical Test of Sensory Interaction in Balance, while instrumented with six Opal inertial measurement units (IMUs; v2; APDM Wearable Technologies Inc.). IMUs were placed on the feet, wrists, around the waist at the level of the lumbar spine, and across the torso at the level of the sternal angle (Fig. 1). Participants stood as still as possible facing a blank white wall for four 30-s trials: (1) eyes open; (2) eyes closed; (3) eyes open, foam surface; and (4) eyes closed, foam surface. Here, we report only on performance during the eyes open condition. We elected to use only the eyes open condition because this condition is most relevant to daily life activities. Furthermore, previous work has reported age differences in postural sway during quiet stance with eyes open (Maki et al. 1990; Baloh et al. 1994), and eyes open postural sway has been shown to predict falls among older adults (Fernie et al. 1982; Maki et al. 1990).

Inertial data were recorded using MobilityLab software (version 2; APDM Wearable Technologies Inc.). After each trial, MobilityLab calculated 25 spatiotemporal features of postural sway (Supplementary Table A1) using the validated iSway algorithm (Mancini et al. 2012). To condense these variables into several summary metrics, we ran an exploratory factor analysis (Supplementary Material A). This procedure yielded four factors: anterior/posterior (A/P) sway path, A/P sway speed and variability, medial/lateral (M/L) sway path, and M/L sway speed and variability. We then calculated a balance composite score for each factor to use in subsequent analyses.

Four-Minute Walk

Although instrumented with the IMUs, participants also completed an overground walk. Participants walked back and forth across a 32-foot room for 4 min at whichever pace they considered to be their “normal walking speed.” Participants were instructed to refrain from talking, to keep their arms swinging freely at their sides, and to keep their head up and gaze straight ahead. Each time they reached the end of the room, they completed a 180° turn and walked the length of the room again.

After the session, the MobilityLab software calculated 14 spatiotemporal gait variables of interest (Supplementary Table B1). The algorithm for calculating these metrics has been validated through comparison to force plate and motion capture data [see internal validation by MobilityLab (https://support.apdm.com/hc/en-us/articles/360000177066-How-are-Mobility-Lab-s-algorithms-validated-) and Washabaugh et al. (2017)]. To obtain summary metrics of gait, we extracted one variable from each of the four gait domains described by Hollman et al. (2011): gait rhythm [cadence (steps/min)], gait phase [stance (% gait cycle)], gait pace (composite score), and gait variability (composite score). See Supplementary Material B for further details regarding the selection and calculation of these summary metrics.

Session 2: MRI Scan

Anatomical Acquisition

MRI was conducted using a Siemens MAGNETOM Prisma 3 T scanner (Siemens Healthcare) using a 64-channel head coil. We first collected a 3D T₁-weighted anatomical image using a magnetization-prepared rapid gradient-echo sequence for MRS voxel placement and tissue segmentation/correction. The parameters for this anatomical image were as follows: repetition time (TR) = 2000 ms, echo time (TE) = 3.06 ms, flip angle = 8°, field of view = 256 × 256 mm², slice thickness = 0.8 mm, 208 slices, voxel size = 0.8 mm³.

MRS Acquisition

In the following sections and in Table 2, we describe all parameters suggested by the MRS quality assessment tool (Peek et al. 2020). We used the universal HERMES sequence to simultaneously detect GSH, γ-aminobutyric acid (GABA), and glutamate + glutamine (Glx) (Saleh et al. 2016, 2019). HERMES is a J-difference editing method that allows for multiple MEGA-PRESS (Mescher et al. 1998) experiments to be conducted simultaneously. Briefly, the HERMES sequence includes four subexperiments containing (A) a dual-lobe editing pulse, ON_GABA = 1.9 ppm, ON_GSH = 4.56 ppm, and three single-lobe editing pulses: (B) ON_GABA = 1.9 ppm, (C) ON_GSH = 4.56 ppm, and (D) OFF_GABA, OFF_GSH = 7.5 ppm. The Hadamard combination A–B + C–D derives GSH-edited spectra, and A + B–C–D derives GABA+- and Glx-edited spectra. Additional HERMES parameters included total acquisition time = 10:48 min, TR = 2000 ms, TE = 80 ms, 20-ms editing pulse duration, averages = 320, 2048 data points, 2 kHz spectral width, and variable power and optimized relaxation delays water suppression. Shimming was performed using the Siemens interactive shim tool and FAST(EST) MAP (Gruetter 1993).

We collected data from two 3 × 3 × 3 cm³ voxels in the medial frontal cortex and bilateral sensorimotor cortex (Fig. 2). We placed the frontal voxel superior to the genu of the corpus callosum on the midsagittal slice. We placed the sensorimotor voxel to align with the lower limb primary sensorimotor cortex. We aligned the center of this voxel with the posterior portion of the motor hand knob in the axial view, then centered the voxel on the midline of the brain, and placed the voxel as superior as possible while still remaining on brain tissue.

Figure 2

MRS voxel placement. (A) Top: Placement of the medial frontal voxel: superior to the genu of the corpus callosum on the midsagittal line. The voxel shown is every participant’s 3 × 3 × 3 cm³ voxel, normalized to standard space and overlaid onto a template brain. Lighter colors indicate areas of more overlap across participants. Bottom: Plot created using PaperPlot.m showing all participants’ spectra overlaid (black) and the GSH model fit (gray) for the frontal voxel. (B) Top: Placement of the sensorimotor voxel: centered with the posterior portion of the motor hand knobs in the axial view (motor hand knobs are outlined in white), then centered on the midline of the brain and placed as superior as possible. The voxel shown is every participant’s 3 × 3 × 3 cm³ voxel, normalized to standard space and overlaid onto a template brain. Lighter colors indicate areas of more overlap across participants. Bottom: Plot created using PaperPlot.m showing all participants’ spectra overlaid (black) and the GSH model fit (gray) for the sensorimotor voxel.

