Clozapine induces astrocyte-dependent FDG-PET hypometabolism

Advances in functional imaging allowed us to visualize brain glucose metabolism in vivo and non-invasively with [18F]fluoro-2-deoxyglucose (FDG) positron emission tomography (PET) imaging. In the past decades, FDG-PET has been instrumental in the understanding of brain function in health and disease. The source of the FDG-PET signal has been attributed to neuronal uptake, with hypometabolism being considered as a direct index of neuronal dysfunction or death. However, 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. With this in mind, we investigated glucose utilization changes after pharmacologically downregulating GLT-1 with clozapine (CLO), an anti-psychotic drug. Adult male Wistar rats (control, n = 14; CLO, n = 12) received CLO (25/35 mg kg−1) for 6 weeks. CLO effects were evaluated in vivo with FDG-PET and cortical tissue was used to evaluate glutamate uptake and GLT-1 and GLAST levels. CLO treatment effects were also assessed in cortical astrocyte cultures (glucose and glutamate uptake, GLT-1 and GLAST levels) and in cortical neuronal cultures (glucose uptake). CLO markedly reduced in vivo brain glucose metabolism in several brain areas, especially in the cortex. Ex vivo analyses demonstrated decreased cortical glutamate transport along with GLT-1 mRNA and protein downregulation. In astrocyte cultures, CLO decreased GLT-1 density as well as glutamate and glucose uptake. By contrast, in cortical neuronal cultures, CLO did not affect glucose uptake. This work provides in vivo demonstration that GLT-1 downregulation induces astrocyte-dependent cortical FDG-PET hypometabolism—mimicking the hypometabolic signature seen in people developing dementia—and adds further evidence that astrocytes are key contributors of the FDG-PET signal.


Introduction
Regulation of brain glucose metabolism is key for supporting high-energy demands during neuronal firing and it involves multiple enzymes and transporters. Beyond neurons, other brain cells are also metabolically active such as the astrocytes. Indeed, astrocytes are known for their highly glycolytic metabolism (Itoh et al. 2003;Bouzier-Sore et al. 2006;Supplie et al. 2017)and positioning of their end-feet 2010; Bittner et al. 2011;Jakoby et al. 2014). Additionally, it was observed a neuronal preference for lactate oxidation and a predominant glycolytic metabolism of glucose by astrocytes in cellular models (Itoh et al. 2003;Bouzier-Sore et al. 2006). In vivo studies have also unraveled that reduced astrocytic glutamate transport diminishes astrocyte glucose uptake (Cholet et al. 2001;Voutsinos-Porche et al. 2003). The glutamate transporter 1 (GLT-1) is the most abundant glutamate transporter in the mammalian brain, in which it has a dominant role in glutamate uptake (around 90%) (Danbolt 2001)and is predominantly located on astrocytes (Zhou and Danbolt 2013). Interestingly, glutamate transport via GLT-1 is considered as a triggering signal for glucose uptake by astrocytes (Pellerin and Magistretti 1994), this biological process being vital for brain energetics. It is important to mention that there are reports challenging the ANLS concept (Dienel 2019), which makes this topic controversial and requiring further experimental evidence.
Based on the aforementioned data, the question whether astrocyte glucose metabolism contributes to functional imaging signals, such as for positron emission tomography (PET) imaging with [ 18 F]fluoro-2-deoxyglucose (FDG) (Reivich et al. 1977;Phelps et al. 1979), has been raised. Specifically, in 1996, Pellerin and Magistretti put forward a hypothesis based on compelling evidence supporting the notion of astrocytes as contributors of FDG-PET signal (Magistretti and Pellerin 1996). In the following years, additional findings have corroborated this proposal Bonvento et al. 2002;Figley and Stroman 2011). Still, for over 40 years, the FDG-PET signal has been used as an index of neuronal activity (Sestini 2007). Recently, however, it has been demonstrated that targeting astrocytes alters the FDG-PET signal. More specifically, deletion of insulin receptors on astrocytes reduces FDG-PET signal (Garcia-Caceres et al. 2016), and activation of glutamate transport via GLT-1 on astrocytes causes widespread but graded increases in FDG-PET signal (Zimmer et al. 2017).
Recent important theoretical articles are already acknowledging the role of astrocytes in the brain FDG-PET signal (Iaccarino 2017;Zetterberg and Bendlin 2021;Perani 2019;Chetelat et al. 2020). Still, whether a reduction of glutamate uptake by astrocytes would cause in vivo FDG-PET hypometabolism-recapitulating what is seen in neurodegenerative disorders-remains unexplored. On the basis of the evidence presented above, we hypothesized that astrocyte GLT-1 downregulation would cause FDG-PET hypometabolism. To further test this hypothesis, we conducted in vivo and in vitro pharmacological challenges with CLO, a well-known anti-psychotic drug that reduces GLT-1 density (Melone et al. 2001)and glutamate transport (Vallejo-Illarramendi et al. 2005), to assess its effects on cerebral FDG-PET signal in rats.

