Proteomics of the astrocyte secretome reveals changes in their response to soluble oligomeric Aβ

Abstract Astrocytes associate with amyloid plaques in Alzheimer's disease (AD). Astrocytes react to changes in the brain environment, including increasing concentrations of amyloid‐β (Aβ). However, the precise response of astrocytes to soluble small Aβ oligomers at concentrations similar to those present in the human brain has not been addressed. In this study, we exposed astrocytes to media from neurons that express the human amyloid precursor protein (APP) transgene with the double Swedish mutation (APPSwe), and which contains APP‐derived fragments, including soluble human Aβ oligomers. We then used proteomics to investigate changes in the astrocyte secretome. Our data show dysregulated secretion of astrocytic proteins involved in the extracellular matrix and cytoskeletal organization and increase secretion of proteins involved in oxidative stress responses and those with chaperone activity. Several of these proteins have been identified in previous transcriptomic and proteomic studies using brain tissue from human AD and cerebrospinal fluid (CSF). Our work highlights the relevance of studying astrocyte secretion to understand the brain response to AD pathology and the potential use of these proteins as biomarkers for the disease.


Recently, plasma GFAP levels have been associated with increased
Aβ pathology (Benedet et al., 2021;Pereira et al., 2021) and are elevated in older individuals at risk of AD (Chatterjee et al., 2021) further highlighting the relevance of astrocytes in the disease.
Astrocytes react to Aβ plaques in AD, however, the specific functional changes that this reactivity causes and the implications in the onset and progression of AD remain uncertain. Exposure to extracellular Aβ induces an astrocytic response in culture, as evidenced by an increase in astrocyte-mediated neurotoxicity through the release of N-SMase (Jana & Pahan, 2010) and soluble inflammatory factors (Garwood et al., 2011), or increased synaptotoxicity mediated by the secretion of factors such as glutamate (Talantova et al., 2013), complement C3 (Lian et al., 2015(Lian et al., , 2016, or CXCL1 (Perez-Nievas et al., 2021). Additionally, Aβ exposure may interfere with astrocytic protective functions as it results in decreased secretion of synaptogenic factors such as TSP-1 (Rama Rao et al., 2013) or TGF-β1 (Diniz et al., 2017) and impairs phagocytosis and degradation of dystrophic neurites (Sanchez-Mico et al., 2021).
Previous proteomics studies have characterized the astrocyte secretome in resting conditions (Dowell et al., 2009;Greco et al., 2010;Han et al., 2014;Lafon-Cazal et al., 2003) and in response to proinflammatory stimuli such as lipopolysaccharide (LPS) or cytokines (Delcourt et al., 2005;Keene et al., 2009;Lafon-Cazal et al., 2003), as well as specific insults including cholinergic stimulation (Moore et al., 2009), angiogenin (Skorupa et al., 2013), mechanic injury (Lai et al., 2013;Thorsell et al., 2008) or in response to endoplasmic reticulum (ER) stressors (Smith et al., 2020). The astrocyte secretome has also been characterized in response to synthetic Aβ42 (Lai et al., 2013). However, much of the existing research characterizing the astrocyte response in AD is limited due to the lack of appropriate tools to mimic in culture the species and concentrations of Aβ found in the AD brain. Synthetic Aβ peptides have been typically used at doses that exceed the concentrations of Aβ found in AD brain by 100-1000 times (Allaman et al., 2010;Diniz et al., 2017;Lai et al., 2013;Varshavskaya et al., 2022). In addition, the nature of the peptides is crucial to determine their underlying effects. Rather than monomers or highly aggregated forms, soluble Aβ oligomers are the species typically associated with synaptic dysfunction and loss of dendritic spines (Arbel-Ornath et al., 2017;Lacor et al., 2004;Shankar et al., 2007Shankar et al., , 2008, increased tau phosphorylation and missorting into dendrites (Zempel et al., 2010), impairment of axonal transport (Sherman et al., 2016;Vossel et al., 2010) or increased reactive oxygen species (ROS) production (Behl et al., 1994).
Investigating how astrocyte secretion is modulated in response to naturally secreted forms of Aβ will better help to understand the contribution of astrocytes to AD pathology.
In this study, we characterized changes in the astrocyte secretome in response to soluble Aβ oligomers, at concentrations and species similar to those present in the human brain, secreted into the culture medium from neurons that express a human APP transgene with the double Swedish mutation (APPSwe, Tg2576 mice (Hsiao et al., 1996)). APPSwe neurons also secrete other human APP-derived fragments (Wu et al., 2010), providing a cellular environment that resembles that one experienced in the brain. Using an optimized methodology to analyze the extracellular media with mass spectrometry while discriminating serum components, we generated a list of proteins whose levels are significantly changed in the astrocyte secretome upon treatment with neuron-derived Aβ oligomers. We used functional and pathway annotation to explore the processes that are altered in astrocytes in these disease-mimicking conditions. Our data identified that Aβ exposed astrocytes show altered secretion of proteins involved in the reorganization of the extracellular matrix and the cytoskeleton, as well as in protective antioxidant and chaperone function responses. Some of the identified proteins are altered in CSF and tissues from AD patients, supporting the idea that astrocyte-secreted proteins can be explored as potential biomarkers for disease and can aid the understanding of functional astrocyte changes in AD. under the control of the hamster prion protein promoter (Hsiao et al., 1996). Tg2576 mice were originally obtained from Taconic farms (Germantown, NY, USA), and were maintained in-house by breeding males with C57Bl/6/SJL F1 (MGI:5655343) females as recommended by the suppliers to minimize aggressive phenotypes.

