Reactive astrogliosis: A friend or foe in the pathogenesis of Alzheimer’s disease

Astrocytes are highly efficient homeostatic glial cells playing a crucial role in optimal brain functioning and homeostasis. Astrocytes respond to changes in brain homoeostasis following central nervous system (CNS) injury/diseased state by a specific defence mechanism called reactive astrogliosis. Recent studies have implicated and placed reactive astrogliosis in the centre of pathophysiology of Alzheimer's disease (AD) and other neurodegenerative disorders. The AD biomarker field is evolving rapidly with new findings providing strong evidence which supports the notion that a reactive astrogliosis is an early event in the time course of AD progression which may precede other pathological hallmarks of AD. Clinical/translational in vivo PET and in vitro postmortem brain imaging studies demonstrated ‘a first and second wave’ of reactive astrogliosis in AD with distinct close‐loop relationships with other pathological biomarkers at different stages of the disease. At the end stages, reactive astrocytes are found to be associated, or in proximity, with amyloid plaque and tau pathological deposits in postmortem AD brains. Several new PET‐tracers, which are being in pipeline and validated at a very fast pace for mapping and visualising reactive astrogliosis in the brain, will provide further invaluable mechanistic insights into AD and other non‐AD dementia pathologies. The complementary roles of microglia and astrocyte activation in AD progression, along with the clinical value of new fluid astrocytes biomarkers in the context of existing biomarkers, are the latest avenue that needs further exploration.

started to learn and explore more about glia cells and were surprised by their deep connection in one or another way in almost all of the CNS physiological functions. As of now, glial cell populations can be subdivided into four major categories (1) Astroglia (2) Microglia (3) Oligodendroglia, which include oligodendrocytes and their precursors NG2 + glia cells, with their percentages varying significantly in different regions of the human brain (Jäkel & Dimou, 2017).
Astrocytes account for approximately 20-40% of the glial cell population in the human CNS and are considered as the key 'homeostatic cells' tiling and supporting the whole brain and the spinal cord for optimal functioning and neuronal information transfer (Fakhoury, 2018;Jäkel & Dimou, 2017;Vasile et al., 2017). The name 'astrocyte' first emerged in the late 19th century to describe star-shaped glial cells, and, despite their electrical silence and non-neuronal attributes, researchers argued that they have much broader and vital functions (Verkhratsky & Butt, 2018;Verkhratsky & Nedergaard, 2018). As we move forward in time, a plethora of evidence highlighted astrocytes involvement in a wide array of physiological functions, such as ho-

meostasis of neurotransmitters and neuromodulators, extracellular
ions and pH, metabolic support, blood-brain barrier (BBB) integrity, synaptic plasticity, connectivity and many more. Morphologically, astrocytes are very diverse and heterogenous and their molecular subtypes can differ significantly depending on their developmental origin and brain regions. Diversity of astrocytes in CNS is very broad (Verkhratsky et al., 2019) but classically, they are categorised into white matter fibrous and grey matter protoplasmic astrocytes Most importantly, astrocytes form a second arm (parenchymal) of defence in the brain following the BBB (which forms the first line of defence against invading pathogens, harmful chemical agents/molecules, and harmful immune cells/mediators (Muldoon et al., 2013)) by responding to neuroinflammation/injuries through an activation process/state called reactive astrogliosis (Figure 1c) (Escartin et al., 2021). They become hypertrophic and show over-expression/high immunoreactivity of the astrocyte marker protein called glia fibrillary acidic protein (GFAP), a cytoskeleton intermediate filaments protein (Carter, Herholz, et al., 2019;. Over the past few decades, several studies have suggested a broad role of reactive astrogliosis in the pathogenesis of Alzheimer's disease (AD) and other neurogenerative disorders (Carter, Herholz, et al., 2019;Gonzalez-Reyes et al., 2017;Matias et al., 2019;Siracusa et al., 2019;Verkhratsky et al., 2010). Findings even showed that reactive astrogliosis could precede early pathological hallmarks of AD, such as amyloidβ (Aβ) and Tau during disease progression ( Figure 2) (Carter et al., 2012Marutle et al., 2013;Rodriguez-Vieitez et al., 2015, 2016Schöll et al., 2015). In this review, we want to explore the role of reactive astrogliosis, specifically in the progression of AD, in a broader sense with regards to astrocyte heterogeneity, recently published in vitro and in vivo studies focused on PET-imaging, plasma, and cerebrospinal fluid (CSF) biomarkers. We will also discuss and present a new perspective on the ongoing debates concerning (1) Astrocytes or microglia-who is the first responder and (2) GFAP being a true marker of reactive astrogliosis in AD pathology.

| A S TRO C Y TE S PATHOLOG I C AL CHANG E S: NON -RE AC TIVE TO RE AC TIVE S TATE
As we started to drift away and challenge the neuron-centric concepts for neuropathological changes, more and more focus is directed towards neuroglia to understand their diverse role in the evolution of several neuropsychiatric disorders. It would not be wrong to say at this point, that we begin to appreciate more and more the multifaceted and rather complex role of astrocytes in these disorders. Astrocytes' pathological changes can be represented either by a (1) non-reactive state or (2) reactive state. The non-reactive state mostly includes transformations, such as astrodegeneration and pathological remodelling, whereas the reactive state is characterised by a distinct phenotypic state/mechanism called reactive astrocytes/reactive astrogliosis. (Verkhratsky et al., 2017).

