Higher fasting brain glucose is associated with lower gray matter volume in healthy adults

Abstract

Glucose hypometabolism is implicated in age-related neurodegeneration, with peripheral markers such as hyperglycemia and insulin resistance linked to increased dementia risk and brain atrophy. However, the degree to which peripheral and cerebral glucose dysregulation are coupled remains unclear. To address this, we used ultra-high-field ¹H MRS to directly quantify fasting brain glucose concentrations in the posterior cingulate cortex of 47 healthy adults from across the lifespan, with concurrent structural MRI and fasting blood glucose measurements. We found that both fasting blood and brain glucose increase with age but follow distinct trajectories: peripheral glucose rises significantly by midlife (40-60 years) and then plateaus, whereas cerebral glucose remains stable until older age (60-80 years) before increasing. Furthermore, brain glucose, but not blood glucose, independently predicts gray matter loss and partially mediates age-related atrophy, with effects strongest in subcortical, GLUT4-expressing regions. Our results provide direct in vivo evidence that higher fasting brain glucose levels are associated with age-related gray matter loss, and suggest a delayed trajectory compared to peripheral glucose elevation, consistent with cerebral metabolic prioritization. Brain glucose measured by MRS represents a promising biomarker of metabolic dysfunction that may enable early detection of neuronal vulnerability and inform interventions targeting brain-specific glucose metabolism.

Introduction

The brain relies on a continuous and tightly regulated glucose supply to meet its energy demands. Increasing evidence suggests that aberrant glucose homeostasis, characterized by hyperglycemia, insulin resistance, and reduced cellular glucose uptake, occurring both peripherally and within the brain itself, play a crucial role in age-related cognitive decline and neurodegeneration. In the periphery, longitudinal and cross-sectional studies have found that even among healthy adults without diabetes, higher normal blood glucose levels are associated with a greater risk of developing dementia ^1,2, Alzheimer’s disease ³, increased brain atrophy ^4-7, and lower cognitive performance ⁸, while decreased peripheral glucose regulation was found to be associated with decreased cognitive performance, memory impairments, and hippocampal atrophy ⁹. Notably, the brain regions affected by elevated blood glucose, including the PCC, frontal cortex, and hippocampus, overlap substantially with areas showing early vulnerability in Alzheimer's disease, highlighting the importance of glycemic control for brain health and the need to better under the impact of glucose metabolism on neurodegeneration.

One major cause of peripheral glucose dysregulation is insulin resistance. Insulin is the primary hormone by which glucose is taken up into insulin-sensitive cells, a process mediated by the GLUT4 glucose transporter in GLUT4-dependent tissues like skeletal muscle, cardiac muscle, and adipose tissue ¹⁰. When insulin signaling is impaired, these cells cannot import glucose from the bloodstream, resulting in hyperglycemia alongside hypometabolism, and eventual atrophy ¹¹. The peripheral oral glucose tolerance test (OGTT) has proven invaluable for detecting impaired glucose regulation ¹². By measuring peripheral glucose accumulation after a standardized glucose challenge, elevated blood glucose levels following the OGTT indicate impaired glucose clearance and utilization rather than merely increased supply. Recent evidence suggests that similar principles may apply to brain glucose homeostasis, where just as peripheral glucose accumulation in the OGTT reflects insulin resistance and metabolic dysfunction, brain glucose accumulation may serve as an indicator of impaired cerebral glucose utilization and neuronal insulin resistance ¹³. The concept of “brain insulin resistance” has emerged as a key mechanism linking metabolic dysfunction to neurodegeneration, with studies showing that brain insulin resistance can be detected long before clinical symptoms appear and may drive early pathological changes ^14,15.

