Amyloid‐β impairs the phagocytosis of dystrophic synapses by astrocytes in Alzheimer's disease

Reactive astrocytes and dystrophic neurites, most aberrant presynaptic elements, are found surrounding amyloid‐β plaques in Alzheimer's disease (AD). We have previously shown that reactive astrocytes enwrap, phagocytose, and degrade dystrophic synapses in the hippocampus of APP mice and AD patients, but affecting less than 7% of dystrophic neurites, suggesting reduced phagocytic capacity of astrocytes in AD. Here, we aimed to gain insight into the underlying mechanisms by analyzing the capacity of primary astrocyte cultures to phagocytose and degrade isolated synapses (synaptoneurosomes, SNs) from APP (containing dystrophic synapses and amyloid‐β peptides), Tau (containing AT8‐ and AT100‐positive phosphorylated Tau) and WT (controls) mice. We found highly reduced phagocytic and degradative capacity of SNs‐APP, but not AT8/AT100‐positive SNs‐Tau, as compared with SNs‐WT. The reduced astrocyte phagocytic capacity was verified in hippocampus from 12‐month‐old APP mice, since only 1.60 ± 3.81% of peri‐plaque astrocytes presented phagocytic structures. This low phagocytic capacity did not depend on microglia‐mediated astrocyte reactivity, because removal of microglia from the primary astrocyte cultures abrogated the expression of microglia‐dependent genes in astrocytes, but did not affect the phagocytic impairment induced by oligomeric amyloid‐β alone. Taken together, our data suggest that amyloid‐β, but not hyperphosphorylated Tau, directly impairs the capacity of astrocytes to clear the pathological accumulation of oligomeric amyloid‐β, as well as of peri‐plaque dystrophic synapses containing amyloid‐β, perhaps by reducing the expression of phagocytosis receptors such as Mertk and Megf10, thus increasing neuronal damage in AD. Therefore, the potentiation or recovery of astrocytic phagocytosis may be a novel therapeutic avenue in AD.

together, our data suggest that amyloid-β, but not hyperphosphorylated Tau, directly impairs the capacity of astrocytes to clear the pathological accumulation of oligomeric amyloid-β, as well as of peri-plaque dystrophic synapses containing amyloid-β, perhaps by reducing the expression of phagocytosis receptors such as Mertk and Megf10, thus increasing neuronal damage in AD. Therefore, the potentiation or recovery of astrocytic phagocytosis may be a novel therapeutic avenue in AD.

K E Y W O R D S
Alzheimer, amyloid-beta, astrocytes, dystrophic synapses, microglia, pathology, phagocytosis
An important feature observed in the brain of AD patients and amyloidogenic mice is the early appearance of dystrophic neurites close to Abeta plaques. Dystrophic neurites, highly swollen axonal/ presynaptic elements, significantly contribute to synaptic and neuronal pathology in preclinical stages of AD (Nixon, 2007;Nixon & Yang, 2011;Sanchez-Varo et al., 2012;Torres et al., 2012). Furthermore, neuritic plaques are surrounded by reactive microglia and astrocytes (Heneka et al., 2015;Meyer-Luehmann et al., 2008;Serrano-Pozo, Betensky, Frosch, & Hyman, 2016), but it is not clear whether astrocytes promote the development and progression of dystrophies or whether, on the contrary, they constitute a physical barrier to limit their growth and protect other cells from the toxicity on the plaque halo, as suggested by several authors (Kraft et al., 2013;Perez-Nievas & Serrano-Pozo, 2018), and are engaged in the phagocytosis of dystrophies. The latter scenario is supported by evidence that astrocytes can phagocytose Abeta (Jones, Minogue, Connor, & Lynch, 2013;Wyss-Coray et al., 2003;Xiao et al., 2014) and synapses (Chung et al., 2013;Chung, Allen, & Eroglu, 2015;, which may represent an attempt to protect other cells, such that alterations in the phagocytic capacity of astrocytes may result in reduced clearance of Abeta and dystrophic synapses. Previously, we demonstrated that astrocytes are capable to enwrap, internalize, and degrade presynaptic dystrophies in the hippocampus of APP mice and AD patients (Gomez-Arboledas et al., 2018). However, the proportion of engulfed dystrophic neurites by astrocytes was below 7% of total dystrophies. This fact, along with the accumulation of these dystrophies during the disease course, suggests that the efficiency of the astrocyte phagocytic process might be limited or impaired. Reduced clearance of accumulated dystrophic neurites may contribute to the progression of the AD pathology.