Animals
Adult male Wistar rats, 90 days old, were maintained under a 12-h light-dark cycle (lights on at 7 A.M.), at a controlled room temperature (22 ± 1 °C) and with free access to food and water. Animals' weight was measured once a week and volume of liquid ingestion was measured three times a week.

Pharmacological treatment
Rats were randomly divided into two groups: control and CLO. Rats in the CLO group received the drug CLO (Cristália, Itapira/Brazil) in free drinking water (25-35 mg kg −1 day −1 ) for 6 weeks, starting at postnatal day (PND) 91, until euthanasia. The CLO solution was prepared as previously described by Terry et al. (Terry et al. 2003). Treatment regimen, dose, and length of treatment followed the procedure described by Melone et al. (Melone et al. 2001). CLO was selected for these experiments because it specifically reduces GLT-1 (without affecting other glutamatergic transporters) (Melone et al. 2003)and substantially crosses the BBB, allowing for oral administration (Cremers et al. 2012;Naheed and Green 2001).

[ 18 F]FDG-microPET scanning
Rats were scanned longitudinally before treatment (Baseline), at PND 90, and after 6 weeks of treatment (Followup), at PND 136, as previously described (Bellaver et al. 2019). Briefly, after overnight fasting, animals received an intravenous injection (0.4 mL) of FDG (mean ± s.d.: 1.03 ± 0.07 mCi) into the tail vein. Then, each rat returned to its home cage for a 40-min period of awake uptake of FDG. After, they immediately underwent a 10-min static acquisition under anesthesia (2% isoflurane at 0.5 L/min oxygen flow) (see supplementary material for detailed methodology).

Metabolic networks
Cross-correlation maps were built as previously described (Bellaver et al. 2019), using a mask containing 14 volumes of interest (VOIs): left (L)-right (R) frontal cortex (FtC), L-R temporoparietal cortex (TmPtC), L-R thalamus (Th), L-R hypothalamus (Ht), L-R striatum (St), L-R hippocampus (Hip), and L-R cerebellum (Cer). No voxel-wise network analysis was conducted due to rat brain size and PET limited resolution. Stable metabolic brain networks, using an innovative multiple sampling scheme (Schu 2021), were constructed by computing Pearson's correlation coefficients based on 2000 bootstrap samples. Graph theoretical measures such as density, global efficiency, assortativity coefficient, average degree, and average clustering coefficient were calculated for each of the bootstrap samples. Networks were corrected for multiple comparisons using the false discovery rate (P< 0.005) (Rubinov and Sporns 2010).

Synaptosomal preparation and L-[ 3 H]glutamate (L-[ 3 H]Glut) potassium-stimulated release
This procedure was performed as previously described (Almeida et al. 2017) using the whole cortex (left hemisphere) from 137 to 140 PND rats (see supplementary material for detailed methodology).

