Ultrastructural evidence of microglial heterogeneity in Alzheimer’s disease amyloid pathology

All experimental procedures were performed in agreement with the guidelines of the Institutional Animal Ethics committees, in conformity with the Canadian Council on Animal Care and the Animal Care Committee of Université Laval. The animals were housed under a 12-h light-dark cycle at 22–25 °C with free access to food and water. Fourteen-month-old APP^Swe-PS1Δe9 mice on a C57Bl/6J background (N = 4) were compared with age-matched wild-type littermate controls (N = 3). The transgenic mice coexpress human presenilin one variant (A246E) and a chimeric mouse/human amyloid precursor protein (APPSwe) [37]. Only males were used in this study considering sex differences in amyloid deposition that were described in APP^Swe-PS1Δe9 mice [42].

APP^Swe-PS1Δe9 mice were injected with Methoxy-X04 (10 g/kg; Tocris Bioscience) 24 h prior to sacrifice as previously described [41]. Methoxy-X04 is a Congo Red fluorescent derivative that binds to β sheets with high affinity. Mice were anesthetized with sodium pentobarbital (80 mg/kg, i.p.) and transcardially perfused with ice-cold phosphate-buffered saline (PBS; 50 mM at pH 7.4) followed by 3.5% acrolein and 4% paraformaldehyde both diluted in phosphate buffer (PB; 100 mM at pH 7.4). Transverse sections of the brain (50-μm-thick) were cut in PBS using a vibratome (Leica VT1000S) and kept in a cryoprotectant solution containing glycerol and ethylene glycol at − 20 °C until further processing [41, 43].

With the aid of a stereotaxic mouse brain atlas [44], brain sections containing the ventral hippocampus CA1 (Bregma − 2.75 to − 3.5) were selected and placed into a 24-well plate. Each section was assigned an individual well. Using an ultraviolet filter with a range of 380–480 nm, an inverted Nikon Eclipse TE300 microscope was used to visualize the methoxy-X04 stained plaques at a magnification of × 4. Pictures of the sections were captured both in bright and fluorescent field modes, naming the images according to the sections position into the 24-well plate [41]. Brain sections from the age-matched controls containing the same level of ventral hippocampus as the Methoxy-X04-labeled sections were selected for comparison.

Hippocampal sections containing the CA1 from two human cases of AD (C-0032, 76 years of age, 24-h post-mortem delay; H-17, 81 years of age, 6-h post-mortem delay) were obtained from the brain bank established at the CERVO Brain Research Center. The two samples are considered typical cases of AD with similar levels of pathology, staged 5 out of 6 on the Braak scale. In the two samples, Bielschowsky staining revealed moderate numbers of fAβ plaques, especially in the parietal and temporal lobes. With regard to the hippocampus, an overall atrophy was noted, together with a moderate number of fAβ plaques. Brain banking and post-mortem tissue handling procedures were approved by the Ethic Committee of the Institut Universitaire en santé mentale de Québec and by Université Laval. Brains were obtained with written consent and the analyses performed in conformity with the Code of Ethics of the World Medical Association (Declaration of Helsinki). Brains were first cut in half along the midline and hemibrains were sliced into 2-cm-thick slabs along the coronal plane. These slabs were fixed by immersion in 4% paraformaldehyde at 4 °C for 3 days. They were stored at 4 °C in PBS (100 mM at pH 7.4) containing 15% sucrose and 0.1% sodium azide. Slabs containing the hippocampus CA1 were cut with a freezing microtome into 50-μm-thick sections that were serially collected in PBS and stored at − 20 °C in a cryoprotectant solution containing glycerol and ethylene glycol until use.