MRS Processing

We analyzed MRS data using Gannet (version 3.1.5) (Edden et al. 2014) in MATLAB (R2019b). First, we ran the GannetLoad.m and GannetFit.m functions, which include (1) coil combination using generalized least squares (An et al. 2013); (2) estimation of the B₀ drift using the creatine signal at 3 ppm; (3) robust spectral registration of each transient to a weighted average reference to minimize subtraction artifacts (Mikkelsen et al. 2020); (4) Hadamard combination of the fully processed HERMES subspectra to generate GSH- and GABA+-edited difference spectra; (5) application of the Hankel singular decomposition water filtering method to remove the residual water signal (Barkhuijsen et al. 1987); and (6) implementation of a weighted nonlinear regression to model the two difference-edited signals; here, the neighboring coedited signals were downweighted to reduce their impact on modeling errors. The GSH-edited spectrum was modeled between 2.25 and 3.5 ppm using a Gaussian to model the GSH signal at 2.95 ppm, four Gaussians to model the coedited aspartyl signals at 2.55 ppm, and a nonlinear baseline.

We used GannetCoRegister.m to create a binary mask of the MRS voxels and register these masks to the T₁-weighted structural image. We then used the Computational Anatomy Toolbox 12 (version 1450) (Gaser and Dahnke 2016) to segment each participant’s T₁-weighted image. We implemented GannetSegment.m to determine voxel tissue fractions (i.e., fractions of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) in the MRS voxel. Using GannetQuantify.m, we computed tissue-corrected values for all statistical analyses (GSH.ConcIU_TissCorr). This adjusts GSH estimates to account for heterogeneous tissue composition in the voxel, as well as for the relative visibility and transverse and longitudinal relaxation times of water in GM, WM, and CSF (Gasparovic et al. 2006; Harris et al. 2015). Metabolite levels, as well as reference signals, differ between GM, WM, and CSF (Harris et al. 2015). Tissue correction is particularly relevant for aging populations (Porges et al. 2017b). For instance, if older adults have less GM due to age-related atrophy in a voxel compared with young adults, the older adults will also present with less metabolite concentration in that voxel. Tissue correction thus permits assessment of whether there are age differences in neurometabolite levels in the tissue that remains in the voxel. Throughout the present work, we report tissue-corrected GSH levels referenced to water (in institutional units).

MRS Exclusions

See Supplementary Table G1 for details on exclusions of MRS datasets. We excluded MRS datasets if the GSH fit error (i.e., GSH.FitError_W) was >20% or if robust spectral registration failed for that dataset. We selected 20% for several reasons: (1) datasets with fit errors <20% passed acceptable visual inspection and (2) fit errors ≥20% were >2.5 standard deviations above the group mean (i.e., >97th percentile). Thus, similar to Saleh et al. (2020), we selected a threshold value for data rejection. Of note, we did not exclude one older adult for whom we used a 20-channel head coil instead of a 64-channel coil due to his large head size. The uncorrected and corrected GSH levels for this individual fell within the range of that of the other older participants (see Supplementary Fig. G1 for details).

Statistical Analyses

We conducted all statistical analyses using R (version 4.0.0) (R Core Team 2013).

Demographic Variables

First, we compared demographic, physical fitness, sleep, and testing timeline variables between the age groups. We tested the parametric t-test assumptions of normality within each group (using shapiro.test) and homogeneity of variances between the groups (using leveneTest in the car package (Fox and Weisberg 2018)). The majority of these variables did not meet parametric t-test assumptions, so we used wilcox.test to conduct nonparametric, independent samples, two-sided Wilcoxon rank-sum tests for age group differences. We report the group medians and interquartile ranges for each demographic variable. We also report nonparametric effect sizes (Rosenthal et al. 1994; Field et al. 2012) (see Supplementary Material C for details on this calculation). To test for differences in the sex distribution within each age group, we conducted a Pearson’s chi-square test using chisq.test.

Age and Voxel Differences in GSH Levels, Bulk Tissue Composition, and MRS Variables

To examine whether GSH levels differed by age group or brain region, we used a linear mixed model approach (lme, Pinheiro et al. 2020). We entered age group, voxel (i.e., frontal or sensorimotor), and the age group*voxel interaction as predictors, and included a random intercept for each subject. We used this same approach to test for age and brain region differences in bulk tissue composition (i.e., voxel GM, WM, and CSF fractions) and MRS metrics (e.g., model fit error).

GSH Relationships with Motor and Cognitive Performance

For the behavioral metrics, we used linear models (lm) to test the relationship between GSH levels and performance for both voxels and age groups. We ran four models in total for each age group. Models 1 and 2 tested for relationships between balance and GSH levels for the frontal and sensorimotor voxels, respectively. These models included as predictors each of the four balance metrics, plus sex and leg length as covariates of no interest (as these affect postural sway, Kim et al. 2010). Models 3 and 4 similarly tested for relationships between gait and GSH levels and included as predictors each of the four gait metrics, plus sex and leg length as covariates of no interest (as these affect gait, Samson et al. 2001; Ko et al. 2011; Kobayashi et al. 2016). Model 5 tested for a relationship between MoCA score and GSH, controlling for sex and years of education (Malek-Ahmadi et al. 2015).

For the significant GSH–performance relationships, we also computed the partial correlation (i.e., the correlation controlling for the covariates listed above) by correlating the residuals from (1) regressing each of the covariates onto the significant performance variable and (2) regressing each of the covariates (but not the significant performance variable) onto GSH concentration. Finally, we used a Fisher r-to-Z transformation to test for age differences in the strength of the partial correlation.

Furthermore, for each model that indicated a significant relationship between GSH levels and behavior, we ran an additional linear model with the significant behavioral measure as the outcome variable and GSH, Glx, and GABA as predictors, controlling for the above covariates. This was to provide further support for the specificity of the relationship between GSH levels and motor performance; that is, we hypothesized that these excitatory and inhibitory neurometabolites would not relate to behavior, and that including these would not influence the significant relationship between GSH levels and motor performance.

Corrections for Multiple Comparisons

We corrected the P values for each model using p.adjust with method = “bh” to apply the Benjamini–Hochberg false discovery rate (FDR) correction (Benjamini and Hochberg 1995). We present these FDR-corrected P values within the tables, and we use these corrected values for all interpretations of the results.

