18F-Fluorodeoxyglucose Uptake in Cerebrospinal Fluid Reflects Both Brain Glucose Demand and Impaired Blood–Brain Barrier Transport in Alzheimer’s Disease

Abstract

Glucose delivery to the brain requires transporters at the blood–brain barrier (BBB), whose downregulation may be associated with neuronal deficits in Alzheimer’s disease (AD). Whether this downregulation is due to reduced demand or primary BBB dysfunction remains unclear. We investigated novel 18F-Fluorodeoxyglucose Positron Emission Tomography (18F-FDG-PET) measures, namely ventricles (FDG_Ventricles) and cortical uptake (FDG_Cortex), and the FDG_Ventricles/FDG_Cortex ratio in 224 patients with AD compared to those in 35 controls (CTRLs). AD patients showed lower FDG_Cortex and FDG_Ventricles and higher cerebrospinal fluid (CSF) lactates than CTRLs. We found a positive correlation between FDG_Cortex and FDG_Ventricles in both groups, although this was less strong in AD patients (AD: r = 0.358; p < 0.001; CTRL: r = 0.516; p = 0.003). Multivariate regression analyses showed that only older age was associated with reduced FDG_Cortex and FDG_Ventricles in CTRLs. Conversely, lower FDG_Cortex was associated with higher Qalb and higher plasma glucose levels within the AD group. Moreover, lower FDG_Ventricles and FDG_Ventricles/FDG_Cortex ratios were associated with elevated CSF lactates in this group. Stratifying AD patients by Apolipoprotein E (APOE) genotype revealed distinct patterns. In APOE ε3 homozygotes, FDG_Cortex showed no associations, while FDG_Ventricles and FDG_Ventricles/FDG_Cortex were negatively associated with CSF lactate. In APOE ε4 carriers, lower FDG_Cortex was linked to higher plasma glucose and QAlb, whereas FDG_Ventricles and FDG_Ventricles/FDG_Cortex were positively associated with CSF p-tau/Aβ42. Our findings suggest that, in patients with AD, FDG_Ventricles and the FDG_Ventricles/FDG_Cortex ratio may reflect alterations in brain metabolism and glucose extraction capacity. These parameters are differently linked with age, BBB integrity, and metabolic dysfunction (CSF lactates), according to APOE genotype.

1. Introduction

Alzheimer’s disease (AD) is characterized by the deposition of amyloid-β (Aβ) plaques and hyperphosphorylation of tau proteins, in parallel with significant impairments in neurovascular regulation [1], blood–brain barrier (BBB) integrity [2], and brain glucose metabolism [3]. Positron emission tomography with [18F]fluoro-2-deoxyglucose (18F-FDG-PET) has become a key tool in AD diagnosis, as it detects a decline in the cerebral metabolic rate of glucose—a phenomenon that can be observed even before the onset of clinical symptoms in individuals who will later develop dementia due to AD [4,5]. Importantly, the decline in glucose metabolism occurs earlier and is more pronounced than the appearance of brain atrophy [6,7]. One potential explanation for this early alteration in cerebral glucose uptake could be an abnormal delivery of glucose to the brain. Indeed, the loss of glucose transporters (GLUTs) from the BBB has been documented to occur before the onset of AD symptoms [8]. Moreover, postmortem studies on brains from patients with AD revealed that the decreased expression of GLUTs was significantly correlated with abnormal tau hyperphosphorylation, suggesting a link between impaired glucose transport across the BBB and the progression of AD pathology [9].

With its highly selective permeability, the BBB separates circulating blood from the brain’s extracellular fluid. While not all patients with AD exhibit an overtly BBB disruption, it has been hypothesized that altered receptor-mediated transport could impair not only the clearance of toxic amyloid and tau proteins [10] but also the transport of glucose from plasma to the brain, contributing to the early metabolic changes observed in AD [11].

Studies examining brain glucose uptake predominantly leverage 18F-FDG-PET imaging. Indeed, 18F-FDG emits a radioactive signal upon entry into the central nervous system—which can be detected in both the parenchyma and cerebrospinal fluid (CSF). Once inside the brain cells, 18F-FDG is phosphorylated to FDG-6-phosphate which cannot be further metabolized [12]. Thus, this method enables the measurement of glucose uptake without fully reflecting subsequent glucose metabolism. Moreover, recent evidence supports the idea that the 18F-FDG signal in CSF spaces—particularly the ventricles—may convey meaningful information about altered glucose transport and metabolism in neurodegenerative diseases. Specifically, Zhou et al. [13] demonstrated that dynamic 18F-FDG PET can quantify ventricular CSF clearance, serving as a surrogate marker for glymphatic function, particularly relevant to amyloid and tau clearance in AD.