The temporal relationship between peripheral and cerebral glucose dysregulation is complex and may reveal a sequence of changes in metabolism that begin well in advance of clinical symptom onset. Longitudinal findings from up to an eight-year period suggest that altered peripheral glucose metabolism in midlife may precede brain-specific changes, with impaired glucose tolerance in midlife leading to decreases in regional cerebral blood flow (rCBF) in the frontal, parietal, and temporal cortices compared to those with normal glucose tolerance ¹⁶. In a shorter-term study, older adults whose fasting glucose increased over one year displayed regional atrophy in the hippocampus and inferior parietal cortex, as well as increased amyloid accumulation in the precuneus, despite remaining cognitively normal ¹⁷. These findings support the idea that peripheral glucose dysregulation may signal emerging brain vulnerability, but the sequence and directionality of this relationship is unclear. There are three possible patterns: 1) brain and peripheral glucose levels tightly track each other, reflecting a direct coupling of systemic and cerebral glucose metabolism, 2) elevations in brain glucose precede elevations in the periphery if buffering mechanisms fail early enough, or 3) elevations in brain glucose could lag behind the periphery, implying that the brain is initially protected and its homeostasis is prioritized. Alterations in blood-brain barrier (BBB) transporters ^18,19 and astrocytic metabolic buffering ²⁰ observed with aging support the latter pattern, suggesting that the brain may preserve glucose homeostasis even under conditions of peripheral hyperglycemia.

Despite extensive research into systemic glucose dysregulation, measured in the periphery, few studies have measured glucose concentrations directly in the brain to test how brain and peripheral levels track across the lifespan. This is a significant limitation in the literature, as glucose transport across the BBB can become decoupled from peripheral glucose levels with age and neurodegenerative disease ^21-23. Studies using ¹⁸F-fluorodeoxyglucose positron emission tomography (FDG-PET) have shown that the cerebral metabolic rate of glucose (CMRglc) is about 25% lower in those with Alzheimer’s disease. These metabolic reductions seem to follow predictable patterns associated with cognitive decline: they first emerge in the hippocampus and posterior cingulate cortex (PCC), then extend to the temporal and parietal lobes, and later progress to the frontal lobes ²⁴. In healthy aging, CMRglc reductions in the precuneus and cingulate gyrus are associated with cognitive decline, especially in the domains of memory and executive function ²⁵. Although there is extensive research using FDG-PET to assess CMRglc across the lifespan and in disease, the modality measures glucose uptake, which is a composite of multiple cellular processes, rather than actual glucose concentrations within brain tissue. It also requires radioactive tracers, limiting longitudinal studies and the populations that can be studied. ¹H magnetic resonance spectroscopy (¹H MRS) offers a unique solution by providing steady-state information on metabolic pools within brain tissue, complementing the kinetic information provided by FDG-PET ²⁶. While PET measures how much glucose enters cellular pathways, ¹H MRS measures how much glucose accumulates in the tissue, both intracellularly and extracellularly, serving as a brain-based equivalent to the peripheral OGTT by detecting impaired glucose clearance and utilization.

The present study used ¹H MRS to quantify fasting brain glucose concentrations and test how brain and blood glucose levels relate to each other, and to atrophy, in healthy aging. The first objective was to test whether changes in brain glucose levels track, precede, or lag behind peripheral glucose levels, providing evidence to help determine the temporal coupling of systemic and brain glucose metabolism. The second objective was to examine the relative contributions of brain versus peripheral glucose concentrations to age-related brain atrophy, and how they mediate the relationship between age and atrophy. To address these objectives, we conducted a cross-sectional study using ¹H MRS to directly measure fasting brain glucose concentrations in the PCC, alongside structural MRI to assess gray matter volumes in healthy adults from across the lifespan (N = 47, aged 22-77 years). By measuring both brain and peripheral glucose levels in the same individuals, our study quantified the specific impacts of brain and peripheral glucose on neuroanatomical vulnerability in healthy aging, helping to disentangle mechanisms of glucose dysregulation in aging.

Methods

Study Population

Structural MRI and ¹H-MRS data were collected from a cohort of healthy adults aged 22 to 77 years. Exclusion criteria included MR contraindications for ultra-high-field imaging, diagnoses of psychiatric and/or neurological disorders, diagnoses of diabetes mellitus, insulin resistance (HbA1c > 5.7%), history of brain injury, recreational drug use, severe alcohol use, use of medications that affect glucose and/or insulin utilization, and current or recent adherence to a low-carbohydrate or ketogenic diet. To confirm the absence of overt cognitive impairment, participants completed two cognitive screening tests: the Mini-Mental State Examination and the CNS Vital Signs battery. Scores were within the normal range for age and education, and no participant met the criteria for cognitive impairment. Detailed clinical and demographic characteristics for the study population can be found in Table 1. The study was approved by the Institutional Review Boards of Massachusetts General Hospital (2015P000652) and Stony Brook University (IRB2019-00208). Participants were recruited from the Boston metropolitan area through online and paper advertisements. The study was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent.