| Animals
Moderate severity phenotypes were monitored, and animals were bred before 6 months of age. Up to 5 animals of the same gender were kept in the same cage. The genotype of the animals was determined by polymerase chain reaction on DNA obtained from the embryos, as previously described (Mitchell et al., 2009). Water and food were available (Picolab rodent diet 20; # 5053; Lab Diet) ad libitum. Animals were housed at 19-22°C, humidity 55%, 12-h:12-h light:dark cycle with lights on at 07:30. A total of 5 CD1 dams were used to obtain pools of 10 pups on average that were used to prepare primary glial cultures. A total of 5 breeding pairs from Tg2576 and C57Bl/6/SJL crosses were used to obtain embryos from which to prepare either wild type or Tg2576 neuronal cultures. No animals were excluded. Mice were sacrificed by cervical dislocation or decapitation according to Home Office-approved protocol.

| Culture of primary cortical neurons and collection of neuronal conditioned media
To obtain oligomeric Aβ enriched conditioned media, we cultured neurons from the cerebral cortex of Tg2576 mice bearing the human APPSwe mutation at embryonic day 15 (E15) as previously described (DaRocha-Souto et al., 2012;Perez-Nievas et al., 2021;Wu et al., 2010 in Neurobasal media with 0.5% B27, to limit the amount of highabundance proteins such as albumin. Equivalent amounts were used to prepare media from control neurons (WT_NCM).

| Culture of primary mouse astrocytes and collection of astrocyte lysates and conditioned media
Primary astrocyte cultures were prepared from the cortex of wildtype CD1 mice on postnatal days 1-3 as previously described (Schildge et al., 2013). Briefly, brains from a littermate were harvested and placed in ice-cold HBSS with HEPES, where the meninges were removed, and the cerebral cortices were dissected, pooled together, and mechanically dissociated. Mixed glial cells were cultured in poly-D-lysine coated flasks and grown in high glucose (Gibco™, cat no. DMEM 21969035) media with 10% fetal bovine serum (Gibco™, cat no. 10500064), GlutaMAX™ supplement, and 100 units/mL penicillin and 100 μg/mL streptomycin at 37°C in a humidified incubator with 5% CO 2 . Media was changed after 2 days and then every 5 days. On days 11-14, microglia cells and oligodendrocytes were removed by overnight shaking at 200 rpm and astrocytes were seeded on poly-D-lysine (Sigma, cat no. P6407) coated 6-multiwell plates and were used after 3 days at 80-90% confluency (>95% of cells were GFAP-positive).

Astrocyte growth medium was washed and changed to
Neurobasal media supplemented with 0.5% B27 medium 24 h prior to treatment. Astrocytes were cultured in Neurobasal media without B27 for 24 h, to collect astrocyte-conditioned media to prepare an in-house library. Astrocytes were cultured in APPswe_NCM and WT_NCM for 24 h, to collect astrocyte conditioned media to analyze by mass spectrometry (APPse_ACM and WT_ACM). Media were filtered through a 0.22 μm spin filter to eliminate any detached cells or debris and concentrated through a 3 kDa molecular weight cutoff concentrating column (Millipore, cat no. UFC500396) at 13 000 g for 30 min, prior to analysis. For validation by western blotting, cells were washed once in PBS and harvested on lysis buffer (20 mM Tris-HCl pH 6.8, 137 nM NaCl, 1 mM EGTA, 1% Triton X-100, 10% glycerol, 1x Roche complete mini protease inhibitor).

