The relationship between the morphological subtypes of microglia and Alzheimer’s disease neuropathology

Abstract

Microglial associations with both the major Alzheimer’s disease (AD) pathognomonic entities, β‐amyloid‐positive plaques and tau‐positive neurofibrillary tangles, have been noted in previous investigations of both human tissue and mouse models. However, the precise nature of their role in the pathogenesis of AD is debated; the major working hypothesis is that pro‐inflammatory activities of activated microglia contribute to disease progression. In contrast, others have proposed that microglial dystrophy with a loss of physiological and neuroprotective activities promotes neurodegeneration. This immunohistochemical study sought to gain clarity in this area by quantifying the morphological subtypes of microglia in the mildly‐affected primary visual cortex (PVC), the moderately affected superior frontal cortex (SFC) and the severely affected inferior temporal cortex (ITC) of 8 AD cases and 15 age and gender‐matched, non‐demented controls with ranging AD‐type pathology. AD cases had increased β‐amyloid and tau levels compared to controls in all regions. Neuronal loss was observed in the SFC and ITC, and was associated with atrophy in the latter. A major feature of the ITC in AD was a decrease in ramified (healthy) microglia with image analysis confirming reductions in arborized area and skeletal complexity. Activated microglia were not associated with AD but were increased in non‐demented controls with greater AD‐type pathology. Microglial clusters were occasionally associated with β‐amyloid‐ and tau‐positive plaques but represented less than 2% of the total microglial population. Dystrophic microglia were not associated with AD, but were inversely correlated with brain pH suggesting that agonal events were responsible for this morphological subtype. Overall these novel findings suggest that there is an early microglial reaction to AD‐type pathology but a loss of healthy microglia is the prominent feature in severely affected regions of the AD brain.

Introduction

The clinical diagnosis of Alzheimer’s disease (AD) is probabilistic and typically supported by an abbreviated form of a neuropsychological evaluation such as the Clinical Dementia Rating (CDR) 36. Confirmation of AD can only be made at autopsy where the most comprehensive diagnostic schema is provided by the National Institute of Aging‐Alzheimer’s Association work‐group 35. This diagnostic schema incorporates previously described semi‐quantitative criteria for (A) β‐amyloid (Aβ) immunostaining 60, (B) Braak staging of neurofibrillary tangles (NFTs) 6 and (C) CERAD neuritic plaque score 33 to generate an “ABC” score. An intermediate or high “ABC” score being considered sufficient to confirm AD in a demented individual. Yet, this pathological diagnosis is also probabilistic caused by the common finding of AD‐type pathology, albeit at lower levels, in similarly aged but non‐demented individuals 4. Aβ pathology appears first in the cortex before spreading more ventrally; whereas neurofibrillary pathology, including NFTs, begins in the medial temporal lobe, then spreads more dorsally and laterally throughout the limbic system and cortex with relative sparing of primary cortices 6.

Along with Aβ plaques and NFTs, the pathology of AD is characterized by neuronal loss and gliosis 59. Microglial activation has been considered a key pathomechanism in AD, with activated microglia associating with neuritic plaques and subsequently producing a neurotoxic milieu that contributes to neuronal degeneration 30, 41, 46. Single‐nucleotide polymorphisms in genes exclusively or largely expressed in microglia are associated with increased AD risk including TREM2, CD33, CR1, ABCA7 and SHIP1 31. Microglial activation in AD appears to be Aβ‐dependent; with Aβ binding to NLR Family Pyrin Domain Containing 3 16, and pattern recognition receptors such as RAGE, scavenger receptors 44 and toll‐like receptors (TLR2, TLR4 and TLR6) 8 being sufficient to cause microglial activation in mouse models of AD.

In the past, microglia were regarded as quiescent cells that only responded to brain injury. It is now known that microglia are physiologically active, surveying their territory by continually extending and retracting their processes 38. Extensive in vitro and animal studies of microglia demonstrate their ability to sample their surrounding microenvironment and are particularly involved in synaptic homeostasis 26 throughout life by pruning of synapses in the developing CNS 47 and mediating remodeling in the adult brain 40. Microglia are functionally dynamic cells as evidenced by the wide range of morphologies observed in the brain under normal and pathological states 11. There have been a number of attempts to categorize microglia into distinct subtypes but there is little consensus on what constitutes each subtype or the nomenclature used to describe them.

Briefly, in studies using human post‐mortem brain tissue, ramified (healthy) microglia (previously called “resting” or “quiescent” microglia) are characterized by equally distributed, long, thin, ramified processes attached to a small, spherical soma 5. The terms “activated,” “reactive,” “bushy” and “hypertrophic” have been used to describe microglia that are responding to injury or potentially injurious events. Activated microglia display significantly hypertrophic soma and processes with some degree of deramification (hence the use of the term “deramified” by some authors) 3, 45. Fully activated microglia, with enlarged soma and lacking processes entirely, are termed “amoeboid” and have enhanced phagocytic capabilities 52. “Dystrophic” microglia are characterized by highly tortuous, asymmetrical and truncated processes and may demonstrate discontinuous or beaded immunostaining. Such dystrophic microglia may also be termed “pseudo‐fragmented” as electron microscopy studies have revealed that apparent discontinuities in their cellular processes represent a redistribution of the selected marker, with the processes remaining contiguous when visualized using multiple markers 61. Other features of dystrophic microglia that occur variably include the formation of spheroids and condensation within the nucleus, possibly indicative of pyknosis 56. Streit and colleagues found that dystrophic microglia preceded neurofibrillary pathology in AD 54 raising the possibility that a loss of microglial function may be as, or more, important than a gain of toxic function in AD pathophysiology.