Astrodegeneration leading to morphological atrophy and loss
of function is prominent in several neurological and psychiatric disorders, such as schizophrenia and major depressive disorder. In amyotrophic lateral sclerosis (ALS), astrocytes atrophy followed by increased cell death can interfere with glutamate uptake causing silencing and degeneration of motor neuronal activity (Rossi et al., 2008), whereas in AD, it has been linked with early synaptic loss and cognitive impairment (Verkhratsky et al., 2010(Verkhratsky et al., , 2015. Clasmatodendrosis, which is a form of astrodegeneration characterised by extreme terminal disintegration, fragmentation, or beading of the distal cell processes, has been also observed in postmortem neuropathological specimen of seizures, cerebral ischaemia as well as in elderly/ageing brain (Balaban et al., 2021). At the same time, the pathological remodelling of astrocytes results in the possession of toxic characteristics/functions which promote the pathophysiology of the disease. For example, in mesial temporal lobe epilepsy, abnormal astrocyte morphological changes and uncoupling in the hippocampus disrupted the K(+) homeostasis and progression of spontaneous seizures (Bedner et al., 2015). This study led to the point that mesial temporal lobe epilepsy could have a more gliabased origin instead of a neuronal one, challenging the established notion of epilepsy being a neuronal syndrome. Another example includes Alexander disease, where gain of new toxic functions by astrocytes and GFAP mutations drove the pathology (Messing et al., 2012).
Reactive astrogliosis, as mentioned above, is the acquisition of 'reactive phenotype/state' by astrocytes in response to a CNS injury (Escartin et al., 2021). From the evolutionary point of view, it is a conserved defence mechanism that regulates neuroinflammation, limits the site of damage by forming a perilesion barrier and compact glial scar with fibromeningeal and other glia cells (i.e. anisomorphic astrogliosis) and promotes post-injury neuronal function recovery (via isomorphic astrogliosis) (Fakhoury, 2018;Guttenplan & Liddelow, 2019;Sofroniew & Vinters, 2010;Verkhratsky et al., 2010Verkhratsky et al., , 2013. There are many names and definitions in the field for reactive astrogliosis, but according to the recently published consensus statement by a group of experts, in order to put forward uniform guidelines for the reactive astrocyte nomenclature and definitions, 'reactive astrogliosis is a process whereby, astrocytes engage in molecularly defined programmes involving changes in transcriptional regulation, as well as biochemical, morphological, metabolic, and physiological remodelling, which ultimately result in gain of new function(s) or loss or up-regulation of homeostatic ones, in response to pathology' (Escartin et al., 2021). The complexity of reactive astrogliosis and the resulting astrocyte heterogeneity is very puzzling and can change from disease to disease. These transformations could range from beneficial to malfunctional/toxic states depending on the disease and the pathways inducing them. For example, in traumatic brain injury (Levine et al., 2016) and motor neuron injury (Tyzack et al., 2014), activation of the transcription factor STAT3 (signal transducer and activator of transcription 3) promoted astrocyte (or astroglial) resilience (which is defined by set of successful astroprotective responses that maintain cell-intrinsic homeostatic functions in neural circuits while promoting both neuronal and astrocyte survival (Escartin et al., 2021)), which prevented necrosis and supported neuronal integrity, whereas in AD mouse models, it was the inhibition of STAT3 that ameliorated the pathology (Reichenbach et al., 2019). AD is one of the key brain disorders where research on F I G U R E 1 Astrocyte and microglia activation in AD. (a) Astrocytes in the brain of postmortem AD patient as demonstrated by GFAP immunostaining (brown). The figure is adapted with permission from (Yu et al., 2005). (b) Astrocytes in close proximity of neuritic plaques as observed with Aβ (brown) and GFAP (light blue) immunostaining in the temporal cortex of AD patient with Swedish mutation and in sporadic AD patient (insert). The figure is adapted with permission from (Nordberg, 2004). (c) Illustrations showing activation of astrocyte and microglia in response to CNS injury and Aβ pathology and their further role in the etiology and progression of AD. The current PET-imaging tracers and fluid biomarkers (blood and CSF) for assessing and monitoring reactive astrogliosis and microglia activation in AD are also highlighted. CXCL1, chemokine (C-X-C motif) ligand 1; DED, deutriuml-deprenyl; GFAP, glial fibrillary acidic protein; IL-8, interleukin-8; PET, positron emission tomography; sTREM2, soluble triggering receptor expressed on myeloid cells 2; TGF-alpha, transforming growth factor alpha; TRAIL, tumour necrosis factor-related apoptosis-inducing ligand astrocyte reactivity 'reactive astrogliosis' has been the central focus in recent times, with several published and ongoing studies aimed at deciphering its underlying mechanisms and role in the AD pathogenesis for future diagnostic and therapeutic interventions.

| RE AC TIVE A S TROG LI OS IS IN AD: TH E RE SILIENT FRONTLIN E SOLDIER
AD is a progressive neurodegenerative disorder and the most common form of dementia leading to acute memory loss and a variety of other cognitive disabilities affecting millions of people worldwide (Atri, 2019;Selkoe, 2011;Winblad et al., 2016). The classical hallmarks of AD pathology include Aβ plaques and neurofibrillary tau-tangles (Selkoe, 2011). According to the National Institute of Aging and Alzheimer's Association (NIA-AA) framework guidelines, which were published in 2011 and updated recently, AD should be defined as a biological construct and diagnosed by its underlying pathological hallmarks and biomarkers (Jack et al., 2011(Jack et al., , 2018. The  (Dubois, 2000;Dubois & Albert, 2004;Dubois et al., 2007Dubois et al., , 2010Dubois et al., , 2014Dubois et al., , 2016.

