Factors associated with development and distribution of granular/fuzzy astrocytes in neurodegenerative diseases

Abstract Granular/fuzzy astrocytes (GFAs), a subtype of “aging‐related tau astrogliopathy,” are noted in cases bearing various neurodegenerative diseases. However, the pathogenic significance of GFAs remains unclear. We immunohistochemically examined the frontal cortex, caudate nucleus, putamen and amygdala in 105 cases composed of argyrophilic grain disease cases (AGD, N = 26), and progressive supranuclear palsy (PSP, N = 10), Alzheimer’s disease (AD, N = 20) and primary age‐related tauopathy cases (PART, N = 18) lacking AGD, as well as 31 cases bearing other various neurodegenerative diseases to clarify (i) the distribution patterns of GFAs in AGD, and PSP, AD and PART lacking AGD, (ii) the impacts of major pathological factors and age on GFA formation and (iii) immunohistochemical features useful to understand the formation process of GFAs. In AGD cases, GFAs consistently occurred in the amygdala (100%), followed by the putamen (69.2%) and caudate nucleus and frontal cortex (57.7%, respectively). In PSP cases without AGD, GFAs were almost consistently noted in all regions examined (90–100%). In AD cases without AGD, GFAs were less frequent, developing preferably in the putamen (35.0%) and caudate nucleus (30.0%). PART cases without AGD had GFAs most frequently in the amygdala (35.3%), being more similar to AGD than to AD cases. Ordered logistic regression analyses using all cases demonstrated that the strongest independent factor of GFA formation in the frontal cortex and striatum was the diagnosis of PSP, while that in the amygdala was AGD. The age was not significantly associated with GFA formation in any region. In GFAs in AGD cases, phosphorylation and conformational change of tau, Gallyas‐positive glial threads indistinguishable from those in tufted astrocytes, and the activation of autophagy occurred sequentially. Given these findings, AGD, PSP, AD and PART cases may show distinct distributions of GFAs, which may provide clues to predict the underlying processes of primary tauopathies.


INTRODUCTION
Currently, the pathological diagnosis of various tauopathies is based on the presence of disease-specific neuronal and glial inclusions in which phosphorylated tau protein is abnormally aggregated, as well as their anatomical distribution. Neurofibrillary tangles (NFTs) are the most common nondisease-specific tau-positive lesions, and historically, several studies have attempted to define several diseases by the anatomical distributions of NFTs alone visualized by early silver impregnation methods. Thereafter, argyrophilic astrocytic lesions demonstrated by a sensitive silver stain, the Gallyas method, that is, tufted astrocytes (TAs) (35,57,58,77) and astrocytic plaques (16,35), were found to have diagnostic value in progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD). The Gallyas method also demonstrated the thorn-shaped astrocytes that are found in cases having various pathological conditions (26). Subsequently, the accumulation of phosphorylated tau in these astrocytic lesions was reported (16,26,57). Besides thorn-shaped astrocytes, another morphological subtype of tau-positive non-disease-specific astrocytic lesions has been reported. The astrocytic lesions have a small amount of tau accumulation in the perinuclear regions of astrocytes, as well as fine or fuzzy granular tau accumulation in astrocytic processes. Botez et al first described this subtype of astrocytic lesions in the amygdala in brains with argyrophilic grain disease (AGD) and called them "bush-like astrocytes" (4). They also noticed that these astrocytic lesions were not stained by the Gallyas method. Thereafter, morphologically and immunohistochemically similar tau-positive astrocytic lesions were also reported using different terms, for example, diffuse granular tau immunoreactivity in astrocytic processes (42), diffuse granular tau immunopositivity along astrocytic processes (41), fine granular tau immunoreactivity in astrocytic processes (17), astrocytes with finely tau immunoreactive processes (76), TA-like astrocytic lesions (27) and granularshaped astrocytes (80). In a recently proposed classification of tau-positive astrocytic lesions, these tau-positive astrocytic lesions were called "granular/fuzzy astrocytes (GFAs)" as a subtype of "aging-related tau astrogliopathy (ARTAG)" (38).
GFAs preferably occur in the gray matter, that is, the cerebral cortex and nuclei rather than the white matter (37,38). GFAs are basically Gallyas-negative, although accumulated tau only in the cytoplasm of astrocytes can rarely show weak argyrophilia (38). GFAs can be found in elderly cases having heterogeneous clinical and pathological bases (17,27,41,42), including PSP (27,44,80), Alzheimer's disease (AD) (44), primary age-related tauopathy (PART) (44), CBD (44), Pick's disease (44) and chronic traumatic encephalopathy (18). It is not always easy to evaluate the effect of each pathology on GFA formation because several tauopathies often coexist in a single case. Especially, given that AGD increases in frequency with age (9) and might be associated with the development of GFAs (4), the potential effect of AGD on the frequency of GFAs in other tauopathies needs to be carefully evaluated. However, although only a few previous studies have examined the relationship between various tauopathies and ARTAG (ie, thorn-shaped astrocytes and GFAs) (43,44), as far as we know, the frequency and distribution of GFAs by specific anatomical regions in pathological conditions common in the elderly, that is, AGD, PSP without AGD, AD without AGD or PART without AGD, have not been examined.
The aims of the study described here were to examine (i) the frequency and distribution of GFAs in AGD cases, as well as PSP, AD and PART cases lacking AGD, respectively, (ii) the impacts of major pathological factors and the age on GFA formation in regions in which GFAs preferably occur and (iii) immunohistochemical findings useful to understand the formation process of GFAs. To address these issues, we first semiquantitatively examined GFAs, as well as TAs, which are potentially related to GFA formation (27), in the frontal cortex, caudate nucleus, putamen and amygdala in AGD, PSP, AD and PART cases, as well as cases of various other degenerative diseases. Then, after performing univariate and multivariate analyses to assess the effects of factors on the development of GFAs by region in all cases, we examined the immunohistochemical features of GFAs in AGD cases using phosphorylation-or conformational change-specific anti-tau, anti-ubiquitin and anti-p62 antibodies as well as the Gallyas method. In this paper, we demonstrate that GFAs may occur with disease-specific distribution patterns in cases of AGD, PSP without AGD, AD without AGD and PART without AGD, respectively, and that age may not be always associated with GFA formation. In addition, we demonstrate that the increase of immunoreactivities of phosphorylation-and conformationdependent anti-tau antibodies in GFAs, the occurrence of Gallyas-positive glial threads in GFAs and the activation of autophagy may occur sequentially in AGD cases.