The exact mechanisms governing neuronal glucose uptake remain poorly understood, raising critical questions about the interpretation of 18F-FDG-PET data. In the context of AD, is the cortical hypometabolism due to the loss of neuronal and glial cells, or is it linked to reduced 18F-FDG transport caused by BBB dysfunction? Moreover, the extent to which 18F-FDG uptake is influenced by alterations in glucose metabolism—potentially contributing to the pathogenesis of late-onset AD—has yet to be fully elucidated [14]. These questions should be examined in the context of Apolipoprotein E (APOE) genotype, the strongest genetic risk factor for late-onset AD. The presence of the APOE ε4 allele not only increases the risk of developing AD but is also linked to a more aggressive disease progression, with more pronounced amyloid and tau pathology [15]. Recent studies have further demonstrated that APOE ε4 affects brain cell metabolism, leading to reduced glycolytic activity and impaired mitochondrial respiration [16,17].

To attempt to address some of these issues, the present study investigated the role of novel 18F-FDG-PET parameters in late-onset AD. We evaluated both the 18F-FDG signal in the cerebrospinal fluid of ventricles (FDG_Ventricles), not directly related to cellular uptake, and the same parameter with respect to the total amount of cortical glucose uptake (FDG_Ventricles/FDG_Cortex). Moreover, we considered the relationship between these parameters, BBB permeability, and glucose metabolism, as well as the possible modulating role played by APOE genotype.

2. Materials and Methods

2.1. Patients Enrolment and Study Design

Between January 2021 and June 2024, we enrolled 362 consecutive outpatients in active follow-up at the Memory Clinic of the University Hospital “Tor Vergata” in Rome. We considered eligible for the study all patients aged 60 to 85 that had obtained a biomarker-based diagnosis of AD—defined as the presence of decreased CSF Aβ42 with or without increase CSF p-tau181.

The criteria for retrospective inclusion were as follows: (1) a complete diagnostic workup, including standardized neurological examination, laboratory testing, magnetic resonance imaging, 18F-FDG-PET scanning, neuropsychological assessment, APOE genotyping, and CSF analysis, and (2) the fulfillment of the diagnostic criteria for dementia [18] or mild cognitive impairment due to AD [19]. Specifically, the neuropsychological battery consists of the following: Mini Mental State Examination (MMSE) to assess global cognitive functions; the Rey Auditory Verbal Learning Test—immediate and delayed recall for verbal episodic memory; the Rey–Osterrieth Complex Figure Test—copy and recall for visuo-spatial memory; Raven Colored Progressive Matrices for abstract reasoning; and the Stroop test and verbal fluency test for attention and executive functions. The exclusion criteria were as follows: (1) a Hachinski scale score > 4 at baseline MRI, considered suggestive of vascular co-pathology; (2) a history of traumatic brain injury within 6 months before lumbar puncture; (3) the use of antipsychotics or antidepressants. These criteria were selected to minimize potential confounding factors that could affect brain glucose metabolism or CSF biomarker levels. Eventually, 224 patients were enrolled in this retrospective study.

Furthermore, 35 age-matched controls (CTRLs) were also enrolled in this study. Specifically, CTRLs were recruited among patients admitted to the Neurology Department of the University Hospital “Tor Vergata” in Rome between January 2021 and June 2024, whose CSF samples were collected in accordance with standard hospital practice. Upon discharge, all 35 subjects had received a diagnosis of either functional neurological disorder (n = 33) or tensive-type headache (n = 2). Residual CSF samples obtained during routine diagnostic evaluations were used for the analyses. Active infections, incidental presence of cognitive impairment, and other primary neurological conditions had been ruled out, including non-specific CSF changes such as an increased CSF cell count (>4 cells/mmc) or altered AD biomarker profile. Subjects underwent also MRI and 18F-FDG-PET scanning for diagnostic purposes. Eventually, only subjects with normal CSF and 18F-FDG-PET findings and without structural or functional brain abnormalities were selected for the CTRL group.

Written informed consent was acquired from all participants or legally authorized representatives. All procedures were performed according to the Declaration of Helsinki. The local ethical committee considered the study protocol an observational retrospective design (57.25CET2PTV).