F I G U R E 2
The 'first (early)' and 'second (late)' wave of reactive astrogliosis in AD pathology. (a) In vivo brain retention of 11 C-PiB, 11 C-DED and 18 F-FDG in pre-symptomatic mutation carriers (pMC) autosomal dominant AD (ADAD) case at baseline and follow-up (after ~3 years). The red and yellow scales correspond to contrasts showing higher retention, and the blue scales correspond to lower retention. Adapted with permission from (Rodriguez-Vieitez et al., 2016). (b) In vivo 11 C-DED micro-PET studies in amyloid precursor protein Swedish mutation (APPswe) and wild-type mice models showing differential time course for reactive astorgliosis in APPswe mice as compared to WT mice. 11 C-DED binding was higher at 6-months of age but begin to decline with age (23-months of age; left panel) whereas no change in 11 C-DED binding was observed in WT mice with age (almost similar binding at 11-and 24-months of age; right panel). All images are presented in a common scale from 0.0 to 0.4 units of non-displaceable binding potential (BP ND ). Adapted with permission from . (c) In vitro large brain section autoradiography showing increase binding of 3 H-BU99008 in sporadic AD case as compared to control at the end stages. The color scale represent binding from low (green) to high (dark red). Adapted with permission from (Kumar et al., 2021). (d) The illustration showing hypothetical model of reactive astrogliosis 'first' and 'second' wave at different stages of AD in context of other pathological hallmarks. ADAD, autosomal dominant AD case; DED, deutriuml-deprenyl, FDG, flurodeoxyglucose; PIB, Pittsburgh Compound B 2001; Siracusa et al., 2019;Verkhratsky et al., 2010). In response to stress/injury conditions, activated astrocytes can upregulate the production of Aβ and the release of pro-inflammatory mediators such as cytokines (Interleuckin-1β and IL-6) and tumour necrosis factor-alpha (TNFα) (Sajja et al., 2016), thereby initiating a detrimental cascade that leads to neuronal dysfunction. Interestingly, Aβ can also recruit astrocytes at the lesion site via the production of chemotactic molecules such as monocyte chemoattractant protein-1 (Smits et al., 2002;Wyss-Coray et al., 2003) and abnormally upregulate astrocyte-mediated nuclear factor-kappa B (NF-kB). It can also complement signalling pathways that can disrupt dendritic morphology and the neuron-glia network/connectivity and thereby brain cognitive functions (Lian et al., 2015). Astrocytes can internalise pathological tau leading to abnormal tau accumulation and propagation during AD pathology (Chiarini et al., 2017;Ikeda et al., 1992;Martini-Stoica et al., 2018;Perez-Nievas & Serrano-Pozo, 2018). The exact underlying mechanism involved in tau transport/propagation through astrocytes is still poorly explored (Narasimhan et al., 2017;Perea et al., 2019). However, the relationship between astrocytes and tau is starting to evolve more and more with a new study showing that extracellular tau oligomers could trigger neuroinflammation by targeting and inducing astrocyte senescence in the brains of AD patients by promoting the release of high mobility group box 1 (HMGB1), a well-known senescence-associated inflammation marker (Gaikwad et al., 2021). The complexity and involvement of reactive astrogliosis are not just limited to key AD pathological hallmarks. It actually goes way beyond, as demonstrated by a recently published study where severe reactive astrogliosis promoted excessive production of hydrogen peroxide (H 2 O 2 ) from MAO-B, aggravating pathological features of AD such as tauopathy, neuronal atrophy, cognitive decline and eventually death in a newly developed animal model of reactive astrocytes (GiD) (Chun et al., 2020).
Another study showed that reactive astrocytes induce memory impairment in AD mice models by abnormal production or release of the inhibitory gliotransmitter GABA. The astrocytic GABA was produced by MAO-B and released through bestrophin1 channels to act on presynaptic GABA receptors, causing impaired spike probability, thereby affecting synaptic plasticity and memory (Jo et al., 2014). Another important aspect where reactive astrocytes seem to be involved in promoting/regulating neuroinflammation, synaptic dysfunction, and glutamate dysregulation/excitotoxicity in AD, is via dysregulation of Ca 2+ signalling leading to hyperactivation and/ or over-expression of calcineurin (CN; a Ca 2+ /calmodulin dependent protein phosphatase) and CN-dependent nuclear factor of activated T-cells (NFAT4) (Abdul et al., 2009;Pleiss et al., 2016). Up-regulation of CN/NFAT4 pathway in turn modulate the expression of glutamate transporter GLT-1 causing imbalance/loss of glutamate buffering capacity in AD mouse models (Sompol et al., 2017). Please refer to the review by Sompol and Norris (Sompol & Norris, 2018) for further reading regarding Ca 2+ dysregulation and reactive astrocytes.
Moreover, astrocyte-mediated neuroinflammation has also been associated with several other neurodegenerative and CNS disorders such as ALS, Parkinson's disease, Huntington's disease, Alexander disease, strokes, epilepsy, multiple sclerosis, neurotrauma, and CNS tumours (Maragakis & Rothstein, 2006;. All these outcomes and attributes suggest a strong correlation not only between reactive astrogliosis and AD but also point toward the involvement of reactive astrogliosis in a broad spectrum of life-threatening neurodegenerative and CNS disorders; thus making astrocytes a very promising target for the development of novel in vivo positron-emission tomography (PET) imaging biomarkers/ probes.
PET-imaging is a powerful in vivo technique that uses selective and specific PET-tracers to monitor specific proteins, their location and density as well as metabolic processes in the body (Fowler et al., 2005). Thanks to recent advances in the field of in vivo biomarkers, different types of PET-tracers, CSF and plasma biomarkers are currently being used to track Aβ plaques, tau neurofibrillary tangles (NFTs), glucose metabolism, and neuroinflammatory processes (microglia and astrocytes activation) in AD and other proteinopathies (Nordberg, 2011(Nordberg, , 2014Perani et