| Secretome analysis
Conditioned media from 5 independent neuronal cultures, with each biological replicate analyzed in duplicate, were analyzed as previously described  by Secret3D workflow . Collected media was filtered on microcon filters with 10 KDa cutoff (Millipore, cat no. MRCPRT010), and buffer was exchanged with 8 M Urea 100 mM Tris pH 8. 50 mg of secreted protein was sonicated with BIORUPTOR (3 cycles: 30 s on/30 s off). Cysteine reduction and alkylation were performed by adding 10 mM TCEP (Thermo Scientific, cat no. 20490) and 40 mM 2-chloroacetamide (Sigma-Aldrich, cat no. C0267) in 8 M Urea 100 mM Tris pH 8 for 30 min at room temperature. By using microcon filters with 10 kDa cutoff (Millipore), the buffer was exchanged by centrifugation at 9300 g for 10 min, and PNGase F (New England Biolabs, cat no. P0708) (1:100 = enzyme: secreted proteins) was added for 1 h at room temperature following manufacturer's instruction. Buffer was again exchanged by centrifugation at 9300 g for 10 min with 50 mM ammonium bicarbonate, and proteins in the solution were digested by Lys-C and trypsin (Kulak et al., 2014). Peptides were recovered on the bottom of the microcon filters by centrifugation at 9300 g for 10 min, adding two consecutive washes of 50 μL of 0.5 M NaCl. Eluted peptides were purified on a homemade C18 StageTip. 1 μg of the digested sample was injected onto a quadrupole Orbitrap Q-exactive HF mass spectrometer (Thermo Scientific). Peptide separation was achieved on a linear gradient from 95% solvent A (2% ACN, 0.1% formic acid) to 55% solvent B (80% acetonitrile, 0.1% formic acid) over 75 min and from 55% to 100% solvent B in 3 min at a constant flow rate of 0.25 μL/min on UHPLC Easy-nLC 1000 (Thermo Scientific) where the LC system was connected to a 23cm fused-silica emitter of 75 μm inner diameter (New Objective, Inc.), packed in-house with ReproSil-Pur C18-AQ 1.9 μm beads (Dr. Maisch Gmbh, Ammerbuch, Germany) using a high-pressure bomb loader (Proxeon).
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (www.ebi.ac.uk/ pride/) (Perez-Riverol et al., 2022) partner repository with the dataset identifier PXD036343.

| MS analysis and database search
MS analysis was performed as reported previously (Geyer et al., 2016). Raw MS files were processed with MaxQuant software (1.6.0.16), making use of the Andromeda search engine (Cox et al., 2011). MS/MS peak lists were searched against the UniProtKB Mouse complete proteome database (uniprot_cp_mouse_2019) in which trypsin specificity was used with up to two missed cleavages allowed. Searches were performed selecting alkylation of cysteine by carbamidomethylation as fixed modification, and oxidation of methionine, N-terminal acetylation, and N-Deamination as variable modifications. Mass tolerance was set to 5 ppm and 10 ppm for parent and fragment ions, respectively. A reverse decoy database was generated within Andromeda, and the false discovery rate (FDR) was set to <0.01 for peptide spectrum matches (PSMs). For identification, at least two peptide identifications per protein were required, of which at least one peptide had to be unique to the protein group. Matching between runs was performed across all samples that are neuron-conditioned media and astrocyte-conditioned media, plus conditioned media from astrocyte without serum that was used as reference library.

| ELISA
A quantitative ELISA kit was used to detect mouse APOE (Abcam, Cat no. Ab215086) in conditioned media according to the manufacturer's instructions.

| Western blotting
Media concentrated through a 3 kDa molecular weight cut-off column and cell lysates were resuspended and boiled in Laemmli buffer.
Samples were subjected to 12% SDS-PAGE and transferred to a ni- Cruz, Cat no. sc-32233), followed by 2 h incubation in the appropriate anti-mouse or anti-rabbit secondary antibodies. Signal was visualized using an Odyssey CLx infrared imager (LI-COR Biosciences) and quantified using ImageStudio Lite (Li-COR) software.

| Quantification and statistical analysis
Sample size was estimated based on previous analysis where changes in astrocyte secretion were determined in response to similar media containing Aβ oligomers (Perez-Nievas et al., 2021) and secretome studies and was further validated by principal Histogram of all the values log2 transformed was performed to confirm if the transformed matrix follows the normal distribution prior to T-test statistical analysis that was performed applying FDR <0.05 or p < 0.05, as reported.

| Bioinformatics analysis
Gene ontology (GO) term over-representation for biological process, molecular function, and cell component were identified using LAGO (Boyle et al., 2004) (https://go.princ eton.edu/). P-values were calculated based on the hypergeometric distribution with the Bonferroni correction for multiple comparisons, with p < 0.05 after the correction considered significant. STRING 11.5 (https://strin g-db. org/) (Szklarczyk et al., 2021) was used to predict protein-protein interactions (PPIs) within the differentially expressed proteins, and local STRING network clusters were used to identify if any of the interacting proteins take part in the same biological process. Pathway enrichment analysis was performed using the Reactome pathway database using the online PANTHER v16.0 software (http://panth erdb.org/) (Mi et al., 2021) using Fisher's Exact test and FDR to correct for multiple comparisons. Statistical significance was defined as FDR <0.05.