MATERIALS AND METHODS Subjects
A total of 1125 autopsy cases were registered in the database at the Department of Neuropsychiatry, Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences as of the end of December 2017. All cases died in psychiatric hospitals or neurological departments of general hospitals. We first selected 202 cases that were recently assessed using modern standardized methods including a panel of immunohistochemistry, modified Bielschowsky silver stain and Gallyas-Braak silver stain from 2001 to 2017 for the selection of AGD, PSP, AD and PART cases. AGD cases (N = 26) that lacked any other primary tauopathies except for PART [ie, AD, PSP, CBD, globular glial tauopathy (GGT), diffuse NFTs with calcification (DNTC), Pick's disease and frontotemporal lobar degeneration with tau gene mutation (FTLD-MAPT)] were selected. Likewise, PSP (N = 10) cases that lacked any other primary tauopathies except for PART were selected. AD (N = 20) cases without any other tauopathies were then selected. Definite PART cases without any other degenerative disease [N = 5; Braak & Braak stage (Braak stage) I-IV, Thal phase 0] were also selected from the 202 cases.
Among the 25 non-tauopathy cases, 13 cases also had PART: nine had definite PART (Braak stage I-IV, Thal phase 0), and four had possible PART (Braak stage I-IV, and Thal phases 1-2). Therefore, we made a PART group (N = 18) by combining these additional 13 PART cases and the first five definite PART cases without any other degenerative diseases (Supporting File S1).
The demographic data of a total of the 105 cases that we eventually examined in the present study are shown in Table 1. In this study, except for AGD and CBD cases, we selected only cases that lacked argyrophilic grains in any anatomical region. Because all of our CBD cases had various degrees of AGD, they were exceptionally included in this study despite the coexistence of AGD. Regarding PSP pathology, several previous studies including ours suggested that AGD cases often had small numbers of NFTs and TAs in the basal ganglia and brain stem nuclei, distribution of which is reminiscent of that in PSP cases (27,64). While TAs are considered to be specific to PSP, whether all TAs observed in cases of other diseases mean an early change of PSP remains unclear. Therefore, in the present study, even when cases had a few Gallyasor AT8-positive TAs, they were not diagnosed as PSP when lacking NFTs that met the pathological criteria of PSP (25). For example, when an AGD case had a few AT8-positive TAs but lacked sufficient NFTs that fit the pathological criteria of PSP, the case was classified into an AGD group.
Autopsies were carried out after informed consent was obtained from family members, and all experiments in this study were approved by the ethical committees of the Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences, the National Hospital Organization Minami-Okayama Medical Center, Zikei Institute of Psychiatry and the Tokyo Metropolitan Institute of Medical Science.
The criteria of each pathological diagnostic category were as follows: AGD was diagnosed based on Gallyaspositive grains in the limbic system with and without the temporal cortex (7) and the absence of any other primary tauopathies except for PART (13) [ie, NFTs with Braak stage I-IV (6,8) and no or minimal Aβ deposits with Thal phase 0-2 (69)]. The diagnosis of PSP was based on the presence of both Gallyas-positive TAs and AT8positive NFTs whose quantity and distribution met the pathological criteria of either typical or atypical PSP (25). The AD group was composed of cases having intermediate or high level AD pathology (Braak stage V-VI, Thal phase 3-5) (52). Cases of CBD met the pathological criteria and had Gallyas-positive AT8-positive astrocytic plaques in the striatum and frontal cortex (15). Cases of GGT (N = 2) had numerous 4-repeat (4R) tau-positive 3-repeat (3R) tau-negative globular glial inclusions that correspond to type I (N = 1) and type II (N = 1) histologies (1). DNTC was diagnosed based on severe NFTs (Braak stage VI), the absence of or minimal Aβ deposits (ie, Thal phase 0-1), Fahr-type calcification and circumscribed temporal lobe atrophy (31,36,66,78). Pick's disease was defined by the presence of 3R tau-positive 4R taunegative Pick bodies (53,79). Previously reported cases of frontotemporal lobar degeneration due to P301L tau mutation (50) and postencephalitic parkinsonism (22) were also included. Because, the case of postencephalitic parkinsonism showed distinct clinical and pathological findings (30), it was not included in the AD group although it had severe NFTs and Aβ deposits (Braak stage V, Thal phase 4). MSA was diagnosed based on Gallyas-positive α-synuclein-positive Papp-Lantos inclusions and tissue degeneration with a distinctive distribution (61,74). Diagnoses of FTLD-TDP (2,12,47,56) and ALS-TDP were based on TDP-43 pathology and neuronal loss with a distinct distribution and the absence of other primary neurodegenerative diseases. FTLD-FUS was composed of two cases of basophilic inclusion body disease and one ALS case with the FUS mutation (12,54,55). The diagnoses of DRPLA and Huntington's disease were pathologically and genetically confirmed. Myotonic dystrophy cases had a diagnosis of DM1 by genetic and/or clinical criteria (21,71).

