Disease-related microglia heterogeneity in the hippocampus of Alzheimer’s disease, dementia with Lewy bodies, and hippocampal sclerosis of aging

Five groups of cases (Table 1) were pathologically-confirmed as either AD (n?=?7), HS-Aging (n?=?7), AD?+?HS-aging (n?=?4), DLB (n?=?12), and NC (n?=?9). HS-aging and DLB cases were included in this study to determine if there is
disease specificity in microglia pathology and to provide the first quantitative analysis
of microglia in HS-Aging. Pure HS-aging cases lacked substantial additional pathologies
AD-type pathology, or Lewy bodies 19]–21], as shown in Table 1. The neuropathological changes associated with neocortical/diffuse Lewy body disease
include, by definition, ?-synuclein immunoreactive neuronal inclusions (Lewy bodies)
and processes in multiple portions of the cerebral neocortex. In pure DLB, there are
low levels of amyloid-? pathology or NFTs, as shown in Table 1.

Primary goals of this study were to assess regional microglia heterogeneity and to
exploit the ability of digital neuropathological quantification to detect in differences
microglial morphometry when cases are stratified according to their neurodegenerative
diseases. Six regions of interest (ROI) were identified by dividing the hippocampal
formation into the dentate gyrus (DG), the cornu ammonis (CA) areas (CA1, CA2/3, and
CA4), the subiculum (sub), and the adjacent white matter (WM) (Fig. 1). Representative examples of the ROIs are shown in Fig. 1.

Fig. 1. Regions of interest used for microglia analysis. A representative hippocampus is shown
for the five neuropathological diagnoses included in this study. The outlines illustrate
the boundaries used in identifying the following brain regions: white matter (WM),
subiculum (sub), the cornu ammonis (CA) areas, CA1, CA2/3, CA4, and the dentate gyrus
(DG). The ROIs shown in the figure are not the actual ROIs used for analysis, as some
of the ROIs (WM and sub) could not be included in the image frame, as the brain region
was larger than the image frame

Pattern of CD68 staining in the hippocampus of autopsy cases

Quantification of the CD68 positive pixels is shown in Fig. 2. By a one-way ANOVA a significant effect of disease status was found sub (Fig. 2c; F_4,38?=?6.3001; p?=?0.0007), CA1 (Fig. 2d; F_4,38?=?8.0944; p??0.001), DG (Fig. 2g; F_4,38?=?5.3332; p?=?0.0019), and in the average of the six regions in the hippocampus formation
(Fig. 2h; F_4,38?=?4.3221; p?=?0.0062). No significant effect was found by a one-way ANOVA in WM (Fig. 2b), CA2/3, (Fig. 2e), or CA4 (Fig. 2f). HS-aging, AD, and AD?+?HS-aging were found to have significantly more CD68⁺ staining in the CA1 region compared to NC or DLB cases (Fig. 2d). However, there was no significant difference among the three disease conditions
(HS-aging, AD, and AD?+?HS-aging) in the CA1 region (Fig. 2d). Interestingly, we found significantly more CD68⁺ staining in the DG of AD cases compared to the other four groups (Fig. 2g). When averaged across the six-hippocampal formation sub regions, the AD cases were
found to have significantly more CD68⁺ staining compared to NC or DLB groups. Overall, the greatest CD68⁺ staining was seen in the WM, as is evident by the heatmap summary of the CD68 positive
pixel analysis (Fig. 2i). A survey of the CD68⁺ staining in the six-hippocampal formation regions illustrates the regional and disease-specific
heterogeneity in the staining (Fig. 3). Of note is a large round cell type that can be found in areas of high density staining
as shown in Fig. 3b-c. Interestingly, just distal to the very intense accumulation of CD68⁺ cells, the CD68⁺ staining was unremarkable, with a few ramified microglia (Fig. 3d). Quantification of the number of large round CD68⁺ cells was done using the nuclear algorithm, by adjusting the algorithm to detect
only the large round cells as shown in Fig. 4. In comparison to design based stereological methods, limitations of the nuclear
algorithm include an inability to provide an estimate of the total number of microglia,
because of a lack of 3-dimensional volume measurements 22], 23]. Limitations notwithstanding, results of the nuclear algorithm were similar to the
positive pixel algorithm, with the HS-aging and AD groups having the greatest number
of CD68⁺ cells (Table 2). As shown by the heatmap, the greatest number of CD68⁺ cells was found in the WM of AD cases (Fig. 4).