Primary cortical neuronal cultures
Neuronal cultures were prepared as previously described (Nunez 2008) with modifications. Newborn (0-1 day old) male Wistar rats' cortices were aseptically resected out. Tissue was dissociated enzymatically-with trypsin 0.05% (Gibco®) and DNase 0.006% (Sigma-Aldrich)-and mechanically (Pasteur pipet). Samples were incubated for 20 min at 37 °C for enzymatic action and again mechanically dissociated for 5 min. After, 20% FBS was added and samples were centrifuged at 150 g for 5 min. The pellet was resuspended in Neurobasal medium (Gibco®) supplemented with 0.5% B27 (Gibco®), 10 mM glutamine, and 0.02% penicillin/streptomycin (10,000 U/mL-Gibco®). Cells were plated at a density of 3-5 × 10 5 cells/cm 2 into 24-well plates pre-coated with poly-l-lysine and cultured at 37 °C in an atmosphere with 5% CO 2 . Half of the medium was replaced 1 h after plating cells. Medium changes, also of half of the medium volume, occurred once every 4 days. After cells reached confluence, neurons were treated with 50 μM CLO (Sigma-Aldrich) for 48 h and used for immunofluorescence or uptake procedures.

D-[ 3 H]
Asp is a substrate for the high-affinity sodiumdependent glutamate transport system; therefore, its uptake was used as a proxy of glutamate uptake (Taxt and Storm-Mathisen 1984;Davies and Johnston 1976). The sodiumdependent D-[ 3 H]Asp uptake was assessed in adult cortical slices (137-140 PND rats) incubated with 100-μM l-glutamate and in primary cultured astrocytes (P1-2 rats) incubated with 10 μM (basal) or 100 μM (stimulated) l-glutamate. The procedures followed protocols previously described (Davies and Johnston 1976;Debernardi et al. 1999) with modifications (see supplementary material for detailed methodology).

RNA extraction and quantitative real-time PCR (qRT-PCR)
The whole cortex (right hemisphere) from 137 to 140 PND rats was used for the qRT-PCR analysis on which the mRNAs of GLT-1 and GLAST were quantified as previously described (Bellaver et al. 2018) (see supplementary material for detailed methodology).

High-performance liquid chromatography (HPLC) procedures
High-performance liquid chromatography (HPLC) procedure was performed as previously described (Souza et al. 2016) using cerebrospinal fluid (CSF) to quantify three amino acids: glutamate (Glut), aspartate (Asp), and glutamine (Gln). Additionally, a HPLC with ultraviolet (UV) Fig. 1 Effects of CLO treatment on brain glucose metabolism assessed by [ 18 F]FDG-microPET. Brain metabolic maps showing the mean standardized uptake value reference (SUVr) of the control (a) and CLO (b) groups at Baseline and after the treatment period. Voxel-wise images representing the percentage of change difference (c) and t-statistical maps (d) between Baseline and Follow-up of control and CLO groups. Brain mask showing the volumes of interest (VOIs) (e). Control and CLO group mean regional FDG SUVr values in the whole brain (f), frontal (g) and temporoparietal cortices (h), hippocampus (i), striatum (j), thalamus (k), hypothalamus (l), and cerebellum (m) at Baseline and Follow-up. Dynamic raster plots displaying individual FDG SUVr regional means in the control (n) and CLO (o) groups. a-d Images are projected into a standard magnetic resonance imaging (MRI) in axial, sagittal, and coronal planes (control, n = 11; CLO, n = 13). Data represented as violin plots indicating the mean, interquartile range, and the minimum and maximum values. Two-way ANOVA with repeated measures, followed by Sidak's detection method was applied to verify CLO presence and stability in the animals' drinking solution (see supplementary material for detailed methodology).

Behavioral analysis
At the end of the 6 weeks of treatment, animals were submitted to the open-field and the novel object recognition (NOR) tasks (see supplementary material for detailed methodology).