In this study, we have examined the phagocytic capacity of astrocytes using synaptoneurosomes (SNs) isolated from APP transgenic mice, which contain dystrophic synapses, compared to those isolated from WT and Tau (P301S) mice, which do not present dystrophic neurites. Impaired phagocytosis was observed in the presence of SNs from APP, but not from Tau mice. Finally, the maintenance of the phagocytic impairment after microglia depletion supports the conclusion that Abeta, and not microglia-released factors, is the factor limiting phagocytosis in astrocytes in AD.
The supernatant was then centrifuged at 14,000g (5 min at 4 C). The final pellet was resuspended in 300 μl SN-Buffer and kept at 4 C.
Supernatants, which comprise both extracellular and cytosolic proteins (S1 fractions), were aliquoted and stored at −80 C until use.
2.5 | Total RNA and protein extraction, reverse transcription and quantitative real-time RT-PCR Total RNA and proteins were sequentially extracted, from cells or mouse hippocampal tissue, using Tripure Isolation Reagent (Roche) (100 mg tissue/1 ml Tripure Isolation Reagent) following manufacture recommendations (Jimenez et al., 2014;Romero-Molina et al., 2018;Sanchez-Mejias et al., 2016). RNA integrity was determined by RNA Nano 6000 (Agilent) (RIN 8.5 ± 0.5). RNA was quantified using Nanodrop 2000 (Thermo Fischer). Proteins were quantified using Lowry. Reverse transcription (RT) (4 μg of total RNA as template) was performed with High To assess microglia dependent-or independent gene expression, primer sequences were obtained from Liddelow et al., 2017 and qPCR was done using FastStart Universal SYBR Green Master (Roche).
For dot blot, 0.1 μg of protein from the different soluble fractions was directly applied to dry nitrocellulose, in a final volume of 3 μl.

| Degradation of synaptoneurosomes (pHRodo chase)
SNs were covalently labeled using 0.3 μM pHRodo (pHrodo TM Red succinimidyl (NHS) ester; Thermo Fisher Scientific) following the manufacturer protocol. Primary astrocytes were incubated for 6 hr with pHRodo-labeled SNs. After this initial incubation, the media were removed and substituted by fresh culture media. Astrocyte cultures were then incubated for different times (0-72 hr after loading), and pHRodo positive cells were identified by flow cytometry (λ ex = 560 nm; λ em = 585 nm) as above. For normalization, the number of pHRodo positive cells at any particular "chase" time was referred to the initial number of phRodo positive cells (6 hr time point).

| Acridine orange staining
The cellular acidic compartments can be visualized by acridine orange staining. Astrocytes were treated with pHRodo labeled SNs for 6 hours. Cells were incubated with medium containing 0.1 μg/ml acridine orange (Invitrogen) for 15 min at 37 C. The acridine orange was removed by washing coverslips with PBS and fresh buffer (NaCl 140 mM, KCl 2.5 mM, MgCl 2 1 mM, CaCl 2 2.5 mM, Glucose 10 mM, HEPES 10 mM) was added. Fluorescent images were taken using a confocal microscope (Leica TCS-SP2-AOBS).

| Electron microscopy
For conventional transmission electron microscopy (EM), 250-μm thick sections from 12-month-old APP brains (previously fixed by perfusion with PLP fixative), the coverslips with primary astrocytes (fixed with 4% PFA) and the isolated SNs from 12-month-old APP and agematched WT mice (fixed with 4% PFA) were postfixed in 1% osmium tetroxide in 0.1 M PB, block stained with uranyl acetate, dehydrated in graded acetone (brain sections) or ethanol (primary cell cultures and isolated SNs) and embedded in Araldite (EMS, USA) or Durcupan (Sigma Aldrich). For immunogold labeling, 50-μm thick vibratome APP sections or coverslips with primary astrocytes were cryoprotected in a 25% sucrose and 10% glycerol solution, followed by freezing at -80 C in order to increase the antibody binding efficiency. Sections were then incubated 48 h in the primary antibody (anti-Abeta42 rabbit polyclonal antibody, 1:5000 dilution, Abcam; anti-Megf10 rabbit polyclonal antibody 1:1000 dilution, Millipore, or anti-AQP4 rabbit polyclonal antibody, 1:1000 dilution, Sigma Aldrich) in a PBS 0.1M/0.1% sodium azide/2% BSA-solution at 22 C. The tissue-bound primary antibody was detected by the incubation with the corresponding 1.4 nm gold-conjugated secondary antibody (1:100, Nanoprobes) overnight at 22 C. After postfixation with 2% glutaraldehyde and washing with 50 mM sodium citrate, the labeling was enhanced with the HQ Silver TM Kit (Nanoprobes), and gold toned.