Results

A total of 37 young and 23 older adults completed cognitive and motor testing, as well as collection of MRS data from voxels placed in the frontal and sensorimotor cortices.

Demographics

There were no significant age differences for most demographic variables, including sex, alcohol use, handedness, or footedness. Importantly, there were also no age differences in the number of days elapsed between the two testing sessions or in the difference in start time for the two sessions. See Table 1 for complete demographic information.

Table 1

Participant demographics, physical characteristics, sleep, and testing timeline

Predictors	Young adult median (IQR)	Older adult median (IQR)	W or χ2	FDR Corrected P	Nonparametric effect sizea
Demographics
Sample size	37b	23b	—	—	—
Age	21.8 (2.5)	72.8 (8.9)	—	—	—
Sex	19 F; 18 M	11 F; 12 M	0.07	0.856	—
Years of education	15.0 (3.0)	16.0 (3.0)	219.0	0.010*	−0.41
Alcohol usec	2 (3.0)	2 (4.0)	480.0	0.586	−0.11
Physical characteristics and fitness
Handedness laterality scored	85.7 (25.0)	100.0 (22.4)	351.0	0.436	−0.15
Footedness laterality scored	100.0 (22.2)	100.0 (133.9)	492.5	0.436	−0.14
Body mass index (BMI)	22.7 (5.6)	26.0 (3.9)	184.5	0.003**	−0.47
Leisure-time physical activitye	46.0 (38.0)	29.0 (21.0)	563.5	0.037*	−0.32
Sleep
Hours of sleep, behavioral session	7.0 (1.5)	7.5 (1.5)	385.0	0.684	−0.08
Sleepiness ratingf, behavioral session	2.0 (1.0)	1.0 (1.0)	594.0	0.018*	−0.36
Hours of sleep, MRI session	7.0 (2.0)	7.0 (1.5)	324.5	0.266	−0.20
Sleepiness ratingf, MRI session	2.0 (1.3)	1.0 (1.0)	585.5	0.018*	−0.37
Testing timeline
Time between behavioral testing and MRI (# of days)	4.0 (7.0)	4.0 (4.5)	415.0	0.878	−0.02
Difference in start time for behavioral testing versus MRI (# of hours)	1.3 (1.5)	1.2 (1.1)	462.5	0.684	−0.07

Note: In the second and third columns, we report the median ± interquartile range (IQR) for each age group in all cases except for sex. For sex, we report the number of males and females in each age group. In the fourth and fifth columns, for all variables except sex, we report the result of a nonparametric two-sample, two-sided Wilcoxon rank-sum test. For sex, we report the result of a Pearson’s chi-square test for differences in the sex distribution within each age group. P values were FDR corrected (Benjamini and Hochberg 1995) across all models included in this table; significant P values are bolded.

^aIn the sixth column, we report the nonparametric effect size as described by Rosenthal et al. (1994) and Field et al. (2012) (see Supplementary Material C).

^bAll subjects (i.e., 37 young and 23 older adults) are included in the comparisons in this table. However, we excluded several individuals from the subsequent analyses involving the MRS and behavioral metrics (see Supplementary Table G1 for details).

^cParticipants self-reported alcohol use on the Alcohol Use Disorders Identification Test (AUDIT) (Piccinelli 1998).

^dWe calculated handedness and footedness laterality scores using the Edinburgh Handedness Inventory (Oldfield 1971) and Waterloo Footedness Questionnaire (Elias et al. 1998).

^eWe assessed physical activity using the Godin Leisure-Time Exercise Questionnaire (Godin and Shephard 1985). Both the young and older adult group medians fell within the “active” range (i.e., scores of ≥24). One older adult did not complete this questionnaire, so n = 22 older adults for this variable.

^fParticipants rated their sleepiness level from 1 to 7 at the start of each session using the Stanford Sleepiness Scale (Hoddes et al. 1972). One younger adult did not complete the sleepiness rating for the MRI session, so n = 36 young adults for this variable.

^*P < 0.05, ^**P < 0.01.

Higher GSH Levels in Older Age

The older adult group exhibited cortical atrophy; across both voxels, older adults had a lower GM fraction and higher CSF fraction compared with the younger adults (Table 2 and Fig. 3).

