Functional Correlates of Microglial and Astrocytic Activity in Symptomatic Sporadic Alzheimer’s Disease: A CSF/18F-FDG-PET Study

Glial and microglial cells contribute to brain glucose consumption and could actively participate in shaping patterns of brain hypometabolism. Here, we aimed to investigate the association between 18F-fluorodeoxyglucose (18F-FDG) uptake and markers of microglial and astrocytic activity in a cohort of patients with Alzheimer’s Disease (AD). We dosed cerebrospinal fluid (CSF) levels of soluble Triggering Receptor Expressed on Myeloid cells (sTREM2), Glial Fibrillary Acidic Protein (GFAP), a marker of reactive astrogliosis, and β-S100, a calcium-binding protein associated with a neurotoxic astrocytic profile. No associations were found between sTREM-2 and 18F-FDG uptake. Instead, 18F-FDG uptake was associated negatively with CSF β-S100 in the left supramarginal gyrus, inferior parietal lobe and middle temporal gyrus (Brodmann Areas (BA) 21 and 40). Increased β-S100 levels could negatively regulate neuronal activity in the temporo-parietal cortex to prevent damage associated with AD hyperactivity, or rather they could reflect neurotoxic astrocytic activation contributing to AD progression in key strategic areas. We also identified a trend of positive association of 18F-FDG uptake with CSF GFAP in the right fronto-medial and precentral gyri (BA 6, 9 and 11), which has been reported in early AD and could either be persisting as an epiphenomenon tied to disease progression or be specifically aimed at preserving functions in the frontal cortex. Overall, CSF markers of astrogliosis seem to correlate with cortical glucose uptake in symptomatic sporadic AD, highlighting the role of astrocytes in shaping regional hypometabolism and possibly clinical presentation.


Introduction
Alzheimer's Disease (AD) is marked by the accumulation of amyloid-β (Aβ) peptides, which according to the amyloid hypothesis triggers tau-hyperphosphorylation and its intraneuronal deposition [1]. Despite these two pathological changes being the main actors of AD, it is widely recognized that many transversal mechanisms participate in its pathophysiology, acting alongside Aβ and tau pathology [2]. For instance, AD is often complicated by comorbidities, including diabetes and other metabolic conditions, which further increase the complexity of disease mechanisms [3,4].
Specifically, unravelling the interplay between cellular neuroinflammatory contributionsi.e., microglial activation and astrocytic reactivity-amyloid pathology and tauopathy has been the object of many efforts [5][6][7], because of their interesting and possibly ambivalent role in the progression of the disease. We obtained written consent from all participants and/or legally authorized representatives. The ethical committee of the Santa Lucia Foundation accounted for the study protocol as an observational retrospective design (Prot. CE/AG4/PROG.392-08).

CSF Sampling and Laboratory Analysis
All lumbar punctures were performed with sterile technique between 8 and 10 am. A 10 mL CSF sample was collected for each patient in polypropylene tubes. A total of 2 mL was used for routine biochemical analysis and we centrifuged the remaining 8 mL at 2000× g at +4 • C for 10 min and aliquoted them in 1 mL portions. The aliquots were frozen at −80 • C for further analysis of CSF, AD, and glial biomarkers. We used commercially available kits for biochemical analysis. CSF Aβ42, p-tau and t-tau concentrations were determined using a sandwich enzyme-linked immunosorbent assay (EUROIMMUN ELISA©), LUMINEX© Multiple assays ELISA was used for CSF GFAP and β-S100 concentrations.
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).