Hippocampal sections from the human cases, wild-type littermate controls, and APP^Swe-PS1Δe9 mice containing Aβ plaques in the CA1 were selected for immunohistochemical (immuno)EM staining. They were washed in PBS, quenched with 0.3% hydrogen peroxide (H₂O₂) in PBS, and washed again. Sections were then incubated with 0.1% solution of NaBH₄ for 30 min at room temperature (RT). After several washes, sections were incubated in a blocking buffer of Tris-buffered saline (TBS; 50 mM at pH 8.0) containing 10% fetal bovine serum, 3% bovine serum albumin, and 0.01% Triton X100 for 2 h at RT. Sections were incubated overnight at 4 °C in rabbit anti-ionized calcium binding adaptor molecule 1 (IBA1) antibody (1:1000 in blocking solution; Wako Pure Chemical Industries) and rinsed in TBS. The sections were afterwards incubated for 1.5 h in goat anti-rabbit IgGs conjugated to biotin (1:200 in TBS; Jackson Immunoresearch) and 1 h in the ABC Vectastain system kit (1:100 in TBS; Vector Laboratories) for immunoperoxidase staining. The staining was revealed using diaminobenzidine (DAB; 0.05%) and H₂O₂ (0.015%) in TBS for 8 min. Sections were post-fixed flat in 1% osmium tetroxide in PB for 30 min and dehydrated in increasing concentrations of ethanol, immersed in propylene oxide, impregnated in Durcupan resin (Electron Microscopy Sciences; EMS) overnight at RT, and polymerized between ACLAR films (EMS) at 55 °C for 72 h. Areas of interest were excised from the embedded sections, glued to resin blocks [41], and sectioned at 70–80 nm using a Leica UC7 ultramicrotome.

Ultrathin sections collected on square mesh grids (EMS) were imaged using a FEI Tecnai Spirit G2 microscope equipped with an ORCA-HR Hamamatsu (10MP) camera. Microglial cell bodies were identified using a series of distinctive features that comprise their electron density, association with extracellular space pockets, characteristic long stretches of endoplasmic reticulum (ER), numerous vacuoles and intracellular inclusions, irregular contours with obtuse angles, and small elongated nucleus delineated by narrow nuclear cisternae [45, 46]. In most cases, they were also identified by their IBA1 immunoreactivity. Our analysis did not distinguish between the yolk-sac derived microglia and circulating myeloid cells that infiltrate the brain [47].

For qualitative analysis, pictures were acquired at magnifications ranging from × 1400 to × 13,000. Both mouse and human hippocampal samples were analyzed qualitatively to investigate microglial interactions with the synaptic neuropil, including contacts with dystrophic neurites and dendritic spines, phagocytosis, and association with extracellular space pockets containing debris. For quantitative analysis in the mouse samples, pictures of microglial cell bodies and processes were randomly acquired, at × 6800 and × 9300, respectively. In the APP^Swe-PS1Δe9 mice, three different subregions were differentiated. The first subregion contained one fAβ plaque, with some rare cases showing two fAβ plaques, accompanied by neuronal dystrophy within a 113 μm × 113 μm grid square; the second subregion displayed neuronal dystrophy notably observed as clusters of swelled neurites containing autophagic vacuoles as well as electron dense bodies but no fAβ plaque within a 113 μm × 113 μm grid square; and the third subregion only comprised ultrastructurally unaffected neuronal and glial cell compartments with intact plasma membranes and organelles, comparable to the wild-type control samples, within a 113 μm × 113 μm grid square. In the third subregion, myelinated axons were also surrounded by compact sheaths of myelin without ultrastructural alteration. While great care was taken in defining these three subregions in our samples, and the plaques were identified in light microscopy prior to ultrathin sectioning, it is possible that neuronal dystrophy was present outside of the imaged section. The same number of microglial cell bodies and processes was imaged in the wild-type controls, ranging from 7–15 microglial cell bodies and 50–70 microglial processes per subregion/animal. To ensure that some of the larger microglial processes were not, in fact, proximal processes containing part of the soma, a 2 μm² area was instituted as an upper bound for microglial process size.