Table 1. Clinical and demographic information of study participants.

Includes clinical marker results collected at screening visit.

Study Population Description
Sample size	47
Sex	24 Male 23 Female
Age (mean ± SD years)	52.7 ± 14.3
Age range	22 - 77
HbA1c (mean ± SD %)	5.20 ± 0.28
HbA1c range	4.5 - 5.7
BMI (mean ± SD kg/m2)	25.2 ± 4.3
BMI range	17.7 - 37.23

Experimental Design

Structural MRI and ¹H MRS data were collected from a cohort of healthy adults across two scanning sessions, conducted on separate days within a 14-day interval. Both sessions were conducted in the morning following an overnight fast, during which participants abstained from all food and drink besides water for at least 8 hours prior to scanning. During the ¹H MRS scans, a white cross on a black background was presented, and participants were instructed to keep their eyes open and let their mind wander. Fasting blood glucose was measured immediately prior to scanning using a Precision Xtra Blood Glucose & Ketone Monitoring System (Abbott Laboratories).

¹H Magnetic Resonance Spectroscopy

Participants were scanned on a 7T Siemens Magnetom MRI Scanner (Siemens Healthineers, Erlangen, Germany) with a 32-channel head coil built at the Massachusetts General Hospital Athinoula A. Martinos Center for Biomedical Imaging. ¹H MRS spectra were collected using a single-voxel stimulated echo acquisition mode (STEAM) sequence with short echo time (TE) = 5.00 ms, repetition time (TR) = 4500 ms, mixing time (TM) = 75 ms, water suppression bandwidth = 132 Hz, averages = 80, voxel size = 20×20×20 mm³, acquisition bandwidth = 4000 Hz, vector size = 2048 points, radiofrequency pulse duration = 3200 ms, and with 3D outer volume suppression interleaved with variable power and optimized relaxation (VAPOR) water suppression. A short TE was selected to avoid underestimating concentration levels of metabolites with reduced T2 relaxation times in older participants ²⁷. Unsuppressed water acquisition was obtained with similar parameters using four averages. Using the T1-weighted (T1w) structural image collected prior to MRS acquisition, the voxel was placed in the posterior cortex, encompassing parts of the median PCC and precuneus (Figure S2B). The PCC was used as the region of interest due to its high baseline metabolic rate ²⁸ and vulnerability to metabolic changes associated with neurodegenerative disease, evidenced by a reduction in glucose metabolism in the PCC in individuals with early Alzheimer’s disease ²⁹.

Metabolite concentrations were quantified using LCModel ³⁰ with a baseline stiffness parameter of 0.25 ppm (Figure S2A). The LCModel basis set used for quantification, described in ³¹, was used in prior works with ultra-high-field, short-TE STEAM spectra ^32,33. Frequency and eddy current corrections were performed prior to data processing. All spectra and fits were visually inspected for quality, and spectra were excluded if they contained large lipid peaks or excessive noise. Due to the relatively low concentration of glucose in the brain we applied a CRLB %SD cutoff of 300% to retain sufficient sample size for analysis, as strict Cramér-Rao lower bound (CRLB) thresholds (commonly 20-50%) can introduce bias by disproportionately excluding low-concentration metabolites and lead to loss of valuable data ³⁴. As a robustness check, we repeated our analyses using additional CRLB %SD cutoffs of 50% and 100%, and our results do not change significantly (Figure S1). Repeating our analyses using Glc+Tau as the metabolite of interest, rather than Glc, with a CRLB %SD cutoff of 20%, also did not significantly change our results. Metabolites were analyzed in absolute concentration levels (relative to water). Using the T1w structural image, segmentation of the MRS voxel to obtain tissue fractions of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) was performed using Gannet (version 3.3.2) ³⁵ and SPM 12 ³⁶. Tissue fractions were used to perform CSF-correction prior to statistical analyses. As the CSF contains a non-negligible amount of Glc ³⁷, we repeated the multiple linear regressions performed in Figure 2A and Figure 3A while including fractional volume of CSF as a covariate. We also repeated the regressions while including fraction volumes of GM and WM as covariates. Doing so did not significantly change our results (Tables S1-S3).