CLO treatment causes brain glucose hypometabolism in vivo
The microPET analysis after CLO pharmacological intervention showed a clear FDG SUVr reduction mainly in cortical brain regions (Fig. 1a, b). Specifically, the CLO group (n = 13) presented a widespread FDG cortical hypometabolism, ranging from 10 to 20% (Fig. 1c). By contrast, the control group (n = 11) presented only a few small clusters of glucose metabolism reduction, comprising values between 1 and 6% (Fig. 1c). A voxel-wise t-statistical analysis showed a significant glucose hypometabolism mostly in the cortical regions of the CLO group, with a peak effect in the parietal cortex (peak, t (23) = 7.62, p < 0.0001; Fig. 1d). On the other side, the control group presented only negligible reductions in a few isolated voxels (Fig. 1d). The two-way ANOVA of the whole brain FDG mean SUVr showed significant time effect (p = 0.0029; F(1, 22) = 11.2; Fig. 1f) but no group (p = 0.86; F(1, 22) = 0.035) or interaction (p = 0.89; F(1, 22) = 0.016), which indicates that time (Follow-up) affected the whole brain metabolism in both groups if compared to their own Baseline. A brain mask with 14 VOIs (Fig. 1e), based on the Paxinos and Watson Rat Brain Atlas, was used to obtain mean regional FDG SUVr values. In the regional analysis, a two-way ANOVA identified a significant effect of time and an interaction in the frontal cortex (time, p < 0.0001; interaction, p = 0.02; Fig. 1g), temporoparietal cortex (time, p < 0.0001; interaction, p = 0.003; Fig. 1h), and hippocampus (time, p < 0.0001; interaction, p = 0.04; Fig. 1l). The striatum (Fig. 1j) and thalamus (Fig. 1k) presented only an effect of time (striatum, p = 0.008; thalamus, p = 0.007). The hypothalamus (Fig. 1l) and cerebellum (Fig. 1m) showed no significant differences in the two-way ANOVA. Following the two-way ANOVA, Sidak's multiple comparisons test showed FDG hypometabolism in the frontal cortex (p < 0.0001; t (22) = 5.8), temporoparietal cortex (p < 0.0001; t (22) = 6.6), and hippocampus (p < 0.0001; t (22) = 5.49) on the follow-up CLO group in comparison to its baseline. By contrast, no regional changes were found in the control group (Baseline vs. Follow-up). The individual dynamic raster plot reinforces the observation that CLO-induced FDG hypometabolism is more pronounced in cortical areas, and that FDG metabolism remains stable in the control group (Fig. 1n, o). Furthermore, a multiple sampling scheme metabolic network analysis was performed to identify brain reorganization patterns in response to CLO. In comparison to the control group (Fig. 2a), the CLO group showed multiple altered connections within the metabolic network, presenting a widespread hyposynchronicity (p < 0.005, FDR-corrected, Fig. 2b). These findings were corroborated by consistent changes in graph measures, including reduced global efficiency (p < 0.0001, Fig. 2c), density (p < 0.0001, Fig. 2d), degree (p < 0.0001, Fig. 2e), and clustering coefficient (p < 0.0001, Fig. 2f). The assortativity coefficient was not significantly different between groups (p = 0.31, Fig. 2g).