| Analysis of the astrocyte secretome using an in-house library shows changes in response to Aβ oligomers
To investigate changes in the astrocyte secretome in response to soluble Aβ oligomers similar to those found in human disease brain (Arbel-Ornath et al., 2017;Shankar et al., 2008;Wu et al., 2010), we exposed primary mouse astrocytes to conditioned media of neurons derived from Tg2576 mouse embryos that express a human APP transgene harboring the double Swedish mutation K670N, Hsiao et al., 1996)  Aβ42:Aβ40 ratios of 1:10 as in human AD brain, and concentrations ranging between 2 and 8 nM, similar to what it has been reported in human CSF (Snider et al., 2009). Conditioned media was diluted to treat primary astrocytes with 2 and 0.2 nM concentration of Aβ40 and Aβ42 oligomers, respectively. Astrocytes were exposed to APPSwe neuron-conditioned media (APPSwe_NCM) or wild type neuron conditioned media (WT_NCM) and the astrocyteconditioned media was collected after 24 h. The conditioned media from stimulated astrocytes (APPSwe_ACM) or control (WT_ACM) were subjected to mass spectrometry, together with APPSwe_NCM or WT_NCM (Figure 1a), to control for the components that were already present in neuron media. Importantly, exposure to Aβcontaining media at the concentration and time used in this study did not result in astrocyte cell death, as we previously reported (Perez-Nievas et al., 2021), and therefore altered cellular viability is not a confounding factor in the proteomic analysis.
An initial analysis using traditional shotgun proteomics identified 109 proteins, 14 of which were significantly altered in the secretome of astrocytes exposed to APPswe neuron conditioned media (APPSwe_ ACM) compared to wild type (WT_ACM) (data not shown). Proteomics of the secretome is usually limited by the high abundance of serum proteins, which creates an excessive dynamic range between albumin and other proteins contained in the media. Our experimental design required the use of serum-like supplement to ensure neuron survival, and this contains albumin and other proteins in high abundance (Brewer et al., 1993). Albumin immunodepletion strategies were not considered to avoid the nonspecific loss of proteins of interest (Bellei et al., 2011). To overcome these limitations, we prepared an in-house library of proteins contained in the conditioned media of astrocytes grown in media without serum for 24 h and used it as a matching library following a previously published protocol (Geyer et al., 2016).
We identified 2095 proteins in the secretome of astrocytes treated with media from 5 independent neuronal cultures (Table S1). Figure 1b shows a volcano plot comparing the secretome from astrocytes treated with media from APPswe neurons (APPswe_ACM) with that of astrocytes treated WT neurons (WT_ACM), which rendered a list of 94 significantly altered proteins (Table S1). Because astrocytes were treated with media that contains other neuron-secreted proteins in addition to Aβ oligomers we excluded those proteins that were found both in APPSwe_NCM or WT_NCM respectively (Table S1). This resulted in a final list of 45 astrocyte-secreted proteins that were affected by exposure to oligomeric Aβ, being differentially expressed in F I G U R E 1 Proteomics analysis of the astrocyte conditioned media in response to Aβ-containing media. (a) Schematic representation of the experimental approach. Conditioned media (NCM) from primary neurons prepared from Tg2576 mice (APPswe) or wild-type littermates (WT) mice were collected at DIV14. Wild-type astrocytes were treated with either APPswe-conditioned media (APPswe_NCM) or WTconditioned media (WT_NCM), and the resultant astrocyte-conditioned media (ACM) was analyzed by mass spectrometry. Traditional shotgun proteomics identified 109 proteins differentially secreted by astrocytes treated with Aβ-containing media while referencing to an astrocyte library increased the yield to 2095 proteins. Created with BioRe nder.com. (b) Volcano plot showing proteins significantly up (red) and downregulated (blue) in the secretome of Aβ-stimulated astrocytes, APPswe_ACM, compared to control, WT_ACM, before proteins contained in neuron media were excluded. The significance threshold was set at ≥1.3 and the absolute fold change threshold was set at >0.5. (c) Heatmap of proteins differentially secreted from Aβ-stimulated astrocytes, after excluding proeins present in NCM. The blue-to-red scale indicates low-to-high protein levels. Differentially expressed proteins were defined based on a p < 0.05.
APPSwe_ACM compared to WT_ACM ( Figure 1C). Of these identified proteins, 23 and 22 proteins were significantly increased or decreased, respectively, when astrocytes were treated with media from APPswe neurons containing Aβ oligomers (Table 1). Among these proteins increased, we identified known markers of reactive astrocytes such as APOE, vimentin, and SERPINA3N (Escartin et al., 2021;Zamanian et al., 2012), whose expression increases in astrocytes in human AD brains (Viejo et al., 2022), demonstrating the disease relevance of the proteins identified. Further analysis through ELISA and Western blotting, validated the changes observed in two of these proteins, APOE and PPIB, in astrocyte media ( Figure S1).