Table 2

Age and voxel differences in GSH, bulk tissue composition, and MRS variables

Mean (SD)		Predictors	Estimates (SE)	CI	t	FDR Corrected P	Fit
GSH (i.u.)a
Young: 1.55 (0.44)	Old: 1.97 (0.71)	Fixed effects
Frontal: 1.56 (0.44)	SM: 1.85 (0.67)	(Intercept)	1.66 (0.09)	1.47 to 1.84	17.80	<0.001***
		Age group (Old)	0.52 (0.15)	0.21 to 0.83	3.38	0.003**
		Voxel (Frontal)	−0.22 (0.10)	−0.42 to (−0.03)	-2.29	0.035*
		Age group (Old) * Voxel (Frontal)	−0.18 (0.16)	−0.51 to 0.14	-1.12	0.267
		Random effects
		σ2	0.16
		τ00 Participant	0.13
		Model fit
		ICC					0.45
		Marginal R2					0.19
		Conditional R2					0.55
Gray matter (GM) fractionb
Young: 0.50 (0.04)	Old: 0.40 (0.04)	Fixed effects
Frontal: 0.46 (0.05)	SM: 0.47 (0.06)	(Intercept)	0.51 (0.01)	0.49 to 0.52	83.14	<0.001***
		Age group (Old)	−0.10 (0.01)	−0.13 to (−0.08)	-10.43	<0.001***
		Voxel (Frontal)	−0.02 (0.01)	−0.03 to 0.00	-2.56	0.018*
		Age group (Old) * Voxel (Frontal)	0.02 (0.01)	0.00 to 0.04	2.22	0.031*
		Random effects
		σ2	0.00
		τ00 Participant	0.00
		Model fit
		ICC					0.46
		Marginal R2					0.63
		Conditional R2					0.80
White matter (WM) fractionb
Young: 0.38 (0.05)	Old: 0.35 (0.05)	Fixed effects
Frontal: 0.38 (0.06)	SM: 0.36 (0.05)	(Intercept)	0.35 (0.01)	0.34 to 0.37	42.77	<0.001***
		Age group (Old)	0.01 (0.01)	−0.02 to 0.03	0.40	0.688
		Voxel (Frontal)	0.05 (0.01)	0.04 to 0.07	6.03	<0.001***
		Age group (Old) * Voxel (Frontal)	−0.07 (0.01)	−0.10 to (−0.04)	-4.97	<0.001***
		Random effects
		σ2	0.00
		τ00 Participant	0.00
		Model fit
		ICC					0.44
		Marginal R2					0.23
		Conditional R2					0.58
Cerebrospinal fluid (CSF) fractionb
Young: 0.12 (0.04)	Old: 0.25 (0.06)	Fixed effects
Frontal: 0.16 (0.08)	SM: 0.18 (0.07)	(Intercept)	0.14 (0.01)	0.12 to 0.16	16.22	<0.001***
		Age group (Old)	0.10 (0.01)	0.07 to 0.13	6.95	<0.001***
		Voxel (Frontal)	−0.04 (0.01)	−0.05 to (−0.02)	-4.20	<0.001***
		Age group (Old) * Voxel (Frontal)	0.05 (0.01)	0.02 to 0.08	3.38	0.001**
		Random effects
		σ2	0.00
		τ00 Participant	0.00
		Model fit
		ICC					0.49
		Marginal R2					0.61
		Conditional R2					0.80
GSH fit errorc
Young: 6.60 (2.46)	Old: 7.57 (2.94)	Fixed effects
Frontal: 7.63 (3.04)	SM: 6.28 (2.07)	(Intercept)	5.93 (0.45)	5.03 to 6.83	13.19	<0.001***
		Age group (Old)	0.95 (0.74)	−0.54 to 2.44	1.28	0.276
		Voxel (Frontal)	1.34 (0.61)	0.12 to 2.56	2.21	0.063
		Age group (Old) * Voxel (Frontal)	0.04 (1.00)	−1.98 to 2.05	0.04	0.970
		Random effects
		σ2	6.06
		τ00 Participant	0.60
		Model fit
		ICC					0.09
		Marginal R2					0.09
		Conditional R2					0.18
H2O FWHMd
Young: 9.33 (1.12)	Old: 9.79 (1.61)	Fixed effects
Frontal: 10.05 (1.49)	SM: 8.96 (0.87)	(Intercept)	8.92 (0.21)	8.50 to 9.34	42.72	<0.001***
		Age group (Old)	0.10 (0.35)	−0.59 to 0.80	0.29	0.796
		Voxel (Frontal)	0.83 (0.29)	0.25 to 1.42	2.85	0.013*
		Age group (Old) * Voxel (Frontal)	0.71 (0.48)	−0.26 to 1.68	1.47	0.197
		Random effects
		σ2	1.40
		τ00 Participant	0.04
		Model fit
		ICC					0.02
		Marginal R2					0.21
		Conditional R2					0.23
B0 frequency offsete
Young: 0.01 (0.01)	Old: 0.02 (0.01)	Fixed effects
Frontal: 0.01 (0.01)	SM: 0.01 (0.01)	(Intercept)	0.01 (0.002)	0.01 to 0.01	5.50	<0.001***
		Age group (Old)	0.01 (0.003)	0.00 to 0.01	1.63	0.110
		Voxel (Frontal)	−0.01 (0.002)	−0.01 to 0.00	-3.57	<0.001***
		Age group (Old) * Voxel (Frontal)	0.01 (0.003)	0.00 to 0.01	2.02	0.065
		Random effects
		σ2	0.00
		τ00 Participant	0.00
		Model fit
		ICC					0.42
		Marginal R2					0.18
		Conditional R2					0.52

Note: On the left, we report the mean (standard deviation) for each outcome variable, split by age group and by voxel. On the right, we report the results of a linear mixed effects model testing for age group, voxel, and interaction effects for each variable. P values were FDR corrected for each model (Benjamini and Hochberg 1995); significant P values are bolded. Marginal R² values consider only the variance of the fixed effects; the conditional R² values consider both the fixed and random effects. In this table, we include only subjects who had data for both the frontal and sensorimotor voxels (see Supplementary Table G1 for details). SD, standard deviation; SM, sensorimotor; SE, standard error; CI, 95% confidence interval; ICC, intraclass correlation coefficient; i.u., institutional units.

^aHere, we report the Gannet variable GSH.ConcIU_TissCorr, which is the GSH concentration within the voxel, corrected for volume fractions of GM, WM, and CSF in the voxel, as well as the relative visibility of water in the GM, WM, and CSF, and the transverse and longitudinal relaxation times of water in GM, WM, and CSF (Gasparovic et al. 2006).

^bHere, we report the Gannet variables fGM, fWM, and fCSF, which indicate the fraction of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) within the voxel.

^cHere, we report the Gannet variable GSH.FitError_W, which is the root sum square of the standard deviation of model residuals normalized to the model amplitude of the GSH and water signals, expressed as a percentage (Edden et al. 2014). We excluded subjects with fit error >20% prior to running any statistical tests (see Supplementary Table G1).

^dHere, we report the Gannet variable H2O.FWHM, which is the full-width half-maximum of the unsuppressed water signal (Edden et al. 2014).

^eHere, we report the Gannet variable AvgDeltaF0, which is the mean difference between the observed frequency of the residual water signal in the prefrequency-corrected subspectra and the nominal water frequency at 4.68 ppm (Mikkelsen et al. 2017).

^*P < 0.05, ^**P < 0.01, ^***P < 0.001.

Figure 3

Higher CSF fraction in older age. Left: CSF fraction within the frontal (left) and sensorimotor (right) voxels for older (top) and young (bottom) adults. In both voxels, older adults had higher CSF concentrations compared with young adults. Right: CSF fraction (white) within the sensorimotor voxel, shown for a single older (top) and a single younger (bottom) participant. The CSF fraction is overlaid onto each participant’s native space T₁-weighted anatomical image.