The primary antibody was omitted in negative control experiments.
Samples for conventional electron microscopy and those processed with the immunogold technique, were visualized first using semithin sections stained with toluidine blue and then selected areas were cut in ultrathin sections and examined with an electron microscope (JEOL JEM 1400).

| Statistical analysis
Statistical analyses were performed using the GraphPad software 7.0.
Data are presented individually or as mean ± standard deviation (SD).
The comparison between two groups was done by two-tailed Stu-dent´s t test. Data with more than two groups were analyzed using one-way ANOVA, followed by Tukey post hoc test. Data with two independent factors were analyzed using two-way ANOVA, followed by Bonferroni post hoc test. The significance was always set at 95% of confidence.

| Deficient phagocytosis and degradation of Abeta-containing dystrophic synapses by astrocytes
The phagocytic capacity of astrocytes was assayed using primary cultures from WT mice and synaptoneurosomes (SNs), isolated from 12-month-old APP (amyloidopathy model), 9-month-old Tau Abeta42 immunogold-positive phagocytic structures (phagosomes) were found within primary astrocytes incubated with SNs derived from APP mice.
Using these SNs preparations, we next quantitatively evaluated the astrocytic phagocytosis and degradative properties. SNs fractions (from WT, APP or Tau mice) were loaded with either TAMRA (uptake experiments; Figure S1b,c) or covalently labeled with the pH sensitive dye pHRodo (degradation kinetics, chase experiments, Figure S1d). Positive cells were quantified by flow cytometry. As shown (Figure 1c,c1), TAMRA-loaded WT-or Tau-derived SNs were efficiently and rapidly internalized by astrocytes. Data were fitted using one-phase association kinetic, and the "half-time" constants were calculated (Figure 1c,c2). No differences between SNs-WT and SNs-Tau were observed. However, in parallel experiments, SNs-APP presented a significantly slower kinetics (Figure 1c,c1) that, consequently, resulted in a significant increase in the uptake half-time constant (Figure 1c,c2, one-way ANOVA F(2, 13) = 21.89; p < .0001; Tukey p < .05). This inhibitory effect in the phagocytic capacity was specific for astrocytes, since primary microglial cells displayed no uptake differences (half-time constants: 0.39 ± 0.05 hr, 0.46 ± 0.15 hr and 0.51 ± 0.10 hr, n = 3, for TAMRA-loaded SNs-WT, SNs-APP or SNs-Tau, respectively; Figure S2).
Next, we evaluated the degradation of the engulfed material using pHRodo-labeled SNs. Control experiments demonstrated that pHRodo positive cargo was internalized by astrocytes and located in acidic compartments ( Figure S1d). Thus, astrocytes were "pulsed" for 6 hr with pHRodo-labeled SNs (from WT, APP, or Tau), washed and "chased" for different times. As shown (Figure 1d,d1 In fact, comparable decay constants were observed for both, WT or Tau-derived proteins, including total Tau (Figure 2a,b,c). Quantitatively, these data were similar to those observed using pHRodo (compare Figures 1d and 2b,c). On the other hand, and in consonance with our previous data, astrocytes loaded with SNs-APP accumulated synaptophysin, PSD95 and, more relevant, Abeta peptides. The quantitative analysis of these experiments (Figure 2b,c) demonstrated the existence of a significant increase in the decay constant for all APPderived SNs proteins (one-way ANOVA F(7, 26) = 11.84, p < .001; Tukey p < .05; Figure 2c). Thus, these data demonstrated that, at least in vitro, primary astrocytes presented clear deficiencies in the capacity to phagocytose and degrade SNs derived from APP mice, not noted in WT or Tau derived SNs.
The deficient degradation of APP-derived SNs internalized by primary astrocytes was also evidenced by confocal triple immunofluorescence of cultures after 24 hr of incubation (Figure 2d). Abeta42 labeled SNs accumulated within astrocytes (GFAP-positive), and did not co-localize with the lysosomal marker Lamp1 (Figure 2d,d1-d5).
Lysosomes were located surrounding the engulfed SNs, however, only few partially contacted them (see Figure 2d,d6 using confocal super-resolution microscopy). This may indicate deficient fusion of phagosomes with lysosomes, however, the mechanism precluding their interaction needs to be elucidated.