In order to image microglia near fAβ plaques in the human tissue, scanning electron microscopy (SEM) with array tomography was used. The same tissue was used for both transmission electron microscopy (TEM) and SEM imaging. However, after ultrathin sections were made, they were collected onto polished silicon chips and treated with 2% osmium tetroxide in 1.5% potassium ferrocyanide solution, followed by 1% thiocarbohydrazide, and 1% osmium, as described in the National Center for Microscopy and Imaging Research serial blockface SEM protocol [48]. Following this post-fixation, the chips were mounted onto pin stubs using double-sided conductive carbon tape and imaged in a Zeiss Crossbeam 540 SEM, using acceleration voltage of 1.4 kV and current of 1.2 nA, at a maximum magnification of 5 nm per pixel.

The analysis of microglial cell bodies and processes comprised several ultrastructural measures of morphology, phagocytic activity, cellular stress, and physiological function. Experimenters were blinded to the experimental conditions throughout the analysis. Size and shape descriptors were determined using ImageJ. For each microglial cell body and process, phagocytic inclusions (appearing electron-lucent and named “empty” versus containing materials and named “non-empty”), lysosomes (primary, secondary, versus tertiary), lipid bodies, fibrillar materials, ER dilation, vacuoles (diameter < 100 nm), and extracellular space pockets containing debris were counted, according to the quantitative code 0, 1, 2, and 3+ (designating 3 and more occurrences). Lysosomes were identified by their dense, heterogeneous contents enclosed by a single membrane [49]. Primary lysosomes possessed a homogenous granular content and their diameter ranged from 0.3 to 0.5 μm [50]. Secondary lysosomes were 1 to 2 μm across, and their content was heterogeneous showing fusion with vacuoles. Tertiary lysosomes ranged in diameter between 1.5 and 2.5 μm, and they were usually fused to one or two vacuoles associated with lipofuscin granules, as well as lipidic inclusions showing signs of degradation [51]. Lipidic inclusions were identified as the clustering of round organelles with an electron dense, either opaque or limpid, cytoplasm enclosed by a single membrane. Lipofuscin granules were identified by their oval or round shapes, finely granular composition, and associated amorphous materials [51]. Extracellular fAβ was identified as densely packed fibrils and filaments, according to previous ultrastructural descriptions [52]. ER dilation was recognized by a swelling of the cisternal space ranging from 50 to 300 nm [53]. Extracellular space pockets containing debris, which could result from “exophagy” (degradation of cellular constituents by lysosomal enzymes released extracellularly), exocytosis (the process of expelling the contents of a membrane-bound vesicle into the extracellular space, often lysosomal and in preparation for phagocytosis; [54]), or pinocytosis (also named bulk-phase endocytosis, by which cells can take up extracellular contents in a non-phagocytic manner; [55]) was defined by the appearance of degraded materials (including cellular membranes and organelles) or debris in the extracellular space juxtaposing the microglia [54].

For microglial processes, their encirclement of neuropil compartments (axon terminals, dendritic spines, synapses between axon terminals and dendritic spines, cellular elements with signs of degradation) was also quantified. Encirclement was defined as microglial interactions with these neuropil compartments that displayed at least two points of contact, sometimes extending over several hundreds of nanometers. They were scored using the quantitative code 0, 1, 2, and 3+ (designating 3 and more elements). The encircled elements were identified according to the following criteria: axon terminals contained synaptic vesicles and were frequently seen branching from axons or making synapses onto dendritic branches and spines, and dendritic spines were identified as extensions from dendrites often forming synapses where a postsynaptic density was observed. Moreover, microglial encirclement of extracellular debris was determined.