| Over two-thirds of Aβ -modulated astrocytesecreted proteins follow non-conventional secretory pathways
Prediction tools for subcellular localization based on experimental evidence such as DeepLoc, LAGO terms, or ProtComp, predicted that only 26-35% of the differentially secreted proteins are typically found in the extracellular space (Figure 2a-c). In line with this, the procollagen C proteinase enhancer, PCEP1 or PCOLCE (Adar et al., 1986). PREP, a prolyl endopeptidase involved in the degradation of collagen was also secreted upon Aβ exposure (Gaggar et al., 2008).
Although generally functioning as intracellular proteins, a high proportion of cytoskeletal components were among the proteins found secreted in response to Aβ containing media. These included three out of the six known actin isoforms, β-and γ-cytoplasmic actins (ACTB and ACTG1) and cardiac actin (ACTC1), as well as the actin-binding proteins tropomyosin α-4 chain and α-parvin, septin 7, and the astrocyte-specific intermediate filament vimentin (Figure 3a,b, Figure S2 and Tables S4-S6). Some of these are important structural components of focal adhesions that mediate contacts with the ECM.  (Chang et al., 2005(Chang et al., , 2006; and prolyl isomerase (PPIA), although cytoplasmic, is secreted as a defense mechanism in response to oxidative stress (Nigro et al., 2013). The observed changes in collagen type I may also result from a transcriptional response to oxidative stress . Similar findings were obtained when using STRING to identify known and predicted physical and functional protein-protein interactions (PPIs) (Szklarczyk et al., 2021), which resulted in a significant PPI enrichment in differentially upregulated proteins (p-value = 3.08 × 10 −11 , n = 23), indicating that the number of interactions in these proteins network is significantly different from random networks. STRING local network clustering analysis revealed proteins involved in collagen biosynthesis and detoxification of reactive oxygen species (Figure 3c and Table S7).

| Comparison of the Aβ -stimulated astrocyte secretome with astrocyte transcriptomics and CSF proteomics studies in AD
In recent years, single-nucleus transcriptomics has identified transcriptional changes that occur specifically in astrocytes in the human AD brain. To put our results in context with human data, we compared the list of proteins that we generated with the astrocyte-specific genes that were found to be differentially expressed in early-vs late-stage human AD brain (Mathys et al., 2019), in AD vs control brain (Grubman et al., 2019;Lau et al., 2020;Sadick et al., 2022) and showing a positive correlation with Aβ or phospho-tau pathology .
We also included a comparison with a study where astrocyte-specific gene clusters were applied to 766 whole brain transcriptomes, including control, mild cognitive impairment (MCI), and demented (AD) cases (Galea et al., 2022) (Table S9). After selecting those genes that TA B L E 1 List of proteins showing significant differences in the secretome of astrocytes treated with WT_NCM versus APPSWE_ NCM (p-value <0.05).

TA B L E 1 (Continued)
encode proteins predicted to be secreted, we searched for those that were differentially regulated by Aβ in our study (Figure 4a). Despite the great degree of variability among published RNAseq datasets, some astrocyte-secreted proteins are consistently found in AD transcriptomic studies. For example, ALDHL1 or PEBP1, whose secretion is increased upon Aβ treatment, were found upregulated in AD in 3 out of 7 studies (Figure 4a). APOE, ACTB, or PFKP are consistently found in nearly all transcriptomic studies, but these were reported as either up-or down-regulated (Figure 4a). BCAN is found in 4 out of 7 comparisons as upregulated in AD, while its secretion from astrocytes is decreased in our study, suggesting a complex role in the regulation of this protein in the AD brain.
To further explore whether changes in protein secretion from astrocytes may reflect changes in CSF in AD, we compared our data with three systematic reviews as well as with three recent proteomic studies, which characterized the proteomic composition of CSF in AD patients compared to controls (Bader et al., 2020;Bastos et al., 2017;Higginbotham et al., 2020;Pedrero-Prieto et al., 2020;Visser et al., 2022;Wesenhagen et al., 2020)

