Reactive astrogliosis is associated with higher cerebral glucose consumption in the early Alzheimer’s continuum

Glial activation is one of the early mechanisms altered in Alzheimer’s disease (AD), probably in response to amyloid-β (Aβ) deposition [1‐5]. Contrary to what was previously thought, glial activation has emerged as an active contributor to AD evolution [6], which may influence not only the propagation of the main pathological hallmarks but also the clinical evolution of cognitive disturbances in patients [7, 8]. Reactive astrocytes can affect several different phenotypes and trigger morphological, functional and molecular changes [9, 10]. This process, named reactive astrogliosis, can be investigated with fluid biomarkers such as glial fibrillary acidic protein (GFAP), S100B, and chitinase-3-like protein 1 (YKL-40) (see for a review [11]).

GFAP is an astrocytic intermediate filament protein, mainly expressed in the brain (https://​www.​proteinatlas.​org/​ENSG00000131095-GFAP/​tissue), and variations in its immunoreactivity reflect changes in astrocyte cytoskeleton [10, 11]. GFAP can be measured in CSF and, more recently, also in blood, which allows more convenient assessments. Several studies in symptomatic AD showed an increase in CSF or blood GFAP [12‐17], although some studies did not find them [18, 19]. Plasma GFAP has showed interesting characteristics compared to its CSF counterpart, as it is elevated already in preclinical AD [20, 21]. In a previous multi-centric study, we showed that plasma GFAP has an early increase in the preclinical stage of Alzheimer’s, which appears to be driven by Aβ pathology [22]. Further, plasma GFAP has also been observed to be elevated in other neurological conditions presumably as a sign of astrocytic response to brain injury [23‐25].

[¹⁸F]fluorodeoxyglucose ([¹⁸F]FDG) is a glucose analog labeled with a positron emitter isotope (¹⁸F) that allows measurement of regional cerebral glucose consumption using positron emission tomography (PET). In symptomatic stages of AD, [¹⁸F]FDG uptake is typically reduced in temporo-parietal brain regions, probably reflecting, at least in part, synaptic dysfunction and neuronal loss [26, 27]. As such, it is considered to be a well-established marker of neurodegeneration in AD [28]. Conversely, higher [¹⁸F]FDG uptake has been reported in cognitively unimpaired individuals with positive AD biomarkers [29‐31], possibly related to neuroinflammatory processes [32]. More specifically, we have shown that higher [¹⁸F]FDG uptake in preclinical AD is associated with higher levels of core AD biomarkers and other markers related to neuroinflammation [31].

However, the specific source of the [¹⁸F]FDG-PET signal has been the focus of extensive debate. Traditionally, it has been only attributed to neuronal uptake, with hypometabolism being considered as a direct index of neuronal dysfunction or death. But other brain cells are also metabolically active, including astrocytes. Based on the astrocyte-neuron lactate shuttle hypothesis, the activation of the glutamate transporter 1 (GLT-1) acts as a trigger for glucose uptake by astrocytes. This hypothesis postulates that, under physiologic conditions, neurons use glucose for the production of ATP. At the same time, in astrocytes, glucose is predominantly used to produce lactate, which is then released and taken up by neurons as an additional source of energy [33, 34]. As consequence, astrocytes might account for a substantial proportion of cerebral glucose utilization [35‐38]. In this line, in two studies with rodents, the pharmacological modulation of the astrocytic GLT-1 activity was linked to the [¹⁸F]FDG PET signal, suggesting a direct association between astrocytic activity and [¹⁸F]FDG uptake [39, 40]. Nonetheless, the interpretation of these results needs further confirmation, especially on its translation to humans (see [41] for a comment).

For this reason, the investigation of the association between astrogliosis and glucose consumption in humans is relevant at this point. Thus, in this work, we studied the associations between plasma and CSF GFAP, markers of astroglial reactivity, with [¹⁸F]FDG uptake in the brain of cognitively unimpaired individuals, many of them early in the Alzheimer’s continuum. As it has been hypothesized that the evolution of glial biomarkers is not monotonic in the course of AD [42, 43], we also investigated interactions as a function of AT stages [44].

