A gray matter volume network in pathologically confirmed cases of argyrophilic grain disease: An exploratory analysis

Abstract

Background

Argyrophilic grain disease (AGD) is a common yet underrecognized tauopathy that often mimics Alzheimer's disease (AD) in clinical and imaging presentations. While regional atrophy in AGD has been reported on magnetic resonance imaging (MRI), network-level structural changes remain poorly understood.

Objective

We aimed to explore a gray matter volume network related to AGD.

Methods

Structural MRI data were collected from 12 patients with pathologically confirmed AGD (age at MRI, 87.7 ± 5.5 years; male, 4), 12 patients with pathologically confirmed AD (83.4 ± 10.0 years; male, 4), and 9 healthy controls (HCs; 82.4 ± 1.9 years; male, 2) at Fukushimura Hospital in Japan. Scaled Subprofile Model with principal component analysis was applied to preprocessed gray matter volume data of AGD and HCs to identify an AGD-related network.

Results

An AGD-related network involving relative reduction in the ambient gyrus, entorhinal cortex, hippocampus, amygdala, and thalamus was identified. Represented by principal components 1, 2, and 3, this network showed significantly higher expression in patients with AGD than HCs (p < 0.0001, permutation test). The expression of the network was also higher in patients with AD than HCs (p < 0.0001, t-test).

Conclusions

This exploratory study identified a gray matter volume network related to AGD, providing a basis for future research of network-based imaging approaches.

Keywords

Alzheimer's disease argyrophilic grain disease dementia gray matter magnetic resonance imaging principal component analysis

Introduction

Argyrophilic grain disease (AGD) is an age-related tauopathy characterized by argyrophilic grain formations and gray matter atrophy, primarily in the limbic and temporal regions.^1,2 Although its neuropathological criteria and staging are well-established,² AGD often remains underrecognized in clinical settings due to symptomatic and neuroimaging overlap with other dementias, such as Alzheimer's disease (AD), for which disease-specific therapies have become available. Identification of AGD in patients with cognitive impairment or dementia is therefore essential for appropriate clinical management. These considerations highlight the need for neuroimaging markers to aid in the recognition of AGD.

Previous MRI studies have identified characteristic regional atrophy in AGD, particularly within the ambient gyrus, entorhinal cortex, amygdala, and hippocampus.^3,4 These studies have advanced our understanding of AGD-associated structural changes. However, most investigations have been limited to regional analyses, leaving whole-brain network-level assessments unexplored.

Network-based imaging approaches, such as the Scaled Subprofile Model (SSM), offer a systems-level perspective to identify disease-specific spatial patterns. SSM has demonstrated success in identifying disease-related metabolic or functional networks in other neurodegenerative conditions.⁵ Furthermore, its application to structural MRI data has revealed a gray matter volume network in AD.⁶ Applying this method to AGD could offer the potential to uncover disease-related networks and facilitate clinical diagnosis. This study aimed to explore a gray matter volume network related to AGD using structural MRI data from pathologically confirmed AGD cases.

Methods

Participants

Twelve patients with AGD, 12 patients with AD, and nine age- and sex-matched healthy controls (HCs) who were recruited from the Fukushimura Brain Dock were enrolled at Fukushimura Hospital, Toyohashi, Japan (Table 1). This study was approved by the Ethics Committees of Fukushimura Hospital and Tokushima University Hospital. Postmortem consent was obtained from the bereaved of all patients with AGD and AD to use the data for academic purposes. This study was carried out in accordance with the Declaration of Helsinki.

Table 1.

Demographic and neuropsychological data.

	AGD	AD	HC
Number of cases	12	12	9
Male/Female	4/8	4/8	2/7
Age at MRI (y)	87.7 ± 5.5	83.4 ± 10.0	82.4 ± 1.9
MMSE	15.5 ± 7.0	9.8 ± 7.0	N.A.
Neuropathological features
Brain weight (g)	1108.1 ± 113.4	1108.6 ± 122.6	N.A.
AGD (Saito) I/II/III	0/2/10	0/0/0	N.A.
NFT (Braak) I–VI	II, 11; III, 1	IV, 4; V, 1; VI, 7	N.A.
CERAD score 0/A/B/C	3/4/4/1	0/0/2/10	N.A.
TDP-43P (Nelson) 0/1/2/3	8/0/4/0	8/0/4/0	N.A.

Data are presented as absolute number or mean ± standard deviation. CERAD: Consortium to Establish a Registry for Alzheimer's Disease; HC: healthy controls; MMSE: Mini-Mental State Examination; MRI: magnetic resonance imaging; N.A.: not applicable; NFT: neurofibrillary tangle; TDP-43P: TAR DNA-binding protein-43 kDa pathology.