Older adults had significantly higher GSH levels across both voxels (Table 2 and Fig. 4). This difference in GSH levels implies that there is an age-related increase in cortical GSH concentration within the tissue that remains in the voxel after accounting for age-related atrophy. Importantly, there was no age difference in GSH fit error, water full width at half maximum, or B₀ frequency offset. Although several factors (e.g., T₂ changes with age, Marjańska et al. 2013, 2017; Deelchand et al. 2020) could have impacted GSH quantification, it is unlikely that such factors would have resulted in higher GSH levels in older age; we elaborate on these considerations in the Discussion.

Higher GSH Levels in Sensorimotor Versus Frontal Cortex

Across both age groups, GSH levels were higher in the sensorimotor voxel compared with the frontal voxel (Table 2 and Fig. 5). The GM and CSF fractions were higher in the sensorimotor voxel, and the WM fraction was higher in the frontal voxel.

GSH Relationships with Motor but Not Cognitive Performance for Older Adults Only

We did not observe relationships between frontal voxel GSH and performance (Table 3) or between GSH and MoCA scores (Supplementary Table E1). However, higher GSH levels within the sensorimotor, but not the frontal voxel, were associated with poorer performance on multiple motor measures for the older adults only.

Figure 4

Higher GSH levels in older age. GSH levels for older (top) and young (bottom) adults in the frontal (left) and sensorimotor (right) voxels. Across both voxels, older adults had higher GSH levels.

Figure 5

Higher GSH levels in sensorimotor versus frontal cortex. Frontal and SM (sensorimotor) GSH levels within each young (left) and older (right) adult. Each line represents one participant. For both age groups, GSH levels (group medians shown in black) were higher in the sensorimotor compared with the frontal voxel.

Table 3

GSH relationships with motor and cognitive performance

Young adults						Older adults
Predictors	Estimates (SE)	CI	t	FDR Corrected P	Fit	Predictors	Estimates (SE)	CI	t	FDR Corrected P	Fit
Model 1: Frontal GSH—Balance
(Intercept)	3.05 (1.43)	0.11 to 5.98	2.13	0.299		(Intercept)	−2.56 (2.59)	−8.21 to 3.09	-0.99	0.400
A/P sway path	−0.09 (0.11)	−0.31 to 0.14	-0.79	0.512		A/P sway path	0.31 (0.15)	−0.03 to 0.64	1.99	0.247
A/P sway speed and variability	−0.03 (0.09)	−0.20 to 0.15	-0.29	0.771		A/P sway speed and variability	−0.37 (0.14)	−0.68 to 0.06	-2.57	0.172
M/L sway path	0.10 (0.09)	−0.09 to 0.29	1.11	0.487		M/L sway path	−0.04 (0.12)	−0.30 to 0.23	-0.31	0.764
M/L sway speed and variability	0.09 (0.09)	−0.10 to 0.28	0.95	0.490		M/L sway speed and variability	0.18 (0.12)	−0.08 to 0.45	1.51	0.277
Sex	−0.25 (0.17)	−0.59 to 0.09	-1.48	0.487		Sex	−0.40 (0.33)	−1.13 to 0.33	-1.20	0.355
Leg length	−0.002 (0.002)	0.00 to 0.00	-1.12	0.487		Leg length	0.01 (0.003)	0.00 to 0.01	1.70	0.268
R2					0.14	R2					0.42
Adjusted R2					-0.05	Adjusted R2					0.13
Model 2: Sensorimotor GSH—Balance
(Intercept)	1.90 (1.87)	−1.94 to 5.74	1.01	0.449		(Intercept)	−4.11 (2.49)	−9.39 to 1.16	-1.65	0.165
A/P sway path	−0.14 (0.13)	−0.40 to 0.12	-1.09	0.449		A/P sway path	0.07 (0.15)	−0.23 to 0.38	0.51	0.619
A/P sway speed and variability	−0.13 (0.10)	−0.34 to 0.09	-1.22	0.449		A/P sway speed and variability	−0.32 (0.15)	−0.63 to 0.01	-2.22	0.097
M/L sway path	−0.01 (0.10)	−0.21 to 0.19	-0.10	0.940		M/L sway path	0.08 (0.13)	−0.20 to 0.36	0.61	0.619
M/L sway speed and variability	0.24 (0.10)	0.03 to 0.45	2.34	0.188		M/L sway speed and variability	0.55 (0.13)	0.27 to 0.84	4.13	0.006**
Sex	−0.23 (0.20)	−0.65 to 0.19	-1.14	0.449		Sex	−0.61 (0.30)	−1.24 to 0.02	-2.04	0.101
Leg length	−0.0001 (0.002)	0.00 to 0.00	-0.08	0.940		Leg length	0.01 (0.003)	0.00 to 0.01	2.63	0.064
R2					0.21	R2					0.64
Adjusted R2					0.04	Adjusted R2					0.50
Model 3: Frontal GSH—Gait
(Intercept)	0.29 (4.84)	−9.63 to 10.22	0.06	0.952		(Intercept)	−12.85 (10.35)	−35.40 to 9.69	-1.24	0.555
Rhythm: Cadence (steps/min)	0.002 (0.02)	−0.04 to 0.04	0.10	0.952		Rhythm: Cadence (steps/min)	0.04 (0.02)	0.00 to 0.08	2.24	0.156
Phase: Stance (%GC)	0.04 (0.06)	−0.09 to 0.16	0.61	0.789		Phase: Stance (%GC)	0.14 (0.15)	−0.20 to 0.48	0.90	0.586
Pace: Composite score	−0.03 (0.05)	−0.14 to 0.08	-0.59	0.789		Pace: Composite score	−0.01 (0.13)	−0.30 to 0.28	-0.10	0.924
Variability: Composite score	−0.05 (0.03)	−0.11 to 0.01	-1.69	0.356		Variability: Composite score	0.17 (0.07)	0.02 to 0.31	2.52	0.156
Sex	−0.31 (0.17)	−0.67 to 0.04	-1.80	0.356		Sex	0.23 (0.28)	−0.37 to 0.84	0.84	0.586
Leg length	−0.001 (0.002)	0.00–0.00	-0.75	0.789		Leg length	0.002 (0.003)	0.00–0.01	0.65	0.613
R2					0.23	R2					0.44
Adjusted R2					0.06	Adjusted R2					0.15
Model 4: Sensorimotor GSH—Gait
(Intercept)	4.74 (6.72)	−9.02 to 18.50	0.71	0.923		(Intercept)	−17.23 (9.93)	−38.28 to 3.83	-1.73	0.239
Rhythm: Cadence (steps/min)	−0.01 (0.03)	−0.06 to 0.04	-0.40	0.923		Rhythm: Cadence (steps/min)	0.03 (0.02)	−0.01 to 0.08	1.48	0.239
Phase: Stance (%GC)	−0.02 (0.08)	−0.18 to 0.15	-0.19	0.923		Phase: Stance (%GC)	0.19 (0.13)	−0.09 to 0.47	1.43	0.239
Pace: Composite score	0.01 (0.07)	−0.14 to 0.15	0.10	0.923		Pace: Composite score	0.13 (0.11)	−0.10 to 0.36	1.18	0.299
Variability: Composite score	−0.03 (0.04)	−0.12 to 0.05	-0.86	0.923		Variability: Composite score	0.28 (0.08)	0.12–0.45	3.71	0.013*
Sex	−0.22 (0.23)	−0.70 to 0.25	-0.97	0.923		Sex	−0.23 (0.29)	−0.85 to 0.40	-0.77	0.455
Leg length	−0.001 (0.002)	−0.01 to 0.00	-0.49	0.923		Leg length	0.005 (0.003)	0.00–0.01	1.66	0.239
R2					0.08	R2					0.60
Adjusted R2					-0.12	Adjusted R2					0.45