A total of 314 participants were included in this study. Their basic demographic characteristics are included in Table 1. In brief, the mean age was 61.1 years old, 62.4% were women, and 54.1% were APOE-ε4 carriers. Regarding the biomarker-based AT classification, 202 participants were A-T-, 88 were A + T-, and 24 were A + T + . These AT groups were significantly differed in age, percentage of APOE-ε4 carriers and CL values, showing higher levels in more advanced AT stages (A + T +  > A + T- > A-T-).

We first compared plasma and CSF levels of GFAP. Plasma GFAP and CSF GFAP were positively correlated (β [95% confidence interval (95%CI)]: 0.40 [0.32,0.49], p < 0.001; Fig. 1A). Afterwards, we compared these two biomarkers taking into account the AT stage. We observed that whereas A-T- and A + T- participants showed a positive correlation between plasma and CSF GFAP levels, A + T + participants did not (Fig. 1B). This resulted in a significant difference in the slope between T- (i.e., A-T- and A + T-) and T + (i.e., A + T +) participants (β [95%CI]: − 0.32 [− 0.57, − 0.06], p = 0.039).

Plasma GFAP (β_plasma [95%CI]: 0.19 [0.09, 0.29], p = 0.002), but not CSF GFAP (β_CSF [95%CI]: 0.08 [− 0.02, 0.18], p = 0.165), correlated positively with [¹⁸F]FDG uptake in the meta-ROI (Fig. 2, Table 2). Sensitivity analyses showed that these associations were not significantly changed when we include CSF Aβ42/40 or CSF Aβ42/40 and CSF p-tau in the models as covariates (Supplementary Table S2 and Figure S2). We also observe that results did not significantly change when we only included subjects with a time difference between [¹⁸F]FDG PET and GFAP biomarkers drawing equal or below six months (Supplementary Table S3).

At the voxel level, the correlation between plasma GFAP and [¹⁸F]FDG uptake was observed in almost all cortical brain regions and was absent only in medial temporal and some occipital regions (Fig. 3A). CSF GFAP also correlated positively with [¹⁸F]FDG uptake but in less spread areas, like temporoparietal areas, including the bilateral temporal poles (Fig. 3B). No negative associations were observed for any of the GFAP biomarkers.

We observed that the interaction between T- and T + participants was significant in the meta-ROI with CSF GFAP but not with plasma GFAP (Fig. 4 and Table 2) so that in T + participants, the association between GFAP biomarkers and [¹⁸F]FDG uptake became negative in these areas. In the voxel-wise analysis, this interaction was significant in left temporo-parietal areas and bilateral thalamus with plasma GFAP with a similar behavior (Fig. 5A). This significant interaction was also observed with CSF GFAP in temporo-parietal areas, although more widespread, including medial regions such as precuneus and posterior cingulate (Fig. 5B). Only one representative cluster scatter plot is shown in Fig. 5 per biomarker, but we present all the others in the Supplementary Material (Fig. S3). The inverse contrast did not present any significant result with any of the GFAP markers. Additionally, we present mean [¹⁸F]FDG uptake maps for each of the AT groups in the Supplementary Material (Fig. S4).

The main finding of this study is that higher plasma GFAP is strongly associated with higher global cerebral glucose consumption early in the Alzheimer’s continuum. Interestingly, this association was more widespread than the one found for its CSF counterpart. The regional cerebral pattern of the association with plasma GFAP included, but was not restricted to, areas of early amyloid deposition, such as the posterior and anterior cingulate cortices. Importantly, these associations were independent of Aβ pathology, as measured by CSF Aβ42/40, and tau pathology, as measured by CSF p-tau. Nonetheless, in specific areas showing hypometabolism in AD [26], such as lateral parietal and middle frontal lobe, this positive association between plasma GFAP and [¹⁸F]FDG uptake occurred in the group of individuals who were still tau-negative (T-) but the correlation turned negative in the group of tau-positive (T +) individuals. To our knowledge, this is the first report of associations between plasma GFAP and [¹⁸F]FDG PET in the Alzheimer’s continuum. Our findings suggest that higher astrocytic reactivity, probably in response to early AD pathological changes, is related to significantly higher glucose consumption. However, the association between astrocytic reactivity and glucose consumption seems to uncouple with the onset of tau pathology, maybe due to a failure in sustaining such elevated energetic demands.