Figure 2. Fasting brain glucose, but not blood glucose, is associated with lower total gray matter volume and partially mediates the effects of aging on brain structure.

(A) Standardized effect sizes (β) from a multiple linear regression model (outcome: total gray matter volume; predictors: age, BMI, fasting brain glucose, and fasting blood glucose) demonstrate that age and fasting brain glucose are significant predictors of total gray matter volume, while BMI and fasting blood glucose are not. Error bars represent 95% CI. (B) Mediation analysis diagram illustrating the indirect pathways from age to total gray matter volume via both fasting glucose measures. Age is associated with increases in both fasting brain glucose and blood glucose, but only the pathway through brain glucose significantly mediates the relationship between age and total gray matter volume (indirect effect: −0.090, 95% CI [−0.245, −0.012], p-values ranging from 0.035 - 0.041), while blood glucose shows no significant mediation (indirect effect = 0.042, 95% CI approximately [−0.047, −0.146], p-values ranging from 0.394 - 0.414).

Figure 3. Effects of fasting brain glucose on subcortical and cortical gray matter volumes: global and regional analyses.

(A) Standardized effect sizes (β) from multiple regression models (outcomes: cortical and subcortical gray matter volumes; predictors: age, BMI, fasting brain glucose, and fasting blood glucose) demonstrate that age significantly predicts both subcortical and cortical volumes, while fasting brain glucose is a significant predictor of subcortical gray volume only. BMI and fasting blood glucose do not significantly predict gray matter volumes. Error bars represent 95% confidence intervals. (B)/(C) Regional subcortical/cortical brain maps show standardized β coefficients for fasting brain glucose predicting regional volumes, controlling for age, BMI, and blood glucose. Strong negative associations appear in the cerebellum, hippocampus, and insula, while ventricular volumes show positive associations consistent with brain atrophy.

Structural MRI Acquisition and Processing

A multi-echo magnetization prepared rapid gradient echo (MEMPRAGE) sequence was used to acquire T1w structural images using a 1 mm isotropic voxel size and four echoes (TE1 = 1.61 ms, TE2 = 3.47 ms, TE3 = 5.33 ms, TE4 = 7.19 ms, TR = 2,530 ms, flip angle = 7°, R = 2 acceleration in the primary phase encoding direction with 32 reference lines), and online GRAPPA image reconstruction, leading to a total volume acquisition time of 6 minutes and 3 seconds.

T1w structural images were preprocessed using fMRIPrep (version 20.2.3, ³⁸). The T1w images were bias-field corrected and skull-stripped using ANTs version 2.3.3 ³⁹. Cortical and subcortical segmentation was performed using FreeSurfer version 6.0.1 ⁴⁰, and FreeSurfer-provided parcellations were used to define regions. Cortical parcellations were based on the Desikan-Killiany atlas ⁴¹, segmenting the cortex into 34 regions per hemisphere, for a total of 68 cortical regions. Subcortical segmentation was based on FreeSurfer’s standard aseg atlas ⁴², segmenting the subcortex into 18 subcortical (9 per hemisphere) and 5 ventricular (2 per hemisphere, plus third ventricle) regions. Regional and total gray matter volumes were analyzed as fractions of intracranial volume (ICV).

Statistics

For all analyses, fasting brain glucose and blood glucose were calculated as an average of the measurement from both scanning days when available, otherwise, measurements from a single day were used. To compare fasting glucose levels in the brain and periphery (Figure 1), values were z-scored within their respective compartments, and Pearson correlations were used to determine overall lifespan trends. Group differences between age brackets were tested with independent t-tests. Test-retest reliabilities of fasting brain and blood glucose measurements were examined using Pearson correlations, intraclass correlations, and Bland-Altman plots (Figure S3).