Since defective phagocytosis in astrocytes in a mouse model of septic shock has been described to be associated with a decrease in the expression of key genes implicated in the phagocytic process, such as Mertk and Mefg10 (Liddelow et al., 2017), we also tested the expression of these two genes. As shown in Figure 2e, their expression was unaltered (as compared with controls) in astrocytes treated Significance was tested by one-way ANOVA (F (7,26) = 11.84. p < .0001), followed by Tukey post hoc test (p < .05). (d) Triple GFAP/Abeta42/Lamp1 confocal immunofluorescence of primary astrocytes (in green) incubated with SNs derived from APP mice revealed multiple Abeta-positive SNs (in red) within astrocytic cytoplasm (d1 and detail in d2). These SNs are surrounded by numerous lysosomes (in blue) but no fusion between both compartments is detected (see d3 to d5). Super-resolution image (d6) of an Abeta42-positive SN (in red) surrounded by Lamp1-positive lysosomes (in green) shows scarce overlap between these two structures. (e) Mertk and Megf10 expression was evaluated in astrocyte cultures (n = 5) treated, in parallel, with SNs isolated from WT, APP or Tau mice. Untreated cultures were used as controls. The expression of both phagocytic receptors decreased exclusively in SNs-APP, as compared with other groups (two-way ANOVA F(3,30) = 11.03, p < .0001; Bonferroni *p < .05). Scale bars, d1, 25 μm; d2, 10 μm; d3-d5, 5 μm; d6, 1 μm [Color figure can be viewed at wileyonlinelibrary.com] with SNs from WT or Tau mice and significantly reduced in astrocytes treated with SNs-APP (two-way ANOVA F(3,30) = 11.03, p = .0002; Bonferroni p < .05).

| Limited astrocytic phagocytosis in aged APP mice
So far, our data demonstrated that SNs isolated from APP mice impaired the astrocyte capacity to engulf and degrade synaptic/ dystrophic elements in vitro. Therefore, we next aimed to verify this finding in vivo, in the hippocampus of 12-month-old APP model. As expected, astrocytes were reactive, as shown by the upregulated expression of pan-astrocytic proteins GFAP, S100b, ALDH1L1 ( Figure 3a, two-way ANOVA F(1,24) = 56.43, p < .0001; Bonferroni p < .05, for 12-month-old APP mice (n = 4) versus 3-month-old APP (n = 4)) in astrocytes surrounding Abeta plaques (Figure 3b). Reactive astrocytic processes were in close contact with the peri-plaque APPpositive dystrophic neurites (Figure 3c and inset). Interestingly, we observed reduced expression of Mertk and Megf10 genes in F I G U R E 3 Limited phagocytic capacity of reactive astrocytes also in 12-month-old APP mice. (a) The expression of classical astrocytic gene markers (such as GFAP, S100b, Aqp4 or ALDH1L1) showed a significant increase (two-way ANOVA F(1,24) = 56.43, p < .0001; Bonferroni post hoc test p < .05) in 12-month-old APP mice, as compared with the pre-pathology stage in 3-month-old APP mice (n = 4/age). (b) Combined confocal GFAP immunofluorescence and thioflavin-S staining in 12-month-old APP mice showing a fibrillar Abeta plaque (green) surrounded by reactive astrocytes (red). (c) Double GFAP/APP confocal immunofluorescence in 12-month-old APP mice revealed the intimate contact of reactive astrocytic processes (green) with peri-plaque dystrophic neurites (red); a higher magnification detail is showed in the insert. (d) Expression analysis of Mertk and Megf10 genes in 3-and 12-month-old hippocampal APP samples (F(1,14) = 19.32, p = .0006, followed by Bonferroni post hoc test *p < .05). (e) Megf10 is expressed by thin distal reactive astrocyte processes in close contact with Abeta fibers from the periphery of Abeta plaques of 12-month-old APP mice; e1, double GFAP/Megf10 confocal immunofluorescence showing abundant reactive astrocytes processes (in green) around an Abeta plaque (not stained; indicated with asterisk); Megf10 does not colocalize with the GFAP-positive processes; instead, the labeling is mostly present in the plaque periphery. e2-e4; Megf10 immunogold electron microscopy precisely identified the location of Megf10 labeling (white arrows in e3 and e4) in the plasma membrane of thin astrocytic processes in contact with Abeta fibers (asterisk in e3). (f) Aquaporin 4 (AQP4) immunogold electron microscopy (f1) in 12-month-old APP mice showing a process (highlighted in yellow) from a reactive astrocyte near an Abeta plaque that contains several dystrophic neurites in the cytoplasm (arrows in the higher magnification detail in the inset). AQP4 labeling is located in the astrocytic plasma membrane, surrounding the internalized dystrophic neurites. Astrocytes containing dystrophies were quantified in 12-month-old APP mice and the percentages of positive astrocytes were calculated (f2, n = 20 plaques from 3 different animals). Scale bars, b and c, 20 μm (inset 5 μm); e1, 20 μm (inset 5 μm); e2, 1 μm; 3, 0.2 μm; e4, 0.5 μm; f1, 5 μm (inset, 0.5 μm) [Color figure can be viewed at wileyonlinelibrary.com] 12-month-old APP mice (Figure 3d, two-way ANOVA F(1,14) = 19.32, p = .0006, Bonferroni p < .05, n = 4/5 for 3 or 12 months old, respectively). The expression of Megf10 in the hippocampus of 12-monthold APP mice was also tested by immunofluorescence confocal microscopy. As shown in Figure 3e (e1 and inset) a faint punctuate Megf10 immunolabeling was observed in the peri-plaque area, where the thin distal astrocyte processes (weak or negative for GFAP labeling) are located. To visualize the precise location of Megf10 labeling we used immunogold EM (Figure 3e,e2-e4). Megf10 immunoreactivity was located at the plasma membrane of astrocytic processes closely associated with bundles of amyloid fibrils radiating from plaque cores.