Studying the relationship between metabolic demand and astrocytic reactivity in early preclinical AD stages is very relevant because, under physiological conditions, the brain accounts for 20–25% of overall glucose-derived energy [33, 53], half of which is presumably consumed by astrocytes [35‐37]. Such energetic consumption is continuous, and its deprivation, even for short durations, can result in neuronal damage and death [54]. Therefore, failure to sustain this metabolic demand already in preclinical AD stages may have important deleterious consequences later in the course of the disease.

Previous literature examined astrocytic reactivity and glucose consumption in sporadic and autosomal dominant AD patients with [¹¹C]deuterium-L-deprenyl ([¹¹C]DED) PET. [¹¹C]DED is a PET radiotracer that has been used to visualize reactive astrocytes and that colocalizes with GFAP immunohistochemistry in autoradiographic post mortem studies [55]. In these reports, higher [¹¹C]DED binding was found in presymptomatic mutation carriers and sporadic Aβ + MCI, but not in sporadic AD dementia patients [56, 57]. This result was suggestive of an early increase of astrocytosis in presymptomatic mutation carriers that decreased in more advanced stages. While no associations were found between [¹¹C]DED and [¹⁸F]FDG uptake at baseline, longitudinal follow-up revealed that uptake of both tracers declined as the mutation carriers converted from presymptomatic to prodromal AD, especially in the precuneus as well as in frontal and cingulate cortices [58]. This suggests that the non-monotonic astrocytic response to early AD pathological changes had a metabolic correlation [59]. Our findings are in close agreement with these previous studies with [¹¹C]DED and [¹⁸F]FDG PET, showing also a direct link between astrocytic response and metabolism. However, the absence of Aβ + cognitively unimpaired participants in these previous studies prevented the extrapolation of their findings to the sporadic preclinical Alzheimer’s continuum.

In this regard, the main contribution of our study to the existing literature is the confirmation of an association between reactive astrogliosis and cerebral glucose consumption early in the Alzheimer’s continuum that, as shown here, is inverted in participants with tau pathology in preclinical stages in specific regions like parietal and middle frontal lobe. This change of behavior may be related to a phenotypic change in astrocytes under pathological conditions. In a physiological situation, astrocytes significantly contribute to [¹⁸F]FDG uptake by the activation of astroglial glutamate transport [39]. However, in the presence of tau pathology, reactive astrocytes expressing higher levels of GFAP consistently express lower levels of genes associated with glutamate/GABA homeostasis [60]. In turn, a breakdown of normal astrocyte function leads to impaired clearance of amyloid and proinflammatory cytokine release, which eventually might lead to neurodegeneration [6]. In this regard, it is worth noting that the negative association between [¹⁸F]FDG and CSF GFAP, and in a lesser extent plasma GFAP, in the A + T + group was found in typical areas of early AD-related hypometabolism, such as the precuneus and parieto-temporal areas [26]. Therefore, it could be thought that such stage-dependent glucose metabolic changes associated with reactive astrocytes might contribute to the hypometabolism observed in clinical AD stages, which is thought to reflect synapse loss and neurodegeneration.

Interestingly, we also found this change of behavior in the thalami when looking at the association between plasma GFAP and [¹⁸F]FDG. Although this is not a typical AD region, we and others have previously found some metabolic alterations in the thalami in the early stages of the Alzheimer’s continuum. For instance, Johnson and colleagues reported increased [¹⁸F]FDG uptake in the thalami in cognitively unimpaired Aβ + subjects and participants with intermediate levels of amyloid-β compared to Aβ- participants [61]. In a more recent study, we also found increased glucose consumption in cognitively unimpaired participants, which we were able to relate to increased CSF core AD biomarkers as well as increased CSF markers of neuroinflammation, including GFAP and YKL-40 [62]. Thus, it seems that glucose consumption in the thalami may be especially altered in these early stages of AD, presumably due to an increase of astrocytic reactivity in relation to early amyloid-β and tau pathologies.

A surprising finding was that the association with [¹⁸F]FDG was more widespread with plasma than with CSF GFAP and only significant in AD-related regions in the case of plasma GFAP. In addition, we and others have shown that plasma GFAP changes with Aβ pathology were stronger with plasma than CSF GFAP [21, 22]. Therefore, our findings also support the idea that plasma GFAP may be an earlier marker than its CSF counterpart. The question that then arises is whether plasma and CSF GFAP come, at least partially, from different sources. The fact that plasma and CSF GFAP have a significant but mild correlation also favors this idea. A preferential release of GFAP to blood flow in response to Aβ pathology might be explained since astrocytes, as part of the blood brain barrier (BBB), have end-feet positioned at the intraparenchymal capillaries that allow them to regulate the delivery of oxygen and glucose to neurons [33, 54]. It could be argued that GFAP found in blood better reflects early BBB dysfunction in response to Aβ pathology than CSF GFAP through direct release of the protein from these end-feet into the bloodstream. This is also supported by the rapid release of GFAP in blood after other types of BBB dysfunction such as traumatic brain injury [63].