Behavioral effects and treatment measures
Total distance traveled (p = 0.17; t (26) = 1.42) and total immobile time (p = 0.13; t (26) = 1.56) during the openfield task were not different between groups (control, uptake results in both conditions (basal and stimulated) for control and CLO groups (d). Uptake values expressed as percentage of control basal results (n = 5 for each group). Data represented as violin plots indicating the mean, interquartile range, and the minimum and maximum values. Two-way ANOVA followed by Sidak's n = 14; CLO, n= 12), therefore, suggesting no sedative side effects of CLO (Prut and Belzung 2003) (Supplementary  Fig.2a-c). Analysis of the time spent exploring objects during the NOR task revealed that both groups (control, n = 7; CLO, n = 5) equally explored the two identical objects during the training session and spent more time exploring the new object during the test session (control, p = 0.0004; t (6) = 7.22; CLO, p = 0.0004; t (4) = 10.86), indicating no impairment of short-term recognition memory ( Supplementary Fig. 2d-f). Animals' body weight was measured once a week during treatment. Results showed time effect, as animals gained weight due to age-dependent normal body growth (p < 0.0001; F(7, 196) = 336.6; Supplementary Fig. 3a), but no group effect (p = 0.35; F(1, 28) = 0.89). The liquid intake showed an increase over time (p < 0.0001; F(11, 121) = 7.85; Supplementary  Fig. 3b), and a lower volume intake for the CLO group (group effect, p = 0.036; F(1, 11) = 11.73), probably due to differences in the drinking water palatability. In future experiments, the addition of small quantities of a compound such as sucrose, in order to make the drinking water containing the experimental drug more palatable, should be considered. Finally, a high-performance liquid chromatography (HPLC) analysis confirmed that the animals' drinking solution preparation was yielding the expected concentration with high reproducibility, presenting a relative standard deviation (RSD) of 1.86% ( Supplementary  Fig. 3c). The experimental drinking solution also showed a high stability, with no noticeable modifications of CLO concentration in a 24-h period at room temperature (Supplementary Fig. 3d).