2.2. CSF Collection and Biomarker Analysis

All lumbar punctures were performed between 8 and 10 a.m. An 8 mL CSF sample was collected for each patient in polypropylene tubes. A total of 2 mL of CSF was used for routine biochemical analysis including the calculation of the Albumin Quotient (Qalb) as the BBB permeability index—considering CSF/serum albumin—and levels of lactates. A second aliquot of 2 mL was used for CSF AD biomarkers. We used commercially available kits for biochemical analysis. CSF amyloid-β 1–42 (Aβ42), phosphorylated-tau (p-tau), and total tau (t-tau) concentrations were determined using a sandwich enzyme-linked immunosorbent assay (EUROIMMUN ELISA©, Waltham, MA, USA). Amyloid groups were defined according to EUROIMMUN guidelines: CSF Aβ42 was Aβ-positive if Aβ < 600 pg/mL or Aβ-negative if Aβ42 ≥ 600 pg/mL.

Blood samples were also drawn in EDTA tubes. The DNA was extracted automatically and APOE genotyping was conducted by allelic discrimination technology with real-time PCR, according to the manufacturer’s instructions (TaqMan; Applied Biosystems, Foster City, CA, USA).

2.3. F-FDG-PET Data

All PET scans were conducted at the Nuclear Medicine Unit of the University Hospital “Policlinico Tor Vergata” in Rome using a General Electric VCT PET/CT scanner (GE Medical Systems, Powell, TN, USA). Participants fasted for at least 5 h before the intravenous administration of 18F-FDG, and serum glucose concentrations were confirmed to be within acceptable ranges, as recommended by the European Association of Nuclear Medicine guidelines [20]. Patients received an intravenous injection of 18F-FDG (dose ranging from 185 to 295 MBq), followed by hydration with 500 mL of saline solution (0.9% sodium chloride). Imaging began 30 min post-injection and lasted for ten minutes. Detailed acquisition and reconstruction parameters were consistent with those described in the cited guidelines [20].

Brain uptake of 18F-FDG was assessed using Statistical Parametric Mapping (SPM) 12 software (Wellcome Department of Cognitive Neurology, London, UK; available at https://www.fil.ion.ucl.ac.uk/spm/software/spm12/, accessed on 1 January 2025) running on MATLAB 2022b (Mathworks, Natick, MA, USA). PET data underwent conversion from DICOM to Nifti format utilizing MRIcron software (accessible at https://www.nitrc.org/projects/mricron/, accessed on 1 January 2025), followed by normalization procedures. To minimize image intensity distortions caused by spatially varying artifacts and enhance the accuracy of automated processing, a bias regularization factor of 0.0001 was applied. Additionally, the Gaussian smoothing kernel’s full width at half maximum (FWHM) was restricted to 60 mm to prevent the algorithm from modeling intensity variations attributed to distinct tissue types. For image processing, a bias regularization factor and a smoothing kernel of an 8 mm full width at half maximum (FWHM) were selected to balance sensitivity and specificity, minimizing noise while preserving regional anatomical accuracy, consistently with standard PET imaging recommendations. The tissue probability map embedded in SPM12 was internally developed from brain 18F-FDG PET scans, including data from 285 Alzheimer’s disease patients and 121 healthy control subjects.

An affine registration with mutual information with the tissue probability maps [21] was used to achieve approximate alignment with the ICBM spatial template European brains [22]. Warping regularization was set with 1 × 5 arrays (0, 0.001, 0.5, 0.05, 0.2); smoothing (to cope with functional anatomical variability not compensated by spatial normalization and to improve signal-to-noise ratio) was set to 5 mm; and the sampling distance (encoding the approximate distance between sampled points in estimating model parameters) was set to 3.

The clusters containing the cortex, ventricles (CSF), and pons (see below) were exported by the means of the WFU PickAtlas tool implemented in SPM 12 (WFU PickAtlas (RRID:SCR_007378); available online: https://www.nitrc.org/projects/wfu_pickatlas/ (accessed on 23 April 2025)) [23] (Figure 1). Specifically, the mean signal intensities calculated from the cortex and ventricles within each subject were normalized to the average intensities of the pons volume of interest. The use of normalization based on activity in the pons, rather than the cerebellum, brainstem, or primary sensorimotor cortex, has been reported to result in greater accuracy in discriminating patients from controls in neurodegenerative diseases [20,24]. Normalization to the pons was chosen due to its demonstrated stability and robustness in AD, as metabolism in this region is relatively preserved even in early stages of neurodegeneration. Previous studies have confirmed that the pons provides reliable reference values for normalization, helping to reduce inter-subject variability and enhance sensitivity to pathological changes in cortical metabolism, which is critical for detecting subtle disease-related effects [20,25].