| IN VITRO AND IN VIVO PE T IMAG ING S TUD IE S
Despite great progress in the PET-imaging field, we are still lacking specific PET-tracers for visualising the astrocyte process/activation, and very few PET-tracers targeting astrogliosis have been studied so far. The most common one is 11 C-deutriuml-deprenyl ( 11 C-DED) (Carter et al., 2012;Fowler et al., 2005;Rodriguez-Vieitez & Nordberg, 2018) which targets the monoamine oxidase B (MAO-B is a flavin-containing enzyme, overexpressed in both astrocytes and serotonergic neurons regulating the concentration of amine neurotransmitters in the brain (Bortolato et al., 2008)).
We pioneered in demonstrating an increased in vivo 11 C-DED binding in mild cognitive impairment patients (MCI; amyloid positive) as compared to controls, MCI (amyloid negative), and AD patients (Carter et al., 2012). Our in vitro multi-PET studies, including 11 C-DED (Kadir et al., 2011;Lemoine et al., 2017;Marutle et al., 2013), highlighted the relationship between astrogliosis, fibrillar Aβ and glucose metabolism in AD. In postmortem studies performed on the first AD subject imaged with 11 C-PIB (Pittsburgh Compound B; a PET-tracer that targets Aβ fibrils) (Klunk et al. 2004), we found no significant regional correlation between 3 H-PIB and 3 H-DED binding in different brain regions, while there was a close positive correlation between 3 H-PIB binding and GFAP levels (Kadir et al., 2011).
The laminar distribution of 3 H-DED binding compared with 3 H-PIB in AD postmortem brain was also different (Marutle et al., 2013).
Furthermore, in a large number of autopsy AD and control brains, a positive correlation between 3 H-DED and 3 H-florbetaben (a PETtracer that targets Aβ fibrils) in the hippocampus, temporal and parietal cortex was observed (Ni et al., 2021). In follow up postmortem autoradiography studies with 3 H-THK5117 (a PET-tracer that targets tau deposits) and 3 H-DED, we observed a similar laminar cortical brain pattern for tau deposits and activated astrocytes in AD and autosomal-dominant AD (ADAD) cases with PSEN1DE9 and AβPParc mutations (Lemoine et al., , 2020. These studies suggested a close-knit link between Aβ, tau and reactive astrogliosis and highlighted that different subtype of reactive astrocytes might be associated with the pathogenesis of AD. In vivo-PET studies in presymptomatic ADAD carriers as well as in vivo micro-PET studies in amyloid precursor protein Swedish mutation (APPswe) transgenic mice models have together demonstrated the important finding that reactive astrogliosis was significantly increased during the initial stages of AD progression preceding the early Aβ deposition in brain. Thus, in a longitudinal in vivo multi-PET tracer study performed in members of different known ADAD families, a significantly elevated reactive astrogliosis (higher 11 C-DED binding) was observed in presymptomatic ADAD carriers compared to non-carriers. The reactive astrogliosis was significantly elevated even 15-20 years before the onset of clinical symptoms followed by a decline with increasing Aβ load and time/ disease progression (Figure 2a) (Rodriguez-Vieitez et al., 2016;Schöll et al., 2015). Furthermore, a positive correlation has also been observed between 11 C-DED binding and cerebral glucose hypometabolism (measured with 18 F-deoxyglucose ( 18 F-FDG) uptake) as well as with cortical microstructure in presymptomatic ADAD cases (Carter, Chiotis, et al., 2019;Vilaplana et al., 2020). A negative correlation between 11 C-DED and 11 C-PIB in presymptomatic stages of ADAD was seen (Carter, Chiotis, et al., 2019), whereas in postmortem brain analyses in advanced sporadic AD cases have been observed with a positive correlation between deprenyl ( 3 H-DED) and amyloid ( 3 Hflorbetaben) (Ni et al., 2021).
In order to investigate the possible time course of reactive astroglisosis in APPswe transgenic mice models, transgenic mice and control animals underwent micro-PET studies with 11 C-deprenyl and 11 C-PIB. The APPswe transgenic mice showed higher 11 C-DED binding indicating elevated reactive astrogliosis at 6-8 months age which preceded the increased Aβ plaque load (increased 11 C-PIB binding) measured at 18-24 months age of the animals (Figure 2b) . Thus, a similar pattern of early increased reactive astrogliosis was observed in the APPswe mice as in presymptomatic ADAD carriers , 2016. The 11 C-DED binding in APPswe transgenic mice declined with age, and a similar behaviour was also observed in presymptomatic ADAD carriers as mentioned above (Rodriguez-Vieitez et al., 2016).
These above-described studies clearly established the reliability of MAO-B as an excellent marker for reactive astrogliosis. However, since the expression of MAO-B is not specific to astrocytes and can also be observed to a great extent in different neuronal families (Lin et al., 1993;Vitalis et al., 2002), this further prompts the need for more specific astrocyte PET-imaging biomarkers/tracers. BU99008 is a recently developed novel astrocyte PET-tracer (Tyacke et al., 2012), which targets the mitochondrial imidazoline 2 binding sites (I 2 Bs) predominantly expressed in the outer mitochondrial membrane of the astrocytes and to a lower extent in neurons (Li, 2017;Tyacke et al., 2018). I 2 Bs represent a highly heterogenous group of proteins that bind with high affinity to 3 H-Idazoxan (Li, 2017). I 2 Bs have been proposed to regulate GFAP expression (Olmos et al., 1994) and have shown to be co-expressed with MAO-B in the human