In this immunohistochemical investigation, microglia were identified by their expression of ionized calcium binding adaptor molecule 1 (IBA1), which is known to be an effective pan‐microglial marker that is independent of “activation” states compared to other commonly used markers 23, 54. In our previous work, we had demonstrated greater densities of dystrophic microglia in severely affected cortical areas of the AD brain, namely the inferior temporal cortex (ITC) and the anterior cingulate cortex 9. Here, this work has been extended by quantifying the morphological subtypes of microglia by manual and automated methods in AD cases and controls with variable amounts of AD‐type pathology in three neocortical areas – the primary visual cortex (PVC), superior frontal cortex (SFC) and ITC. AD pathology and particularly tau pathology are known to spread in a stereotypical fashion through the cortex 2, 6, 13, 60. Furthermore, the density of NFTs is inversely correlated with surviving neurons in the AD cortex 14. These three regions were chosen to reflect the reported spread of tau pathology, with the PVC being one of the last regions to be affected 6. This is also reflected by a volumetric study where the severely affected ITC undergoes ~35% atrophy and the PVC ~ 15% in AD cases compared to gender‐matched controls 17, a scenario that potentially allows regional comparisons to model disease progression within cases 58. The aim here was to gain a greater understanding of spatiotemporal relationships between the morphological subtypes of microglia and AD progression.

Methods

This immunohistochemical study of microglial subtypes using human post‐mortem brain tissue from AD patients and non‐demented controls was approved by the University of Sydney’s Human Research Ethics Committee (HREC #2015/477). All tissue samples for this study along with demographics, clinical and pathological diagnosis were supplied by the New South Wales Brain Tissue Resource Centre (NSW BTRC) and the Sydney Brain Brank (SBB), collectively referred to as NSW Brain Banks (NSWBB) following clearance from their Scientific Advisory Committee. APOE ε4 genotyping was also carried out as previously described 32. NSWBB measure brain pH on a frozen (‐80^oC) segment of the lateral cerebellar hemisphere which has previously been shown to be representative of the whole brain pH 51 and may be used as a marker of the length and severity of events in the pre‐mortem period 12, 34. The demographic and clinicopathological characteristics of the cohort are listed in Table 1.

Table 1.

Cohort details.

Case ID (n = 23)	Age	Sex	Status	Cause of death	Disease duration (years)	PMI (hours)	Brain pH	Fixation (months)	CDR	ABC score	AD likelihood b	APOE genotype
M01	69	Male	Control	Cardiac failure	–	16	6.6	97	–	A3 B1 C3	Low	E3/E4
M02	74	Female	Control	Cancer	–	20	6.59	19	–	A3 B1 C3	Low	E3/E4
M03	76	Female	AD	Cardiorespiratory failure	11	3	5.98	35	3	A3 B2 C3	Intermediate	E3/E4
M04	77	Male	AD	Aspiration pneumonia	9	26	6.34	12	2	A3 B3 C2	High	E4/E4
M05	78	Female	Control	Respiratory failure	–	11	6.3	78	–	A0 B0 C0	Low	E3/E3
M06	78	Female	Control	Toxicity	–	45	6.05	58	–	A0 B1 C0	Low	E3/E4
M07	81	Male	Control	Cardiac failure	–	29	6.57	26	–	A0 B2 C0	Low	E3/E3
M08	80	Male	Control	Respiratory failure	–	12	6.5	9	–	A0 B0 C0	Low	E3/E3
M09	80	Female	AD	Alzheimer's disease	10	32	6.54	25	3	A3 B3 C3	High	E3/E4
M10	82	Female	Control	Respiratory failure	–	7.5	6.45	102	–	A3 B1 C2	Low	E3/E3
M11	83	Male	AD	Cerebrovascular	5	25	6.26	70	3	A3 B3 C3	High	E4/E4
M12	83	Female	AD	Uraemia	7	3	5.88	61	3	A3 B3 C2	High	E3/E4
M13	84	Female	AD	Aspiration pneumonia	13	6	6.32	64	3	A3 B3 C3	High	E3/E4
M14 a	85	Male	Control	Cancer	–	9	6.57	61	–	A2 B2 C3	Intermediate	E3/E3
M15	85	Female	Control	Respiratory failure	–	10	6.63	34	–	A2 B1 C0	Low	E3/E3
M16	85	Female	Control	Pneumonia	–	23	6.44	73	0	A2 B0 C1	Low	E3/E3
M17	85	Female	AD	Cardiorespiratory failure	5	10	5.91	15	3	A3 B3 C3	High	E3/E3
M18 a	87	Female	Control	Cancer	–	5	6.38	34	–	A2 B2 C0	Intermediate	E3/E4
M19 a	92	Female	Control	Pancytopenia	–	5	6.08	34	0	A2 B2 C2	Intermediate	E3/E3
M20	93	Female	Control	Cardiac failure	–	21	6.96	81	0.5	A1 B0 C0	Low	E3/E3
M21 a	87	Female	Control	Acute peritonitis	–	24	–	–	0	A3 B2 C1	Intermediate	–
M22	98	Female	AD	Cerebrovascular	6	11	6.11	18	3	A3 B3 C2	High	E3/E3
M23 a	102	Female	Control	Acute renal failure	–	5	5.92	35	0	A2 B2 C2	Intermediate	E2/E3

^aHigh‐pathology controls.

^bAs per NIA‐AA diagnostic criteria 35.