ROIs for ventricles specifically included lateral ventricles as predefined by the WFU PickAtlas. Cortical ROIs were defined as a composite cortical mask encompassing frontal, parietal, temporal, and occipital lobes using the anatomical label atlas in WFU PickAtlas [23]. A dataset of normalized 18F-FDG values relevant to the cluster under study was exported. To determine whether the normalized 18F-FDG values for the studied cluster were Gaussian distributed, the D’Agostino K-squared normality test was applied (with the null hypothesis being normal distribution).

2.4. Statistical Analysis

Continuous variables are presented as the mean ± standard deviation (SD) if normally distributed and as the median (Interquartile range_IQR) if not normally distributed. Categorical variables are expressed as percentages (%). Patients with AD were assigned to the APOE ε4 subgroup when carrying at least one ε4 allele (ε4/ε4 or ε3/ε4), while all the remaining patients exclusively carried APOE ε3 alleles (ε3/ε3).

Statistical differences in continuous variables between groups were tested using t tests or the Mann–Whitney U test in case of a non-normal distribution. Pearson’s X2 squared was used for categorical variables. Spearman correlation tests were performed to analyze the relationships between non-parametric variables. To compare the strength of independent correlation coefficients between groups, we used Fisher’s r-to-z transformation (http://vassarstats.net/rdiff.html, accessed on 23 April 2025). To test if factors including Qalb, p-tau/Ab42, and CSF lactates were linked to changes in the 18F-FDG-PET parameters, we performed different multivariate regression analyses. The analyses were performed in both AD and CTRL groups. We then used separate models, considering only AD patients stratified according to APOE genotype.

Statistical analysis was performed via JASP© (Version 0.18.3-Computer Software-JASP TEAM 2020, https://jasp-stats.org/, accessed on 30 April 2025) and GraphPad Prism© (Version 9.5.0, GraphPad Software, San Diego, CA, USA, www.graphpad.com, accessed on 8 April 2025).

All results were computed with two-tailed tests of significance; p-values < 0.05 were considered statistically significant.

3. Results

3.1. Participant Characteristics

Statistically significant differences in age and the prevalence of the APOE ε4 genotype were observed between the CTRL and the all AD group (

Table 1. Comparisons of demographic features and CSF dosages between CTRLs and patients with AD.

	CTRL (n = 35)		All AD (n = 224)		p
	Mean ± SD	Median (IQR)	Mean ± SD	Median (IQR)
MMSE	28.60 ± 1.22	29.00 (2.22)	22.73 ± 2.79	23.50 (3.00)	<0.001
CSF Aβ42 (pg/mL)	1028.22 ± 290.34	1005.00 (334.00)	391.57 ± 140.25	390.50 (182.20)	<0.001
CSF p-tau (pg/mL)	35.75 ± 15.19	32.35 (23.00)	84.57 ± 44.37	73.50 (53.00)	<0.001
CSF t-tau (pg/mL)	212.05 ± 133.80	169.00 (158.95)	616.98 ± 351.76	555.81 (457.65)	<0.001
CSF p-tau/Aβ42	0.036 ± 0.018	0.034 (0.020)	0.252 ± 0.201	0.200 (0.147)	<0.001
Qalb	7.21 ± 2.15	6.98 (2.42)	6.72 ± 2.87	6.05 (3.54)	<0.001
CSF lactates (pg/mL)	1.63 ± 0.40	1.50 (0.69)	1.79 ± 0.35	1.80 (0.40)	<0.001
Age (y)	67.09 ± 7.33	67.00 (11.50)	70.25 ± 6.83	71.00 (9.00)	0.085
	%		%
Sex (F)	40.0		53.5		0.007
APOE (ε4)	14.3		44.3		0.014
	mean ± SD		mean ± SD
FDGVentricles	0.94 ± 0.22		0.72 ± 0.13		<0.001
FDGCortex	1.28 ± 0.15		1.20 ± 0.11		<0.001
FDGVentricles/FDGCortex	0.71 ± 0.11		0.60 ± 0.10		<0.001

CTRL: control; AD: Alzheimer’s disease; n: number; p: p-values; SD: standard deviation; IQR: interquartile range; y: years; F: female; CSF: cerebrospinal fluid; APOE: Apolipoprotein E; Qalb: Albumin Quotient; FDG: fluorodeoxyglucose. Continuous variables are presented as mean ± SD if normally distributed and as median (IQR) if not normally distributed. Categorical variables are expressed as percentages (%). Bold p-values denote statistical significance.