Fig. 2. Digital neuropathological quantification using positive pixel algorithm of CD68⁺ immunostaining in the hippocampus of autopsy cases. Representative example of (a) CD68 staining and a digitally generated mark-up showing the ability of the positive
pixel algorithm to detect the staining. Digital neuropathological quantification of
the CD68 staining using the positive pixel algorithm is shown for the (b) WM, (c) sub, (d) CA1, (e) CA2/3, (f) CA4, (g) DG, and for the (h) average of the hippocampal formation. Circles represent an individual case, with
mean and SEM shown for the group. Statistical comparisons: *p??0.05 compared to AD cases. ^§p??0.05 compared to HS-aging cases. ^‡p??0.05 compared to AD?+?HS-aging cases. (i) Heatmap summarizes the results shown in (b-h) (also see Table 2)

Fig. 3. Survey of CD68⁺ staining in the hippocampus of autopsy cases. (a) Representative examples of CD68⁺ staining pattern in the brain regions analyzed by digital neuropathological analysis.
(b) Low power photomicrograph of hippocampus of a DLB individual (case #36) highlights
an area of intense staining (blue arrow) shown in (c), and an area of low CD68 staining (black arrow) in a nearby region (d)

Fig. 4. Digital neuropathological quantification using nuclear algorithm of number of large
round CD68⁺ cells in the hippocampus of autopsy cases. Representative example of (a) CD68 staining and a digitally generated mark-up showing the ability of the nuclear
algorithm to detect the staining of the large round cells, but not smaller cells or
processes. Digital neuropathological quantification of the CD68 staining using the
nuclear algorithm to detect the staining is shown for the (b) WM, (c) sub (F_4,38?=?2.9934; p?=?0.0321), (d) CA1, (e) CA2/3, (f) CA4, (g) DG (F_4,38?=?4.3393; p?=?0.0061), and for the (h) average of the hippocampal formation. Circles represent an individual case, with
mean and SEM shown for the group. Statistical comparisons: *p??0.05 compared to AD cases. (i) Heatmap summarizes the results shown in (b-h) (also see Table 2)

Digital quantification of IBA1 staining in the hippocampus of autopsy cases

Quantification of the number of IBA1⁺ cells by the nuclear algorithm is shown in Fig. 5. A representative example of the ability of the algorithm to detect individual cells
is shown in Fig. 5a. By a one-way ANOVA, a significant effect of disease status was found in the CA1
region (Fig. 5d; F_4,38?=?3.9914; p?=?0.0092), CA2/3 region (Fig. 5e; F_4,38?=?5.8525; p?=?0.0011), and in the CA4 (Figure 5f F_4,38?=?2.6929; p?=?0.0473). No significant effect was found by a one-way ANOVA in WM (Figure 5b), sub (Fig. 5c), DG (Fig. 5g), or in the average of the six regions in the hippocampus formation (Fig. 5h). In the CA1 region, the HS-aging had an increased number of IBA1⁺ cells compared to NC, AD or DLB. As shown by the heatmap summary, a similar pattern
of increased number of IBA1⁺ microglia was found in the HS-aging and AD?+?HS-aging groups compared to the NC,
AD, or DLB groups (Fig. 5i). Quantification of the IBA1 positive pixels (Table 2) also showed a similar pattern of increased IBA1⁺ staining in the HS-aging and AD?+?HS-aging groups compared to the NC, AD, or DLB
groups (Fig. 6).

Fig. 5. Digital neuropathological quantification using nuclear algorithm of number of IBA1⁺ cells in the hippocampus of autopsy cases. Representative example of (a) IBA1 staining and a digitally generated mark-up showing the ability of the nuclear
algorithm to detect six stained cells. Digital neuropathological quantification of
the IBA1 staining using the nuclear algorithm is shown for the (b) WM, (c) sub, (d) CA1, (e) CA2/3, (f) CA4, (g) DG, and for the (h) average of the hippocampal formation.. Circles represent an individual case, with
mean and SEM shown for the group. Statistical comparisons: ^§p??0.05 compared to HS-aging cases. (i) Heatmap summarizes the results shown in (b-g) (also see Table 2)