frontal cortex during the process of aging (Ballesteros et al., 2000). I 2 Bs are implicated in several neurological and neurodegenerative disorders including AD. Most importantly, postmortem studies have shown that I 2 Bs density is increased by 63% in AD (Ruiz et al., 1993) and could thereby be an ideal surrogate marker to visualise and monitor astrocyte reactivity in the brain during normal and disease/injured states. Initial studies (preclinical and clinical) in animal models, primates and in healthy human participants, demonstrated good specificity, brain uptake, biodistribution and radiation safety profile for the tracer (Kealey et al., 2013;Parker et al., 2014;Tyacke et al., 2012Tyacke et al., , 2018. Building on these studies, we demonstrated for the first time that 3 H-BU99008 could detect/target reactive astrogliosis in AD postmortem brains with good specificity and selectivity ( Figure 2c) (Kumar et al., 2021). These findings seem to replicate in vivo in AD patients as well, as demonstrated by Calsolaro and colleagues in a pilot study published shortly after our postmortem study using 11 C-BU99008 (Calsolaro et al., 2021). Further, they also observed similar findings as DED in vivo with BU99008, namely that reactive astrogliosis was higher in MCI cases as compared to AD cases (Livingston et al., 2021). Interestingly, frozen large brain section autoradiography showed differences in the binding behaviour of 3 H-BU99008 and 3 H-DED, pointing towards the possibility that they might be either targeting a different sub-population or specific subtypes/ phenotypes of astrocytes (Kumar et al., 2021). Please refer to the review by Escartin and colleagues (Escartin et al., 2021) for further reading regarding reactive astrocytes nomenclature and definitions (states vs. phenotypes). Higher binding of 3 H-BU99008 and 3 H-DED was also observed in different AD brain regions specifically in the hippocampus as compared to control. This observation complemented previous (Marutle et al., 2013) and current postmortem studies (Ni et al., 2021), where higher 3 H-DED binding in different AD brain regions (prominent in hippocampus), as compared to control, at the end stages along with high GFAP immunoreactivity was observed.
There is a clear trend in astrocyte reactivity at different stages of AD based on DED binding dynamic changes and GFAP immunoreactivity observed in our presymptomatic to symptomatic cases in vivo-PET, APPswe transgenic mice models micro-PET (Carter et al., 2012;Rodriguez-Vieitez et al., 2015, 2016Schöll et al., 2015) and in vitro end stage AD postmortem studies (Figure 2d) (Kadir et al., 2011;Marutle et al., 2013;Ni et al., 2021;Rodriguez-Vieitez et al., 2015). We can observe an increased DED binding at preclinical / prodromal stages (reflected by 'first wave' in Figure 2d), lower DED binding (at AD dementia) and then, at advanced/end stage of AD dementia (postmortem), increase again in DED/BU99008 binding (represented by 'second wave' in Figure 2d). This most likely highlights the complexity of astrocyte heterogeneity and points towards different astrocyte states or subtypes/phenotypes (e.g. A1, A2, A (n) and A(n+1) at different stages of AD progression (Escartin et al., 2021;. From the data that we have collected from both in vivo PET as well as in vitro postmortem brain tissue studies, we can hypothesise that astrocytes act as the first ' frontline soldiers' who respond to initial injury by soluble Aβ species (Allaman et al., 2010;Narayan et al., 2014;Tong et al., 2017;Wyssenbach et al., 2016)  the preclinical/prodromal and initial AD dementia stages but were working in the background to prevent the brain from complete failure. They in turn could undergo reactive transformations and remodelling (leading to scar formation) at the advanced AD dementia stages-in order to give a last fight in attempting to rescue existing brain functions. The second wave represents several possibilities.
One possibility is that it most probably reflects the astrocytes efforts to limit the Aβ lesion/injury region through glial scar formation via remodelling (please refer to Sofroniew, 2009;Sun & Jakobs, 2012) for further reading). This could also somewhat explain why we observed a positive correlation between GFAP positivity/overexpression and 3 H-PIB and 3 H-florbetaben binding at the end stages of the disease. Another possibility is that these astrocytes represent a sub-population of ageing astrocytes in AD brains which become reactive with age and display high GFAP expression as reported by several studies (please refer to the review by Palmer and Ousman of disease-associated astrocytes with high GFAP-state, which they termed as DAAs, in an AD mouse model (Habib et al., 2020). DAAs seem to be upregulated during the initial phases of the disease and become more prominent and abundant as the disease progresses and are found to be in close proximity to Aβ. Moreover, they also observed DAAs in aged WT mice and in ageing postmortem human brains which are aligned with the second described possibility. These assumptions, based on published studies, need further exploration.
Ongoing studies in our group and larger patient studies along with the development of new promising PET-tracers (such as SMBT-1 Harada et al., 2020;Villemagne et al., 2020), which is in the pipeline) will shed more light on these observations and help to better understand astrocyte reactivity and heterogeneity as demonstrated here by 'first' and 'second' wave in AD pathogenesis.