However, no immunostaining was observed in astrocytic processes in contact with axonal dystrophies. Furthermore, quantitative analysis of EM experiments (Figure 3f) of 12-month-old APP mice demonstrated that only 1.60 ± 3.81% (n = 20 plaques from three different animals) of the astrocytes directly contacting Abeta plaques displayed engulfed dystrophies within their cytoplasm. That is, astrocytes might also present, in vivo, limited phagocytic capacity of dystrophic elements. In addition, the ultrastructural analysis of reactive astrocytes in tagged APP mice showed the presence of partially digested inclusions (see Figure S3), suggesting also reduced degradative capacity.

| What causes inefficient phagocytosis in astrocytes? Abeta versus microglia
So far, our results demonstrated limited capacity of astrocytes to engulf and degrade Abeta-containing synapses including dystrophic presynaptic elements. This fact could be due to a direct action of Abeta or an indirect effect of microglial factors since: (a) microglia are present in vivo as well as in the primary astrocyte cultures (see below), and (b) it has been reported that astrocytes incubated with a cocktail of microglial cytokines, mimicking a septic shock milieu, show defective pHRodo uptake in vitro (Liddelow et al., 2017).
To evaluate the cross-talk between microglia and astrocytes, we first examined whether genes induced by microglial factors (septicshock like astrocytic phenotype; Liddelow et al., 2017) such as Gbp2, Ggta, Psmb8, Srgn, Gbln, H2-d1, H2-t23, and Serping were up-regulated in cultured astrocytes exposed to SNs from WT, tauopathy or amyloidopathy mice, as well as in APP mice. Importantly, these genes were also up-regulated in astrocytes isolated from APP/PS1 mice (Orre et al., 2014), further supporting their relevance in AD. As shown  et al., 2018) in APP mice, it is plausible that reactive astrocytes express microglia-dependent markers. As shown in Figure 4c, the expression of 6 out 8 genes tested showed a significant increase in 12 months versus 3 months APP mice (two-way ANOVA F (1,48) = 62.15; Bonferroni p < .05, n = 4 per gene and age), whereas most microglia-independent genes were not altered in the same samples. Although we cannot rule out that microglia-dependent genes were expressed as well in cells other than astrocytes in vivo, taken together these data support a local influence of microglia in the phenotype of reactive astrocytes in AD.
On the other hand, in vivo reactive astrocytes are located surrounding Abeta plaques and, therefore, exposed to soluble oligomeric Abeta (oAbeta) peptides, released from the plaque periphery (Izuo et al., 2019;Sanchez-Varo et al., 2012;Serrano-Pozo et al., 2016). In consequence, soluble Abeta peptides may directly affect the phenotype of astrocytes in AD. Therefore, we next evaluated whether soluble oAbeta (S1 fractions) derived from 12-month-old APP-mice also exert a similar effect on primary cultures as APP-derived SNs.
