Characterization of astrocytes throughout life in wildtype and APP/PS1 mice after early-life stress exposure

C57Bl/6 J and bigenic APPswe/PS1dE9 hemizygous animals on a C57Bl/6 J background were used. APP/PS1 mice express mutated versions of the APP and PS1 proteins under the mouse prion promoters, containing a chimeric mouse/human amyloid precursor protein with the Swedish mutation (K595N/M596L) and the presenilin-1 protein with a deletion on exon 9 [46, 47]. Experimental animals were bred in house as described previously [48]. In total, three cohorts of male WT C57Bl/6 J mice of different ages (P9, P30, and 6 months) and two cohorts of male WT/APP/PS1 mice of different ages (4 months and 10 months) were used for experiments. All mice were kept under standard housing conditions (temperature 20–22 °C, 40–60% humidity level, chow/water ad libitum, 12/12 h light/dark schedule). Experiments were approved by the Animal Experiment Committee of the University of Amsterdam and performed in accordance to European Union (EU) directive 2010/63/EU.

The ES paradigm consisted of limiting nesting and bedding material from P2 to P9 as described previously [48]. Briefly, at P2, litters were randomly assigned to control (CTL) or ES condition. CTL litters received standard nesting and bedding material. ES litters were placed on a fine-gauge stainless-steel mesh positioned 1 cm above the cage floor with half the amount of nesting material. At P9, pups were moved to standard cages or sacrificed (P9 cohort). This ES model has been shown to lead to decreased body weight gain, increased adrenal weights, and decreased dentate gyrus volume in pups at P9 [48]. We assessed all of these also in our current cohort to confirm the effectiveness of the ES exposure. Dentate gyrus volume was measured using the Cavalieri principle, by multiplying the total area measured in ImageJ with the thickness of the slice, as well as the number of parallel series created per brain.

Mice were sacrificed via transcardial perfusion (P9, P30, 4, 6, and 10 months) or rapid decapitation (P9, 4, and 10 months), and tissue was harvested for either immunohistochemical or gene expression analyses, respectively. To collect brain material for immunohistochemical purposes, mice (P9: WT-CTL n = 7, WT-ES n = 6; P30: WT-CTL n = 4, WT-ES n = 6; 4 months: WT-CTL n = 10, WT-ES n = 11, APP/PS1-CTL n = 7, APP/PS1-ES n = 7; 6 months: WT-CTL n = 4, WT-ES n = 8; 10 months: WT-CTL n = 7, WT-ES n = 8, APP/PS1-CTL n = 5, APP/PS1-ES n = 4) were anesthetized via an intraperitoneal injection of pentobarbital (120 mg/kg Euthasol®) and transcardially perfused with 0.9% saline followed by 4% paraformaldehyde in phosphate buffer (PB 0.1 M, pH 7.4). The brains were harvested as previously described [14], sliced in 40 μm thick coronal sections (4 parallel series for P9, 6 parallel series for adult) and stored in antifreeze solution (30% Ethylene glycol, 20% Glycerol, 50% 0.05 M PBS) at − 20 °C until further processing. To collect the brain material for gene expression analyses, mice (P9: WT CTL n = 8, WT ES n = 5; 4 months: WT-CTL n = 7, WT-ES n = 6, APP/PS1-CTL n = 5, APP/PS1-ES n = 4; 10 months: WT-CTL n = 6, WT-ES n = 11, APP/PS1-CTL n = 6, APP/PS1-ES n = 6) were sacrificed via rapid decapitation and hippocampi were quickly dissected and snap-frozen on dry ice. RNA was extracted from hippocampi using the TRIzol method (TRIzol, Invitrogen). Reverse transcription of RNA to cDNA was performed using SuperScript® III Reverse Transcriptase (Invitrogen), and cDNA samples were stored at − 20 °C until further processing.

Free-floating brain sections were incubated with primary (polyclonal rabbit anti-GFAP, 1:10000, Dako) and secondary antibody (goat anti-rabbit, 1:200, Vector biotinylated), followed by 90 min incubation with avidin-biotin complex (ABC elite kit, 1:800, Vectastain, Brunschwig Chemie). Subsequently, for chromogen development, sections were incubated for 20 min with 0.5 mg/mL 3,3′-diaminobenzidine (DAB) with 0.01% H2O2 in 0.05 M TB. After DAB staining, sections were mounted on pre-coated glass slides (Superfrost Plus slides, Menzel) followed by coverslipping. Volume estimations of the P9 DG were obtained by measuring DG tracings of 6 bilateral sections to estimate the total surface area in square micrometers, which was multiplied by the number of series (4) and section thickness (40 μm).