Immunohistochemistry

Immunohistochemistry (IHC) was performed on free‐floating 45‐μm fixed sections from the PVC, SFC and ITC of cognitively normal individuals with variable AD‐type pathology (CDR = 0–0.5; n = 15) and AD patients (CDR 2–3; n = 8) to quantify the morphological subtypes of microglia. Heat‐induced epitope retrieval was performed by baking sections in a 60^oC oven overnight using a 0.95 mmolL^‐1 sodium citrate (pH 8.5) buffer. Sections were washed with 50% ethanol followed by endogenous peroxidase block using 0.9% H₂O₂ (50% ethanol diluent). Standard blocking was performed for 30 minutes in 10% normal goat serum (NGS; Gibco, Life Technologies Australia Pty Ltd., Mulgrave, Australia) diluted using a 0.05 molL^‐1 tris‐buffered saline (1 × TBS; pH 7.4) wash buffer solution with additional Triton‐X₁₀₀ detergent (0.1%). Sections underwent primary antibody (IBA1; 1:1000, rabbit monoclonal; 019‐1974, Wako Pure Chemical Industries, Japan) incubation at 4^oC overnight. All incubations were followed by three washes with gentle agitation for five minutes using 1 × TBS. Negative control sections incubated without primary antibodies were run concomitantly during each staining procedure. Secondary antibody incubation (biotinylated anti‐rabbit IgG (H+L), 1:200; BA‐1000, Vector Laboratories, Burlingame, CA, USA) was performed at room temperature (RT) for one hour followed by an hour in avidin‐biotin‐peroxidase complex solution (1:100; Vectastain Elite ABC, Universal). DAB (0.7 mg/mL; SigmaFast^TM 3, 3’‐diaminobenzidine tablets, Merck & Company, Inc., Kenilworth, NJ, USA) in an aqueous solution containing urea hydrogen peroxide tablet (0.67 mg/mL; H₂O₂ equivalence, 0.24 mg/mL) was applied for 15 seconds. Regressive counterstaining involving sequential immersion into hematoxylin 28, followed by acid alcohol (two second immersion) and Scott’s blueing solution (45 seconds) was performed prior to cover slipping using DPX mounting media (06522‐Sigma Pharmaceuticals, Rowville, VIC, Australia).

Quantification of microglia

Microglia were quantified from images acquired by an Olympus VS‐120 slide scanner following an adapted stereological approach 27. Briefly, images of three 500‐μm wide strips of the full cortical thickness were acquired from areas where the gray matter was at its thinnest and strictly parallel to the gray matter‐white matter boundary. An eyepiece graticule at 200 × magnification (0.25 mm²) with defined inclusion and exclusion lines was used as an unbiased counting frame. IBA1 + cells were counted if a nucleus and surrounding soma could be identified. Inter‐rater reliability was established between two experienced raters on 10% of sections with >95% concordance for both microglial and neuronal counts. All counts were expressed as densities and adjusted to account for cortical atrophy in AD by multiplying raw densities by the fraction of the mean cortical width of each case by the mean cortical width of all controls for each region. Total microglia were enumerated and microglia also subdivided into one of three subtypes based on their morphology: ramified, activated or dystrophic. Ramified cells were defined on the basis of their thin, highly branched processes and spherical soma. Activated microglia were defined by somal enlargement with either hyper‐ramification or a reduced number of thickened processes. Microglia that were ameboid in shape, large and lacking processes were counted as activated cells as they occurred infrequently. Dystrophic cells were identified by spheroidal swellings, fragmentation of cellular processes and reduced ramification 57. In addition, microglial clusters were also counted; arbitrarily defined as three or more cell bodies that could fit within a 50 μm² graticule sub‐region.

Immunofluorescence

Immunofluorescent (IF) double labeling was performed to investigate relationships between microglia and Aβ‐ and tau‐positive pathology. Antigen retrieval was as described above, followed by a 12‐minute formic acid (90%) incubation at RT to optimize Aβ immunoreactivity. All incubations were followed by three washes for five minutes with gentle agitation using 0.01molL^‐1 PBS (pH 7.4). Sections were co‐incubated in antibodies for IBA1 (1:1000) and Aβ (1:500; mouse monoclonal; M0872, Dako, Santa Clara, CA, USA) at 4^oC overnight. Sections were co‐incubated in secondary antibodies (1:200; AlexaFluor 488, goat anti‐mouse IgG (H+L); A11001, Invitrogen, Waltham, MA, USA); (1:200; AlexaFluor 568, goat anti‐rabbit IgG (H+L); A11011, Invitrogen) for 72 hours with gentle agitation at 4^oC in the dark and wrapped in aluminum foil. Autofluorescence caused by lipofuscin and aldehydes was minimized by washing sections for 10 minutes in 0.1% Sudan Black B (B.D.H Laboratory Chemicals Group) (70% ethanol diluent). Sections were mounted with DAPI Fluoroshield^TM medium (F6057; Merck & Company, Inc., Kenilworth, NJ, USA) before coverslipping. Tau immunolabeling was performed as previously described 9. Briefly, Heat‐induced epitope retrieval was performed with sodium citrate (pH 6.0), before permeabilizing, blocking with BSA and overnight incubation in primary antibodies: total tau (1:500; Dako, K9JA) and IBA1 (1:50; Millipore, MABN92). Sections were co‐incubated in secondary antibodies (1:200; AlexaFluor 555 and 647, anti‐mouse and anti‐rabbit IgG (H+L), respectively; A21424 and A21244, respectively, Invitrogen). The total tau antibody has been previously shown to immunostain NFTs, neuritic plaques and neuropil threads and is comparable to phospho‐tau (Ser262; 12E8) immunoreactivity 42. Hoechst 33342 was added to label nuclei and sections were mounted in Prolong Gold (Invitrogen).

Aβ and tau imaging

Researchers were blind to section status throughout the staining and imaging process. Aβ IF slides were imaged using a Zeiss LSM 510 Meta confocal microscope at the Advanced Microscopy Facility (AMF), Bosch Institute, The University of Sydney. Aβ was quantified in three cortical strips using a 20×/0.8 objective. These regions of interest (ROIs) were located at random while visualizing DAPI staining. Cortical strips were constructed of serial images (450 μm²) from the surface to the gray‐white boundary; moving inward sequentially. Individual images were maximum intensity projections (MIPs) of three z‐slices (z‐step = 5 μm) centered at the area with the most intense level of staining and were thresholded manually. Widefield images of total tau IF slides were captured using an Olympus VS‐120 slide scanner. Whole section overviews using DIC and fluorescence were taken with the 10 × objective to clearly identify gray matter and overall level of staining. Four ROIs of 500 μm² (totaling 1 mm² per section) were systematically mapped out within mid‐cortical layers (II‐V) distributed throughout the entire section. A 6‐μm deep z‐stack comprised of seven z‐slices of each ROI was obtained using a 40×/0.95 objective. ROIs chosen were representative of the overall level of tau staining throughout the gray matter, allowing for an accurate assessment of tau pathology within the cortex. All slide scanner image files (.vsi) were opened in Fiji using the BioFormats Importer. A MIP was produced and saved as a .tif file for subsequent thresholding and positive pixel analysis. Tau‐positive plaques and tangles within all ROIs of each case were manually counted.