18 F-FDG-PET Data Acquisition
All acquisitions were performed at the Nuclear Medicine Unit of the University Hospital "Policlinico Tor Vergata" in Rome with a General Electric VCT PET/CT scanner (GE Medical Systems, Tennessee, USA). All subjects fasted for at least 5 h before i.v. injection of FDG, and serum glucose levels were in range according to European Association of Nuclear Medicine guidelines [36]. Patients were injected intravenously with 18 F-FDG (dose range 185-295 MBq) and then hydrated with 500 mL of saline (0.9% sodium chloride), according to a previous similar report of our group in this field [37]. The scan started 30 min after the injection and lasted ten minutes. Acquisition and reconstruction parameters are reported elsewhere and were followed according to the report cited above. [37].
Upon imaging evaluation, all 27 subjects showed typical findings of brain cortical hypometabolism in key regions compatible with a diagnosis of classical AD (parietal and posterior cingulate cortices, precuneus, or a combination of the above) [38].
PET data were converted from DICOM to Nifti format using the MRIcron software (available at https://www.nitrc.org/projects/mricron/, accessed on 1 January 2023) and then subjected to normalization. Bias regularization was applied (0.0001) to limit distortions due to smooth, spatially varying artifacts that can modulate image intensity and interfere with automated image processing. The FWHM of Gaussian smoothness of distortion (to prevent the algorithm from trying to model out intensity variations due to different tissue types) was set to a limit of 60 mm; the tissue probability map implemented in SPM12 was used (TPM.nii). An affine registration with mutual information with the tissue probability maps [39] was used to achieve approximate alignment with the ICBM spatial template-European brains [40,41]. Warping regularization was set with the following 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; sampling distance (encoding the approximate distance between sampled points in estimating model parameters) was set to 3.
We used an 8 mm isotropic Gaussian filter to blur individual variations (especially gyral variations) and increase the signal-to-noise ratio. Prior to regression analysis, the following parameters and post-processing tools were used: global normalization (which brings the images to a global value) = 50 (using proportional scaling); masking threshold (which helps identify voxels with acceptable signal in them) was set to 0.8; transformation tool of statistical parametric maps to normal distribution; correction of SPM coordinates to match Talairach coordinates, subroutine implemented by Matthew Brett (http://www. mrc-cbu.cam.ac.uk/Imaging, accessed on 1 January 2023). Brodmann surfaces (BA) were determined within a range of 0 to 3 mm from the corrected Talairach coordinates of the SPM output isocenter using a Talairach client (available at http://www.talairach.org/index.html, accessed on 1 January 2023). As suggested by Bennett et al. [42], sPM t-maps were calculated for multiple comparisons using the false discovery (p < 0.05) and corrected for multiple comparisons at the cluster level (p < 0.001).
The level of significance was set at 100 (5 × 5 × 5 voxels, i.e., 11 × 11 × 11 mm) contiguous voxels. The voxel-based analyses were performed in the subgroup of patients using regression analyses. We designed independent models assessing the effects of each CSF biomarker (levels of GFAP, sTREM-2 and βS100) as regression factors-independent variables-on cortical 18 F-FDG uptake, using age and sex as covariates, as well as t-tau levels in order to adjust for another marker of neurodegeneration. Positive and negative associations were tested in all cases.
The cluster obtained in the regression analysis was then exported by means of the WFU Pickatlas tool implemented in SPM 12. Specifically, the mean signal intensities calculated from each cluster 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 whole-brain counts as in the reference region, has been reported to result in greater accuracy in discriminating patients from controls in neurodegenerative diseases. As previously suggested by Pagani and colleagues [43]), a dataset of normalized 18 F-FDG values relevant to the cluster under study was exported. To determine whether the normalized 18 F-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). We then performed linear univariate regression analyses using GraphPad Prism© version 9.3.1 for Windows (GraphPad Software, San Diego, California USA, www.graphpad.com, accessed on 1 January 2023) to visualize significant findings, namely the association between CSF glial biomarkers and normalized continuous data representing metabolism in the specific Broadman Areas of interest.

Results
First, we retrieved a trend of positive association between 18 F-FDG uptake and CSF levels of GFAP in the right frontal-medial and precentral gyri (see Figure 1), with peaks in BA 6, 9 and 11 [see Table 2].
ations were tested in all cases.
The cluster obtained in the regression analysis was then exported by means of the WFU Pickatlas tool implemented in SPM 12. Specifically, the mean signal intensities calculated from each cluster 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 whole-brain counts as in the reference region, has been reported to result in greater accuracy in discriminating patients from controls in neurodegenerative diseases. As previously suggested by Pagani and colleagues [43]), a dataset of normalized 18 F-FDG values relevant to the cluster under study was exported. To determine whether the normalized 18 F-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). We then performed linear univariate regression analyses using GraphPad Prism© version 9.3.1 for Windows (GraphPad Software, San Diego, California USA, www.graphpad.com, accessed on 1 January 2023) to visualize significant findings, namely the association between CSF glial biomarkers and normalized continuous data representing metabolism in the specific Broadman Areas of interest.