Note: We present the results of the multiple linear regression models testing the relationships between GSH and mobility metrics, controlling for the relevant covariates. P values were FDR corrected for each model (Benjamini and Hochberg 1995); significant P values are bolded. SE, standard error; CI, 95% confidence interval; GC, gait cycle.

^*P < 0.05, ^**P < 0.01.

Greater M/L sway speed and variability (i.e., greater postural instability) was correlated with higher GSH levels only for the older adults (Table 3 and Fig. 6). The young adults had a weak but nonsignificant positive association between M/L sway speed and variability and GSH levels. The partial correlation strength was significantly different between young and older adults (young adult partial r = 0.40; older adult partial r = 0.72; Z = 1.67; P = 0.048).

Figure 6

GSH relationships with motor but not cognitive performance for older adults only. Partial correlations of GSH levels with M/L postural sway and gait variability for young (left) and older (right) adults. Partial correlations are accounting for the effects of the covariates included in each model. In each of these cases, there was a significant relationship between higher sensorimotor GSH levels and poorer motor performance for the older but not the younger adults.

Greater gait variability was correlated with higher sensorimotor GSH levels for the older adults only (Table 3 and Fig. 6). No relationship emerged between gait variability and GSH levels for the young adults. The partial correlation strength was significantly different between young and older adults (young adult partial r = −0.16; older adult partial r = 0.68; Z = 3.48; P < 0.001).

To further test the specificity of the identified relationships between GSH and motor function for older adults, and not global shifts in metabolite concentrations, for the significant relationships above, we ran an additional linear model including as predictors GSH, plus the two other neurometabolites edited by HERMES: the excitatory neurochemicals Glx and the primary inhibitory neurotransmitter within the brain, GABA. We also included sex and leg length as covariates of no interest in these models. All relationships between physical function and GSH remained when including Glx and GABA as additional predictors; for older adults, the relationships remained significant between sensorimotor GSH levels and M/L sway speed/variability (P = 0.007) and gait variability (P = 0.003). There were no significant relationships between Glx or GABA levels and these motor metrics.

Of note, it is likely that the residual water signal differs between age groups due to age-related changes in bulk tissue composition impacting water T₁ relaxation. Although the current consensus suggests that sufficiently suppressed water is unlikely to impact metabolite quantification (Tkáč et al. 2020), to further ensure that age differences in residual water did not impact our findings, we reran statistical analyses for our significant findings including water suppression efficiency (calculated as per Kreis et al. 2020) as a covariate. Inclusion of this metric in our significant models did not change any findings (Supplementary Material H).

Discussion

We identified higher frontal and sensorimotor GSH levels for older compared with younger adults when accounting for age-related cortical atrophy. Across both age groups, we identified higher GSH levels for the sensorimotor compared with the frontal voxel. For the older adults only, we identified multiple relationships between higher sensorimotor GSH levels and poorer motor performance.

One potential explanation for higher brain GSH levels for older adults is that higher levels of GSH occur as a response to mitigate age-related increases in oxidative stress and maintain regional redox homeostasis within the brain. That is, perhaps in normal aging, in some regions of the brain, GSH antioxidant levels increase in response to increasing oxidative stress that occurs during aging. Past in vivo human studies have found higher MRS-measured GSH levels in MCI compared with age-matched controls (Duffy et al. 2014), but lower GSH levels in AD compared with controls (Mandal et al. 2015). Given the association between cognitive impairment and ROS production (Brawek et al. 2010), these findings could be interpreted as an ROS-induced upregulation of GSH in the early stages of cognitive decline. Similarly, past evidence suggests that MRS-measured GSH levels are higher in early schizophrenia (Wood et al. 2009), but lower after full symptoms emerge (Matsuzawa et al. 2008). Higher MRS-measured GSH levels have also been reported in post-traumatic stress disorder (Michels et al. 2014) and early psychosis (Godlewska et al. 2014). Furthermore, pharmacologically induced GSH depletion in the brain has been shown to result in cognitive decline in rodents (González-Fraguela et al. 2018). Therefore, high GSH levels could be associated with high levels of underlying cellular stress (e.g., ROS emissions) until reaching a threshold that exceeds the hormetic response capabilities of the cell.

The precise mechanisms that dictate GSH regulation in the aging brain remain unknown. Nonetheless, in theory, an increase in GSH could reflect either upregulation of GSH production or downregulation of GSH catabolism. For instance, one study (Mythri et al. 2011) found significantly less γ-glutamyl transpeptidase activity in the cortex and striatum of Parkinson’s disease patients, suggesting that lower rates of GSH breakdown were contributing to the increased GSH seen in these tissue samples.