Neuronal staining and quantification

Neurons were quantified on 10‐μm formalin‐fixed paraffin‐embedded (FFPE) sections stained with 0.1% cresyl violet acetate. Sections were dewaxed in xylene and hydrated through a series of graded ethanol solutions. Cresyl violet solution was applied for eight minutes followed by a deionized water rinse. Differentiation was achieved by briefly washing sections in absolute ethanol. The appropriate level of staining was confirmed by light microscopy before sections were cleared in xylene for nine minutes and mounted with DPX for cover slipping. Neuronal counts were performed using an Olympus BX50 microscope at 200 × magnification.

Image analysis

Image analysis was performed using Fiji (NIH, Bethesda, MD, USA). Aβ and tau areal fraction, a more accurate indicator of pathology load than quantifying individual entities because of size variation, was determined by expressing a positive pixel count as a percentage of total pixels (Supplementary Figure S1). However, to enumerate the percentage of Aβ‐immunopositive plaques (Aβ plaques) that were associated with a microglial cluster, the density of Aβ plaques was also calculated for each region. Given the subjective nature of microglial subtype classification, automated morphometric image analyses were also performed to complement traditional, qualitative cell counts. Cortical strip images of IBA1‐stained IHC sections were used and underwent color deconvolution and manual thresholding where two primary parameters were measured: (i) “arborised area” which was determined by summing the area of all convex hull polygons overlaying each cell mask and (ii) “structural complexity” which was assessed using the Skeletonize 2D/3D ImageJ plugin 1 which provided total branch length, total number of branches and total number of junctions (Supplementary Figure S2).

Statistical analyses

The normality of all data was tested using the Shapiro‐Wilk test. Group differences were determined using either a Welch’s T test or a Mann–Whitney U test where data were not normally distributed, or chi‐squared test. Regional differences were compared using either one‐way analysis of variance (ANOVA) with Games‐Howell test for pairwise comparisons or Kruskal‐Wallis test with Dunn’s test for pairwise comparisons using SPSS statistics 24 (SPSS Inc., Chicago, IL, USA). P‐values < 0.05 were considered statistically significant. Potential confounders including brain pH, APOE genotype, post‐mortem interval (PMI) and fixation period along with age and sex were investigated by univariate analysis. Factors that were significantly associated with microglial subtypes, arborized area or skeletal analysis parameters were included in multivariate models along with case‐control status using JMP 10 (SAS Institute Inc., Cary, NC, USA). Microglial parameters that survived multivariate testing were further investigated by calculating Pearson’s r or Spearman’s ρ for relationships with tau or Aβ. All graphs presented here show significant findings following multivariate testing and were produced using GraphPad Prism 7.00 (GraphPad Software Inc., La Jolla, CA, USA).

Results

Clinicodemographic characteristics

Pathologically confirmed AD cases and non‐demented controls were matched for age and sex. The AD cases were more likely to carry the APOE ε4 allele (P = 0.04), had greater AD‐type pathology according to the ABC scoring system 35 and Braak staging 7 and had lower brain pH (P = 0.03) and brain weight (P = 0.04) (Table 2).

Table 2.

Demographic, clinical and pathological characteristics of cohort. ^a

	Control (n = 15)	AD (n = 8)	P‐value
Sex (male/female)	4, 11	2, 6	0.9
Age (years)	83.9 ± 8.1	83.3 ± 6.8	0.9
Disease duration (years)	–	8.3 ± 3.0	–
ABC score (Low, intermediate, high)	10, 5, 0	0, 1, 7	<0.0001
Braak stages (0/I/II, III/IV, V/VI)	11, 4, 0	0, 2, 6	0.0002
Brain pH	6.4 ± 0.27	6.2 ± 0.24	0.03
PMI (hours)	16.2 ± 11.1	14.5 ± 11.5	0.8
Fixation (months)	52.9 ± 29.7	37.5 ± 23.9	0.2
Brain weight (g)	1241 ± 150.7	1086 ± 163	0.04
APOE ε4 (no/yes)	10, 4 b	2, 6	0.04

^aMean ± standard deviation throughout.

^bAPOE genotype of one control case missing (M21).