Results
First, we retrieved a trend of positive association between 18 F-FDG uptake and CSF levels of GFAP in the right frontal-medial and precentral gyri (see Figure 1), with peaks in BA 6, 9 and 11 [see Table 2].  Table 2).   Table 2). In the Cluster level section (left), the number of voxels, the uncorrected and corrected p-value of significance and the cortical region where the voxel is found are reported for each significant cluster. In the Voxel level section (right), all coordinates of the correlation sites (with z-score of the maximum correlation point) and the corresponding subcortical region are reported for each significant cluster. L: left; R: right; FEW: Familywise error; FDR: False discovery rate; BA: Broadman Area; p unc: uncorrected p, * = p < 0.05. Moreover, a significant negative relationship between brain glucose consumption and CSF levels of β-S100 was also found in a wide cluster that included the left supramarginal gyrus, the inferior parietal lobe and middle temporal gyrus (see Figure 2), with peaks in Brodmann Areas 21 and 40 [see Table 3]. We did not find any significant relationships between 18 F-FDG uptake and CSF levels of sTREM-2.
unc: uncorrected p, * = p < 0.05. Moreover, a significant negative relationship between brain glucose consumption and CSF levels of β-S100 was also found in a wide cluster that included the left supramarginal gyrus, the inferior parietal lobe and middle temporal gyrus (see Figure 2), with peaks in Brodmann Areas 21 and 40 [see Table 3]. We did not find any significant relationships between 18 F-FDG uptake and CSF levels of sTREM-2.

Figure 2.
Three-dimensional brain rendering showing clusters obtained in SPM regression analysis for the negative correlation between CSF β-S100 and brain 18 F-FDG uptake (see Table 3). Table 3. Multiple regression analysis showing the CSF β-S100 related areas of decreased 18 F-FDG brain uptake (negative correlation), corrected for age, sex and CSF levels of t-tau.

Analysis Cluster Level Voxel Level
Negative correlation (β-S100) Lastly, we performed linear regression analyses to assess the association between glial CSF biomarkers and metabolism in the specific Broadman Areas of interest. Figure 3a shows regressions between β-S100 and 18    Three-dimensional brain rendering showing clusters obtained in SPM regression analysis for the negative correlation between CSF β-S100 and brain 18 F-FDG uptake (see Table 3). Table 3. Multiple regression analysis showing the CSF β-S100 related areas of decreased 18 F-FDG brain uptake (negative correlation), corrected for age, sex and CSF levels of t-tau.

Cluster p (FDR-corr)
Cluster extent Cortical region z-score of maximum Talairach  Lastly, we performed linear regression analyses to assess the association between glial CSF biomarkers and metabolism in the specific Broadman Areas of interest. Figure 3a shows regressions between β-S100 and 18   . Scatter plots demonstrating the linear relationship between CSF β-S100 and 18 F-FDG brain uptake in the left BA40 (blue), left BA21 (black), and the enlarged hub in the homolateral parietal lobe (PL, pink) (a); the relationship between CSF GFAP and 18 F-FDG brain uptake in the right BA9 (orange), BA6 (blue) and BA11 (grey) (b).

Discussion
Despite its widespread use in both clinical settings and basic research, the identity of the cell types contributing to 18 F-FDG-PET signal is still debated. Traditionally, 18 F-FDG-PET signal has been attributed only to neuronal uptake, with hypometabolism being con-