Fig. 6. Digital neuropathological quantification using positive pixel algorithm of IBA1⁺ immunostaining in the hippocampus of autopsy cases. Representative example of (a)
IBA1 staining and a digitally generated mark-up showing the ability of the positive
pixel algorithm to detect the staining. Digital neuropathological quantification of
the IBA1 staining using the positive pixel algorithm is shown for the (b) WM, (c) sub, (d) CA1 (F_4,38?=?5.0943; p?=?0.0025), (e) CA2/3 (F_4,38?=?4.8888; p?=?0.0032), (f) CA4, (g) DG,, and for the (h) average of the hippocampal formation (F_4,38?=?3.0201; p?=?0.0311). Circles represent an individual case, with mean and SEM shown for the
group. Statistical comparisons: *p??0.05 compared to AD cases. ^§p??0.05 compared to HS-aging cases. ^‡p??0.05 compared to AD?+?HS-aging cases. (i) Heatmap summarizes the results in (b-h) (also see Table 2)

IBA1⁺ microglia morphology in the hippocampus of autopsy cases

An examination (Fig. 7a) of the IBA1⁺ microglia in the six ROIs in the five neuropathologic groups showed remarkable heterogeneity
in microglia density, as captured by the digital neuropathological quantification.
There was also heterogeneity in IBA1⁺ microglia morphology, which was underappreciated in the digital neuropathological
analysis, as microglia density and cell number were measured irrespective of the microglia
morphology. For example, a striking pattern of IBA1⁺ microglia morphology is the rod-shaped microglia, which were readily apparent in
a subset of cases. As shown in Fig. 7b-c, rod-shaped microglia are characterized by a narrow cell body with a few planar processes.
The rod-shaped microglia could be found as individual cells (Fig. 7b), or as long and thin groups of cells that may have fused (Fig. 7b and c and Fig. 8). The appearance of microglia with polarized and parallel processes suggested that
the microglia could be following neurites—possibly, degenerating axons or neurons
themselves. To test the possibility that microglia could be surrounding degenerating
neuronal processes, double label immunofluorescence was performed for microglia (IBA1)
and NFTs (PHF1). Fig. 8a shows abundant PHF1⁺ staining and IBA1⁺ rod-shaped microglia in the CA1 region of an AD individual (case #23). We found no
evidence of systematic overlap of PHF1⁺ neurites and IBA1⁺ rod-shaped microglia, as shown in Fig. 8b. Rather, long trains of rod-shaped microglia could sometimes be seen to run parallel
to and between PHF1⁺ neurons but did not co-localize with the PHF1⁺ staining (Fig. 8c). In this example, the tip of the rod-shaped microglia was near (but not within)
a PHF1⁺ structure, and the IBA-1 immunoreactive structure appeared to be a fusion / cluster
of multiple cells with 5 clearly visible DAPI⁺ nuclei (Fig. 8d).

Fig. 7. Survey of IBA1⁺ staining in the hippocampus of autopsy cases. (a) Representative examples of IBA1⁺ staining pattern in the brain regions analyzed by digital neuropathological analysis(b) A low powered photomicrograph shows the widespread distribution of rod shaped microglia
in the CA1 region of a DLB individual (case #34). Long trains of microglia (highlighted
by blue arrows) are shown at higher magnification in (c).

Fig. 8. Lack of localization of IBA⁺ rod-shaped microglia to PHF1⁺ neurons in an AD individual (case #23). (a) A low powered photomicrograph shows the distribution of rod-shaped microglia next
to PHF1⁺ cells. (b) A linear group of rod-shaped microglia is shown at a higher magnification. (c) A second example of rod-microglia, where the microglia run parallel and between
PHF1+ neurons. (d) Of note, the polar end of the rod-microglia (white arrow) was found to have 5 DAPI⁺ nuclei

Another pattern of microglia morphology observed was the dystrophic / degenerating
microglia, which overlapped morphologically with cells that have been described to
have processes that are spheroidal, beaded, de-ramified, or fragmented 24], 25]. Examples of dystrophic / degenerating microglia are shown in Fig. 9. In AD (Fig. 9a) and DLB (Fig. 9b), for example, the dystrophic / degenerating microglia had very thin processes that
are beaded and fragmented. In HS-aging (Fig. 9c) and AD?+?HS-aging (Fig. 9d), dystrophic microglia morphology was more striking, and the processes of the microglia
were beaded and tortuous.