Conventional neuropathological examination and immunohistochemistry
A standardized neuropathological evaluation was done on all subjects. Brain tissue samples were fixed post-mortem with 10% of formaldehyde and embedded in paraffin. Tenµm-thick sections from the frontal, temporal, parietal, occipital, insular and cingulate cortices, hippocampus, amygdala, basal ganglia, midbrain, pons, medulla oblongata and cerebellum in the left hemisphere were prepared. For the standardized neuropathological assessment, sections were stained with hematoxylin-eosin and Klüver-Barrera stains, and selected regions stained with modified Bielschowsky silver and Gallyas methods. Then, formalin-fixed paraffin sections were cut at six μm thickness from selected regions from all cases for the standard immunohistochemical evaluation of representative neurodegenerative changes. Deparaffinized sections were incubated with 1% of H 2 O 2 in methanol for 30 minutes to eliminate endogenous peroxidase activity, and washed in phosphate-buffered saline (PBS, pH 7.4). After blocking with 10% of normal serum, sections were incubated overnight at 4°C with primary antibodies ( Table 2). When using antitau, anti-ubiquitin, anti-p62, anti-α-synuclein and anti-TDP-43 antibodies, sections were autoclaved for 10 minutes in 10 mM of sodium citrate buffer at 121°C for antigen retrieval. When using 12B2, RD4, RD3 and T22, sections were autoclaved for 10 minutes in 10 mM of sodium citrate buffer at 121°C and treated with 70% of formic acid for 10 minutes. After three 5-min washes in PBS, sections were incubated in biotinylated secondary antibody for 30 minutes, and then, in avidin-biotinylated horseradish peroxidase complex (ABC Elite kit, Vector, Burlingame, CA, USA) for 60 minutes. After three 5-min washes in PBS, the peroxidase labeling was visualized with 3,3'-diaminobenzidine (DAB) as the chromogen. Sections were lightly counterstained with hematoxylin.
AGD was diagnosed based on the presence of Gallyas-, AT8-and RD4-positive argyrophilic grains, and the distribution was assessed using sections stained with the Gallyas method and the staging system reported by Saito et al (63). Other major degenerative changes, that is, NFTs (AT8, Thal phase (69), median (63), median  (12,54,55) were also evaluated in all subjects using established criteria, respectively.

Semiquantitative assessment of GFAs and TAs
It was originally reported that GFAs were frequently found in the amygdala in AGD cases (4). Our previous study revealed that GFAs may be also frequent in the frontal cortex and striatum in AGD and PSP cases (27). In addition, it was reported that these regions may be potentially significant sites to understand the progression pattern of GFAs (44). Therefore, in the present study, we semiquantitatively evaluated the density of GFAs in the superior frontal gyrus, caudate nucleus, putamen and amygdala in all subjects using the staging system: stage 0, no lesion; stage 1, one or more lesions in the anatomical region but less than one lesion per × 200 visual field; stage 2, one lesion per × 200 visual field; stage 3, 2 to 10 lesions per × 200 visual field; stage 4, over 11 lesions per × 200 visual field. The grades were determined separately by two investigators (TM and OY) who did not know whether the case had any pathological data. If there was a discrepancy between the results, the two researchers reviewed the tissues and decided the final grade by discussion. In our previous study, some AGD cases had a few astrocytic lesions that bore both tau-positive granular components like GFAs and Gallyas-positive glial threads like TAs (27). In PSP cases, TAs were reported to preferentially develop in the frontal cortex and striatum (3,24,29). Therefore, in this study, to examine the relationship between GFAs and TAs, TAs were additionally evaluated in the superior frontal gyrus, caudate nucleus, putamen and amygdala in all cases on AT8-and Gallyas-stained sections, respectively, using the staging system for GFAs.
To secure the reproducibility of results, we regarded taupositive astrocytic lesions as GFAs only when the lesions had the typical morphological features of GFAs ( Figure 1A-G) originally reported in AGD cases (4) and shown in the consensus paper on ARTAG (38). In addition, because it is difficult to sharply distinguish "GFAs with Gallyas-positive glial threads" from "Gallyas-positive TAs" on Gallyas-stained sections, we counted astrocytic lesions having Gallyas-stained glial threads as TAs irrespective of the quantity of threads.

Double immunohistochemistry and double staining of AT8 immunohistochemistry and Gallyas method
To examine whether GFAs occur in association with Aβ deposits, double immunostaining with AT8 and 12B2 was done in all cases that had Aβ deposits in the frontal cortex, striatum and/or amygdala. Primary antibody labeling in the first cycle (AT8) was detected in the same way as single staining using the DAB reaction. Then, primary antibody labeling in the second cycle (12B2) was also detected in the same way as single staining, and the reaction was visualized with Vector Blue Alkaline Phosphatase Substrate (Vector Laboratories, Inc.) as the chromogen. Double staining using the Gallyas method and AT8 immunohistochemistry was also done in representative AGD and PSP cases to examine whether GFAs actually have TA-like Gallyas-positive glial threads. Sections were first stained by the Gallyas method, followed by immunostaining with AT8. The peroxidase labeling was visualized with Vector Red Alkaline Phosphatase Substrate (Vector Laboratories, Inc.) as the chromogen. Sections were lightly counterstained with hematoxylin.