Discussion
Despite its widespread use in both clinical settings and basic research, the identity of the cell types contributing to 18 F-FDG-PET signal is still debated. Traditionally, 18 F-FDG-PET signal has been attributed only to neuronal uptake, with hypometabolism being considered as a direct index of neuronal dysfunction-loss of neuropil, synapse, or functional impairment-or death. However, evaluating glucose consumption in AD is made complex by other additional factors, including changes in the expression of glucose transporters (GLUT) (e.g., a reduction of GLUT-3), insulin/insulin-growth-factor 1 axis dysregulation and defects due to the disruption of the neurovascular unit [44][45][46]. Moreover, other metabolically active cells are thought to contribute to glucose consumption in the brain [26,30].
Firstly, we observed a strong negative association between levels of CSF β-S100 and 18 F-FDG cortical uptake in a wide cluster including the left supramarginal gyrus, the parietal lobe, and the middle temporal gyrus, with peaks in Brodmann Areas 21 and 40. To our knowledge this association is a novel finding, since relationships between this component, CSF β-S100, and cortical metabolism have never been described in symptomatic AD. Of note, one previous work reported no association between these two variables in a group of around 89 subjects (of which 25 A+T+) from the ALFA+ cohort, which includes presymptomatic and very early symptomatic patients with AD [31]. Thus, this result seems to be tightly bound to disease mechanisms that apply to late stages of the disease, after the onset of cognitive decline.
As a calcium binding protein, CSF β-S100 plays an active role in excitatory neurotransmission, as its exogenous form increases calcium concentrations in both cultured neurons and astrocytes [47]. By affecting intracellular Ca2+-dependent processes, extracellular β-S100 modulates synaptic plasticity, especially long term potentiation [48], and is linked to both hippocampal and non-hippocampal cognitive symptoms of AD [49].
Extensive data confirms that AD is marked by the presence of excitatory/inhibitory imbalance supported by Aβ-induced hyperexcitability [50,51] and built on pyramidal neuron hyperexcitability-that is also unsupervised due to GABAergic dysfunction [52]-and inhibition of glutamate reuptake [53]. Indeed, both AD-derived astrocytes and neurons cultured with AD astrocyte conditioned medium show aberrant intracellular Ca2+ dynamics in response to glutamate [54]. In light of this, speculations can be made that the increase of CSF β-S100 could reflect an attempt to negatively regulate regional neuronal activity, with the beneficial effect of preventing the neuronal damage associated with hyperactivity, which would result in decreased cortical metabolism and reduced 18 F-FDG uptake. Alternatively, increased β-S100 could account for the local presence of pro-apoptotic neurotoxic astrocytes, favoring disease progression in areas that are typically involved in AD. In either scenario, our findings seem to point to the presence of increased regional vulnerability and/or incipient local neurodegeneration tied to astrocytic activation, with the increase of β-S100 acting either as a partial buffer or inducing neurotoxicity.
Moreover, since the association has been retrieved in key strategic areas of the dominant hemisphere, such as left BA40 and BA21, it could have repercussions on clinical presentation. Both areas belong to linguistic hubs, and are involved with language comprehension, semantic processing, and sentence generation (BA21, middle temporal gyrus) and with phonological abilities and verbal creativity (BA40, inferior parietal lobe, supramarginal gyrus) [55]. BA40 also plays a key role in several other functions, among which are working memory, motor functions (executive control of behavior and imitation, visuomotor transformation/motor planning), deductive reasoning, social perception and empathy [56]. This implies the need for a better understanding of the relationship between CSF levels of β-S100 and cognitive profile, to explore its possible role as a specific biomarker of these dysfunctions. Indeed, subtle impairment in both comprehension and strategic thinking, as well as all other cited cognitive domains, are often underrecognized in AD. Moreover, the identification of specific regional hypometabolic changes/hypoactivity suggests that targeted therapeutic approaches, such as the restoring of excitatory/inhibitory balance, could be beneficial to these symptoms.