Figure 1. While fasting glucose levels rise with age in both the brain and periphery, the timing of increases differs between compartments.

(A) Fasting blood glucose shows a significant positive correlation with age (r = 0.50, p < 0.001, N = 47), consistent with progressive peripheral glucose dysregulation across adulthood. (B) Fasting brain glucose, measured in the posterior cingulate cortex using ¹H-MRS, also increases with age (r = 0.38, p = 0.01, N = 47). (C) Despite both measures increasing with age, fasting blood and brain glucose levels are not correlated within individuals (r = −0.01, p = 0.92, N = 47). (D) When grouped by age bracket (20-39, 40-59, and 60-70 years), mean fasting glucose levels increase significantly between the youngest and oldest age group in both the blood (t = 5.17, p < 0.001) and the brain (t = 3.27, p = 0.003). Mean fasting blood glucose increases significantly from young to middle-age (t = 4.61, p < 0.001), with no further rise in older age (t = 0.34, p = 0.73). In contrast, mean fasting brain glucose levels remain stable from young to middle age (t = 1.29, p = 0.21) and rise significantly from middle- to older age (t = 2.24, p = 0.03), suggesting that age-related increases in peripheral glucose measures manifest earlier in adulthood than increases in brain glucose.

Multiple Linear Regression

Multiple linear regression models were used to examine the associations between fasting brain glucose and brain volume measures, controlling for age, BMI, and fasting blood glucose. All variables were standardized (z-scored) prior to analysis to facilitate comparability of effect sizes. Separate regression analyses were performed for each outcome variable to capture effects at different anatomical scales. First, total gray matter volume was modeled as the dependent variable with all predictors included simultaneously (Figure 2A). Next, cortical and subcortical gray matter volumes were modeled separately to assess differential effects across these broad tissue classes (Figure 3A). Finally, individual regional analyses were performed for each subcortical (Figure 3B) and cortical (Figure 3C) structure obtained from FreeSurfer’s segmentation outputs. This series of models allowed identification of global, tissue-specific, and localized associations between fasting brain glucose and gray matter volume. Results are reported as standardized effect sizes (β) for each predictor, along with their respective p-values and 95% confidence intervals. When comparing across fasting metabolites (e.g., fasting brain Glc+Tau, Glc, and Tau in Figure S1) or individual subcortical and cortical structures (Figure 3B and Figure 3C), correction for multiple comparisons was applied using the Benjamini-Hochberg false discovery rate (FDR) procedure.

Mediation Analysis

To test whether fasting brain glucose or blood glucose mediated the association between age and total gray matter volume (Figure 2B), a parallel mediation model was implemented ^43,44. Prior to the analysis, all variables (age, brain glucose, blood glucose, and total gray matter volume) were z-scored to standardize their scales. Brain glucose, blood glucose, and gray matter volume were residualized to correct for BMI. As a result, all reported mediation coefficients reflect standardized relationships adjusted for BMI. To ensure robustness, the mediation model was repeated across 10 random seeds, each with 10,000 bootstrap samples. Effect sizes for each pathway were equal across seeds, and the ranges of 95% confidence intervals and p-values for each effect size are reported.

Results

Age-related divergence in peripheral and brain glucose trajectories suggests prioritization of cerebral glucose homeostasis

Fasting blood and brain glucose levels were compared across the lifespan using z-scores to allow for a direct comparison between modalities with different units (Figure 1). Both fasting blood and brain glucose levels significantly increase with age (blood: r = 0.50, p < 0.001, N = 47; brain: r = 0.38, p = 0.01, N = 47). Despite both measures increasing with age, fasting blood and brain glucose levels are not correlated within individuals (r = −0.01, p = 0.92, N = 47), and this lack of correlation persisted even when correcting for age (partial r = −0.25, p = 0.09, N = 47).