GFAP+ astrocyte coverage (global and clustered GFAP), cell density, and individual cell complexity were analyzed in immunostained sections. For coverage and cell density analyses, sections were imaged with a 10x objective on a Nikon Eclipse light microscope. To ensure representative analyses of the whole hippocampus, we took rostro-caudal sections located between the bregma levels − 1.34 to − 3.80. We used 6 slices in the P9 samples and 7–8 slices in the P30, P180, and 10-month samples. In analyzing the EC in 4-month and 10-month samples, we took 2 sections between the bregma − 3.64 to − 4.16. Whole HPC and hippocampal subregions dentate gyrus (DG), stratum lacunosum-moleculare (SLM), and cornu ammonis (CA), (CA1, CA2, CA3 separately for P9 cohort) and the EC (4 months and 10 months WT/APP/PS1) were traced using ImageJ software. After tracing, images were converted to 8-bit black-and-white images in ImageJ. A fixed threshold was determined for each cohort to determine the percentage of immunoreactive stained area (coverage). The total thresholded signals and areas of interest across different bregma points analyzed were averaged, producing a single coverage datapoint per animal. In order to measure clustered GFAP, thresholded images were processed in ImageJ using the Analyze Particles function. GFAP signal with a surface area below 1400 μm² was filtered out to obtain clusters of GFAP, and the coverage analysis was repeated on the processed images. Data are presented as percentage coverage.

Astrocyte density was obtained by performing manual cell counts in 4 selected frames of 6744 μm² in the hippocampal subregions of interest (hilus, molecular layer (ML) of the DG, SLM, and CA1). Data are presented as number of GFAP+ cells per surface area. For cell complexity measurements, Z-stack images were obtained with a 40x objective. For each animal, 6 cells in 4 coronal sections (bregma levels − 1.70 until − 2.80) were used for cell complexity measurements, resulting in a total of 24 analyzed cells per animal. The outline of individual cells was traced to obtain a 2D cell surface. Sholl analysis was performed on traced cells using ImageJ [51, 52]. Virtual concentric circles where drawn with a 2 μm radius interval from the soma for a total distance of 74 μm, and the number of intersections was counted. The number of primary processes was counted manually.

Immunohistochemical stainings of GFAP in 4 and 10-month WT/APP/PS1 mice were done on parallel series from the same brains used in our previous study [14], where we characterized amyloid pathology and the microglial markers Iba1 and CD68. This allowed us to normalize the GFAP data to measures of amyloid. In addition to further explore interactions between astrocytes, microglia, and amyloid pathology in the HPC of WT and APP/PS1 mice, we created correlation matrices with the data obtained from different immunostainings (GFAP, CD68, Iba1, 6E10) in 4-month and 10-month-old mice. These descriptive correlations aimed to explore the relation between astrocytes, microglia, and amyloid.

Data were analyzed using SPSS 20.0 (IBM software), Graphpad Prism 5 (Graphpad software), and R 3.5.1 [53]. Data are expressed as mean ± standard error of the mean (SEM) and were considered statistically significant when p < 0.05. As multiple mice from the same litter were included in our experiments, the contribution of litter was tested in a mixed model with litter included as a random factor. Litter effects were corrected when present. Data with only condition (CTL/ES) as predictor variable was analyzed with unpaired Student’s t test when passing parametric assumptions and Mann-Whitney U test otherwise. Data with both condition (CTL/ES) and genotype (WT/APP/PS1) as predictor variables were analyzed using two-way ANOVA. In case of significant interaction effects, post hoc analyses were performed using Tukey’s post hoc test. For data from the Sholl analyses, where multiple cells from one animal were analyzed, the contribution of animal was tested and corrected for as well. Differences between areas under the curve (AUC) and the number of primary processes were analyzed by fitting into a linear model while correcting for nested within animal effects using the nlme package in R [54]. Sholl analyses bar graphs and correlation matrices were generated using the ggcorrplot package [55] in R. Pearson correlations were calculated based on complete pairwise cases. Correlation coefficients were tested against critical values on a two-tailed distribution (alpha = 0.05) based on the number of complete cases per comparison. Correlation plots were generated via the ggplot2 package [56] in R.

ES led to physiological signs of stress in the pups, including decreased body weight gain between P2 and P9 (t (7) = 4.596, p = 0.002), increased adrenal gland weight (t (9) = − 3.591, p = 0.006), and decreased DG volume at P9 (t (11) = 2.368, p = 0.037) (data not shown).