). As expected, the cognitive status (MMSE) as well as all CSF biomarker levels were also significantly different between groups. Both FDG_Cortex and FDG_Ventricles were significantly lower in AD patients with respect to CTRLs, with a larger effect size for FDG_Ventricles (FDG_Cortex r = −0.362; p < 0.001; FDG_Ventricles r = −0.774; p < 0.001) (Figure 2A and 2B). Accordingly, FDG_Ventricles/FDG_Cortex was also significantly lower in AD patients (FDG_Ventricles/FDG_Cortex r = −0.679; p > 0.001) (Figure 2C). No difference was found in Qalb, but higher CSF lactate levels were observed in AD patients with respect to CTRLs (r= 0.282; p = 0.007).

Stratifying AD patients according to APOE genotype, we did not retrieve any significant difference between APOE ε3 (n = 122) and APOE ε4 (n = 97) in clinical–demographical characteristics and CSF biomarker levels, nor in 18F-FDG parameters (p > 0.05 for all comparisons) (Table S1).

3.2. Correlation Analysis Between 18F-FDG Parameters

A significant strong correlation was found between FDG_Cortex and FDG_Ventricles in the CTRL group (r = 0.516; p = 0.003). When considering AD patients, we retrieved a less strong but still significant correlation between the two parameters (r = 0.358; p < 0.001) (Figure 3). No significant differences were observed between the two correlations (z-score 1.038; p = 0.299).

3.3. Regression Analyses

We performed multivariate regression analysis to evaluate which factors among BBB integrity (Qalb), the burden of AD pathology (CSF p-tau/Aβ42), and mitochondrial dysfunction (CSF lactates) were associated with variation in FDG_Cortex, FDG_Ventricles, and FDG_Ventricles/FDG_Cortex. We added in the model possible confounding factors, namely age, sex, and plasma glucose levels (

Table 2. Multivariate regression analyses in CTRLs and AD patients.

	CTRL		All AD
	β	p	β	p
FDGCortex
Age	−0.455	0.044	−0.013	0.854
Sex	n.a.	0.532	n.a.	0.480
Plasma glucose (mg/dL)	−0.080	0.689	−0.147	0.036
Qalb	0.098	0.588	−0.158	0.033
CSF p-tau/Aβ42	0.037	0.869	−0.071	0.295
CSF lactates	0.082	0.698	0.011	0.876
Adjusted R2	0.019		0.036
FDGVentricles
Age	−0.608	0.004	−0.255	<0.001
Sex	n.a.	0.188	n.a.	0.053
Plasma glucose (mg/dL)	0.013	0.940	−0.002	0.978
Qalb	0.226	0.173	−0.067	0.333
CSF p-tau/Aβ42	0.254	0.207	0.075	0.238
CSF lactates	0.026	0.893	−0.207	0.002
Adjusted R2	0.212		0.148
FDGVentricles/FDGCortex
Age	−0.200	0.397	−0.264	<0.001
Sex	n.a.	0.481	n.a.	0.100
Plasma glucose (mg/dL)	0.081	0.707	0.082	0.213
Qalb	0.152	0.439	0.010	0.885
CSF p-tau/Aβ42	0.213	0.375	0.136	0.034
CSF lactates	−0.047	0.838	−0.214	0.002
Adjusted R2	−0.139		0.144

CTRL: control; p: p-values; FDG: fluorodeoxyglucose; CSF: cerebrospinal fluid; Qalb: Albumin Quotient. Bold values denote statistical significance.

).

In the CTRL group, both lower FDG_Cortex (β = −0.455; p = 0.044) and lower FDG_Ventricles (β = −0.608; p = 0.004) were associated with older age.

In the AD group, FDG_Cortex was negatively associated with glucose plasma levels (β = −0.147; p = 0.036) and Qalb (β = −0.158; p = 0.033). Lower FDG_Ventricles was associated with older age (β = −0.255; p < 0.001) and more CSF lactates (β = −0.207; p = 0.002). A lower FDG_Ventricles/FDG_Cortex ratio was again associated with older age (β = −0.264; p < 0.001) and more CSF lactates (β = −0.214; p = 0.002) but also with lower CSF p-tau/Aβ42 (β = 0.136; p = 0.034).