It should also be noted that in another recent study with the same participants, we also investigated the association between [¹⁸F]FDG uptake and CSF YKL-40, another marker related to astrogliosis [62]. As previously stated, in that study, we found that a combination of increased core and neuroinflammation CSF markers were related to increased glucose metabolism in thalami but also in the striatum, and the inferior and transverse temporal gyri. Interestingly, this combination of increased CSF markers was also related to larger grey matter volumes in bilateral insula, bilateral inferior temporal, bilateral supramarginal, and inferior occipital cortices, which suggests that inflammation processes related to early amyloid and tau alterations have an effect not only in brain metabolism but also in its structure. Altogether, it warrants for further investigation in other cohorts to try to decipher the specific relationships between astrocytic reactivity and metabolic and structural brain alterations in early stages of the Alzheimer’s continuum.

Our study is observational and cross-sectional. Therefore, we cannot disentangle any causal relationships. It could be thought that astrocytes react to early Aβ dysmetabolism, but others have postulated that astrocytic metabolic dysfunction might trigger the amyloidogenic processing of the amyloid precursor protein [64]. Still, our findings support conducting longitudinal and interventional studies to better understand the causal inter-relationships among these alterations early in the Alzheimer’s continuum. A better understanding of the pathophysiological pathways affected in preclinical stages is fundamental for the rational design of preventive interventions. Another limitation in our study is that we assume GFAP to be a marker of reactive astrocytes. In this regard, it must be noted that GFAP is not exclusive to astrocytes in the central nervous system and that not all astrocytes produce GFAP [65]. Still, GFAP is the best well-established and commonly used astrocytic marker in the literature [11]. Finally, since astrocytes are not the only metabolically active non-neuronal cells in the brain, it could be thought that the observed increase in glucose consumption is driven by other processes such as microglial activation. In this regard, a recent study shown that microglial activation can also be a significant contributor to the [¹⁸F]FDG PET signal [66]. Given that astrocytes can directly trigger microglial activation [67], it could be possible that the increase of glucose consumption is indirectly mediated or significantly contributed by microglial activation. Nonetheless, in a recent study of our group with the same participants of this study, we did not find the same kind of relation with sTREM2 [62], a marker of microglial activation, and [¹⁸F]FDG uptake as the one found here with plasma or CSF GFAP. Thus, this result does not support to the microglial involvement in the association here presented.

Our study has several strengths. The presence of biomarkers of Aβ and tau pathology has enabled us to stage a large cohort of cognitively unimpaired individuals within the AT framework. In addition, we used the same precise and validated assay to measure GFAP both in plasma and CSF, thus avoiding potential methodological confounders associated with using different analytical platforms in the two biofluids. This assay is commercially available, and, therefore, other research groups can replicate our results in their cohorts. Finally, we could compare head-to-head the associations between [¹⁸F]FDG uptake and CSF and plasma GFAP given that we measured both fluid biomarkers moment in the same time.

In summary, we found that higher plasma GFAP, a marker of reactive astrogliosis, is associated with widespread higher cerebral glucose metabolism as measured with [¹⁸F]FDG in the preclinical Alzheimer’s continuum. Of note, the correlation with glucose metabolism was also observed with GFAP levels measured in the CSF, although less widespread and in different regions. This finding suggests that early Alzheimer’s pathologic changes trigger an astrocytic reaction that is linked to greater energetic demand. With the onset of tau pathology, this coupling between astrogliosis and cerebral glucose consumption is reversed in regions of early AD-related hypometabolism. This suggests a failure to sustain such an elevated energetic consumption due to a disruption of glucose metabolism in astrocytes, already in preclinical AD stages. Our findings contribute to a better understanding of the metabolic requirements of the neuroinflammatory response to early AD pathology, which will be useful for the rational design of preventive interventions targeting inflammatory mechanisms and/or energetic dysfunction in preclinical stages of AD.