When grouped by age bracket (young: 20-39 years, N_blood = 29, N_brain = 9; middle-aged: 40-59 years, N_blood = 37, N_brain = 21; old: 60-79 years, N_blood = 28, N_brain = 17), both blood and brain glucose were significantly higher in the older group compared to the younger group (blood: t = 5.17, p < 0.001; brain: t = 3.27, p = 0.003). However, the trajectories diverged in their timing. For fasting blood glucose, levels were significantly higher in the middle-aged group compared to the young group (t = 4.61, p < 0.001), while no further increase was seen from middle-aged to old (t = 0.34, p = 0.73). In contrast, for fasting brain glucose, levels were significantly higher in the older group compared to the middle-aged group (t = 2.24, p = 0.03), while there was no significant difference between the young and middle-aged groups (t = 1.29, p = 0.21). These findings suggest that the rise in blood glucose may precede the rise in brain glucose, with peripheral glucose elevations appearing in midlife and cerebral glucose elevations appearing later, in older age.

Elevated fasting brain glucose, but not peripheral glucose, predicts reduced gray matter volume independent of age

To evaluate the impact of fasting glucose levels on brain structure, we examined associations between fasting blood glucose, fasting brain glucose (measured by ¹H-MRS), and total gray matter volume in our healthy cohort (N = 47, Figure 2A). In a multiple linear regression model controlling for age and BMI, both age (β = −0.59, p < 0.001) and fasting brain glucose (β = −0.28, p = 0.04) were significant predictors of total gray matter volume, whereas fasting blood glucose (β = 0.15, p = 0.28) and BMI (β = 0.12, p = 0.36) were not. Thus, higher fasting brain glucose was independently associated with lower total gray matter volume, even after accounting for age and peripheral glucose, while fasting blood glucose showed no such relationship.

To further examine the unique contributions of brain and blood glucose, we conducted a mediation analysis testing the indirect effects of age on total gray matter volume via both glucose measures (Figure 2B). Age was associated with increases in both fasting brain glucose (a₁ = 0.28, p = 0.05) and blood glucose (a₂ = 0.43, p = 0.002). Across multiple bootstrap seeds (10 seeds, n = 10,000 bootstraps per seed), only the pathway through brain glucose showed a consistently significant association with total gray matter volume (indirect effect = −0.090, 95% CI approximately [−0.245, −0.012], p-values ranging from 0.035 - 0.041), whereas the indirect effect through blood glucose was consistently non-significant (indirect effect = 0.042, 95% CI approximately [−0.047, −0.146], p-values ranging from 0.394 - 0.414). Collectively, these results indicate that elevated fasting brain glucose, but not fasting blood glucose, is a significant partial mediator of age-related reductions in gray matter volume.

Global and regional associations of fasting peripheral glucose with gray matter volume

In our multiple linear regression analyses (N = 47, Figure 3), fasting brain glucose was found to be a significant predictor of subcortical gray matter volume (β = −0.41, p = 0.003), alongside age (β = −0.46, p = 0.008). In contrast, cortical gray volume was significantly predicted only by age (β = −0.57, p = 0.001), with brain glucose showing a non-significant trend (β = −0.21, p = 0.13). Both models controlled for fasting blood glucose and BMI, neither of which were found to be significant predictors in either model.

At the regional level, exploration of standardized beta coefficients for brain glucose demonstrated spatially heterogeneous associations across subcortical and cortical structures. For each region, two significance values are reported: the uncorrected p-value (p_uncorr), which is the raw statistical significance, and the corrected p-value (p_corr), which is adjusted for multiple comparisons. The strongest negative associations, indicating smaller regional volumes with elevated fasting brain glucose, were observed in subcortical regions including the right cerebellum cortex (β = −0.46, p_uncorr < 0.001, p_corr = 0.09), right hippocampus (β = −0.42, p_uncorr = 0.005, p_corr = 0.16), and left cerebellum cortex (β = −0.404, p_uncorr = 0.007, p_corr = 0.178), along with cortical areas such as the left insula (β = −0.39, p_uncorr = 0.01, p_corr = 0.18) and right superior temporal gyrus (β = −0.35, p_uncorr = 0.03, p_corr = 0.22). In contrast, ventricular volumes, including the left lateral ventricle (β = 0.41, p_uncorr = 0.003, p_corr = 0.13), right lateral ventricle (β = 0.34, p_uncorr = 0.01, p_corr = 0.18), and third ventricle (β = 0.37, p_uncorr = 0.01, p_corr = 0.18), showed positive associations consistent with ventricular enlargement commonly reflecting brain atrophy. However, none of these regional associations survived correction for multiple comparisons, underscoring the exploratory nature of these findings.