Finally, we stratified AD patients according to APOE genotype and performed separate multivariate regression analyses considering, all factors which were significantly associated with FDG_Cortex, FDG_Ventricles, and FDG_Ventricles/FDG_Cortex in the all AD group analysis (

Table 3. Multivariate regression analyses in APOE ε3 and APO ε4 patients.

	APOE ε3		APOE ε4
	β	p	β	p
FDGCortex
Age	−0.126	0.900	−0.032	0.747
Plasma glucose (mg/dL)	−0.059	0.544	−0.268	0.010
Qalb	−0.138	0.166	−0.240	0.019
CSF p-tau/Aβ42	−0.096	0.311	−0.071	0.468
CSF lactates	0.141	0.151	−0.085	0.431
Adjusted R2	−0.001		0.126
FDGVentricles
Age	−0.247	0.008	−0.289	0.002
Plasma glucose (mg/dL)	0.061	0.511	−0.080	0.395
Qalb	−0.032	0.742	−0.218	0.020
CSF p-tau/Aβ42	−0.057	0.531	0.253	0.006
CSF lactates	−0.201	0.035	−0.142	0.155
Adjusted R2	0.069		0.261
FDGVentricles/FDGCortex
Age	−0.238	0.010	−0.303	0.002
Plasma glucose (mg/dL)	0.089	0.341	0.071	0.457
Qalb	0.041	0.669	−0.120	0.202
CSF p-tau/Aβ42	0.015	0.866	0.334	<0.001
CSF lactates	−0.260	0.006	−0.118	0.243
Adjusted R2	0.086		0.248

APOE: Apolipoprotein E; p: p-values; FDG: fluorodeoxyglucose; CSF: cerebrospinal fluid; Qalb: Albumin Quotient. Bold values denote statistical significance.

).

Considering the APOE ε3 group, we did not retrieve any significant association with FDG_Cortex, but we confirmed negative associations between FDG_Ventricles and both age (β = −0.247; p = 0.008) and CSF lactates (β = −0.201; p = 0.035). Similarly, we confirmed the negative associations of both age (β = −0.238; p = 0.010) and CSF lactates (β = −0.260; p = 0.006) with FDG_Ventricles/FDG_Cortex.

In the APOE ε4 group, we retrieved negative associations between both plasma glucose (β = −0.268; p = 0.010) and Qalb (β = −0.240; p = 0.019) and FDG_Cortex. As for FDG_Ventricles, we confirmed a significant association with age (β = −0.289; p = 0.002), but we also found a significant negative association with Qalb (β = −0.218; p = 0.020) and a positive association with CSF p-tau/Aβ42 (β = 0.253; p = 0.006). No association was found between CSF lactates and FDG_Ventricles in this subgroup. Finally, the regression analysis for FDG_Ventricles/FDG_Cortex confirmed the association with age (β = −0.303; p = 0.002) and CSF p-tau/Aβ42 (β = 0.334; p < 0.001) but, again, not with CSF lactates.

4. Discussion

Our study highlights the potential utility of two newly introduced 18F-FDG-PET parameters—FDG_Ventricles and the FDG_Ventricles/FDG_Cortex ratio—in exploring metabolic dysfunctions in late-onset AD.

In our cohort, as expected, no difference was found in BBB permeability, evaluated as Qalb, between the AD and CTRL groups. While the literature does not support the notion of a fully disrupted BBB in AD, it has been shown that early BBB disfunction occurs and exacerbates protein misfolding and neurodegeneration [5]. On the other hand, our AD patients showed higher levels of CSF lactates, supporting the hypothesis that changes in energy metabolism—a hallmark of normal aging processes—may also be involved in late-onset AD [26]. Indeed, when neurons experience an increased demand for oxidative substrates, particularly lactate, astrocytes may supply this demand through their high glycolytic capacity leveraging the Astrocyte–Neuron Lactate Shuttle [27].