GraphPad Prism 7 was used for statistical analysis. The APP^Swe-PS1Δe9 mice were compared to age-matched wild-type littermate controls using a non-parametric Mann-Whitney test. Microglial cell bodies and processes were further compared between plaque-associated, neuronal dystrophy associated, and ultrastructurally healthy tissue of APP^Swe-PS1Δe9 mice using a non-parametric Kruskal-Wallis test without matching, followed by Dunn’s multiple comparisons post-hoc test. Data are expressed as mean ± standard error of the mean (SEM) [56, 57]. The sample size (n) refers to individual microglial cell bodies or processes as previously performed for quantitative ultrastructural analyses [56, 58‐60]. Statistically significant differences are indicated by *,~,#p < 0.05, **,~~,##p < 0.01, and ***,~~~,###p < 0.001.

We characterized at nano-scale resolution microglial ultrastructure among subregions containing fAβ plaques, displaying neuronal dystrophy, or appearing healthy, among the hippocampus CA1 strata radiatum and lacunosum-moleculare of 14-month-old APP^Swe-PS1Δe9 mice versus wild-type littermates (Fig. 1). Microglial cell bodies displayed, in all four conditions, characteristic bean-shaped nuclei containing patches of dense heterochromatin, while their cytoplasm showed variable levels of immunoEM reactivity against the marker IBA1 (Fig. 1a–d). In line with previous descriptions of microglial ultrastructure [46], their cell bodies contained numerous mitochondria often in close apposition to characteristically long stretches of ER, together with occasional Golgi apparati and phagocytic vesicles (Figs. 1d, 2a–c, 3a, b, 4a–c, and 5a, c). When studying whole-region changes in microglial ultrastructure, cell body area was significantly different between wild-type and transgenic mice, showing increase in APP^Swe-PS1Δe9 animals (Fig. 2d, Table 1). Similar findings were obtained by separating the APP^Swe-PS1Δe9 investigation into subregions (ultrastructurally healthy, dystrophic, and plaque-associated with dystrophy) (Fig. 2e, Table 2).

Microglial cell bodies of wild-type and APP^Swe-PS1Δe9 mice were significantly different in size, but their morphology remained largely unchanged (Table 3). Numerous in vivo two-photon imaging studies revealed that microglial processes are incredibly dynamic as they survey the brain neuropil via continuous extension and retraction [61]. Microglial processes are long, thin, often ramified, discontinuous from cell bodies in ultrathin sections, and are identified by their immunoEM staining for IBA1, which diffuses throughout their cytosol (Figs. 2c, 3c–e, 4h–j, and 5b, d). They often accumulate phagocytic vesicles, and more proximal processes also contain mitochondria, occasional ER and/or Golgi apparati, as well as lysosomes. While the area of microglial processes did not differ between genotypes (Table 3), processes from APP^Swe-PS1Δe9 mice had significantly rounder and more solid shape descriptors (Fig. 2f, g and Table 3).

Although their gross ultrastructural features were not significantly different between wild-type and APP^Swe-PS1Δe9 mice, microglial cell bodies and processes interacted differently with their surrounding neuropil depending on the genotype. Strikingly, microglial cell bodies and processes in APP^Swe-PS1Δe9 mice reduced their association with extracellular space pockets containing debris (Fig. 3a, b, g, i). This debris contained cellular membranes that sometimes formed vesicles (see Figs. 3b, c and 4a). In the healthy adolescent brain, pockets of extracellular space are tightly associated with microglial cell bodies and processes, without noticeable accumulation of materials [45]. Whether this accumulation of debris observed here is associated with exophagy, exocytosis, or macropinocytosis events [54, 55, 62] warrants further investigation. In addition, differences were observed based on microglial cell body location among subregions of fAβ plaques, neuronal dystrophy, or apparent health (Fig. 3h). Microglial processes in APP^Swe-PS1Δe9 mice similarly reduced their association with extracellular debris (Fig. 3c, i, Table 3). Microglial processes in APP^Swe-PS1Δe9 animals nevertheless encircled more dendritic spines and axon terminals than processes in wild-type animals (Fig. 3e, f, j–m). This increase in encirclement was also noted in conjunction with degraded structures (Fig. 3d), though the trend did not reach statistical significance (Table 3). However, in the subregions of fAβ plaques, microglial processes were less likely to encircle synaptic elements, while more likely to encircle fibrils of amyloid (Fig. 3m, n). This data indicates that microglia respond to global changes (i.e., increased interactions with synapses in AD model), but still halt normal surveillance duties when in the immediate presence of fAβ.