As reported previously, the soluble S1 fractions, isolated from APP mice brain, contained Abeta oligomeric forms (see Jimenez et al., 2008;Kayed et al., 2003Kayed et al., , 2007Moreno-Gonzalez et al., 2009;Torres et al., 2012;Yang, Li, Xu, Walsh, & Selkoe, 2017). We corroborated the presence of soluble oAbeta in the S1 extracts (by dot blots and anti-oligomeric specific antibodies, that is, OC, fibrillar, and A11, prefibrillar oligomers). As shown (Figure 5a), S1 fractions from 12-month-old APP mice were enriched (as compared with agematched WT) in OC-and A11-positive oligomers (protein loading controls were also showed in Figure 5a, lower panel). Thus, we next assessed, in vitro, whether treatment with S1 could affect the capacity of astrocytes to phagocytose and degrade WT-derived SNs. As shown (Figure 5b), the presence of S1 fractions produced a clear and significant inhibition of both TAMRA uptake (Figure 5b,b1, half-time constant comparison, two-tailed t test, p < .05, n = 4) and pHRodo decrease (Figure 5b,b2, n = 3). In fact, no decay in the number of pHRodo positive cells was observed in the time assessed, so no halfdecay constant could be calculated. In any case, these data indicated that soluble S1 fractions produced a similar effect to that observed using intact APP-derived SNs.
As shown in Figure 5c,c1, extracellular soluble Abeta peptides induced an increase in the expression of 5 out 8 of the microgliadependent genes (two-way ANOVA F(2,83) = 172.40, p < .0001; Bonferroni p < .05), whereas only 2 out of 8 microglia-independent genes were affected. Therefore, Abeta-containing SNs and soluble S1 fractions derived from APP mice produced a similar effect in astrocyte cultures. Although the data were compatible with a direct Abetamediated effect on both phagocytosis and expression of microgliadependent genes in astrocytes, considering that 7.2 ± 0.82% (n = 4) of the total cells in the astrocyte cultures used were microglia, the possibility existed that the actions of Abeta on astrocytes were indirect, mediated by the activation of residual microglia.
In order to discriminate between direct and indirect effects of Abeta, we first tested whether an immortalized astrocyte cell line, WJE (Morikawa et al., 2005), presented similar deficiencies in SNs-APP phagocytic/degradative capacity (TAMRA-or pHRodo experiments) as the primary astrocyte cultures. Indeed, we observed an increase in the uptake half-time constant of SNs-APP, as compared with SNs-WT (0.53 ± 0.25 hr vs. 1.15 ± 0.29 hr, n = 5, p = .002 two tailed t test, for SNs-WT-or SNs-APP, respectively, Figure S4a). Also, in agreement with the observations in primary cultures, WJE-cells displayed a significant delay in the pHRodo degradation (7.59 ± 2.69 hr, n = 6, vs. 46.39 ± 15.82 hr, n = 4, of WT-and APP derived SNs, respectively; two tailed t test p = .004, Figure S4b). Therefore, even in absence of microglial cells, SNs from APP-mice caused significant impairment of astrocytic phagocytic/degradative capacity.
In a second set of experiments, contaminant microglial cells were eliminated using Clodronate liposomes (Kumamaru et al., 2012). After three days of treatment, only 0.63 ± 0.15% (n = 3) of total cells in the culture were from microglial origin (tested by flow cytometry). Using these astrocyte cultures virtually free from microglial contamination, we assessed the effect of APP-derived S1 fractions on pHRodolabeled WT-SN uptake and degradation. As shown (Figure 5d), S1 fractions produced a clear inhibition in pHRodo-labeled WT-SNs uptake (first 24 hr incubation) and degradation in microglia-depleted cultures. In fact, no time-dependent decrease in the pHRodo positive cells was detected and, in consequence, no half-decay constant could be calculated. Clodronate treated astrocytes (in absence of S1 fraction) efficiently phagocytosed and degraded SNs-WT (compared Figure 5b,b2,d). In consequence, the inhibition of the astrocytic phagocytosis was dependent on oAbeta and independent of microglia.
We also used microglial-free cultures to test the expression of microglia-dependent and independent genes (Figure 5c,c1,c2). Noteworthy, S1 fractions did not induce the expression of any of the tested genes in microglia-depleted astrocyte cultures (Figure 5c).
These data suggest that the expression of these genes in astrocyte cultures contaminated with microglia was due to Abeta-mediated activation of microglial cells, whereas the deficiencies in either the uptake or degradation of APP-derived SNs were independent of microglia.
Accordingly, S1 produced no effect on the expression of microgliadependent genes in WJE cultures ( Figure S4c).