Neuropathology

Cortical thickness differed between AD cases (2.1 ± 0.3 mm) and controls (2.6 ± 0.5 mm, P = 0.01) in the ITC, whereas cortical thinning was not observed in the PVC (2.2 ± 0.2) or SFC (2.8 ± 0.4) of AD cases compared to controls (PVC = 2.2 ± 0.2, P = 0.6; SFC = 2.7 ± 0.4, P = 0.8). Neuronal loss was observed in the SFC (145.8 ± 63.5 cells) and ITC (132.9 ± 29.6) of AD cases compared to controls (SFC = 207.2 ± 35.1; ITC = 220.7 ± 69.2) (Figure 1A). Aβ areal fraction in the PVC (1.2 ± 0.5%), SFC (1.6 ± 1.5%) and ITC (1.8 ± 0.6%) of AD cases was greater compared to controls (PVC = 0.3 ± 0.5%, P = 0.0004; SFC = 0.5 ± 0.6%, P = 0.02; ITC = 0.6 ± 0.8%, P = 0.002) (Figure 1B). Similarly, the density of Aβ plaques in the PVC (22.2 ± 17.8 plaques/mm²), SFC (37.1 ± 28.6) and ITC (40.6 ± 19) of AD cases was higher compared to controls (PVC = 6.6 ± 12.8, P = 0.003; SFC = 11.5 ± 15.8, P = 0.01; ITC = 14.4 ± 20.9, P = 0.008) (Figure 1C). Tau areal staining in the mid‐cortical layers was significantly greater in AD cases (PVC = 1.4 ± 0.6%; SFC = 1.9 ± 0.8%; ITC = 1.8 ± 0.6%) compared to controls (PVC = 0.01 ± 0.03%, P < 0.0001; SFC = 0.07 ± 0.2%, P < 0.0001; ITC = 0.1 ± 0.2%, P < 0.0001) across all three regions (Figure 1D). The density of cortical tau‐positive neuritic plaques was significantly higher in the PVC (18.8 ± 12.5 plaques/mm²), SFC (8.4 ± 6.0) and ITC (7.4 ± 9.4) of AD cases compared to controls (PVC = 0.2 ± 0.8, P < 0.0001; SFC = 0.4 ± 1.4, P = 0.0001; ITC = 0.4 ± 0.8, P = 0.002) (Figure 1E). Cortical NFT density was significantly greater in AD cases (PVC = 9.4 ± 8.7 NFTs/mm²; SFC = 25.4 ± 13.9; ITC = 21.9 ± 20.4) compared to controls (PVC = 0.2 ± 0.6, P < 0.0001; SFC = 0.8 ± 3.1, P < 0.0001; ITC = 0.7 ± 1.9, P < 0.0001) in all regions (Figure 1F). Pathological load did not differ significantly between any of the three regions in AD or control cases.

Figure 1.

Neuropathological characterization. Box and whisker plots show mean values for: cortical neuronal density (A), cortical Aβ areal fraction (percentage of positive pixels stained) (B), Aβ plaque density (C), cortical tau areal fraction (D), tau‐positive neuritic plaque density (E) and NFT density (F). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Error bars indicate SD throughout.

Loss of healthy microglia in the ITC of AD brains

Total IBA1‐immunopositive microglia and commonly defined morphological subtypes: ramified (Figure 2A), activated and dystrophic were quantified in the three cortical regions of AD cases and controls. A range of phenotypes within the activated (Figure 2B,C) and dystrophic (Figure 2D,E) subtypes were seen within all individuals as well as clusters made up of activated microglia in those individuals with AD pathology (Figure 2F). Total microglia were lower in the ITC of AD cases compared to controls. This was exclusively caused by a reduction in the number of ramified microglia in the ITC of AD cases (Table 3). In contrast, neither activated nor dystrophic microglia differed in total number, or as a percentage of total microglia, between AD cases and controls in any of the three regions. Microglial clusters were relatively rare in all regions but were seen more frequently in the PVC of AD cases compared to controls. There was no difference in the density of clusters in the SFC or ITC between AD cases and controls.

Figure 2.

Microglial morphologies. Representative photomicrographs from the ITC show exemplars of microglial subtypes quantified here. A. Ramified microglia seen in a control case (M20) have small, spherical soma and symmetrically distributed, thin, highly branched processes. B. Activated microglia in an AD case (M09) that are hypertrophied and have reduced branching compared to ramified microglia. Arrows indicate ramified microglia for comparison. C. Activated microglia that are ameboid in morphology (arrows) are seen among other activated cells (arrow heads) in another AD case (M11). D. Dystrophic microglia that are pseudo‐fragmented with discontinuous IBA1 staining of processes in the same AD case (M11). E. A dystrophic microglia that is deramified and has spheroidal swellings at the end of its processes (arrows) from an AD case (M13). F. Clusters of activated microglia seen in an AD case (M09). Nuclei counterstained with hematoxylin. Scale bar = 50 μm.

Table 3.

Quantification of the morphological subtypes of microglia.

	Control (cells/mm2)	%	AD	%	P‐value
			PVC
Total Microglia	140.7 ± 45.2		139.6 ± 43.1		0.95
Ramified microglia	55.3 ± 34	39.3	39.5 ± 25	28.3	0.26
Activated microglia	55 ± 29.3	39.1	64 ± 27.4	45.8	0.48
Dystrophic microglia	30.4 ± 17.3	21.6	36 ± 19.9	25.8	0.48
Microglial clusters	0.7 ± 1.9		2.8 ± 2.3		0.03
			SFC
Total Microglia	188.4 ± 59.6		187.1 ± 62		0.96
Ramified microglia	70.8 ± 36.9	37.6	49.1 ± 23.8	26.2	0.15
Activated microglia	81.6 ± 47	43.3	90.4 ± 42.9	48.3	0.67
Dystrophic microglia	36 ± 23.4	19.1	47.6 ± 24.3	25.4	0.28
Microglial clusters	0.8 ± 1.1		3.2 ± 3		0.05
			ITC
Total Microglia	205 ± 66		128 ± 52.2		0.01
Ramified microglia	77.3 ± 47.2	37.7	13.8 ± 8.9	10.8	0.001
Activated microglia	86.6 ± 42.4	42.2	70.6 ± 60.4	55.2	0.46
Dystrophic microglia	41.1 ± 23.3	20.0	43.7 ± 30.5	34.1	0.82
Microglial clusters	1.1 ± 1.3		1.9 ± 2.1		0.28

Non‐demented, higher pathology‐controls have increased activated microglia

The control cases were sub‐divided into five “high‐pathology control cases” (HPCs), whose levels of AD‐type pathology were similar to the AD cases and 10 “low‐pathology control cases” (LPCs) with low or no AD pathology (Table 2). ITC gray matter of HPCs was thicker compared to LPCs (P = 0.01; Figure 3A). Aβ positive immunostaining (P = 0.02; Figure 3B) and cortical tau immunostaining (P = 0.01; Figure 3C) was also increased compared to LPCs. This also coincided with more total (P = 0.02) and activated microglia (P = 0.01) in HPCs compared to LPCs (Figure 3D). Group differences presented here survived multivariate testing (see Supplementary tables ST1 and ST2 for univariate and multivariate tests that included data for control cases only).