F I G U R E 3 Bioinformatics analysis of proteins differentially secreted in astrocytes treated with Aβ-oligomers containing media. (a)
Functional annotation clustering analysis of differentially upregulated proteins using Database for Annotation, Visualization and Integrated Discovery (DAVID). The analysis included gene ontology (GO) and pathway (KEGG pathway, Reactome, and WikiPathways) terms.
Only clusters with an enrichment score of ≥1.3 were considered significant. (b) Table summarizing the functional categories of proteins differentially secreted by Aβ-treated astrocytes. The categories were chosen and assigned based on the bioinformatics analysis and manual protein annotation. Arrows in the brackets next to the gene names indicate whether the protein was found differentially up-(↑) or downregulated (↓). (c) Protein-protein interaction network for differentially upregulated proteins generated using STRING. Lines represent protein-protein associations (including but not limited to physical binding) and different thicknesses represent confidence in the interaction. Thicker lines represent higher confidence in the interaction. Circles represent individual proteins and circles of the same color represent proteins that were identified to be involved in the same pathway using STRING local network cluster analysis.
TNR, and SPOCK2), appeared as modified in AD CSF in previous studies (Bastos et al., 2017;Higginbotham et al., 2020;Pedrero-Prieto et al., 2020;Visser et al., 2022;Wesenhagen et al., 2020). For many of the identified proteins, changes are not always consistent across CSF analyses and increased or decreased levels have been simultaneously reported in different studies. APOE is consistently found in nearly all CSF proteomic studies but, similar to the transcriptomic analysis, its levels are found up or downregulated depending on the study. PEBP1, PTGDS, SOD3, CHGA, and SCG5 have been repeatedly identified (4 out of 10 studies) when monitoring for changes in CSF in AD relative to controls (Figure 4b). In fact, chromogranins and secretogranins have been proposed as AD biomarkers with decreased levels of CSF in AD (Abdi et al., 2006;Brinkmalm et al., 2018;Hölttä et al., 2015). Our data suggest that a proportion of the proteins that are commonly found in CSF in AD might have an astrocytic origin.
We used DisGeNET, an online database that compiles information from curated resources, GWAS catalogs, animal models, and published papers (Piñero et al., 2021) to estimate the gene-disease association (GDA) score for genes encoding differentially secreted proteins in our study. 21 out of 44 (47.7%) genes were found to be associated with AD ( Figure S4 and Table S10). Only APOE showed a strong association, with a GDA score of 0.7, while the rest of the genes had GDAs ≤0.1, consistent with the notion that, with the ex-

| DISCUSS ION
Astrocytes are secretory cells that, in response to brain injury, disease, or signals derived from other cell types, release a myriad of factors that modulate neuronal and non-neuronal cells and lead to multiple functional responses (Sofroniew, 2020;Verkhratsky et al., 2016). In Alzheimer's disease, astrocytes become reactive, partly due to their exposure to different Aβ species present in the extracellular environment (reviewed in Perez-Nievas & Serrano-Pozo, 2018). To our knowledge, only one study has examined the secretory response of astrocytes to Aβ and they used synthetic Aβ42 in the μM range (Lai et al., 2013). Our work sought to understand how the secretory profile of astrocytes is altered in response to naturally occurring soluble oligomeric Aβ, in similar nanomolar concentrations and oligomeric species to those found in AD brain (DaRocha-Souto et al., 2012;Hudry et al., 2012;Wu et al., 2010).
While we cannot exclude additional effects due to other human APP-derived fragments or to proteins secreted from APPswe expressing neurons, we believe that our experimental setting closely mimics the brain environment to which astrocytes are exposed in AD. We found changes in 45 proteins, of which 23 showed higher secretion and 22 lower secretion upon exposure of astrocytes to a mixture of 1:10 Aβ42:Aβ40. Our study shows some similarities with Lai et al. (2013), with two proteins, APOE and prostaglandin H2 Disomerase (PTGDS), identified in both datasets. This suggests a very likely implication of these two proteins in the response to Aβ. In line with this, we found that both proteins are commonly identified in CSF, with PTGDS also known as β-Trace protein (BTP), one of the main constituents of CSF (Hoffmann et al., 1993;Urade, 2021). The gene-disease association analysis found APOE to have the strongest link with AD, which is to be expected given that the APOEɛ4 variant is the strongest genetic risk factor for sporadic AD . In the brain, APOE is predominantly secreted by astrocytes and, to a lesser extent, by microglia .
APOE is involved in cholesterol metabolism and, recently, saturated lipids contained in lipoparticles such as APOE have been suggested as mediators of astrocyte-induced toxicity (Guttenplan et al., 2021).
PTGDS catalyzes the conversion of prostaglandin H2 to prostaglandin D2, an arachidonic acid metabolite that modulates the inflammatory response (Joo & Sadikot, 2012) and whose levels are affected in AD brain (Iwamoto et al., 1989;Wong et al., 1992). Moreover, PTGDS-mediated synthesis of prostaglandin D2 in astrocytes modulates in turn the expression of inflammatory mediators in microglia (Choi et al., 2019) and PTGDS secretion has been reported from oligodendrocytes (Pan et al., 2023) and astrocytes (Giacomelli et al., 1996). Besides these two proteins, our secretory profile differed significantly from that of astrocytes treated with synthetic Aβ42 (Lai et al., 2013). These differences are not unexpected. Unlike synthetic Aβ, which generates oligomers and fibrils easily within minutes to hours, especially when used at high concentrations, naturally secreted Aβ peptides assemble into low molecular weight species (dimers to tetramers). These small soluble oligomeric Aβ species are the most bioactive and have synaptotoxic activity (Shankar et al., 2008;Yang, Li, et al., 2017), and astrocytes treated with these are more likely to reflect their responses in AD brain.
A high proportion of proteins identified in this study are important for the organization of the extracellular matrix (ECM).