Discussion

In this cross-sectional study of healthy adults across the lifespan, we found that fasting brain glucose (measured in the PCC using ¹H-MRS) and fasting blood glucose levels both increased with age. Descriptively, age-group comparison suggested that elevations in peripheral glucose were more apparent by midlife, whereas differences in fasting brain glucose were more evident in older age. This divergence is consistent with the third pattern of blood/brain glucose relationship outlined earlier, where the brain prioritizes and maintains glucose homeostasis even when systemic regulation deteriorates. Using multiple linear regression models, we found that fasting brain glucose was negatively associated with total gray matter volume, independent of age and BMI, while fasting blood glucose was not. Mediation analyses further suggest that brain glucose, but not blood glucose, partially mediates the effect of age on gray matter volume. Lastly, we explored the regional effects of these relationships, finding that the association between increased brain glucose and total gray matter volume is driven by loss in the subcortical regions, primarily the cerebellum and hippocampus.

These findings reveal a critical distinction between peripheral and cerebral glucose homeostasis in brain aging. The different age-related trajectories observed in fasting brain and peripheral glucose implies that peripheral dysregulation may precede cerebral metabolic alterations, with midlife blood glucose elevation potentially driving later brain-specific changes. The delayed rise in brain glucose supports the hypothesis that the cerebral glucose homeostasis is initially prioritized and buffered against systemic metabolic stress through protective mechanisms not available to the periphery. An example of such a protective mechanism is the astrocytic network, which synthesizes and stores glycogen to be used as energy source by the brain (via glycogenolysis) in the event of a discontinuity in cerebral glucose supply ^45,46. Another is the ability of the brain to shift to alternative fuel sources such as ketone bodies when glucose metabolism is impaired, using the upregulation of monocarboxylate transporters (MCTs), to shuttle ketones into neurons and provide energy that does not rely on insulin signaling ^47,19. As a result, this glucose sparing effect decreases CMRglc ⁴⁸, reducing the drive for transport of glucose across the BBB ⁴⁹. The periphery lacks these specialized buffering systems, offering a possible explanation as to why peripheral fasting glucose rises earlier in life than fasting brain glucose. Lastly, the brain differs from the periphery in that its glucose uptake relies predominantly on insulin-independent transporters: GLUT1 at the BBB and in astrocytes, and GLUT3 in neurons. GLUT4 is expressed primarily in subcortical brain regions, including the cerebellum, hippocampus, basal ganglia, and hypothalamus, and has limited expression in the cortex ^50,51. Consequently, insulin resistance in the brain will selectively target those GLUT4-expressing regions, which may explain why the association between increased brain glucose and reduced gray matter volume in our study is strongest in subcortical regions.

While previous studies have consistently demonstrated associations between elevated peripheral glucose levels and brain atrophy, our results show that when both brain and peripheral glucose are measured simultaneously, brain glucose emerges as the mechanistically relevant predictor of brain structure. This suggests that prior associations found between blood glucose and brain atrophy may reflect the brain’s impaired ability to utilize glucose rather than simply peripheral glucose elevations. This aligns with previous research using FDG-PET associating reduced CMRglc with increased atrophy, as elevated glucose levels measured using ¹H-MRS reflect accumulation, rather than uptake of glucose, pointing to a potential decrease of glycolysis in the brain ^52,53. This raises the possibility that glucose homeostasis in the brain may serve as a more sensitive indicator or mediator of brain atrophy than peripheral blood glucose, suggesting that interventions targeting brain glucose metabolism may be more effective than focusing solely on peripheral glycemic control for preserving brain structure in aging and disease.