Figure 3.

Comparisons between low pathology controls (LPCs) and high‐pathology controls (HPCs). A. HPCs had thicker cortices, elevated Aβ (B) and tau areal staining (C) compared to LPCs in the ITC. D. Total and activated microglia were also higher in HPCs compared to LPCs.

Reduced microglial arborized area and structural complexity in AD

A reduction in the arborized area was observed in the SFC and ITC of AD cases compared to controls, but not in the PVC (Figure 4A; Table 4). Similarly, a standard positive pixel analysis showed these same effects (Figure 4B). A skeletal analysis showed that the total branch length, the total number of branches and the total number of junctions were significantly reduced in AD cases compared to controls in the ITC (Table 4). Total junctions were also significantly reduced in the SFC of AD cases compared to controls, but there were no changes in any of these parameters in the PVC.

Figure 4.

Microglial arborized area. Box and whisker plots demonstrating the area of tissue covered by microglial processes as determined using the convex hull analysis (A) and IBA1 positive pixel analysis (B) in the PVC, SFC and ITC of AD cases and controls.

Table 4.

Microglial structural complexity.

	Control	AD	P‐value
Arborized area (%)
PVC	29.3 ± 16.7	19.7 ± 9.4	0.09
SFC	39.6 ± 18.5	24.8 ± 9.4	0.02
ITC	35.8 ± 13.4	17.5 ± 8.1	0.0006
Positive pixels (%)
PVC	12.2 ± 5.7	9.2 ± 3.6	0.2
SFC	15.8 ± 6.2	11.3 ± 3.3	0.03
ITC	14.8 ± 4.4	8.8 ± 3.1	0.001
Summary of skeletal analyses (mm2)	PVC
Total branch length (mm)	47.2 ± 33.8	32.9 ± 29.3	0.06
Total #branches	3873 ± 1406	3042 ± 805	0.14
Total #junctions	1114 ± 524.2	756.1 ± 312.1	0.09
	SFC
Total branch length (mm)	78.6 ± 68.6	44.6 ± 36	0.1
Total #branches	5385 ± 1873	3949 ± 1139	0.06
Total #junctions	1762 ± 854.5	1031 ± 426.2	0.03
	ITC
Total branch length (mm)	65.7 ± 50.9	21.9 ± 8.8	0.0002
Total #branches	5480 ± 2091	2665 ± 1110	0.002
Total #junctions	1685 ± 757.6	625.2 ± 386.6	0.001

Microglial clustering was not a major feature of the AD cortex

Clusters of microglia (three or more cell bodies within a 50‐μm² graticule field) were occasionally found near cored and diffuse Aβ plaques and tau‐positive plaques in both the control (Figure 5A–F) and AD groups (Figure 5G–O). These microglia were predominantly activated or, less commonly, dystrophic in morphology (Figure 5G–I). There was no difference between the percentage of Aβ plaques associated with a microglial cluster between the AD and control groups in any of the three regions: PVC (AD = 16.8 ± 14.4%, control = 12.2 ± 23.4; P = 0.15), SFC (AD = 12.4 ± 12.7, control = 5.8 ± 5.5; P = 0.28) and ITC (AD = 5.4 ± 6.5, control = 17.2 ± 19.9; P = 0.25) (Figure 5J–L) or tau‐positive plaques (Figure 5M–O) in either the AD cases or controls.

Figure 5.

Associations between microglia and AD plaques. Microglia were found to only occasionally cluster around tau‐positive plaques and Aβ‐immunopositive diffuse and cored plaques in both AD and control cases. Photomicrographs of immunofluorescent labeling for Aβ (A), IBA1 (B) and the merged image (C) from a control case with elevated AD‐type pathology (M23) demonstrates a cluster of ameboid microglia associated with a cored Aβ plaque. D–F. Ramified microglia and no microglial activation associated with a cored plaque of another control case (M19). Similarly, for AD cases, microglia were found to only occasionally cluster around Aβ plaques; as seen in the photomicrographs from an AD case (M09) demonstrating both significant clustering around diffuse and cored Aβ plaques (G–I), and an absence of microglial clustering (J–L). M–O. Tau‐positive plaques from the same AD case demonstrating an absence of microglial clustering. All photomicrographs were acquired using a Zeiss LSM 800 confocal microscope (40×/1.3 Oil). Scale bars = 50 μm.

Morphological changes of microglia correlate with tau pathology

Potential confounders of microglial parameters including brain pH, PMI and fixation time were investigated by univariate analyses (Supplementary table ST3) and those that were or approached significance were included in multivariate analyses with case status for the three regions. Neuronal loss in the SFC and ITC, and reduced arborized area, IBA1 positive pixels, total branch length, total number of branches and number of ramified microglia in the ITC, as well as increased density of microglial clusters in the PVC remained significantly different following multivariate testing (Table 5; refer to Supplementary table ST4 for all multivariate statistics).

Table 5.

Summary of multivariate analyses (case‐control differences).

	PVC		SFC		ITC
	R 2	P‐value	R 2	P‐value	R 2	P‐value
Neurons			0.26	0.01	0.36	0.003
Arborized area					0.36	0.003
IBA1 positive pixels					0.35	0.004
Total branch length					0.22	0.03
Total #branches					0.12	0.04
Ramified microglia					0.39	0.002
Microglial clusters	0.2	0.04

Those microglial parameters associated with AD status were further investigated for associations with either tau or Aβ pathology (Table 6). In the ITC, neurons, arborized area, IBA1 positive pixels and ramified microglia were all inversely correlated with the level of tau staining. There was also an inverse correlation between total branch length and Aβ in the ITC. Microglial clusters were positively correlated with both Aβ and tau staining in the SFC, and only tau staining in the PVC.