Confocal laser scanning microscopy
Double-labeling immunofluorescence was performed with the combination of phosphorylation-dependent anti-tau antibody (AT8, mouse, monoclonal, 1:500) and anti-4R tau antibody (anti-4R tau, rabbit, polyclonal, 1:300, Cosmo Bio Co., Tokyo, Japan), and the combination of phosphorylation-dependent anti-tau antibody (AT8, mouse, monoclonal, 1:500) and anti-p62-C antibody (p62-C, guinea pig, polyclonal, 1:200, Progen, Heidelberg, Germany). Sections from the frontal cortex and caudate nucleus in representative AGD and PSP cases having TAs and GFAs were autoclaved for 10 minutes in 10 mM of sodium citrate buffer at 120°C. Following washing in PBS, nonspecific antibody binding was blocked with a 5% of skimmilk buffer, and sections were incubated overnight with a mixture of the two primary antibodies at 4°C. After washing in PBS, sections were incubated for 1 h at room temperature with a mixture of two fluorescence-labeled secondary antibodies as follows: Alexa Fluor 488 goat anti-mouse IgG

Tau immunoblotting
Frozen brain tissue from the caudate nucleus and putamen of the right hemisphere in two AGD and two PSP cases was available. In addition, frozen tissue from the frontal or temporal cortex in one PSP case, one CBD case and one AD case was examined as disease controls. These samples were used for Western blotting according to methods described previously (14,48). Brain samples (0.5 g) from patients were individually homogenized in 10 mL of homogenization buffer (HB: 10 mM Tris-HCl, pH 7.5, containing 0.8 M of NaCl, 1 mM of EGTA and 10% sucrose). Sarkosyl was added to the lysates (final concentration: 2%), which were then incubated for 30 minutes at 37°C and centrifuged at 20 000 g for 10 minutes at 25°C. The supernatant was divided into tubes (each 1.3 mL) and centrifuged at 100 000 g for 20 minutes at 25°C. The pellets were further washed with sterile saline (0.5 mL/tube) and centrifuged at 100 000 g for 20 minutes. The resulting pellets were used as the sarkosyl-insoluble fraction (ppt). The sarkosyl-ppt was sonicated in 50 μL (/tube) of 30 mM Tris-HCl (pH 7.5) and solubilized in 2 × sample buffer. Samples were run on gradient 4-20% polyacrylamide gels and electrophoretically transferred to PVDF membranes. Residual protein-binding sites were blocked by incubation with 3% of gelatin (Wako) for 10 minutes at 37°C, followed by overnight incubation at room temperature with primary anti-tau antibodies [T46, mouse, monoclonal, 1:2000. pS396, rabbit, monoclonal, 1:2000 (23,68)]. The membrane was incubated for 1 h at room temperature with anti-mouse IgG (BA-2000, Vector Lab) or anti-rabbit IgG (BA-2000, Vector Lab), then incubated for 30 minutes with avidin-horseradish peroxidase (Vector Lab), and the reaction product was visualized by using 0.1% of DAB and 0.2 mg/mL of NiCl 2 as the chromogen.

Statistical analysis
The Mann-Whitney U test and Fisher's exact test were used to compare the variables between two groups. When comparing the variables between three groups, Kruskal-Wallis and Steel-Dwass tests were used. Spearman rank order correlation analysis was applied for univariate correlations between two variables. To assess the effects of predictor variables on the formation of GFAs, we constructed a multivariate ordered logistic regression model. The stages of GFAs (stages 0 to 4) were categorized into three levels (stage 0, stage 1 and stages 2 to 4) by combining categories in which the number of cases was small, and the data were submitted as the dependent variable in the model. Independent variables tested for their association with the dependent variable consisted of the age at death, Braak stage (stages 0 to 6), Thal phase (phases 0 to 5), PSP (presence or absence) and AGD (presence or absence). Because univariate analyses demonstrated the possibility that clinical and pathological factors may have different effects on GFA formation by anatomical region, the multivariate analysis was carried out after stratification by anatomical region (ie, the frontal cortex, caudate nucleus, putamen and amygdala). Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated after controlling simultaneously for potential confounders. A P-value < 0.01 was accepted as significant. Statistical analysis was performed using Bell Curve for Excel 2.15 (Social Survey Research Information Co., Ltd., Tokyo, Japan).