Finally, we determined whether the effect of Abeta-containing S1 fractions in the expression of Mertk and Megf10 genes was also dependent on microglia (Figure 5e). The expression of these genes was consistently decreased in primary astrocytes treated with APPderived SNs and in vivo in 12-month-old APP mice. In fact (Figure 5e), the expression of both genes was also significantly decreased in F I G U R E 4 Increased expression of microglia-dependent genes in both SNs-APP treated astrocytes and 12-month-old APP hippocampus. The expression of microglia-dependent (a and c) and independent genes (b and d) was measured in astrocyte primary cultures (a and b) and in APP mice (c and d). (a, b) Astrocyte cultures (n = 4) were treated, in parallel, with SNs isolated from WT, APP or Tau mice. Untreated cultures were used as controls. Two-way ANOVA was used for comparison between different groups (microglia-dependent genes, two-way ANOVA F

| DISCUSSION
Astrocyte reactivity is inherent to the AD pathology (Perez-Nievas & Serrano-Pozo, 2018). Reactive astrocytes are predominantly located at Abeta plaques (Serrano-Pozo et al., 2011), but the exact roles of these reactive astrocytes are currently unknown (Arranz & Strooper, 2019). Among many different physiological functions, evidence implicates astrocytes in phagocytic processes in the normal brain (Xiao et al., 2014). Here, we present evidence supporting our prediction based upon in vivo data in amyloidogenic mice and AD F I G U R E 5 Microglia-independent inhibition of astrocytic phagocytosis by soluble (S1 fractions) oligomeric Abeta forms. (a) The presence of soluble Abeta oligomers was confirmed using dot blots and anti-oligomeric Abeta antibodies (OC and A11). Soluble proteins from 12-month-old WT or APP brains were doted (0.1 μg) and incubated with OC or A11. Parallel dots were incubated with anti-GAPDH as loading controls. (b) S1 fractions (10 μg/100 μl culture media) produced a clear inhibition of WT-SNs uptake (b1) or degradation (b2), as compared with untreated controls. TAMRA uptake and pHRodo chase experiments were analyzed as described in Figure 1. Half-time constants are shown in the figure.
Data are means ± SD, n = 4. Significance between the different time points was analyzed using two-way ANOVA (F(1,36) = 63.28, p < .0001, Bonferroni post hot test *p < .05) in b1. Differences between half-time constants (TAMRA uptake) were compared using two tailed t test (p < .05) and are indicated in the panel (c). Microglia-dependent and independent gene expression was induced by APP-derived S1 fractions, and was inhibited after microglial elimination using Clodronate liposomes. Significant differences between SNs-WT, SNs-WT + S1APP treated astrocytes, in the presence or in the absence of clodronate, were assessed by two-way ANOVA, followed Bonferroni post hoc test (microglia-dependent genes F(2,83) = 172.40, p < .0001; microglia-independent genes F(2,83) = 127.1, p < .0001) *p < .05. (d) APP-derived S1 fractions inhibited both uptake and degradation of pHRodo-labeled SNs-WT in the absence of microglial cells. Microglia were eliminated with Clodronate liposomes and the uptake/degradative capacity of astrocytes was assessed using pHRodo labeled SNs-WT, in the presence or in the absence of APP-derived S1 fractions. Data are means ± SD of four experiments. (e) S1 fractions reduced the expression of phagocytosis receptors in the presence or in the absence of microglial cells. The expression of phagocytic receptors (Mertk, Megf10, AXL, and LRP1) was quantified in astrocyte cultures treated with SNs-WT (n = 6) and S1-APP (n = 3) or S1-APP plus clodronate (n = 4).  (Chung et al., 2013). In consonance with these findings, we found a significant decrease in the expression of both receptors in presence of Abeta, even in 12-month-old APP mice, which, moreover, showed discrete expression of Megf10 in reactive astrocytes surrounding Abeta plaques. In this sense, immunogold EM demonstrated that Megf10 was preferentially located in astrocyte processes in close contacts with Abeta plaque fibers, rather than in those engulfing dystrophies. Thus, not only the decrease in the expression, but also the preferential localization of Megf10 in astrocytes contacting Abeta fibers rather than with dystrophies, might explain the inefficiency of astrocytic phagocytosis of dystrophic neurites.
Abeta affected not only phagocytosis in astrocytes, but also their capacity to degrade phagocytosed materials. This defect was demonstrated using three independent approaches: pHRodo experiments, western blots of Abeta and synaptic proteins, and confocal immunofluorescence, which showed a clear decrease in the degradation of phagocytosed material exclusively when Abeta was present. Although not investigated in detail, we indeed observed the accumulation of lysosomal vesicles (Lamp1-positive) surrounding, but not fused to, Abeta-positive phagocytosed SNs. Confocal super-resolution analysis confirmed this observation and, more relevant, partially digested phagocytic vesicles were observed in APP astrocytes. Therefore, in vivo and in vivo evidence suggest that Abeta containing dystrophic synapses inhibit both phagocytosis and the ensuing degradation process in astrocytes. A similarly inefficient astrocytic Abeta degradation was reported previously (Söllvander et al., 2016).