Increased oxidative stress with aging could also explain increases in GSH. Increases in cellular antioxidants in response to oxidative insults have been well documented. Multiple animal model and cell culture studies have shown an upregulation of GSH in response to oxidative stress (Ong et al. 2000), including exposure to toxins such as mercury (Hoffman et al. 2005), radiation (Di Toro et al. 2007), neonatal alcohol (Smith et al. 2005), and methamphetamine (Harold et al. 2000), as well as neurological diseases such as models of Parkinson’s disease (Rodríguez Navarro et al. 2007; Aluf et al. 2010), Huntington’s disease (Tkáč et al. 2007), and AD (Tchantchou et al. 2005).

Although aberrant ROS production can be detrimental to cellular health, ROS also serve as signaling molecules capable of regulating transcriptional events in the cell (Powers et al. 2020). The Kelch ECH-associated protein 1 / Nuclear factor erythroid 2-related factor 2 (Keap1/Nrf2) signaling pathway is a canonical oxidative stress sensor in the cell. Nrf2 is a key protein responsible for increasing antioxidant enzymes by regulating transcription of antioxidant-related genes (Itoh et al. 1999). The endogenous protein, Keap1, suppresses Nrf2 activity under basal conditions by facilitating its removal from the cell. However, oxidative modification of Keap1 by ROS removes its inhibitory effects on Nrf2 and allows for Nrf2-mediated transcription of antioxidant-related genes (Sekhar et al. 2010). Importantly, Nrf2 regulates transcription of the rate-limiting enzyme responsible for synthesizing GSH, γ-glutamate-cysteine ligase (Yang et al. 2005). Therefore, increased ROS emissions with brain aging may result in increased GSH via the Keap1/Nrf2 signaling pathway. There is some literature support for this idea; for instance, exposure of astrocytes in cell culture to high levels of ROS induces transcription of genes responsible for synthesizing GSH (Sagara et al. 1996; Gegg et al. 2003).

It could also be that the observed GSH changes relate to changes in cell type abundance within the aging brain. As GSH is present in higher concentrations in glia compared with neurons (Rice and Russo-Menna 1997), increasing GSH levels could be associated with the increased gliosis that occurs with brain aging (Tong et al. 2011). Stereological cell counting in postmortem human brain suggests that the abundance of astrocytes, one of the predominant producers of brain GSH, remains constant throughout aging, whereas the abundance of other cell types (e.g., oligodendrocytes) decreases (Pelvig et al. 2008). This may be particularly important given that astrocytes are also more capable of inducing the antioxidant defense response via Nrf2 signaling compared with other neuronal cell types (Baxter and Hardingham 2016).

This finding of higher GSH levels for older compared with younger adults is in line with the results of Tong et al. (2016). This group identified GSH increases across the lifespan (i.e., 1 day–99 years old) in postmortem frontal cortex. However, this finding is in contrast to the results of Emir et al. (2011), who reported lower occipital cortex GSH levels for older compared with younger adults using edited MRS at 4T. There are several key differences between our work and this study. Emir et al. (2011) examined a different brain region; this likely contributed to their differing results. More recent work by this group using nonedited MRS at 7T found no age differences in posterior cingulate or occipital cortex GSH levels (Marjańska et al. 2017); however, again, this study tested different brain regions compared with our work. The lack of GSH age differences in posterior brain regions (but not frontal or sensorimotor cortex) could also be due in part to the well-established posterior to anterior shift of brain activity with aging (Davis et al. 2008; Jockwitz et al. 2019). It could be that reduced neural signaling within posterior brain areas leads to lower regional GSH levels; that is, upregulation of GSH may have ceased in these posterior regions as the hormetic response capabilities of these cells have been exceeded.

Given the limited age range in the present study, it is unknown whether the apparent age-related increase in brain GSH presented here may abate in extreme conditions of oxidative stress, such as neurological disease or very old age, as the antioxidant response is overwhelmed (e.g., as recycling or de novo synthesis mechanisms are compromised). Future longitudinal studies and enrollment of much older adults would clarify this.

Some work has reported regional differences in cortical GSH levels (Srinivasan et al. 2010; Tong et al. 2016; Nezhad et al. 2017). Here, we found higher GSH levels in the sensorimotor compared with the frontal voxel across both age groups, as well as tissue composition differences between the two voxels. We found higher GM and CSF fractions in the sensorimotor voxel, and a higher WM fraction in the frontal voxel. These findings fit with one past study reporting higher GSH concentrations in voxels with more GM than WM (Srinivasan et al. 2010). However, another study (Nezhad et al. 2017) reported conflicting findings of higher GSH concentrations in the cortical region with less GM (i.e., anterior cingulate vs. occipital cortex). We suspect that [as discussed by Rae and Williams (2017)] GSH levels likely vary across brain region, but in a more complex manner than that which reflects only GM and WM differences. This notion is further supported by recent work suggesting that human primary motor and somatosensory cortices show proportionally steeper trajectories of volume, myelin, and iron declines with advancing age compared with other brain regions (Taubert et al. 2020). It could be that the sensorimotor cortex structure and neurochemical composition is affected more or earlier by oxidative stress compared with other brain regions.

There were no associations between GSH levels and cognitive performance (i.e., MoCA scores). Our past work (Porges et al. 2017a) suggests that MoCA scores are sensitive enough to identify associations between MRS-measured neurometabolites and cognitive status. In contrast to our previous work (n = 93 older adults; mean age = 73.2 ± 9.9), here we included fewer participants, although our older adult ages were similar. In addition, participants in the present sample had higher MoCA scores compared with our previous work (mean = 25.5 ± 2.5). It could be that, among this higher functioning older adult cohort, we did not have enough variation in MoCA scores to identify a significant association. Furthermore, the limited past work in normal aging has failed to find any relationships between GSH levels and cognitive status (Emir et al. 2011; Chiang et al. 2017); such a relationship has previously been identified only in pathological conditions such as MCI and AD (Mandal et al. 2012, 2015; Oeltzschner et al. 2019). Thus, it could be that GSH–cognition relationships only emerge in cases of more severe cognitive decline, when brain resources (such as antioxidant availability) have substantially declined.