Morphological and immunohistochemical features of GFAs in all subjects
All the AGD cases had GFAs in at least one region examined ( Figure 1A-G). GFAs were composed of small tau-positive granules that were scattered around astrocytic nuclei or often radially arranged around the nuclei, and often in contact with vessel walls. These lesions were labeled with a series of anti-tau antibodies except for RD3. Some GFAs were also p62-positive ( Figure 1H). In AGD cases, the immunoreactive pattern of TAs was similar to that of GFAs ( Figure 1I-O); however, no GFA having small argyrophilic granules was noted on Gallyas-stained sections. All PSP cases also had both GFAs and TAs in which immunoreactivity and morphological features were similar to those in AGD cases, respectively (Figure 2A-E). In the frontal cortex in AGD and PSP cases, GFAs tended to be distributed in the ridge of the convolution rather than the depth of the sulci, especially when the number of lesions was small. Nine of 20 AD cases had a small number of GFAs in at least one region ( Figure 2F-J). Some GFAs in AD cases were observed to be spatially associated with a few short thick dystrophic neurites that may have originated from neurons ( Figure 2G, an arrow). Even in AD cases, the overlap between GFAs and Aβ deposits was rare ( Figure 2F-J). Seven of all 18 PART cases had GFAs at least in one region examined ( Figure 3A-D).
Of seven CBD cases, one had a small number of GFAs in all regions examined ( Figure 2K,L), while all CBD cases had many unclassifiable astrocytic lesions reminiscent of classic astrocytic plaques in which granular, oval or rod-like tau accumulations tended to be distributed in the peripheral region ( Figure 2M,N) in addition to classic astrocytic plaques ( Figure 2O). One case of postencephalitic parkinsonism had a few GFAs in the frontal cortex, caudate nucleus and putamen in addition to a larger number of distinct astrocytic lesions having very fine taupositive granules ( Figure 3E-H). Three of four cases of myotonic dystrophy had a few GFAs in the frontal cortex, caudate nucleus and/or amygdala ( Figure 3I,J). All cases of myotonic dystrophy had characteristic tau-positive astrocytic lesions in which crinkly or twisted thread-like structures were distributed in the peripheral rather than the central portion ( Figure 3K-O). One case with DNTC had a few GFAs in the putamen.
Frequency and distribution patterns of GFAs and TAs in AGD, PSP, AD and PART groups AGD cases, PSP cases without AGD, AD cases without AGD and PART cases without AGD showed distinct distribution patterns of GFAs ( Figure 4A-D). In AGD cases, GFAs were consistently noted in the amygdala (100.0%), followed by the putamen (69.2%) and caudate nucleus and frontal cortex (57.7%, respectively) and the high density of GFAs of stage 2 or over was frequently found in the amygdala (54.2%) rather than the frontal cortex, caudate nucleus and putamen (19.2%, 15.4% and 11.5%, respectively) ( Figure 4A). In PSP cases without AGD, in contrast to AGD cases, a large number of GFAs were consistently noted in the amygdala, caudate nucleus and putamen (100.0%, respectively), and also frequently in the frontal cortex (90.0%) ( Figure 4B), and the high density of GFAs of stage 2 or over was frequently found in the frontal cortex, caudate nucleus, putamen (80%, 70% and 70%, respectively) rather than the amygdala (33.3%) ( Figure 4B). In AD cases without AGD, although far less frequent compared with AGD and PSP groups, GFAs were noted in the putamen (35.0%), followed by the caudate nucleus (30.0%), frontal cortex (20.0%) and amygdala (5.3%) ( Figure 4C). In PART cases without AGD, GFAs were most frequently noted in the amygdala (35.3%), followed by the frontal cortex (16.7%) and the caudate nucleus and putamen (5.6%, respectively), being similar to the distribution pattern observed in AGD rather than AD cases ( Figure 4D). The age at death, sex, brain weight, Braak stage, Thal phase, the proportion of cases with Lewy body disease (LBD) and the proportion of cases with TDP-43 pathology were not significantly different between PART cases with and without GFAs (Mann-Whitney U test and Fisher's exact test; Supporting File S1).
In AGD, PSP, AD and PART groups, the distribution pattern of GFAs was similar to those of AT8-positive TAs and Gallyas-positive TAs, respectively ( Figure 4A-D).
In the comparison between AGD, PSP and AD groups, the densities of GFAs in the amygdala in an AGD group were significantly larger than those in the AD group, respectively (P < 0.001, Kruskal-Wallis and Steel-Dwass tests). The densities of GFAs in all regions in the PSP group were significantly larger than those in the AD group (P < 0.001, respectively, Kruskal-Wallis and Steel-Dwass tests). The densities of GFAs in the frontal cortex, caudate nucleus and putamen in the PSP group were significantly larger than those in the AGD group (P = 0.006, 0.007 and 0.004, respectively, Kruskal-Wallis and Steel-Dwass tests). According to recently proposed distribution patterns of ARTAG (44), the distributions of GFAs in AGD and PART cases were similar to pattern 1 (amygdala first pattern), while the distribution of GFAs in AD cases was similar to pattern 2 (striatum first pattern), respectively (Supporting File S2).

Age at death and pathological features by region in all cases with or without GFAs
We then compared the age at death and pathological features of all cases with and without GFAs to explore factors potentially associated with GFA formation ( Table 3). The age at death in cases having GFAs in the caudate nucleus and putamen was significantly higher than that in cases lacking GFAs, respectively (P = 0.005 and <0.001, respectively, Mann-Whitney U test; Table 3). The AGD stage in cases having GFAs in the caudate nucleus, putamen and amygdala was significantly higher than that in cases lacking GFAs (P = 0.004, 0.002, and <0.001, respectively, Mann-Whitney U test). The frequency of PSP cases was significantly higher in cases having GFAs than in cases lacking GFAs in all regions, respectively (P < 0.001, respectively, Fisher's exact test). Braak stage, Thal phase and the frequencies of LBD and TDP-43-positive lesions did not differ between the two groups.