It could be argued that the limited or insufficient phagocytic/degradative capacity just reflects the normal reduction of the phagocytosis by adult astrocytes. In fact, astrocytes are highly active removing synapses and neurons at early stages of brain development (Chung et al., 2015), whereas the phagocytic activity has been suggested to decrease in the adulthood (Chung et al., 2013). However, it has been recently reported that adult astrocytes may become highly phagocytic in CNS pathologies, such as brain ischemia (Morizawa et al., 2017), multiple sclerosis and Parkinson's disease, where astrocytes respectively phagocytose myelin (Ponath et al., 2017), alpha-synuclein and dopaminergic cells (Morales, Sanchez, Rodriguez-Sabate, & Rodriguez, 2017;Tremblay et al., 2019). In AD, our data point to impaired phagocytic capacity, which may be aggravated due to the expression of ApoE4, as recently demonstrated (Chung et al., 2016;Lin et al., 2018). Malfunctional astrocytes, together with dysfunctional microglial responses Sanchez-Mejias et al., 2016), may, in turn, contribute to the accumulation of presynaptic/axonal dystrophies in AD. This may, in turn, exacerbate progression of AD, because dystrophic neurites have been implicated in the accumulation of phosphorylated Tau protein (He et al., 2018;Sanchez-Varo et al., 2012;Torres et al., 2012) and in the initiation of Tau pathology (He et al., 2018). Along these lines, rapid progressive AD cases might be attributed, in part, to defective clearance of dystrophic neurites as a consequence of the reportedly inefficient astrocyte reactivity (Drummond et al., 2017). An important observation of the present study is that Abeta, but not Tau, impairs astrocyte phagocytosis. Thus, it is tempting to speculate that the reason why dystrophic neurites resulting from the damage inflicted by Abeta plaques to neurons are observed in APP mice, as reported here and in Gomez-Arboledas et al., 2018), but appear to be absent in Tau models (our unpublished results), is not just that there are no Abeta plaques in the latter, but also that astrocyte phagocytosis might be preserved in Tau models.
Finally, an important finding of the present study is that the deficient phagocytosis in AD by astrocytes is directly caused by Abeta, and not by microglia. It has been described that astrocytes exposed to a cocktail of microglia-released cytokines such as TNF-alpha and IL-1alpha, and the complement factor C1q, display, among other features, a reduction in phagocytosis and a decrease in Mertk and Megf10 expression (Liddelow et al., 2017). Since reactive astrocytes and microglia are both located in close contact with Abeta plaques, it is possible that Abeta (either monomeric and/or oligomeric) in senile plaques, or released from them, induced the production of those factors by the peri-plaque microglia. Indeed, we found induction of expression of microglia-dependent genes in three different conditions: in astrocytes exposed to APP-derived SNs; in 12-month-old APP transgenic mice, and in astrocytes in presence of APP-derived S1 soluble fractions (OC and A11 positive). The presence of Abeta oligomers is a constant in the three cases (Figures 1 and 4), as it is the presence of microglial cells, since our primary astrocyte cultures contain a 7% of them, and astrocytes and microglial cells are located together in Abeta plaques. Thus, the concomitant partial expression of microgliadependent genes and decreased phagocytosis were compatible with an indirect effect of Abeta on astrocytes through reactive microglia.
However, SNs-APP also produced the same effect in the immortalized astrocyte cell line WJE, in the absence of microglial cells. Moreover, microglia depletion abolished the observed Abeta-induced expression of microglia-dependent genes, while Abeta still inhibited both the phagocytosis and the degradative process in astrocytes and, furthermore, reduced the expression of Mertk and Megf10 (and other phagocytosis receptors). Therefore, Abeta seems to have two different actions on astrocytes: a direct inhibition of phagocytosis, and the indirect induction of transcriptional programs through microglial cells.
In support of the latter, the microglia-dependent genes are also overexpressed in astrocytes isolated from an AD transgenic model (Orre et al., 2014). These data suggest a similar cross-talk between microglia and astrocytes in AD and septic shock. We thus conclude that the acquisition of a distinct phenotype in peri-plaque astrocytes depends on multiple signals, including those from Abeta and local microglia.
Further studies will clarify the cause of astrocyte malfunction in AD, and its consequences on the pathological and clinical hallmarks of the disease.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding authors upon reasonable request.