Our second finding was the trend of positive association between CSF GFAP levels and glucose consumption in the right medial frontal gyrus, with maximum z-score of correlation peaking in BAs 9, 6 and 11. Since astrocytic reactivity is characterized by both morphological changes-resulting in cellular hypertrophy-and upregulation of GFAP expression and production, our finding supports the presence of active astrogliosis and higher metabolic demands in those areas of the frontal lobes. Previous studies focused on early AD, reported the presence of the same positive association between CSF GFAP and regional metabolism in both the frontal and the temporo-parietal lobes [32]. The loss of this linear association in the temporo-parietal and left frontal regions, which are among the first hubs to be damaged in AD [57], could be due to disease progression in our cohort of patients with symptomatic AD. Conversely, the presence of GFAP-related astrogliosis in the frontal lobe could either be an epiphenomenon resulting from astrocytic dysfunction in other areas-incapable of sustaining higher metabolic demands-or rather be specifically aimed at preserving functions in the frontal cortex in the course of symptomatic AD, supporting the hypothesis of its delayed involvement.
Interestingly, a previous work reported that higher CSF t-tau levels correlate with hypometabolism in the right frontal cortex [58], and the concordance between these and our findings suggests that the frontal lobe is a hot spot for the progression of AD. Astrocytic reactivity might sustain but also be exacerbated by tau/neurodegenerative changes in the frontal lobe, with increased GFAP production and higher metabolic demands possibly trying to sustain damaged circuitry.
In a work from Salvadó et al. (2022) [32], the positive GFAP/glucose consumption association turned negative when switching from isolated amyloidopathy (A+T-) to patients with full-blown AD (hence, with tauopathy: A+T+), which they interpreted as an uncoupling of astrogliosis from metabolism due to failure to sustain elevated energetic demands when tauopathy sets in. Our results suggest that the positive association between CSF GFAP and glucose uptake could uncouple in regions with higher damage but could also be partially maintained in the frontal lobes in symptomatic AD, even in the presence of tau pathology. In light of this, it would be very interesting to explore stage-dependent glucose metabolic changes associated with reactive astrogliosis separating symptomatic A+T-from A+T+, to solve this discrepancy.
Finally, in our cohort, we did not retrieve any significant correlation between CSF sTREM-2 and cortical glucose uptake. This finding contrasts with previous literature showing a link between sTREM-2 and FDG-PET signal in AD patients with Mild Cognitive Impairment (MCI). Biel and colleagues reported that sTREM-2 positively associates with FDG-PET hypermetabolism in patients with CSF findings of amyloidopathy but negative amyloid-PET, while, in case of amyloid-PET positivity-reflecting higher levels of fibrillary Aβ-sTREM-2 associates with hypometabolism instead [33]. Nevertheless, this discrepancy could be due to our limited sample size and to the different clinical and AT profiles of our cohort.
We are aware that this study has limitations. First, our study design is observational and cross-sectional, therefore no assumptions on direct causal relationships can be made, and it is not possible to completely decipher the mechanisms underlying these associations. Also, widening the study cohort could be useful to confirm and to strengthen our results, by repeating the assessments in A+T-and A+T+ patients. In addition, a larger sample size would allow the assessment of direct repercussions on specific cognitive domains.
Nonetheless, our results open several interesting future directions. The difference between our own findings and previous literature on cognitively unimpaired patients with AD stimulates the urge for follow-up longitudinal data, to add meaningful information on the burden of glial and microglial contributions on disease progression. Since it seems to be more sensitive to amyloid changes and have widespread correlation with cortical hypometabolism in early AD [32], it would also be interesting to use plasma GFAP instead of its CSF counterpart. Lastly, implementing the use of novel PET radiotracers such as TSPO for microglial activation could add more information on the bond between microglial inflammation and reactive astrogliosis.

Conclusions
Our results suggest that astrogliosis (CSF GFAP) in the frontal lobe is associated with higher local metabolic demands, while a negative association of CSF β-S100 with cortical metabolism was found in the parieto-temporal lobe, likely reflecting regional vulnerability to incipient damage. No microglial involvement with cortical glucose uptake has been identified in our study.
Overall, these findings represent glucose metabolic changes associated with reactive astrogliosis, adding evidence to the role of astrocytes in shaping 18 F-FDG-PET signal in vivo. This also fuels the need to implement measures of glial and microglial activity in the stratification and profiling of patients, to allow a tailored understanding of pathophysiology and to better account for their contribution in the biological evolution of AD.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Santa Lucia Foundation (Prot. CE/AG4/PROG.392-08).

Informed Consent Statement:
We obtained written consent from all participants and/or legally authorized representatives. Data Availability Statement: Data are available upon reasonable request.

Conflicts of Interest:
The authors declare no conflict of interest.