Considering the 18F-FDG-PET parameters, both cortical and cerebrospinal fluid 18F-FDG uptake (FDG_Cortex and FDG_Ventricles) were significantly lower in patients with AD compared to CTRLs. Reduced 18F-FDG uptake in cortical gray matter is widely recognized as a key feature of AD and a valuable marker for predicting cognitive decline [3]. The underlying concept is that neurodegeneration leads to the loss of neurons and glial cells responsible for glucose uptake or to a reduction in their metabolic efficiency. However, FDG_Ventricles does not represent cellular uptake itself. Its reduction in patients with AD may potentially correspond to a reduced glucose demand from the brain. Indeed, we found a significant correlation between the FDG_Ventricles and the FDG_Cortex levels in both CTRLs and patients, supporting the hypothesis of a physiological “brain pull effect” in which the brain’s glucose demand influences its CSF levels [28]. In this context, it has been shown that GLUT expressions at the BBB and blood–CSF barrier in the choroid plexus are strictly regulated by brain metabolic needs and can be differently modulated in several pathological conditions [29,30]. Interestingly, we noted a weaker correlation between FDG_Ventricles and FDG_Cortex in patients with AD and a more pronounced decrease in FDG_Ventricles compared to that in FDG_Cortex. As a result, the FDG_Ventricles/FDG_Cortex ratio was also significantly lower in patients with AD. This finding may suggest that in such a neurodegenerative condition, while there is a global reduction in cortical glucose uptake, there is an extra rate of glucose reduction in the CSF. This glucose could be extracted and directed to astrocytes, in an attempt to compensate for impaired neuronal energy metabolism, to enhance their glycolytic activity and boost lactate production [31]. Thus, the lactate increase that we observed in the CSF could be linked to the neuronal metabolic impairment reflected by 18F-FDG-PET hypometabolism, as has already been observed in AD pathology [32]. Alternatively, the downregulation of GLUT1 occurring in AD could be different at the cortex and the choroid plexus levels [33], explaining the discrepancy between FDG_Ventricles and FDG_Cortex. Indeed, while bioenergetic failure in aging and neurodegeneration results from increased mitochondrial dysfunction, deficiencies in glucose transport may also be involved. Recent studies in mice suggest that GLUT1 deficiency in the endothelial cells triggers BBB disruption and accelerates the progression of AD neuropathology [9], and reduced GLUT expression has also been detected at the BBB and in the cerebral cortex of patients with AD [11]. Overall, the weaker correlation between 18F-FDG uptake in the cortex and in CSF compared to that in controls could indicate that in AD, the mechanisms regulating glucose transport across the blood–brain and blood–CSF barriers are compromised, leading to a greater dysregulation of glucose homeostasis in the CSF relative to the cortex. These disruptions may reflect underlying pathological changes in the brain’s ability to meet its metabolic demands, potentially contributing to the neurodegenerative process.

To investigate which factors were associated with our 18F-FDG-PET parameters, we conducted a multivariate regression analysis considering BBB permeability (Qalb), AD-related pathology (p-tau/Aβ42), and brain metabolic state (CSF lactates) and including potential confounders like age, sex, and basal plasma glucose levels. In CTRLs, none of these factors significantly influenced the 18F-FDG-PET parameters, except for age, with older individuals showing reduced glucose uptake in both FDG_Cortex and FDG_Ventricles. In contrast, within the AD group, lower FDG_Cortex was not linked to age but rather to higher BBB permeability. In a previous study, we already observed a similar inverse relation between BBB permeability and glucose consumption in the temporal lobes of patients with AD [34]. It is possible that progressive BBB dysfunction may be associated with reduced GLUT expression, causing an impairment in glucose transport that could exacerbate cerebral metabolic dysfunction. Further, lower FDG_Ventricles was associated with increased BBB permeability, as well as with older age and elevated CSF lactates. The relationship with advanced age may be attributed to neuronal mitochondrial failure, which is common with aging and induces astrocytes to extract glucose from both blood and CSF to produce lactate. Additionally, reduced 18F-FDG uptake in the CSF could be linked to a weakened “pull effect” from the brain due to mitochondrial energy dysfunction, which is further associated with elevated CSF lactate levels. Interestingly, a lower FDG_Ventricles/FDG_Cortex ratio was also associated with older age, more CSF lactates, and also with lower AD-related pathology (p-tau/Aβ42). We hypothesize that this parameter may reflect the excess of “glucose extraction” from CSF in an attempt to support brain metabolism, an ability that could be progressively lost as the burden of AD pathology increases. These multivariate analyses may partly explain the weaker correlation between FDG_Cortex and FDG_Ventricles we found in the AD group. Indeed, while only age was associated with these parameters in the CTRLs, in patients with AD, the relationship was influenced by several additional factors, including BBB permeability, the burden of amyloid and tau pathologies, and bioenergetic dysfunctions, which may disrupt the normal metabolic coupling between the cortex and the CSF. Given the cross-sectional design of our study, no causal inferences can be drawn regarding the directionality of the observed associations—i.e., whether BBB dysfunction precedes or follows cerebral metabolic changes. However, our findings raise the hypothesis that alterations in the FDG_Ventricles/FDG_Cortex ratio may reflect early pathophysiological processes in AD. Future longitudinal studies are warranted to explore whether dynamic changes in this ratio over time are predictive of cognitive decline or disease progression.