In the brain, impaired glucose metabolism and subsequent excess of glucose may lead to atrophy through a variety of mechanisms. Elevated fasting brain glucose is associated with elevated myo-inositol (Ins) independent of age, suggesting that glucose itself may be associated with inflammatory processes ³³. Impaired insulin signaling and hyperglycemia elevate the expression and activity of NF-κB, a mediator of inflammatory responses and regulator of apoptosis ⁵⁴. Hyperglycemia, especially if chronic, also increases the production of advanced glycation end-products, which induce oxidative stress and inflammation ^55,56. Although these studies are primarily focused on individuals with diabetes, impaired glucose metabolism can manifest before a formal diagnosis and may exert similar effects. Even subtle impairments in glycemic regulation, measured in the periphery, in healthy, cognitively normal people have been linked to region-specific atrophy and decreased cognitive performance, supporting the idea that higher brain glucose may contribute to neurodegenerative processes independent of clinical diabetes ⁹.

MRS uniquely captures brain glucose accumulation that FDG-PET cannot. While FDG-PET quantifies tracer uptake and phosphorylation kinetics, it does not measure intra- and extra-cellular glucose pools. In contrast, MRS provides steady-state concentrations of glucose and related metabolites in vivo, revealing additional details of the relationship between glucose metabolism and aging. Prior work using J-modulated Point-Resolved Spectroscopy (J-PRESS) found a significant elevation of Glc in the PCC of Alzheimer’s patients relative to both younger and older controls, directly implicating impaired glucose metabolism in disease ⁵². This specificity enables detection of early metabolic impairment before they manifest as reduced uptake on PET, offering a complementary method for comprehensive metabolic profiling.

Although our study offers a unique perspective on the link between glycemic dysregulation and atrophy, some limitations should be acknowledged. Accurate measurement of brain glucose using ¹H MRS at 7T is technically challenging due to spectral overlap with other metabolites such as Tau ⁵⁷, leading many studies to report composites like Glc+Tau for greater quantification stability. Nevertheless, in vivo measurements under physiological conditions have been successfully made at multiple field strengths using short echo times in both humans and rats for over 30 years ^49,57-60. Our main findings were robust to this methodological challenge, as we were able to reproduce our results using both Glc and the summed Glc+Tau signal, with Tau alone having non-significant findings (Figure S1). Future studies could employ J-PRESS sequences or, if measuring other low-concentration metabolites simultaneously is a goal, employ short-TE STEAM sequences but quantify glucose using the H1-α-glucose peak at 5.23 ppm ⁶¹. Furthermore, our cross-sectional study precludes causal inference, and replication using a longitudinal study design, perhaps using both MRS and FDG-PET, would be needed to chart the relationship between brain glucose metabolism and atrophy over time. In addition, although our cohort is large for ultra-high-field ¹H MRS, it remains modest for precisely characterizing age-related trajectories, particularly when dividing participants into age brackets. As a result, differences in the timing of peripheral versus cerebral glucose level changes across the lifespan should be considered exploratory, and larger longitudinal samples will be required to rigorously test these trajectory patterns. Finally, our MRS voxel was placed in the PCC, a cortical region which relies mainly on the insulin-independent glucose transporters GLUT1 and GLUT3. Future studies could expand on this work by placing MRS voxels in GLUT4-rich subcortical areas, which may allow even earlier detection of hypometabolism, while including both GLUT4- and GLUT1/GLUT3-dependent regions would allow for direct comparison of their metabolic trajectories in brain aging.

Our study provides the first direct evidence that fasting brain glucose accumulation, distinct from peripheral glucose elevation or CMRglc, is associated with reduced gray matter in healthy aging, and that the brain maintains glucose homeostasis later in life than the periphery, a pattern consistent with cerebral metabolic prioritization. Using brain glucose levels as a novel biomarker of cerebral metabolic dysfunction opens new therapeutic and diagnostic avenues, as early detection of brain glucose dysregulation via non-invasive ¹H-MRS could enable timely interventions to preserve neuronal and gray matter integrity and slow cognitive decline. Importantly, our findings emphasize that future therapeutic development should prioritize targeting the brain directly, aiming to enhance neural insulin sensitivity, support energy availability buffering via the astrocytic network, and utilize alternate fuel sources to protect against age-related neurodegeneration.

Data Availability Statement

The structural MRI data are available on OpenNeuro (https://openneuro.org/datasets/ds005405/versions/1.0.1). ¹H MRS data, the participants of which are a subset of the structural MRI cohort, are available upon request.