Table 6.

Neuropathological correlations. ^a

		PVC		SFC		ITC
	Pathology	R 2	P‐value	R 2	P‐value	R 2	P‐value
Neurons	Tau b					0.27	0.01
Arborized area	Tau					0.2	0.03
IBA1 positive pixels	Tau					0.18	0.04
Total branch length	Aβ b					0.2	0.03
Ramified microglia	Tau					0.27	0.01
Microglial clusters	Aβ b			0.4	0.001
	Tau b	0.27	0.01	0.12	0.04

^aAll correlations presented in this table include variables that are significantly different between groups and remained so following multivariate analysis (excluding microglial clusters in the SFC, where P = 0.05).

^bIndicates most significant correlation determined by multivariate analysis that included both Aβ and tau areal fractions (following univariate analyses demonstrating correlations with both pathologies).

Brain pH is an important effector of microglial dystrophy

Among the potential confounders, brain pH was inversely correlated with a number of different microglial parameters; most notably the presence of dystrophic microglia (Table 7). The exception was ramified microglia in the SFC where a positive correlation was observed. A combined analysis of cases and controls showed that dystrophic microglia were inversely correlated with brain pH in all three regions (Figure 6).

Table 7.

Brain pH effects on microglial morphology.

	PVC		SFC		ITC
	R 2	P‐value	R 2	P‐value	R 2	P‐value
Arborised area	0.24	0.02	0.37	0.003
IBA1 positive pixels	0.22	0.03	0.4	0.002
Total #branches					0.44	0.0008
Total #junctions			0.4	0.002	0.45	0.0006
Ramified microglia			0.64	<0.0001
Dystrophic microglia	−0.56 a	<0.0001	−0.55 a	0.008	−0.56 a	0.006

^aSpearman’s ρ.

Figure 6.

Dystrophic microglia and brain pH. Scatter plots and regression lines show the inverse relationships between dystrophic microglia and brain pH in the PVC (A), SFC (B) and ITC (C).

Discussion

This study undertook to provide clarity on how microglia relate temporally to the progression of Alzheimer’s disease by comparing morphological subtypes in differentially affected areas of post‐mortem brain tissue. Unlike previous studies, our findings showed that activated microglia were only increased in high‐pathology controls while clustering of microglia with Aβ plaques was uncommon and highest in the mildly affected PVC (~17%), suggesting that there is an early microglial reaction to Aβ but this is prior to the onset of dementia. Whereas only 12% and 5% of Aβ plaques were associated with a microglial cluster in the SFC and ITC, respectively. Second, the severely affected ITC, was characterized by a reduced number of ramified (healthy) microglia. Third, dystrophic microglia were not associated with AD status or pathology, but rather brain pH.

As expected, AD cases had greater quantities of Aβ and tau‐positive plaques and neurofibrillary pathology compared to non‐demented controls. The ITC showed a reduction in the number of neurons and AD‐related cortical thinning, consistent with a more advanced stage of AD than the SFC, which had reduced neurons but no cortical atrophy and the PVC in which neither of these measures was affected. Therefore, changes seen in the PVC can be viewed as representative of earlier disease mechanisms; with previous pathological and molecular studies also suggesting that the amount of tau pathology in the PVC has minimal effects on the neurons in this region 22, 28. Intra‐regional differences in cortical pathology was not as extensive as predicted here from staging schema and volumetric studies. The commonly used ABC staging criteria for AD is semi‐quantitative in nature, stressing the spread of pathology rather than the quantity of pathology, specifically as it pertains to each defined phase or stage of the criteria 21. Indeed, it is quite rare to find studies that have systematically quantified AD pathology across cortical regions, with those that have, describing significant heterogeneity in the accumulation of plaques and NFTs within regions and limited regional differences 37, 64.

The working hypothesis for how microglia contributes to AD pathogenesis has been that activated cells associate with plaques and produce a neurotoxic milieu that culminates in neuronal degeneration 30, 41, 46. Others have suggested that astrocytes may be equally or more prominent in this area 48, 49, 63, while Liddelow and colleagues recently showed that the interplay between microglia and astrocytes is critical by demonstrating that microglia‐derived cytokines induce a neurotoxic phenotype in astrocytes 29. In contrast, work begun by Streit, Graeber and Kreutzberg 55 and continued by Stevens, Barres and colleagues opined that a loss of microglial function and particularly a deficit in their role in synaptic maintenance may be more important in AD pathogenesis 18. Streit and colleagues suggested that dystrophic microglia were an early and prominent feature of AD 54, an idea which may also be relevant to other neurodegenerative diseases such as Dementia with Lewy bodies 3.

Yet, rather than confirming the predominance of either activated or dystrophic microglia in the AD cortex, the major finding here was a reduction in ramified microglia in the ITC only, and that this loss occurred in the absence of a commensurate increase in either the activated or dystrophic subtypes. In contrast, activated microglia were significantly greater in controls with sufficient AD pathology 35 to be pathologically diagnosed with AD. If these “high pathology” controls are representative of the proposed preclinical AD phenotype 53 then microglial activation may be an early driver of the disease. However, the fact remains that microglial reactivity seen in this group was not associated with dementia and so could alternatively be considered a protective mechanism against the neurotoxic effects of relatively high AD‐type pathology, an idea consistent with transition events of protective disease‐associated microglia (DAM) occurring at early stages of the disease process 25. Furthermore, as DAM are known to localize closely around plaques, the possibility that this activation is associated with neuroprotective mechanisms is further strengthened by the observation of increased microglial clustering in PVC, an area that may be considered representative of earlier stages in the disease process caused by the milder effects of AD observed here.