At P30, no differences in GFAP coverage were found in the whole HPC or hippocampal subregions (HPC: t (8) = − 0.995, p = 0.349; DG: t (8) = − 0.721, p = 0.492; CA: t (8) = − 1.070, p = 0.316).

At 6 months, there were no differences in the HPC and its subregions in GFAP coverage (Fig. 1g (CTL) and Fig. 1h (ES) HPC: t (10) = 2.054, p = 0.067; DG: t (10) = 1.909, p = 0.085; CA: t (10) = 2.127, p = 0.059 Fig. 1i; SLM: t (10) = 2.128, p = 0.059 Fig. 1j). A decrease in GFAP+ cell density after ES was found in the hilus (t (10) = 3.169, p = 0.010) but not in other hippocampal subregions (CA1: t (10) = − 0.956, p = 0.362; SLM: t (10) = 1.472, p = 0.172 Fig. 1k; DG: t (10) = 0.773, p = 0.457). Furthermore, no alteration in cell complexity was found at 6 months (number of intersections AUC: t (9) = − 0.711, p = 0.495 Fig. 1l; primary processes: t (9) = − 1.753, p = 0.114).

Gene expression analyses show that ES increased Fasn gene expression levels at 4 months without affecting the other genes that were analyzed (see Table 2).

Normalized GFAP coverage to cell-associated amyloid is increased in the DG but not CA region in 4 months APP/PS1 mice (DG: Mann-Whitney U = 5.0, p = 0.0221, Fig. 5b, CA: t (11) = 1.532, p = 0.1538, not shown). In the 10 months APP/PS1 mice clustered, but not global, GFAP signal in the DG when normalized to extracellular amyloid plaque-load is decreased, (masked GFAP DG: t (6) = 4.691, p = 0.0034; masked GFAP CA: t (6) = 0.4, p = 0.703; global GFAP CA: t (7) = 0.603, p = 0.566; global GFAP DG: t (7) = 2.096, p = 0.0743). To further explore interactions between the stainings for astrocytes (GFAP), microglia (Iba1, CD68) and Amyloid pathology (6E10), we created correlation matrices of the data derived from the stainings, shown in Fig. 5d (4-month-old APP/PS1 mice) and Fig. 5e (10-month-old APP/PS1 mice). Correlation coefficients and matrices were also calculated and created for WT mice (Supplementary Fig. 1).

To our knowledge, this is the first study addressing effects of ES on GFAP expression ES at multiple different ages across the lifespan in the HPC using mice and the limited nesting model. The effects of ES on GFAP expression in WT mice are age and brain region dependent. At P9, a significant increase in GFAP immunoreactivity was present following ES exposure specifically in the SLM but not in other hippocampal subregions. Strikingly, the existing literature shows that ES in the form of maternal deprivation or separation leads to an acute reduction in GFAP [36, 37, 57, 58], followed by an increase in GFAP when a longer period (4–30 days) between the stressor and the analysis was applied [38, 59, 60]. This discrepancy with our current results could possibly be explained by the form of ES exposure. Notably, the limited nesting and bedding ES model applied in this study have a more chronic nature, as it is a continuous exposure to an impoverished environment for 7 days, versus a single episode or intermittent exposure in maternal deprivation and separation models respectively. Thus, the differences in intensity and length of exposure could determine the final effect on astrocytes.

When looking at more lasting effects of ES exposure, GFAP expression in the HPC appeared unaffected at 4 and 6 months of age, while reduced in 10-month-old mice exposed to ES. This finding is in line with the only other report showing that 12 months old rats exposed to maternal deprivation exhibit reduced GFAP [35]. Together, these results suggest that ES programming of astrocytes might only become apparent during aging. Intriguingly, our data suggest that the response of GFAP expression to ES exposure changes over time from an acute upregulation in a specific hippocampal subregion to a long-term downregulation. Such temporal dynamics of GFAP expression, even though on a shorter timescale, have also been observed previously in rats exposed to daily maternal deprivation from P1 to P10. In these rats, no initial changes were observed at P10, but a reduction in GFAP was observed at P20, followed by increased GFAP at P40 [60]. This direction of GFAP modulation at the different ages is not in line with the changes observed in the present study. This could be related to differences in frequency and intensity of the ES model, as well as potential species-specific effects, and further stresses the relevance of the specifics of the ES model. Despite the discussed discrepancies, both studies suggest complex temporal dynamics of GFAP expression in response to ES and show that the effect of ES on GFAP expression is not only lasting but also depends on the neurodevelopmental or lifetime stage.