Correlation analyses of GFA stage and clinical and pathological variables by region in all subjects
The correlations between the GFA stage and clinical and pathological variables in all subjects were analyzed using Spearman rank order correlations (Supporting File S3). The AGD stage was significantly correlated with the GFA stage Twisted fine threads are often radially arranged, appearing to be reminiscent of threads in astrocytic plaques in CBD. However, the threads of astrocytic lesions in MD cases tend to be finer and longer than those usually observed in astrocytic plaques in CBD cases (see Figure 2O). The frontal cortex. All scale bars = 30 μm. were almost consistently noted in all regions examined, and the proportions of cases having the high density GFAs (stage 2 or over) in the striatum and frontal cortex were higher than that in the amygdala. The distribution pattern of TAs was similar to that of GFAs. C. In AD cases (N = 20), GFAs were more frequent in the striatum compared with the frontal cortex and amygdala, although less frequent than in AGD and PSP cases. D. In PART cases (N = 18), GFAs were the most frequent in the amygdala, whose distribution pattern was similar to that in an AGD rather than AD and PSP groups. *Significant differences between PSP and AD groups (P < 0.01). †Significant differences between AGD and AD groups (P < 0.01). ‡Significant differences between AGD and PSP groups (P < 0.01). Kruskal-Wallis and Steel-Dwass tests. in the putamen and amygdala, respectively (ρ = 0.271 and 0.530; P = 0.005 and <0.001). On the contrary, the age at death was significantly correlated with the GFA stage only in the caudate nucleus and putamen (ρ = 0.267 and 0.358; P = 0.006 and <0.001).
The GFA stages in the frontal cortex, caudate nucleus, putamen and amygdala were significantly correlated with the density of Gallyas-positive TAs in the same regions (ρ = 0.647, 0.507, 0.534 and 0.332; P < 0.001, respectively). Neither Braak stage nor Thal phase was significantly correlated with the GFA stage in any region.

Effects of age, Braak stage, Thal phase, PSP and AGD on GFA formation by region
Ordered logistic regression analyses were used to examine whether the age at death, Braak stage, Thal phase, presence or absence of PSP and presence or absence of AGD were independently associated with the increase of GFA stage by region in all cases (Table 4).
In contrast, in the amygdala, the pathological diagnosis of AGD had the strongest association with GFA formation (OR 118.14; 95% CI 23.5-594.4), followed by PSP (OR 52.95; 95% CI 9.3-300.9). The age at death, Braak stage or Thal phase was not significantly associated with GFA formation in any region examined.

Relationship between change of immunoreactivity of GFAs and TAs and progression of argyrophilic grains in AGD cases
Based on the significant correlation between the stages of GFAs and AGD in all regions demonstrated by univariate and multivariate analyses (Table 4 and Supporting File S3), to explore findings associated with the formation process of GFAs, the change of immunoreactivity of GFAs in the frontal cortex and putamen in AGD cases was examined using a panel of anti-tau, ubiquitin and p62 antibodies and the Gallyas method after stratification by AGD stage (Figure 5A,B). In this examination, astrocytic lesions having Gallyas-positive glial threads were regarded as TAs regardless of the quantity of argyrophilic structures. In both regions examined, with the progression of AGD stage, phosphorylation-and conformation-dependent anti-tau antibodies increasingly labeled GFAs and TAs, respectively ( Figure 5A,B). Gallyas-positive TAs were noted only in cases with AGD stages II or higher. p62-positive GFAs occurred only in cases with AGD stage III.  PART cases without any other tauopathies and non-tauopathies. The amygdala was not available in one of these PART cases.

Double staining with AT8 immunohistochemistry and Gallyas method
Double staining with AT8 immunohistochemistry and the Gallyas method demonstrated astrocytic lesions having fine GFA-like granular tau accumulation, and like TAs, a few to a moderate number of Gallyas-positive glial threads in AGD cases ( Figure 6A-D). Some lesions had thicker Gallyaspositive threads ( Figure 6D), being indistinguishable from TAs in PSP cases ( Figure 6F). Similar astrocytic lesions having both fine tau-positive granules and Gallyas-positive threads were noted in PSP cases also ( Figure 6E).

Confocal laser scanning microscopy
In AGD cases, double labeling with AT8 and anti-4R tau antibodies showed the accumulation of 4R tau in GFAs ( Figure 7A-F). Some TAs were intensely stained with anti-4R tau antibody in AGD cases ( Figure 7G-I). The accumulation of p62 was revealed in AT8-positive GFAs and TAs with relatively thick glial threads ( Figure 7J-O).

Tau immunoblotting in representative AGD and PSP cases
Immunoblot analysis of the sarkosyl-insoluble, urea-soluble fraction prepared from the right hemisphere with T46 and pS396 demonstrated approximately 68-and 64-kDa bands in all AGD, PSP and CBD cases ( Figure 8A, lanes 1-8 and Figure 8B, lanes [10][11][12][13][14][15][16][17]. All bands and smears tended to be more strongly stained with pS396 than T46. Low molecular weight tau fragments of approximately 33-and 37-kDa were also noted in PSP (lanes 4-7 and 13-16) and CBD (lanes 8 and 17), respectively. Of two AGD cases, one case (AGD1) showed approximately 33-to 37-kDa tau fragments in the putamen and caudate nucleus (lanes 1, 2, 10 and 11). The immunointensity of the fragments was less than those in PSP cases. The other AGD case (AGD2) showed very weakly stained 33-to 37-kDa tau fragments in the putamen (lanes 3 and 12). The density of GFAs in corresponding regions in the left hemisphere was comparable between these AGD and PSP cases (table of Figure 8B). On the contrary, GFAs and TAs tended to be stained with a larger number of anti-tau antibodies in PSP cases (lanes 13-15) than AGD cases (lanes [10][11][12]. In these cases, GFAs and TAs stained with anti-p62 antibodies and/or Gallyas method were noted only in PSP cases (lanes [13][14][15].