One of microglias' most pivotal roles in the development and maintenance of the brain is their capacity to recognize and actively phagocytose, and degrade extraneous synapses and apoptotic cells. Microglia often contain phagocytic inclusions in their cell bodies and processes. Using immunoEM, the cargos contained within the phagocytic vesicles of microglia can be identified (Fig. 4a–c). Microglial cell bodies in APP^Swe-PS1Δe9 contained fibrillar materials (Fig. 4c, d), only when they were located nearby plaques (Fig. 4f). Empty inclusions appear as electron lucent, while non-empty inclusions are electron dense, and sometimes can be recognized as surrounding membranes, vesicles, or myelin sheaths (Fig. 4a–c). Microglial cell bodies in APP^Swe-PS1Δe9 mice contained significantly fewer empty inclusions than wild-type animals (Fig. 4e). In contrast, microglial cell bodies located near fAβ plaques were more likely to contain non-empty inclusions as compared with ultrastructurally healthy subregions (Fig. 4g).

Microglial phagocytosis begins with a microglial membrane wrapping and engulfing its phagocytic target. The vesicle is then trafficked to the soma via the phagolysosomal pathway, undergoing degradation along the way. As expected, the ratio of phagosomes containing non-empty to empty inclusions was higher within microglial processes than microglial somas (Tables 1, 2, 3, and 4). Upon investigation, we found that microglial processes in APP^Swe-PS1Δe9 mice were not less phagocytic than the processes from wild-type animals (Table 3). However, their total number of inclusions (both empty and non-empty) were significantly reduced in numbers in APP^Swe-PS1Δe9 mice (Fig. 4k, l). Comparing subregions, fibrillar materials were observed only in processes near plaques (Fig. 4m).

To provide additional insights into the alteration of microglial degradation activity, we next investigated the prevalence and determined the stage of lysosomes in microglial cell bodies and processes (Tables 1, 2, 3, and 4). While primary lysosomes are not commonly found in microglial somas (0.167 per cell body, Table 1), secondary and tertiary lysosomes containing phagocytic or degraded materials were often present in both wild-type and APP^Swe-PS1Δe9 animals (between 0.40 and 0.76 per cell body, Table 1). Interestingly, total lysosomes as well as specifically secondary lysosomes increased in microglial processes from plaque-associated areas (Figs. 4n, 5d, Table 4). Although the number of fibrillar materials inclusions was elevated in the APP^Swe-PS1Δe9 animals (Table 1), the numbers of primary, secondary, and tertiary lysosomes per cell body and process were not significantly different with respect to amyloid pathology (Tables 1 and 2).

The best characterized ultrastructural marker of cellular stress is ER stress, read out at high spatial resolution as a lumen dilation of ER and Golgi apparatus cisternae. It has been described in several neurodegenerative diseases, including amyotrophic lateral sclerosis and AD [63, 64]. ER dilation is also characteristic of the dark microglia, a phenotype upregulated in disease contexts including AD pathology [36]. In the current study, we were interested in investigating intermediate stages between the typical microglia and dark ones. Various hallmarks of dark microglia were examined in the ventral hippocampus CA1 strata radiatum and lacunosum-moleculare of wild-type and APP^Swe-PS1Δe9 animals, including the presence of dilated ER among microglial cell bodies and processes, the alteration to organelles including mitochondria, the condensation of cytoplasmic and nucleoplasmic contents (thought to result in a dark/electron-dense appearance), and the loss of heterochromatin pattern.