Alterations in ECM composition have been reported in several
neurodegenerative diseases, including AD (Freitas et al., 2021;Ma et al., 2020). In relation to astrocytes, a transcriptomic analysis comparing AD brains with controls identified specific clusters of astrocytes with upregulation of genes involved in ECM organization . Astrocytes participate in the formation of the extracellular matrix by secreting several factors, including proteoglycans and tenascins (Anwar et al., 2021) and, when reactive or upon Aβ treatment, astrocytes have been implicated in ECM degradation through the release of matrix metalloproteinases (Deb et al., 2003;Muir et al., 2002). Our data also support this active role of astrocytes as modifiers of the ECM composition and function in AD. Of particular interest is the decreased secretion of two of the main components of the perineuronal net (PNN), brevican and tenascin R. PNNs are lattice-like assemblies of ECM proteins that surround neurons to protect them and modulate neuronal activity (Reichelt et al., 2019;Wen et al., 2018).
Evidence from post-mortem brain studies suggests loss of PNNs in AD (Baig et al., 2005), although other studies found preservation of these structures in disease (Brückner et al., 1999;Lendvai et al., 2013;Morawski et al., 2010). Our data suggest that astrocytes exposed to pathological Aβ may decrease the secretion of PNN components, which can lead to a reduced stabilization of synapses and enhanced exposure to oxidative stress and other toxic molecules, including Aβ itself (Miyata et al., 2007;Suttkus et al., 2012).
Many of the up-regulated proteins found in this study have antioxidant activity and/or are involved in the response to F I G U R E 4 Comparison with transcriptomic and cerebrospinal fluid (CSF) Alzheimer's disease studies. (a) Venn diagrams show the numbers of proteins that overlap between this study and other transcriptomic studies of human AD samples. The number of proteins from transcriptomic studies only includes the proteins that were predicted to be extracellular or secreted using the combination of localization prediction tools used in this study. When changes in gene expression in AD brain compared to controls were in the same direction as the secretion of proteins in response to Aβ oligomers relative to control, names are shown in green; when changes were in the opposite directions, names are shown in red; when genes were reported to be both up-and downregulated, the names are shown in orange. (b) Table displaying the proteins that were identified in this study and that were also differentially expressed in CSF from AD compared to controls.
oxygen-containing compounds. Sings of oxidative damage are observed in aged and AD brains (Wang et al., 2014), starting in early phases of the disease (Butterfield et al., 2006;Keller et al., 2005).
Some of the proteins we found dysregulated in this study have been previously linked to oxidative stress and AD. Peroxiredoxin-1, which catalyzes the reduction of hydrogen peroxide and similar compounds, was found up-regulated in the cortex of AD patients (Szeliga, 2020), and a recent large-scale proteomic study discovered that peroxiredoxin-1 levels were elevated in both AD brain tissues and CSF . Moreover, peroxiredoxin-1 conferred resistance to Aβ toxicity in PC12 cells, SH-SY5Y cells, and rat primary hippocampal neurons (Cimini et al., 2013;Cumming et al., 2007). Extracellular superoxide dismutase 3 (SOD3) is another antioxidant enzyme (Wang et al., 2018) that ameliorates the oxidative damage exerted by Aβ25-35 in SH-SY5Y (Yang, Wei, et al., 2017), and its levels in CSF are frequently changed in AD (Bader et al., 2020;Perrin et al., 2011;Ringman et al., 2012;Visser et al., 2022). In cultured astrocytes, Aβ triggers the production of reactive oxygen species (ROS) (Abramov et al., 2004;Askarova et al., 2011). The astrocyte response we report here may represent a self-defense mechanism to protect themselves against the oxidative stress caused by Aβ or a mechanism to provide antioxidant support to neurons (Jiwaji & Hardingham, 2022).
We also report an increased secretion of molecular chaperones.
While typically considered intracellular proteins, chaperones can be secreted through classical and non-classical secretion mechanisms with important roles in extracellular proteostasis (Chaplot et al., 2020). PTGDS is found in plaques in AD brain and its Aβchaperone activity has been reported, through both inhibiting Aβ aggregation and promoting fibril disaggregation (Kanekiyo et al., 2007;Kannaian et al., 2019). PDIA3 is a disulfide isomerase with a thiol oxidoreductase activity that promotes protein folding in the ER (Chichiarelli et al., 2022). It was detected in human CSF bound to Aβ (Erickson et al., 2005) (Brehme et al., 2014), secretogranin 5, also known as 7B2, whose levels are reduced in our Aβ-exposed astrocyte secretome, has also been suggested to function as a chaperone, since it colocalizes with Aβ plaques in human AD brain and prevents fibrillation of Aβ40 and Aβ42 in vitro and blocks their cytotoxic effect in culture (Helwig et al., 2013). Singlecell RNAseq and immunohistochemical studies have frequently reported that genes involved in proteostasis and chaperonemediated responses are up-regulated specifically in astrocytes in AD (Grubman et al., 2019;Lau et al., 2020;A. M. Smith et al., 2021;Viejo et al., 2022) which, together with our study, may imply that astrocytes are a source of extracellular chaperones to protect the disease brain environment.