| G FAP: A TRUE MARK ER OR A S HAM?
The glial fibrillary acidic protein (GFAP) was first isolated by Lawrence Eng and colleagues in 1971 and characterises the main intermediate filament in astrocytes (Eng et al., 1971). Abnormal GFAP up-regulation can indicate changes in astrocytes cytoskeleton integrity and is being acknowledged as one of the most trusted markers of astrocyte reactivity/ reactive astrogliosis (Yang & Wang, 2015). The current view of astrocyte reactivity in AD suggests that these cells can be very heterogeneous, assuming different phenotypes that seem to be region-specific and vary throughout disease progression (Escartin et al., 2021). Indeed, GFAP immunoreactivity was initially demonstrated to have a strong co-localisation with Aβ plaques and NFTs of tau in postmortem analyses of human brain tissue (Hanzel et al., 1999). Further studies led by Serrano-Pozzo, provided a reconceptualisation on how GFAP-reactive astrocytes associate with AD pathology (Serrano-Pozo et al., 2011). The authors showed that Aβ deposition reached a plateau before the onset of symptoms, while reactive astrocytes over-expressing GFAP, positively correlated with disease progression and increased number of NFTs (Serrano-Pozo et al., 2011). A recent systematic review encompassing 306 qualitative, semi-quantitative, and quantitative studies of immunohistochemical analyses of postmortem human brain tissue, established increased GFAP immunoreactivity as the most observed marker of reactive astrogliosis in AD (Viejo et al., 2021). However, detecting GFAP levels in the brain tissue is only possible after the demise of the patient. In this context, more refined methods for detecting GFAP in vivo have been developed.
Currently, plasma and CSF biomarker analyses offer an unprecedented way for assessing the influence of GFAP in AD progression in living patients. A meta-analysis by Bellaver et al. summarised the main findings of works that investigated GFAP levels in CSF and plasma (Bellaver et al., 2021). They showed that among 11 cohorts, including 239 AD patients and 205 cognitively unimpaired (CU) subjects, CSF GFAP was increased in AD compared to CU individuals (Bellaver et al., 2021). Yet, it is important to keep in mind that CSF GFAP may lack the desired specificity to detect AD, as it can also be altered in normal ageing and other neurodegenerative diseases (Michel et al., 2021;Nichols et al., 1993;Si et al., 2004). Supporting this view, a very recent study evaluated 504 individuals from the Swedish BioFINDER-2 cohort, including individuals which are CU, MCI/prodromal, AD, and non-AD disorders (Pereira et al., 2021) and showed that, even though CSF GFAP can predict cognitive decline, non-AD individuals had higher levels of GFAP in the CSF compared to MCI, AD and CU groups (Pereira et al., 2021). At the same time, plasma GFAP in AD may predict Aβ deposition, as measured by Aβ-PET, and cognitive decline (Cicognola et al., 2021;Pereira et al., 2021).
Very interestingly, plasma GFAP levels were unaltered in individuals diagnosed with non-AD dementias, such as frontotemporal dementia and Parkinson's disease (Pereira et al., 2021). Even though high variability to detect GFAP in the plasma was previously observed (Bellaver et al., 2021), these new lines of evidence could support the application of plasma GFAP for the detection of AD in its early stages (Cicognola et al., 2021;Pereira et al., 2021)  .
Keeping in mind the complexity of astrocytic heterogeneity reiterated in above sections, it is also important to mention that GFAPpositive cells (these cells could be further differentiated into high and low GFAP-state as demonstrated by Habib et al. (2020) thereby adding another level of complexity) may present a single population of reactive astrocytes, which seems to be highly dependent on the brain region. For instance, in these studies, careful analysis of rodent brains indicates that ~80% of astrocytes in the hippocampus are GFAP-positive, meanwhile, at least 40% of cells which are S100Bimmunoreactive are GFAP-negative (Bushong et al., 2002;Ogata & Kosaka, 2002), but further studies are still needed. Therefore, GFAP levels may not provide a full overview of reactive astrogliosis in the mammalian brain. However, following in the footsteps of published and ongoing studies in CSF and plasma, we can all agree that GFAP can provide initial overview of reactive astrogliosis in the brain and combining with other existing (discussed below) and new emerging markers, it may provide a more precise diagnosis of AD pathology in the future (in context of astrocyte reactivity).