When examining the specific characteristics of GFAP leading to the modulation of coverage, notably, no alterations in the number or complexity of GFAP+ cells were found as a consequence of ES in any of the age groups. This indicates that ES-induced alterations in GFAP result from intracellular GFAP changes, rather than a change in general astrocyte numbers or complexity [52, 61]. For the full interpretation of our findings, it is important to note that GFAP does not label all astrocytes in the brain, although a large part of astrocytes in the hippocampus is GFAP-positive [62]. Additionally, ES-induced changes in astrocytes can also occur in the absence of GFAP alterations [63]. Next to GFAP protein levels, gene expression analyses showed that ES did not affect gene expression levels of astrocyte-related genes (i.e., Aldh1l1, Aqp4, Glast, Glt1, GluS, Gfap, Vimentin) at P9, 4, or 10 months, with the exception of Fasn mRNA levels, which were increased by ES at 4 months. Fasn is a key enzyme for de novo lipogenesis [64], a process in which astrocytes are highly active [27, 65, 66]. Changes in Fasn gene expression have been implicated in hippocampal changes in neurogenesis [67] and could indicate disrupted lipid metabolism as a result of ES, which might have implications for synapse development [27]. However, further investigations will need to clarify what the functional consequences of this change might be.

APP/PS1 overexpression in mice did not lead to alterations in astrocytes at early-pathological stages but did alter GFAP expression in the HPC and EC, as well as the HPC expression of several astrocyte-related genes, in 10-month-old-mice. In AD patients, GFAP+ reactive astrocytes are particularly associated with Aβ plaques (i.e., astrogliosis) [18, 20–22]. The lack of astrogliosis in the HPC and EC at 4 months is thus in line with the fact that only few Aβ plaques are present during this early-pathological stage [14].

The EC presents with the earliest signs of Aβ pathology in patients [68]. Here, we presented that GFAP immunoreactivity and GFAP cell complexity in the EC was increased in 10-month-old APP/PS1 mice. Similarly, injections with Aβ peptide in the EC caused increased GFAP intensity and larger GFAP cell somas [23], streptozotocin-induced AD in rats resulted in increased numbers of GFAP+ cells in the EC [24], and numbers of GFAP+ cells were increased in the EC of APP/Tau transgenic mice [69]. In contrast, it was reported that astrocytes in the EC appeared atrophied in a triple transgenic AD mouse model (3xTG), an effect that became less pronounced at a more advanced pathological stage (9 months) [70]. In this study, the use of a different transgenic line possibly explains the discrepancy with our current findings, as e.g., age-related epigenetic changes differ between different mouse models of AD [71]. In addition, it is important to note that more research points towards a heterogeneous response of astrocytes in response to AD pathology, as both astrocyte reactivity and atrophy occurs in AD (reviewed in [72]).

In contrast to the EC, we observed a global decrease in GFAP expression in the HPC, while most research demonstrates increased GFAP coverage in the HPC of APP/PS1 mice ranging from 5 to 14 months of age [73–75]. However, importantly, clustering of GFAP was observed both in the EC and HPC of 10-month-old APP/PS1 mice, and coverage of this clustered GFAP in the HPC was increased as expected. Considering that clustered GFAP was more prominent in APP/PS1 mice, and we observed a similar pattern of Aβ plaque load in parallel series of the same brain stained for Aβ, clustering of GFAP is probably due to accumulation of GFAP around Aβ plaques [76]. The reduction in global versus the increase in clustered GFAP coverage is in line with a previous study where general astroglial cytoskeletal atrophy was observed, while GFAP+ astrocytes surrounding Aβ plaques were hypertrophic in the HPC of 18-month-old 3xTG AD mice [77]. Given that astrocytes do not migrate to amyloid plaques [78], current data suggests that astrocytes near plaques are induced to upregulate GFAP expression, while astrocytes in the rest of the HPC downregulate GFAP. As many extracellular factors, e.g., inflammatory factors, can affect transcription and translation of GFAP [79, 80], microenvironmental changes induced by plaque accumulation may lead to local dynamic alterations in GFAP expression.