DISCUSSION
This is the first study that demonstrated distinct distributions of GFAs in AGD, PSP, AD and PART cases and comprehensively evaluated the impacts of patient age and major pathological factors on GFA formation by anatomical region. In AGD cases, GFAs were consistently noted in the amygdala, and to a lesser frequency, in the striatum and frontal cortex (57-69%). In PSP cases, GFAs were almost consistently noted in all regions, and the proportion of cases having the high density of GFAs was higher in the striatum and frontal cortex than in the amygdala. In contrast, only 35% of AD cases showed a few GFAs in the striatum, followed by the frontal cortex (20%) and amygdala (5%). Although the PART cases lacked AGD or any other primary tauopathies, 35% had a few GFAs with a distribution more similar to that in AGD than in AD cases. In multivariate ordered logistic regression analyses, PSP was the strongest independent factor of GFA formation in the frontal cortex, caudate nucleus and putamen, while AGD was the strongest independent factor of GFA formation in the amygdala. In contrast, neither age nor AD pathology was significantly associated with GFA formation in any region examined. In AGD cases, GFAs labeled with phosphorylation-and conformation-dependent anti-tau antibodies, Gallyas-positive glial threads in GFAs and the activation of autophagy sequentially occurred parallel to the progression of AGD. These findings suggest that GFAs may develop with a distinct distribution pattern associated with underlying tauopathies rather than age. Cases of primary tauopathies often had not only diseasespecific astrocytic lesions, but also not negligible numbers of morphologically unclassifiable tau-positive astrocytic lesions (39,43,44). Indeed, in our cases of PSP, CBD, myotonic dystrophy and postencephalitic parkinsonism, unclassifiable tau-positive astrocytic lesions having both morphological features of GFAs and disease-specific astrocytic lesions were