To further explore these associations, we stratified AD patients according to APOE genotype, which highlighted the presence of distinct patterns of association. In the APOE ε3 group, no significant link was found between any of the variables and FDG_Cortex. This lack of association was reflected by the very low adjusted R² observed for this model. However, negative associations with both age and CSF lactates were confirmed for FDG_Ventricles and for the FDG_Ventricles/FDG_Cortex ratio. In contrast, in the APOE ε4 group, lower FDG_Cortex was significantly associated with higher plasma glucose levels and BBB permeability. These findings are consistent with the literature, which has demonstrated a strong link between cortical hypometabolism and insulin resistance [35], as well as a higher degree of vascular abnormalities which can impact AD pathology and cognitive decline in APOE ε4 carriers [16,36,37]. In this subgroup of patients, lower FDG_Ventricles was associated with older age, higher BBB permeability, and higher degree of AD-related pathology, while no association was observed with CSF lactates. Finally, regression analysis of the FDG_Ventricles/FDG_Cortex ratio in the APOE ε4 group confirmed significant associations with age and AD-related pathology, but again, no relationship was found with CSF lactates. Collectively, these findings suggest that APOE genotype may modulate the association between amyloid pathology, vascular changes, and brain metabolic dysfunction. Specifically, while brain glucose metabolism in APOE ε3 patients appears to be primarily associated with bioenergetic failure (e.g., elevated CSF lactates), in APOE ε4 carriers, it seems to be influenced by the combined effects of BBB dysfunction and AD-related pathology. Metabolic failure appears to be a common feature among all AD patients, but it is likely more pronounced in APOE ε4 carriers [16], in whom additional factors—such as a well-documented higher burden of amyloid and tau pathology [15], as well as greater microvascular damage accompanied by reduced GLUT expression [9]—may further exacerbate metabolic dysfunction.

A key strength of this study lies in the introduction and evaluation of novel 18F-FDG-PET parameters related to CSF glucose dynamics, which, to the best of our knowledge, have not been previously described in the literature. However, there are notable limitations to consider. CSF regions were defined using the WFU PickAtlas toolbox, which allows the identification of intraventricular spaces (lateral, third, and fourth ventricles), but does not include subarachnoid spaces. As such, our findings specifically reflect intraventricular FDG dynamics and cannot be extrapolated to extracerebral CSF compartments. Future studies employing dedicated segmentation pipelines or dynamic imaging may provide a more comprehensive assessment of global CSF glucose metabolism. Additionally, the selection of the control group, consisting of individuals with functional neurological disorder or tension-type headache, represent a potential limitation. Nonetheless, these subjects were carefully screened to exclude any abnormal CSF findings or structural/metabolic brain alterations. Additionally, the cross-sectional nature of this study limits our ability to assess causal relationships or temporal dynamics. In particular, the lack of longitudinal follow-up prevented us from determining whether the 18F-FDG-PET alterations observed were associated with a more severe trajectory of disease progression. Future longitudinal studies involving larger and more diverse populations, such as those with preclinical AD, as well as consideration of other physiological variables (e.g., diet, physical activity) that may influence brain bioenergetics and glucose transport, will be crucial to validate and expand upon these findings.

5. Conclusions

Overall, our findings suggest that FDG_Ventricles and the FDG_Ventricles/FDG_Cortex ratio may reflect both metabolic dysfunction and altered glucose dynamics in patients with AD. Furthermore, these measures are associated with key factors such as age, BBB integrity, and mitochondrial dysfunction, with distinct patterns observed across different APOE genotypes.

While our cross-sectional design does not allow for causal inference, these associations are consistent with the hypothesis that bioenergetic imbalance plays a role in the pathophysiology of AD. If validated in longitudinal studies, biomarkers sensitive to early metabolic shifts could contribute to identifying individuals at increased risk of AD. Moreover, given that several therapeutic strategies for AD aim to target metabolic dysfunction, such parameters might prove useful in identifying treatment-responsive individuals and in monitoring therapeutic efficacy.