Three ubiquitously expressed actin isoforms are among the secreted proteins in response to Aβ. Different roles of actin as an extracellular protein have been reported (Sudakov et al., 2017), including its function as a "danger-associated molecular pattern" (DAMP) (Ahrens et al., 2012;Srinivasan et al., 2016). DAMPs trigger an inflammatory response by binding to receptors such as Toll-like receptors, which are expressed mostly in microglia and astrocytes in the brain (Sofroniew, 2020). Other proteins found in our study have been suggested as DAMPs, including peroxiredoxin 1 (PRDX1), which binds to TLR-4 after secretion from cancer cells (Liu et al., 2016;Riddell et al., 2010), or prolyl endopeptidase (PREP) that cleaves collagen to produce proline-glycine-proline (PGP), which acts as a DAMP (Patel & Snelgrove, 2018) and participates in the recruitment of neutrophils (Weathington et al., 2006).
In addition, other proteins upregulated in the Aβ-induced astrocyte secretome have been directly implicated in inflammation. PTGDS synthesizes prostaglandin D2, which functions as a modulator of the inflammatory response (Joo & Sadikot, 2012); peptidylprolyl isomerase A (PPIA) is mainly cytoplasmic but it can also be secreted in response to different inflammatory stimuli such as LPS (Hoffmann & Schiene-Fischer, 2014), and inhibition of extracellular PPIA reduces neurotoxicity and neuroinflammation markers such as NF-kB activation in a mouse model of amyotrophic lateral sclerosis (ALS) (Pasetto et al., 2017). Altogether, a large proportion of the secreted proteins identified in response to Aβ oligomers may be involved in further amplifying or diminishing the inflammatory response initiated by Aβ. Surprisingly, we did not identify any cytokines or chemokines. Cytokines and chemokines are typically secreted from astrocytes, and their secretion profile is altered in the AD brain and more specifically upon treatment with Aβ (González-Reyes et al., 2017), including our own studies using a similar experimental setting (Perez-Nievas, 2021). Cytokine and chemokine secretion is typically detected by immunoassay, while its detection through mass spectrometry is challenging due to their low molecular weight and low abundance relative to other proteins in the media (Kupcova Skalnikova et al., 2017), which may explain their absence in this dataset.
Proteins showing reduced secretion in this study constitute a more heterogeneous group. Among these proteins, we identified proteins related to mitochondrial integrity and function (NDUFAD and CHCHD3), glucose metabolism (H6PD and PFKP), energy sensors (PRKAA1 and SARNP), RNA binding proteins (HNRNPH2), intracellular trafficking (RAB9A and SNX5), as well as secreted neuroendocrine peptides (CHGA and SCG5).
Astrocyte proteins with increased secretion in response to Aβ largely overlapped with proteomics studies conducted in AD CSF.
Increased levels of astrocytic proteins in CSF, such as GFAP, have been associated with Aβ deposition, levels of phospho/total tau, and with markers of synaptic dysfunction (Milà-Alomà et al., 2020;Salvadó et al., 2021). In fact, a CSF proteomic study in the APPPS1 mouse model of amyloidosis highlighted increased levels of proteins related to glial activation and identified APOE and SERPINA3 as CSF biomarkers (Eninger et al., 2022). Our data suggest that some of the proteins that are commonly found in CSF in AD, might be of astrocytic origin and that changes in astrocytes can be reflected in the CSF. This may be the case for some proteins already proposed as AD biomarkers such as chromogranins (Abdi et al., 2006;Brinkmalm et al., 2018;Hölttä et al., 2015).
To conclude, our study shows that in response to human Aβ oligomers, in similar concentrations and species to those found in AD brain, astrocytes present a distinct profile encompassing ECM molecules, antioxidant proteins, components of the cytoskeleton, chaperones, and regulators of inflammatory responses. These changes mimic some of those identified in AD transcriptomic and CSF studies and contribute to a better understanding of how astrocytes play a pivotal role in the molecular response to Aβ in AD and can be relevant for the identification of novel biomarkers.

AUTH O R CO NTR I B UTI O N S
VM performed the mass spectrometry analysis and database Transition Support Award (MR/V036947/1) (MJ-S).

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare no conflict of interest to declare that are relevant to the content of this article.

O PE N R E S E A RCH BA D G E S
This article has earned an Open Data badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at: www.ebi.ac.uk/pride/ archi ve/proje cts/PXD03 6343

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly avail- Beatriz Gomez Perez-Nievas https://orcid.org/0000-0003-1860-012X Maria Jimenez-Sanchez https://orcid.org/0000-0002-2287-1582