823
observed. Therefore, it is considered that where the boundary in the morphological evaluation is set may affect the frequency of GFAs in each study. In this study, to secure the reproducibility of results, only astrocytic lesions which showed morphological features of classic GFAs originally reported in AGD cases were regarded as GFAs (4). The previously reported higher frequency of GFAs in CBD cases (e.g., about 90% in the frontal cortex and striatum in the CBD cases (44) vs. 14.3% in the sites in our CBD cases) may be at least partially explained by such methodological issues, it being difficult to simply compare the frequency of GFAs between studies.
Although there are several previous studies regarding the immunoreactivity of GFAs (17,45,60), the change of Figure 5. Changes of immunoreactivity of GFAs and TAs parallel the progression of the AGD stage in the frontal cortex (A) and putamen (B) in AGD cases. About 35% of cases with AGD stage I had GFAs labeled by several anti-tau antibodies in the frontal cortex and putamen. Neither Gallyas-nor p62-positive lesions were noted in these cases. Gallyaspositive TAs were first noted in cases with AGD stage II, and the proportion of AGD cases with TAs increased with AGD stage. p62positive GFAs were first noted in cases with AGD stage III. In this study, when an astrocytic lesion had Gallyas-positive glial threads, we operationally regarded it as a TA irrespective of the quantity of Gallyaspositive threads.
immunoreactivity related to the process of GFA formation is hardly known. In our study, because the AGD stage was significantly correlated with the density of GFAs (Table 4), we considered that the change of immunohistochemical features of GFAs associated with the progression of the AGD stage may reflect the formation process of GFAs, at least in AGD cases. We selected the frontal cortex and putamen as regions of interest for this examination because the number of GFAs in these sites varied between cases. As a result, it was demonstrated that the increase of GFAs labeled with phosphorylation-and conformation-dependent anti-tau antibodies and the activation of autophagy occurred in the regions parallel to the progression of AGD ( Figure 5). It was also noteworthy that, with the increase of GFAs, Gallyas-positive glial threads indistinguishable from those observed in TAs in PSP cases gradually increased in GFAs in AGD cases ( Figure 5), and that the distribution of TAs was consistent with that of GFAs ( Figure 4). Further, Gallyas-AT8 double staining demonstrated that some GFAs in AGD cases actually had a few fine Gallyas-positive threads in addition to Gallyas-negative tau-positive fine granular structures as features of GFAs. Based on these findings, it is plausible that at least some GFAs may precede changes of full-blown Gallyas-positive TAs. On the contrary, at present, TAs are considered to be disease-specific lesions of PSP. Therefore, regardless of the quantity, the presence of TAs is often regarded as an early change of PSP. In this study, Gallyas-positive TAs were observed in about 40% of AGD cases as well as all PSP cases. However, because the distribution pattern of TAs, like that of GFAs, evidently differs between PSP and AGD cases (Figure 4), we consider that whether a small number of Gallyas-positive TAs observed in AGD cases always means the coexisting early change of PSP remains to be elucidated. Interestingly, in our results of tau immunoblotting, although the number of cases examined was small, the band pattern of approximately 33-to 37-kDa low-molecular weight tau fragments might differ between AGD and PSP cases even when the same anatomical region (ie, striatum) was examined ( Figure 8). It is also unclear whether the inconsistency reflects distinct pathophysiologies in AGD and PSP, respectively, or reflects a continuum of GFAs and TAs that is shared by two tauopathies. Further comparative studies between immunohistochemical and biochemical findings using a larger number of cases are necessary to clarify the pathogenic significances of GFAs and TAs in tauopathies, especially in AGD and PSP.
Kovacs et al (44) statistically explored potential progression patterns of ARTAG in a large case series including cases of CBD, PSP, Pick's disease, PART, AD, MSA, LBD and ALS. AGD coexisted in some cases, although detailed information, for example, the number of cases having AGD and the severity of AGD, was not shown. In their study, they statistically extracted two potential progression patterns of GFAs, that is, patterns 1 and 2 (44). In pattern 1, GFAs may first develop in the striatum and extend to the frontotemporal cortex [the striatum first pattern (37)]. Interestingly, this pattern appears to be consistent with that of GFAs observed in our AD cases ( Figure 4C). Considering that Kovacs's 528 non-FTLD tauopathy cases included 243 AD cases (46.0%), it is plausible that pattern 1 mainly reflects the distribution of GFAs in their AD cases. Regarding the distribution of GFAs in AD cases, a recent study examined the frequency of GFAs in the frontal, temporal, cingulate, angular gyri, putamen, amygdala and entorhinal cortex (59). Although GFAs were comparable in density in these regions, in their study, unlike ours, AD cases with AGD were not excluded because of their high frequency. However, given our findings, such a selection might increase the frequency of GFAs in the amygdala in AD cases. In addition, multivariate analyses demonstrated that AGD was not associated with the presence of GFAs, being inconsistent with our results. However, the effect of AGD on GFA formation was not evaluated by anatomical region. Therefore, their findings cannot be compared with our results. On the  contrary, in pattern 2 reported by Kovacs et al (44), GFAs may first develop in the amygdala and subsequently in the subcortical nuclei and frontotemporal cortex [the amygdala first pattern (37)], being consistent with the distribution pattern of GFAs observed in our AGD cases ( Figure 4A). Somewhat unexpectedly, previously reported findings regarding the detailed distribution and frequency of GFAs in AGD cases are limited. In a previous study including 54 cases of degenerative diseases with coexisting AGD, it was demonstrated that GFAs were most frequent in the medial temporal region including the hippocampus, amygdala and inferior temporal gyrus, and the proportion of cases having GFAs at least in one of these regions was 90.4% (43). In that study, only three cases of AGD without any other primary tauopathy were examined; of these, one case had GFAs at least in one of the regions, although the detailed distribution was not described. In another study, only the effect of grains as coexisting lesions in the hippocampus pyramidal layers, dentate gyrus, inferior temporal gyrus and/ or amygdala on GFA formation in the amygdala in a case series of diverse degenerative diseases was examined, and a significant relationship between grains in these regions and GFAs in the amygdala was noted (44). Based on these previous findings together with our results, when GFAs are predominantly distributed in the amygdala, the possibility should be considered that the case may have the pathogenic process associated with the formation of argyrophilic grains.
It is noteworthy that about 35% of our PART cases had GFAs, the distribution of which was consistent with that in AGD cases, although they lacked argyrophilic grains. In a previously reported series of PART cases as well, a similar distribution pattern of GFAs with the highest frequency in the amygdala (29%) was observed (44). These results led us to consider the possibility that at least some PART cases with amygdala-predominant GFAs actually have not only the pathological process of PART, but also the very early glial change of AGD, even when lacking argyrophilic grains on sections stained with the Gallyas method. This view may be supported by previous statistical analyses using the conditional probability that demonstrated that GFA Figure 8. Tau immunoblot analysis of the sarkosyl-insoluble, ureasoluble fraction prepared from the right hemisphere of representative AGD (AGD1 and AGD2), PSP (PSP1 and PSP2) and disease control cases (PSP3, CBD and AD) with T46 (A) and pS396 (B). All bands and smears tended to be more strongly stained with pS396 rather than T46. Approximately 68-and 64-kDa bands are seen in all AGD (lanes 1-3, 10-12), PSP (lanes 4-7, 13-16) and CBD cases (lanes 8 and 17). Cases of PSP (lanes 4-6, 13-15) and CBD (lanes 8 and 17) show 33-and 37-kDa mass tau fragments, respectively. One of two AGD cases showed less intensely stained 33-to 37-kDa tau fragments in the putamen and caudate nucleus (lanes 1, 2, 10 and 11). In the other AGD case, the immunoreactivity of 33-to 37-kDa tau fragments in the putamen was very weak (lanes 3 and 12). On tau immunohistochemistry in corresponding regions of the left hemisphere, the density of GFAs were comparable between AGD and PSP cases (Table of Figure 8B). However, TAs tended to be stained with a larger number of anti-tau antibodies in PSP cases than AGD cases. Further, in these cases, GFAs and TAs stained with anti-p62 antibodies and/or Gallyas method were noted only in PSP cases. Abbreviations: AGD = argyrophilic grain disease; AD = Alzheimer's disease; Caud = caudate nucleus; CBD = corticobasal degeneration; F = frontal cortex; PSP = progressive supranuclear palsy; Put = putamen; n.e. = Gallyas-positive GFAs were not evaluated in PSP cases because many Gallyas-positive dot-like structures were scattered, and thereby made it difficult to count Gallyas-positive GFAs; T = temporal cortex.