We found several associations between higher GSH levels and poorer motor performance (i.e., greater M/L postural sway and greater gait variability). These motor performance variables have functional significance. Greater M/L postural sway (Maki et al. 1994; Stel et al. 2003) and greater gait variability (Hausdorff et al. 2001) associate with a greater risk of falling for older adults. Although relationships of GSH levels with motor performance in normal aging have not been previously investigated, these findings fit with past work that identified relationships between GSH levels and movement disorders (Srinivasan et al. 2010; Weiduschat et al. 2014; Choi et al. 2015; Doss et al. 2015; Weerasekera et al. 2019). Together, these findings may indicate that GSH is a response to increasing oxidative stress and related tissue damage in the normally aging brain. Higher sensorimotor versus frontal cortex GSH levels could suggest that the sensorimotor cortex is disproportionately affected by oxidative stress in older age (and thus requires the largest GSH antioxidant response). This regionally heightened oxidative stress may then be contributing to these age-related declines in motor function.

Importantly, we identified relationships between motor performance and sensorimotor but not frontal GSH levels. This suggests regional specificity for these GSH relationships, rather than poorer motor performance being a consequence of increased oxidative stress throughout the brain. Moreover, we found no relationship between cortical Glx or GABA levels and these motor performance metrics, again supporting the specificity of this GSH relationship with motor behavior for older adults and not result of more general age-related shifts in metabolite concentrations.

There are several limitations to the present work. Our cross-sectional approach precluded us from assessing how GSH levels alter with aging or how changes in GSH levels across the lifespan relate to declines in motor performance. There is also some evidence that diet may influence MRS-measured GSH levels. One study (Choi et al. 2015) found an association between dairy consumption and brain GSH levels among older adults. In the present work, we did not record food intake or restrict diet prior to the MRI scan; future studies should characterize any effects of diet on brain GSH levels. Similarly, future work should consider how other factors related to accelerated aging (e.g., heavy alcohol use and human immunodeficiency virus status) may contribute to individual variation in GSH levels (Britton et al. 2020).

There are also several general limitations of MRS, such as the large voxel size required, which are currently unavoidable with this method. We assume that any partial volume effects due to the chemical shift displacement are not largely impacting the data. Particularly, for the two voxels in this work (which are both located along the midline), the bulk tissue concentrations are likely similar in the brain regions immediately adjacent to these two voxels. Moreover, if the chemical shift displacement artifact resulted in increased water visibility, the net result would likely bias the analyses against our reported finding of elevated GSH levels in aging.

In addition, we used the HERMES sequence instead of other options [e.g., short TE to minimize the susceptibility to T₂ changes, as done by Emir et al. (2011)]. Although T₂ changes with age could potentially confound quantification of GSH (Marjańska et al. 2013, 2017; Deelchand et al. 2020), it is likely that this would result in reduced visibility and lower estimations of GSH concentration for older adults (i.e., meaning that our finding of elevated GSH in aging may be an under estimation of the true effect). Moreover, other potentially confounding factors, such as asymmetric tissue composition in aging, have been shown to result in reduced estimates of GSH for older adults (Emir et al. 2011), again supporting that our finding of elevated GSH in aging is likely a true reflection of physiology. Furthermore, we demonstrated these age difference and brain–behavior relationships using tissue-corrected GSH levels. Tissue-corrected GSH accounts for tissue type volume fractions, as well as water visibility, and both transverse and longitudinal relaxation times of water in each tissue compartment (Gasparovic et al. 2006); thus, this correction accounts for several factors beyond only age-related atrophy and yields the same findings, further supporting that these results are not due to artifacts or age differences in noise.

These results provide insight into the association between brain aging and oxidative stress. We demonstrate higher GSH levels with normal aging, suggesting a GSH response to increased oxidative stress with older age. We report higher GSH levels in the sensorimotor cortex compared with the frontal cortex for both age groups, as well as associations between sensorimotor GSH levels (but not GABA or Glx levels) and poorer balance and gait. Together, these results suggest that MRS-measured GSH could be a neural response to increased oxidative stress with brain aging and also a marker of poorer motor performance. These results could stem from greater or earlier effects of oxidative stress on the sensorimotor compared with the frontal cortex.

Authors’ Contributions

K.H. participated in initial study design, collected all data, processed the MRS data, conducted the statistical analyses, created the figures, and wrote the manuscript. H.H. contributed to manuscript writing and results interpretation. P.A.J. assisted with data collection, data processing, and manuscript preparation. M.M. advised on MRS data processing and methods, in addition to contributing to manuscript preparation. C.H. consulted on the design and analysis of the motor performance tests. R.E. advised on MRS data acquisition, processing, interpretation, and manuscript preparation. R.S. and E.P. oversaw study design and led the interpretation and discussion of results. All authors participated in revision of the manuscript.

Funding

During completion of this work, K.H. was supported by the National Science Foundation Graduate Research Fellowship (Grant no. DGE-1315138, DGE-1842473) and by National Institute of Neurological Disorders and Stroke training (grant T32-NS082128). M.M. receives salary support from National Institutes of Health (grant K99 EB028828). E.P. was supported by National Institute on Alcohol Abuse and Alcoholism (grant K01 AA025306) and the McKnight Brain Research Foundation; the Center for Cognitive Aging and Memory at the University of Florida. A portion of this work was performed in the McKnight Brain Institute at the National High Magnetic Field Laboratory’s Advanced Magnetic Resonance Imaging and Spectroscopy Facility, which is supported by National Science Foundation (Cooperative Agreement No. DMR-1644779) and the State of Florida. This work was also supported in part by the National Institutes of Health award (S10OD021726) for High End Instrumentation. This study applied tools developed under National Institutes of Health (grants R01 EB016089, R01 EB023963, P41 EB015909).

Notes

The authors wish to thank Aakash Anandjiwala, Justin Geraghty, and Alexis Jennings-Coulibaly for their help in participant recruitment and data collection. The authors also wish to thank all of the participants who volunteered their time, as well as the McKnight Brain Institute MRI technologists, without whom this project would not have been possible. Conflict of Interest: None declared.