Gene expression analyses revealed that Gfap and Vimentin mRNA expression was not altered at 4 but increased at 10 months old. A similar increase in Gfap mRNA was previously shown in AD mice [74] and is in line with the appearance of reactive astrocytes [81]. While the exact function of reactive astrocytes is still not fully understood, genetic deletion of Gfap and Vimentin increased plaque load in APP/PS1 mice [82], suggesting that acquiring a reactive phenotype supports astrocytes in limiting amyloid pathology. Interestingly, Gfap mRNA in a full hippocampal homogenate was increased while we observed decreased global hippocampal GFAP protein coverage and increased clustered GFAP coverage. This suggests that, while indeed overall GFAP levels seem associated with plaque load and Aβ probably drives astrogliosis [83, 84], GFAP dynamics might be differentially affected by the microenvironment based on proximity to plaques. Aqp4 mRNA expression was also increased in APP/PS1 mice at 10 months. Aqp4 is the major water channel expressed in the CNS, and its expression is mostly restricted to astrocytes. Accumulating evidence suggests that Aqp4 plays a role in AD pathology [85] and the glymphatic system [86, 87] and may mediate clearance of Aβ [85, 88]. In line with our findings, multiple studies have found increased levels of Aqp4 in AD patients [89–91]. Furthermore, a decrease in Aldh1l1 was present in APP/PS1 mice. The decrease in Aldh1l1 could suggest a global loss of astrocytes, however, previous research revealed no difference in cortical astrocyte number between healthy and AD brains [92, 93], suggesting that this is probably not the case. Further research is necessary to understand the implication of the reduction of Aldh1l1.

At 10 months, both ES and APP/PS1 overexpression independently decreased global GFAP levels in the HPC. We see in the CTL APP/PS1 mice a distinction between global and clustered GFAP signal, and our analyses show that the two can be disentangled when accounting for aggregation of GFAP signals. Interestingly, neither the global nor the clustered signal is further affected by ES. However, we know from our previous study in the same cohort of mice that ES exposure had age-dependent effects on hippocampal amyloid load, and we in fact find ES-induced differences in GFAP coverage when normalizing to the appropriate pathology readout (Fig. 5a–c). Thus, it seems that ES exposure might still lead to some latent differential astroglial response to Aβ. This would be in line with decreased microglial accumulation observed in this cohort [14].

There is now a growing body of literature showing that astrocytes and microglia interact. Microglia are known to activate astrocytes during neuroinflammation [94, 95] and astrocytes can influence microglial activity, e.g., by releasing factors that increase phagocytosis [96]. Furthermore, emerging evidence is available for the synergistic effect of astrocytes and microglia in the progression of AD pathology [97, 98]. At 4 months, cell-associated amyloid is not correlated to either GFAP or Iba1 but negatively correlates to the microglial phagocytic marker CD68. This is in line with the fact that at this stage mice display mostly cell-associated amyloid-β and only few extracellular plaques; thus, no strong increase in activated microglia or astrogliosis would be expected. Still, Aβ, even in its cell-associated form, is considered somewhat inflammatory [99], and the negative correlation between cell-associated amyloid and CD68 further supports that the earlier reported ES-induced increase in phagocytic microglia might contribute to the reduction in cell-associated amyloid in APP/PS1 mice exposed to ES [14].

At 10 months, we find no correlations between plaque load and total amyloid with GFAP, Iba1, or CD68. This could be due to a saturated response, i.e., as Aβ continues to accumulate, the response of the glial cells does not parallel the increase in plaques. This is in line with the idea that neuroinflammation in AD shifts from being driven to respond to pathology (e.g., via initial Aβ buildup) to being a driving, self-perpetuating factor in disease progression due to the chronic neuroinflammatory environment it has created [100]. Interestingly, at 10 months, plaque load in the DG and CA do not correlate. Similarly, we did not detect correlations in APP/PS1 mice of global and clustered GFAP coverage between the two subregions, supporting the notion that amyloid pathology is changing the dynamics of GFAP expression within the HPC and leading to localized astrocytic responses that might not be reflected in a global analysis. When researching the relationship between astrocyte and microglia in AD pathology, we detected a significant relationship between CD68 and clustered GFAP coverage in the CA region. This might partly reflect the response of both CD68 and clustered GFAP signal to amyloid plaque buildup, perhaps pointing towards coordination of microglial and astrocytic phagocytosis, as reactive astrocytes are also capable of internalizing Aβ [101]. There were no other correlations detected between GFAP and Iba1 or CD68 in 10-month-old mice. More studies are necessary to unravel how precisely these two cell types interact in the response to Aβ pathology.

Additional file 1: Figure S1. Relationship between astrocytes and microglia in wild type mice. (A-B) Pearson correlation coefficients visualized from − 1 (blue) to + 1 (red), with the size of the circles representing the correlation coefficients. Circles without crosses are correlation coefficients significant at p < 0.05. (A) Correlation matrix of staining data from 4mo WT mice. (B) Correlation matrix between staining data from 10mo WT mice. (EPS 3753 kb)