Given the turnover of microglia in the adult human brain 43, the loss of microglia could represent a decrease in proliferation or increase in cell death. Alternatively, it may represent a loss of IBA1 immunoreactivity caused by the downregulation of this calcium binding protein. The latter has important implications for microglial functionality as reduced IBA1 would result in impaired mobility and phagocytosis 24. Additional pan‐microglial markers are required to differentiate between these two scenarios. Indeed, previous investigations have observed significant increases in other markers, such as MHC II, with at least one identifying an IBA1^‐/MHC II⁺ subpopulation 48. However, this phenotypic subpopulation represented a minority of cells in that study. Alternatively, this question could be approached with proliferative and apoptotic markers but our experience with the proliferative marker Ki‐67 10 and the apoptotic marker, cleaved caspase 3 (unpublished data) are that these are rare events in the adult brain and a single “point in time” assay may not be informative for a chronic disease such as AD.

As Aβ is widely regarded as a precursor to tau pathology, further temporal information on the role of microglial subtypes in AD pathogenesis was sought by determining the relative associations with Aβ or tau load. Microglial clusters were correlated with both Aβ and tau pathology in the PVC and SFC (univariate testing), representing early microglial reactivity that may be associated with neuroprotective processes and potentially corresponding with DAM that have been investigated elsewhere 25. Microglial arborized area, IBA1 positive pixels and the number of ramified microglia were inversely correlated with the level of tau in the ITC, which is indicative of the neurodegenerative stage of the disease and is consistent previous reports of microglial degeneration following phagocytosis of AT8 and/or AT100‐postive phospho‐tau species 45.

The other major finding here was an inverse correlation between brain pH and dystrophic microglia in all three regions. A low brain pH is often indicative of a severe and prolonged agonal period 12, 34, common in patients with neurodegenerative diseases. The antemortem factors and events that comprise a patient’s agonal period impact a patient’s brain pH. These factors include: duration of hospitalization, coma, respiratory illness and need for ventilation, systemic inflammation and duration of terminal phase. Unlike the work of Streit and colleagues, there were no differences in the number of dystrophic microglia, including in a sub‐cohort excluding samples with pH < 6.0. It is unclear whether previous studies proposing a link between microglial dystrophy and progression of AD considered differences in brain pH 54; however, the loss of healthy microglia in the ITC here still supports a loss of function hypothesis for the role of microglia in AD. More widely the prominence of immune signaling pathways in omic studies using post‐mortem brain tissue could similarly represent the influence of brain pH rather than a general inflammatory mechanism in brain disease 15, 50, 62, 65.

This investigation had three important limitations. Namely, the subjective categorization of the morphological subtypes of microglia, the use of a single microglial marker, IBA1 and the small cohort size. Microglial activation does not follow a monophasic (“all‐or‐nothing”) process, rather these cells respond to changes in their surroundings in a calibrated manner, thereby allowing for a wide range of morphological presentations that can be difficult to categorize. Microglia are rapidly responsive cells operating in time scales of hours and days unlike AD, the archetypal chronic brain disease. This response time may well account for the relationship between dystrophic microglia and brain pH, a proxy for agonal period. Although IBA1 is recognized as an effective pan‐microglial marker 23, more subtle changes in activation including phagocytic status may have been missed. For example, other human post‐mortem studies have demonstrated upregulation of major histocompatibility complex class II (MHC II) and cluster of differentiation 68 (CD68), markers of activation and phagocytosis, respectively; see review 19.

Study strengths include the use of well‐characterized tissue and a previously verified stereological approach that ensured a quantitative exploration across the entire cortical strip for pathology, residual neurons and microglia 27. This stereological approach ensured that the sampling area is sufficient to generate data that is representative of the tissue section under investigation and by selecting cortical strips at areas where the cortical thickness is at its thinnest, oblique cuts made during tissue sectioning do not inflate thickness measurements. Furthermore, the largely subjective morphological characterization of microglia was corroborated by automated “skeletal” analysis which validated our previous observations of reductions in structural complexity and arborized area in the more severely affected ITC 9. A validated approach for normalization of cortical atrophy 39 was also employed, although this may still have underestimated the true loss of cells in the ITC of AD patients as atrophy in the rostrocaudal dimension is not taken into consideration. Reassuringly, a previous study has indicated that shrinkage in this dimension is minimal 20.

Overall, the novel findings here suggest that ramified microglia are lost in more advanced stages of AD. The loss of the surveillance and support of microglia may cause neurons to be more susceptible to tau‐related degeneration. While activation, as seen in higher pathology controls and as an increase in the clustering of microglia in the PVC of AD cases, may be seen as key processes in preventing conversion to dementia. Future studies on morphological changes in microglia in human AD brain tissue must account for the patient’s brain pH as a potential confounder for dystrophic microglia. Further immunohistochemical studies with functional markers will be required to determine the more nuanced aspects of microglia dysfunction in AD, but the current findings suggest that strategies aiming to augment microglial function are likely to be more effective as a treatment than those attenuating their pro‐inflammatory activities, particularly in earlier phases of the disease.

Data Accessibility Statement

All data pertinent to this study is contained within the manuscript or supplementary tables and figures.

Supporting information

Figure S1. Quantification of Aβ and tau areal fraction. A. Confocal photomicrograph showing Aβ positive immunofluorescent plaques in the ITG of an AD case (M09). B. The corresponding digitally generated mark up used to calculate areal fraction with an ImageJ positive pixel algorithm. Scale bars = 50 μm.

Figure S2. Quantification and analysis of microglial territoriality and skeletal structure. A. Photomicrograph demonstrating a single field of view with IBA1 + microglia and hematoxylin counterstained nuclei from an AD case (M09). ImageJ color deconvolution plugin was used to separate DAB (B) and hematoxylin (C) stains; with the third component (D) representing the remainder of the  subtractive mixing  algorithm. The image of the isolated DAB staining was manually thresholded, with the digitally generated mark up (E) allowing for the convex hull area to be calculated by summing all polygons bounded in green (F) and the production of tagged skeletal frames of each cell mask (G) that could be used to calculate total branch length, total number of branches and total number of junctions. Scale bar = 100 μm.