Healthy ER is characterized by long stretches of parallel membranes with very little space between them. On the other hand, dilated ER appears with non-parallel membranes containing bloated stretches of electron-lucent materials (Fig. 5a, b). When comparing the prevalence of dilated ER among microglial cell bodies from APP^Swe-PS1Δe9 animals and wild-type controls, there was no difference between genotypes (Table 1). After separating our analysis into healthy, neuronal dystrophy, or fAβ plaque subregions, it became clear that an increase in dilated ER was present, but only in plaque-associated microglial cell bodies (Fig. 5e).

In addition to studying dilated ER in microglial cell bodies and processes, we investigated the presence of lipidic inclusions in microglia from WT and APP^Swe-PS1Δe9 animals. Lipidic inclusions are dynamic clusters of lipid droplets, made up of a cholesterol and triglyceride core surrounded by a phospholipid monolayer, which house numerous signaling proteins, and are visible in electron microscopy as circular droplets of homogeneous electron density (Fig. 5a, c, d). Lipid droplets are actively formed in microglial cells that respond to pro-inflammatory cytokines and may represent a marker of cellular metabolic stress or an early hallmark of neuroinflammation [56]. The number of lipidic inclusions significantly increased in microglial processes from APP^Swe-PS1Δe9 animals, especially nearby plaques (Fig. 5f, g).

In addition, microglial cell bodies often displayed reduced IBA1 immunoreactivity in the APP^Swe-PS1Δe9 animals, across the three subregions (see Figs. 5a, c and 1), while the dark microglia were previously shown to downregulate IBA1 in the same hippocampal region of 14-month-old APP^Swe-PS1Δe9 mice [36]. Microglial cell bodies endowed with dilated ER, as observed around the plaques mainly, sometimes appeared darker than the typical microglia (see Figs. 1c and 5a), suggesting slight condensation of their cytoplasmic and nucleoplasmic contents. These darker cells however had a clearly defined heterochromatin pattern, as compared with the dark microglia.

In an effort to determine the clinical relevance of our microglial ultrastructure studies conducted in mouse hippocampus, we investigated tissue gathered from the hippocampi of two post-mortem AD cases. Fifty-micrometer-thick sections containing the hippocampus were stained against IBA1 to identify microglial cell bodies and processes. Occasional fAβ plaques were observed (Fig. 6a, b), with abundant dystrophic neurites (Fig. 6a–d, h, i), recognized by their accumulation of small, regularly packed dense bodies, and numerous autophagic vacuoles [65]. This identification of fAβ plaques and dystrophic neurites in human samples is consistent with the literature [65‐68]. As such, the tissues we investigated from AD individuals closely paralleled the subregions of fAβ plaques and neuronal dystrophy that were analyzed in APP^Swe-PS1Δe9 mice.

Microglia in human AD cases were found to be immunoreactive for IBA1 in both their processes and cell body, and they displayed the same characteristic nuclei described in mice (Fig. 6c, f). Microglial processes were often seen to contact dystrophic neurites and dendritic spines (Fig. 6d, g–j). Interestingly, microglial processes were observed making several contacts and nearly completely encircling multiple sides of a dystrophic neurite containing an accumulation of autophagic vacuoles (Fig. 6g). Microglia in human tissue from patients who suffered from AD were also phagocytic, with many processes displaying one or more phagocytic inclusions (Fig. 6h–j). These phagocytic inclusions appeared either empty or non-empty, consistent with the inclusions described in mouse microglia. Lysosomes were also frequently observed in the human microglial processes, showing intact granular structure and often juxtaposing vesicular inclusions, suggesting a probable fusion to produce a phagolysosome (Fig. 6j). In addition, human microglial cell bodies and processes frequently associated with pockets of extracellular space that contained cellular debris, at various stages of degradation, supporting microglial involvement with the removal of extracellular debris (Fig. 6c–j). Microglia in human AD cases thus recapitulate several ultrastructural characteristics seen in the mouse model, particularly their intimate interactions with fAβ plaques, dystrophic neurites, and dendritic spines.