Discussion
In this study, we demonstrated that CLO causes in vivo brain glucose hypometabolism and alters the brain metabolic network. We also found that CLO decreases GLT-1 immunocontent and mRNA expression along with functional changes, including a reduction in glutamate uptake in cortical slices, and a decline in glucose and glutamate uptake in astrocyte cortical cultures. Importantly, no differences were found in cultures of cortical neurons.
Animals treated with CLO exhibited a significant widespread reduction in FDG-PET signal in the cortex (~ 20%) and hippocampus. Consistently, those brain regions are the ones presenting a high content of GLT-1 (Lehre et al. 1995). Additionally, we found that CLO selectively reduced cortical GLT-1 protein levels, but did not alter GLAST levels, confirming the effect originally seen by Melone et al. (Melone et al. 2003). We also identified a reduction in cortical GLT-1 mRNA expression, suggesting that CLO may act at the transcriptional level. In parallel, CLO reduced GLT-1 and GLAST protein levels in cortical astrocyte cultures. Quite importantly, we corroborated these findings with the functional reduction of astroglial glutamate transport, measured by D-[ 3 H]Asp uptake, in cortical slices and astrocyte primary cultures. We did not find changes in synaptosomal glutamate release from cortical preparations, or in glutamate, aspartate, and glutamine concentrations in the CSF, which suggests that the glutamate-glutamine recycling process remains functional.
The in vitro evaluation using primary astrocyte cultures treated with CLO also showed a reduction of glucose uptake, while no difference was found in cultures of cortical neurons. These results suggest that astrocytes, not neurons, are the main target cells responsible for the phenomenon observed in vivo, i.e., CLO-induced FDG-PET signal reduction. Importantly, the reduction of [ 3 H]2DG uptake by CLO in cortical astrocyte cultures was only observed when stimulated by glutamate. Astrocytes increase their glucose uptake in situations of neuronal activation (Pellerin and Magistretti 2012), i.e., during high energetic demand. Accordingly, a similar outcome was also demonstrated in vivo in a twophoton microscopy study in mice, which also showed that glucose uptake is elevated in astrocytes during activation but remains close to basal levels in neurons (Chuquet et al. 2010). Our data showing that reducing GLT-1 expression decreases activity-dependent [ 3 H]2DG accumulation is also in agreement with another study performed with GLT-1 KO mice showing that invalidation of GLT-1 prevents the whisker-stimulated increase in [ 14 C]-2-deoxyglucose accumulation in the somatosensory cortex of developing animals (Voutsinos-Porche et al. 2003).
It is important to emphasize that the FDG-PET uptake phase in our experiments was conducted in freely moving animals and, thus, the FDG-PET signal here represents glucose metabolism in an awake brain. Our present data should be also put in perspective with previous results obtained with a completely different drug. Ceftriaxone is an antibiotic that was shown to cause an increase in astrocyte GLT-1 expression and activity (Rothstein et al. 2005). Treatment of primary cultures of cortical astrocytes with ceftriaxone was shown to enhance both glutamate transport and glutamatestimulated glucose uptake through its effect on GLT-1 (Zimmer et al. 2017). But quite strikingly, ceftriaxone increased brain FDG-PET signal in rats proportionately to the density of GLT-1 expression in different brain regions (Zimmer et al. 2017). These results are simply a mirror image of the data obtained with CLO. Altogether, results with CLO and ceftriaxone reinforce the notion that astrocytes, through glutamate-stimulated glucose uptake, represent an important source of the FDG-PET signal in vivo, as postulated exactly 25 years ago (Magistretti and Pellerin 1996). Additionally, a recent study demonstrated that the modulation of microglial cells also contributes to the FDG-PET signal (Xiang 2021). Since microglial cells account for less than 10% of brain cells in mammals, the microglia-induced effect on FDG-PET probably has the astrocytic and/or the neuronal compartment as the cellular source. In fact, it was previously demonstrated that Il-1α, IL-1β, and TNFα, classically secreted by activated microglia, enhance glucose uptake in astrocytes (Gavillet et al. 2008).
Besides its effects on glutamate transport presented above, it is also known that CLO interacts with serotonin, muscarinic, dopamine, and histamine receptors (Miyamoto et al. 2005;Stepnicki 2018). However, the direct contribution of these neurotransmission systems to brain energetics is small (Sibson et al. 1998;Scremin and Jenden 1993). Therefore, it is very unlikely that changes in glucose uptake result from these interactions. In fact, there are several studies showing that brain energy metabolism is driven predominantly by glutamatergic and, to a minor extent, GABAergic systems (for review, see (Rothman et al. 1999;Rothman et al. 2003)), with contributions of other neurotransmitters being so small that it can be disregarded in the overall scenario. However, we could not ignore the possibility of a modulatory effect on neuronal metabolism. To rule out a direct effect on neuronal energy metabolism, we tested CLO impact on neuronal cortical cultures under resting and stimulated conditions. Clearly, CLO did not change neuronal glucose metabolism in both situations, thus, strengthening the conclusion of astrocytes being at the origin of the observed hypometabolism. It is important to mention that cortical FDG-PET hypometabolism is a signature of neurodegenerative disorders such as Alzheimer's disease and frontotemporal dementia, which is interpreted as an index of neurodegeneration (Chetelat et al. 2020). Since CLO does not induce neuronal death (Chen and Nasrallah 2019;Lundberg 2020;Bastianetto et al. 2006;Stanisavljevic et al. 2019), one could argue we recapitulated cortical FDG-PET hypometabolism seen in people with dementia by only reducing glutamate transport in astrocytes.
A potential limitation to this conclusion would pertain to behavioral changes such as sedation or freezing that could have contributed to the effects observed with CLO and might jeopardize our interpretation of the cellular origin of the FDG-PET signal (Berti et al. 2014). However, we did not observe changes either in the open-field task or in a more complex behavioral task such as the NOR. Thus, it seems that CLO-induced FDG hypometabolism is not related to major behavioral changes.
In summary, our results provide microPET evidence that a treatment with CLO causes a reduction in FDG-PET signal. Primary culture data indicate that astrocytes, not neurons, are the cells responsible for this phenomenon. These results corroborate the notion that the FDG-PET signal might reflect not only neuronal activity but also astrocyte metabolism. These findings reinforce the need for a renewed interpretation of FDG-PET data in brain imaging studies. Finally, from a clinical perspective, CLO is an important anti-psychotic drug widely used in patients with schizophrenia who are either intolerant or refractory to classical neuroleptics (Conley and Kelly 2001). CLO mechanisms of action are yet not fully understood; therefore, the effect on brain energetics observed in our study might be of importance and should be better studied in this perspective. As a matter of fact, a FDG-PET hypometabolism in the frontal cortex of CLO-responsive patients has been reported a few years ago (Cohen et al. 1997;Molina et al. 2007;Molina et al. 2005). We believe this phenomenon deserves further investigation in clinical research.