Fig. 9. Dystrophic IBA1⁺ microglia in the hippocampus. Examples of IBA1⁺ dystrophic microglia in the CA1 region of AD individual (a; case #20), DLB individual
(b; case #33), HS-aging individual (c; case #15), and AD?+?HS-aging individual (d;
case #27). Scale bar is 25 ?m

The remarkable diversity in the microglia morphology led us to carefully review and
categorize the morphological appearances of the microglia into five distinct classes
(Fig. 10a), to allow measurement of changes in the microglia classes associated with the five
neurodegenerative disease groups. The five classes of microglia morphologies included:
1) ramified microglia, which have a ‘surveying’ non-reactive microglia morphological
appearance, with thin highly branched processes 5], 26]; 2) hypertrophic microglia (often called activated microglia), which have become
enlarged, hyper-ramified or may have short thick processes 5], 26]; 3) dystrophic microglia, with processes that are spheroidal, beaded, de-ramified,
or fragmented 24]–26]; 4) rod-shaped microglia, characterized by a narrow cell body with a few planar processes
27], 26]; and 5) amoeboid microglia, with an enlarged cell body with few to no processes 5], 26]. CD68 staining could clearly identify cells with an amoeboid morphology, and to a
lesser extent cells with a ramified morphology. In contrast, IBA1 staining was useful
to identify all five microglia morphologies. Therefore, IBA1 staining was used to
quantify the distribution in the microglia morphology according to these five subtypes
of microglial shapes. Focusing on the CA1 region of the hippocampus, the number of
each of the five morphological classes of IBA1⁺ microglia was counted in five randomly placed and evenly distributed 250×250?m regions
of interest (ROI). HS-aging cases had fewer ramified microglia than NC (p?=?0.0091) or AD (p?=?0.0027) cases (Fig. 10b, Table 2). HS-aging and AD?+?HS-aging had the most hypertrophic microglia. AD?+?HS-aging cases
had more hypertrophic microglia than NC (p?=?0.0132), AD (p?=?0.0270), or DLB (p?=?0.0044) cases, and HS-aging cases had more hypertrophic microglia
than DLB (p?=?0.0140) cases (Fig. 10c, Table 2). HS-aging cases had more dystrophic microglia than NC (p?=?0.0005), AD (p?=?0.0193), or DLB (p?=?0.0225) cases (Fig. 10d, Table 2). Quantification of rod-shaped microglia identified a subset of cases with abundant
rod-shaped microglia; however, the cases were not specific to a disease group (Fig. 10e, Table 2). AD?+?HS-aging cases had more amoeboid microglia than NC (p?=?0.0428), or DLB (p?=?0.0085) cases (Fig. 10f, Table 2). The total number of microglia in the CA1 region, regardless of morphology, was
greatest in HS-aging and AD?+?HS-aging. AD?+?HS-aging cases had more total microglia
than NC (p?=?0.046), or DLB (p?=?0.0035) cases. HS-aging cases had more total microglia than NC (p?=?0.0072) or DLB (p?=?0.0048) cases (Fig. 10g, Table 2). As the total number of microglia was found to be altered in the different groups,
each of the five microglia classifications was plotted as a percentage of the total
number of microglia (Fig. 10h) to help visualize the microglia morphology distributions within and among the different
diseases.

Fig. 10. Disease specific patterns in IBA1⁺ microglia morphology. (a) Representation of microglia morphologies seen in the hippocampus of aged individuals.
The number of microglia was quantified at 40x magnification in five 250 x 250 ?m regions
of interest (ROIs) that were randomly placed and evenly spaced in the CA1 region.
Following the classification shown in (a), IBA⁺ microglia were classified as either (b) ramified (F_4,38?=?5.3533; p?=?0.0019), (c) hypertrophic (F_4,38?=?5.5082; p?=?0.0016), (d) dystrophic (F_4,38?=?5.7249; p?=?0.0012), (e) rod-shaped, or (f) amoeboid (F_4,38?=?3.9836; p?=?0.0093). (g) The number of microglia (F_4,38?=?7.2694; p?=?0.0002) in the five classifications was summed to get the total number of microglia.
The gray circles in (b-g) represent the average number of microglia per mm² for an individual case, with mean and SEM shown for each group (see also Table 2). Statistical comparisons: ^§p??0.05 compared to HS-aging cases. ^‡p??0.05 compared to AD?+?HS-aging cases. (h) As the total number of microglia significantly varied by group, the number of microglia
in each of the five classifications was plotted as a percent of the total number of
microglia to illustrate the disease-related patterns in microglia morphology (also
see Table 2)