| A S TRO C Y TE S OR MI CROG LIA : WHO IS FIR S T ?
Glial cells have crucial roles in protecting the CNS. Historically, astrocytes and microglial cells have been suggested as the main orchestrators of the neuroinflammatory symphony that seems to prevent the brain from evolving into a disordered state. Recent studies focused on understanding the early stages of AD suggest that astrocytes and microglia may contribute to disease progression, leading the brain into a perpetual inflammatory state (Greenhalgh et al., 2020). Noteworthy, the different states/phenotypes reactive astrocytes (as described above in detail) and activated microglia cells assume could differentially impact AD development (illustration in Figure 1c) (Escartin et al., 2021;Kumar et al., 2021;Nguyen et al., 2020;Ransohoff, 2016).

The microglial cells are resident immune cells in the CNS and
constitute 5-10% of total brain cells (Aguzzi et al., 2013). A simplistic view of microglia proposes the existence of pro-inflammatory (M1) or anti-inflammatory (M2) phenotype (Mills et al., 2000), a categorisation that has been constantly challenged Ransohoff, 2016). In fact, it seems that microglial cells may assume a wide range of phenotypes in response to toxins or pathological insults in the CNS . In 1992, Hardy & Higgins proposed the amyloid cascade hypothesis, suggesting that Aβ deposition leads to NFTs formation and cell death (Hardy & Higgins, 1992). The concept of Aβ driving the development of AD has been constantly debated and the failure of many Aβ-targeting clinical trials jeopardised the amyloid cascade hypothesis (Makin, 2018).
Interestingly, the recent (and controversial) approval of Aducanumab (Clinical trial ID: NCT02484547), a monoclonal antibody that activates microglial cells and reduces the concentration of Aβ plaques in the brain of AD patients, highlighted a key role of microglia in AD (Haeberlein et al., 2020;Sevigny et al., 2016).
A study led by Keren-Shaul characterised a specific type of microglia in a mouse model of AD using sub-tissue single-cell RNA-seq that undergo changes in morphology, function, and distributiontermed disease-associated microglia (DAM) (Keren-Shaul et al., 2017). Among the microglial genes differentially expressed in DAM, the triggering receptor expressed on myeloid cells 2 (TREM2) and Tyrobp, which are closely related to Aβ clearance, are upregulated (Keren-Shaul et al., 2017). The TREM2 is selectively expressed in microglial cells and TREM2 knock out in AD mice models seems to induce tau seeding and spreading around Aβ plaques (Gratuze et al., 2021). These insights from animal models of AD suggest that TREM2 is essential for microglial phagocytic function against Aβ plaques and to prevent tau pathology (Leyns et al., 2019) Alzheimer Network (DIAN) cohort corroborates these findings, indicating that longitudinal increase in CSF sTREM2 is associated with slower deposition of Aβ (Christian et al., 2021). Whether plasma sTREM2 levels are expressed differently in AD patients must be further explored, and the influence of different sTREM2 variants (e.g., the R47H TREM2 variant) should be considered (Ashton et al., 2019;Park et al., 2021;Vilalta et al., 2021). In fact, it seems that microglial activation could display a biphasic profile, in which different phenotypes (either beneficial or detrimental) might distinctly respond to the core pathology of AD, a behaviour very similar to astrocytes. In this context, Ising and colleagues showed that fibrillar forms of Aβ directly affect microglia following NLRP3 inflammasome activation which, in turn, induces tau pathology in an animal model of tauopathy, characterising a detrimental profile of microglia activation (illustrated in Figure 1c) (Ising et al., 2019). Intriguingly, our in vitro autoradiography also demonstrated that microglial activation (measured by [ 3 H]PK11195 binding) was closely related in older animals with high Aβ deposition (Ni et al., 2014). In accordance with these findings, a study by Pascoal et al. (2021)  In keeping with these findings, defining whether microglia serve as a friend (for instance inducing Aβ clearance) or foe (such as driving tau pathology) should be further evaluated. Based on the results discussed above, one could suggest that neuroinflammatory changes related to microglia may depend on Aβ deposition in AD (Joshi et al., 2014;Jung et al., 2015;Serrano-Pozo et al., 2011). On the opposite, reactive astrogliosis seems to be a key component in the early stages of AD as demonstrated by the 'first wave' in the previous section, prior to the formation of Aβ plaques (Calsolaro et al., 2021;Carter, Herholz, et al., 2019;Kumar et al., 2021;Rodriguez-Vieitez et al., 2015).
Among the potential markers of reactive astrogliosis (disregarding GFAP, which has been broadly discussed in the previous section) CSF YKL-40 and blood S100B are increased in AD patients (Bellaver et al., 2021). The YKL-40 is a glycoprotein, primarily expressed in astrocytes, that seems to be altered in inflammatory-related diseases, such as multiple sclerosis (Malmestrom et al., 2014) and AD (Llorens et al., 2017). Interestingly, changes in CSF levels of YKL-40 are also proposed as an early event in AD. Indeed, YKL-40 levels are increased in MCI patients compared to healthy controls. In addition, the ratio between CSF YKL-40 and Aβ42 is a predictor of cognitive impairment in normal subjects (Craig-Schapiro et al., 2010). Further experimental evidence has been recently provided to interpret the involvement of YKL-40 in AD (Lananna et al., 2020). It was shown that deletion of the YKL-40 promoter gene in APP/PS1 mouse, reduced the formation of insoluble Aβ aggregates in the mouse hippocampus (Lananna et al., 2020). In a similar fashion, S100B, a calcium sensor protein mostly localised in astrocytes, seems to be highly expressed in the blood of AD patients in the early stages (Peskind et al., 2001). The over-expression of human S100B in APP/PS1 mice led to an increased burden of Aβ (Mori et al., 2010). Moreover, higher levels of β-secretase were also highlighted in these animals, suggesting that S100B may accelerate Aβ release from APP, favouring the amyloidogenic pathway in AD (Mori et al., 2010). Thus, it is likely that reactive astrogliosis , as indexed by fluid levels of S100B and YLK-40, may impact AD pathology, favouring the formation and deposition of Aβ peptide. In addition to fluid biomarkers, reactive astrogliosis in AD can also be evaluated with PET-imaging using specific PET-tracers as presented in detail in above section. Combining experimental findings and clinical evidence from both fluid and imaging biomarker studies discussed above, the concept of high heterogenicity of reactive astrocytes reactivity in AD is again reinforced (Rodriguez-Arellano et al., 2016).
Whereas additional in vivo PET imaging studies are warranted to support the role of reactive astrocytes in early AD, cellular models of AD could help to explore astrocyte-related mechanisms of toxicity in the pre-amyloid phase (Fontana et al., 2020). Soluble Aβ oligomers (AβOs) have been widely demonstrated to affect astrocyte morphology and function in vitro (Allaman et al., 2010;Narayan et al., 2014;Tong et al., 2017;Wyssenbach et al., 2016). A potential mechanism that could pinpoint reactive astrogliosis as a trigger for amyloidosis in AD involves the internalisation of Aβ peptides in astrocytes (illustrated in Figure 1c). Specifically, it seems that prior to the formation of Aβ plaques, AβOs and Aβ protofibrils may interact with transporters/receptors in the membrane of astrocytes, such as the α7 nicotinic acetyl choline receptor (α7nAChR) (Xiu et al., 2005) and GLT-1 (Tong et al., 2017). Combined, AβOs and Aβ protofibrils can be internalised in astrocytes and following a high intracellular accumulation, macrovesicles release Aβ aggregates into the extracellular space, favouring the formation of Aβ plaques (Nagele et al., 2003;Sollvander et al., 2016). Thus, differentially from microglial activation, which could be a consequence of Aβ deposition, reactive astrogliosis via different mechanisms may favour amyloidosis in AD ( Figure 1c). We think the debate regarding who comes first should be investigated with a new perspective directed more towards how microglia and astrocytes complement each other (via protective or detrimental signalling pathways) in the pathogenesis of AD in response to internal and external triggers.

| CON CLUDING REMARK S
The AD biomarkers field is moving at a very fast pace with several new PET-tracers and fluid biomarkers being in the pipeline and continuously validated. These advances have surely shed much needed light on the role of astrocytes and reactive astrogliosis in the pathogenesis of AD and improved our understanding of their behaviour during different stages of disease progression. The in vivo PET-and postmortem brain tissue studies have provided convincing evidence that there are two waves of astrocyte reactivity during the whole cycle of AD pathogenesis, i.e the 'first wave' during presymptomatic/ prodromal stage and the 'second wave' at the advanced stages of AD dementia. At this point, we do not know if the 'two waves' are unique for AD but we can speculate that it is based on our studies.
If we talk about GFAP as a trusted marker for reactive astrogliosis, increasing evidence has shown that GFAP could provide the initial footprint of astrocyte reactivity in the brain. However, GFAP alone is not enough, and we need a combinatorial approach involving other biomarkers to properly understand the complexity and heterogeneity of astrocytes around AD pathology. It is safe to say that there is still a big divide when it comes to microglia or astrocyte in the pathology of AD. The studies have pointed towards a more synergistic behaviour between microglia and astrocytes in AD. In this regard, the future approaches focused on understanding the complementary role of microglia activation and astrocyte reactivity in AD progression will provide valuable insights that will assist in developing future therapeutic targets and interventions and benefit in better understanding the disease mechanisms of several other neurological disorders. To answer the big question of whether reactive astrogliosis is a friend or foe in AD, we think, based on the published studies and evidence, that reactive astrocytes are very resilient 'friendly' soldiers, that perform whatever functions/tasks are needed to maintain the brain homeostasis in normal and disease state till the end. However, under severe stress conditions they become non-functional and even harmful. If we look from a broader perspective, it seems like reactive astrogliosis is keeping the fight going till the end by undergoing transformations and remodelling to limit brain injury and promote cell survival. Nevertheless, these conclusions will require further investigation and exploration.

ACK N OWLED G EM ENTS
This study was financially supported by the Swedish Foundation for Strategic Research (SSF) (RB13-0192), the Swedish Research Council

CO N FLI C T O F I NTE R E S T S
The authors declare no conflict of interests.

AUTH O R CO NTR I B UTI O N
AK and AN conceptualised the review. AK, IF and AN wrote the first draft of the manuscript. All authors approved the final version of the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable since no new data were generated for this Review article.