Pathological analysis

We stained all 5-µm thick sections from the 24 formalin-fixed one-sided cerebral hemispheres with conventional hematoxylin and eosin, Klüver–Barrera, and Gallyas–Braak methods. We also performed immunohistochemical staining using antibodies against amyloid-β (monoclonal, clone 4G8, 1:10,000; Signet Pathology Systems, York, United Kingdom), phosphorylated tau (monoclonal, clone AT8, 1:1000; Innogenetics, Ghent, Belgium), phosphorylated TDP-43 (polyclonal, pS409/410, 1:3000; Cosmo Bio Co., Ltd, Tokyo, Japan), and phosphorylated α-synuclein (monoclonal, clone pSyn#64, 1:10,000; Fujifilm Wako, Osaka, Japan) in selected sections. The analysis confirmed that the 12 cases of AGD had neuropathological findings of AGD and lacked significant neuropathological changes of AD, and the 12 cases of AD had neuropathological findings of AD and lacked significant neuropathological changes of AGD, as determined by internationally accepted neuropathological criteria.^2,7,8

Image acquisition

The participants underwent MRI studies using a 1.5 Tesla scanner MAGNETOM Aera (Siemens Healthcare, Erlangen, Germany) equipped with a 20-channel head/neck coil. We obtained three-dimensional T1-weighted imaging with the magnetization-prepared rapid gradient-echo (MP-RAGE) sequence and the following parameters: repetition time, 1700 ms; echo time, 4.2 ms; inversion time, 800 ms; flip angle, 19°; field of view, 230 mm; matrix, 256 × 256; 1.25-mm-thick gapless sections; voxel size, 0.898 × 0.898 × 1.25 mm³.^4,9

Preprocessing

We processed the structural images using SPM12 in conjunction with Computational Anatomy Toolbox 12¹⁰ run on MATLAB R2021a. All three-dimensional T1-weighted images were segmented, normalized into the Montreal Neurological Institute space, and modulated, as described previously.⁴ We then made a gray matter mask by applying the Masking toolbox¹¹ to the gray matter images of the 33 participants, as in the previous study.⁶ Finally, all images were smoothed with a Gaussian kernel of 8 × 8 × 8 mm³ full width at half maximum.

Network analysis

We performed the SSM using the Generalized Covariance Analysis version 1.2 (Columbia University Medical Center, New York) run on MATLAB R2019a and SPM12. We used the preprocessed gray matter images of the 12 AGD cases and 9 HCs to derive an AGD-related network. The mean effects were subtracted from both the rows and columns of the subject × voxel data matrix. Principal component analysis (PCA) produced a set of 20 principal components (PCs) along with the fraction of the variance by each PC. This operation also generated subject scores reflecting the degree to which a subject expressed the associated topography in each PC. A subset of the top 5 PCs, corresponding to a cumulative variance of 50.5%, was examined to obtain a relevant topography.⁵ A set of PCs that yielded the lowest value in Akaike's information criterion was selected.¹² The selected PCs were then combined into a single vector. The optimal coefficients for the linear combination of the selected PCs were determined by logistic regression of the corresponding subject scores using JMP Pro 18 (JMP Statistical Discovery LLC, Cary, North Carolina). A produced covariance pattern was transformed to a z-score map standardized by the standard deviation. The subject scores were computed using the topographic profile rating, and differences in the subject scores between the groups were evaluated using permutation test (RStudio).

The voxels with Z ≥ 1.96 were used for determining the brain regions involved in the network. The reliability of the voxel weights that significantly differed from zero was estimated using a bootstrap algorithm. The PCA procedures were repeated for 1000 iterations by resampling the original sets of gray matter images with replacements, and a voxel map of the inverse coefficient variation (ICV) was produced. The significance level for voxel weight reliability was set at an ICV threshold of Z = 1.96 corresponding to p < 0.05.¹³

Results

A gray matter volume network distinguished patients with AGD from HCs. Using SSM on the sample of 12 patients with AGD and 9 HCs, we identified a pattern that showed significantly higher subject expressions in patients with AGD than HCs (p < 0.0001, permutation test; Figure 1). This pattern was represented by a linear combination of PCs 1, 2, and 3, with coefficients of 0.615, −0.544, and −0.571, respectively, which accounted for 13.7% of the overall subject × voxel variance. The network, termed the AGD-related gray matter volume network, was characterized by relative reductions in regions such as the ambient gyrus, entorhinal cortex, cornu ammonis, dentate gyrus, subiculum, amygdala, and thalamus (Table 2, Figure 2). On the other hand, the expression of the AGD-related gray matter volume network was also significantly higher in patients with AD than HCs (p < 0.0001, t-test).

Figure 1.

Expression of the argyrophilic grain disease (AGD)-related gray matter volume network. Subject scores reflecting the degree of AGD-related gray matter volume network expression were significantly higher in patients with AGD than in healthy controls (HCs) (p < 0.0001, permutation test). Error bars indicate mean ± standard deviation.

Figure 2.

Spatial distribution of the AGD-related gray matter volume network. The AGD-related gray matter volume network showed significant relative reductions (Z > 1.96, corresponding to p < 0.05) in the ambient gyrus, entorhinal cortex, cornu ammonis, dentate gyrus, subiculum, amygdala, and thalamus.

Table 2.

Brain regions involved in the AGD-related gray matter volume network.

Brain regions	Peak MNI coordinates (mm)			Voxels	Z max
Decrease
Ambient gyrus/Entorhinal cortex/Cornu ammonis/Dentate gyrus/Subiculum/Amygdala L/Thalamus LR	−27	−10.5	−18	8595	5.86
Ambient gyrus/Entorhinal cortex/Cornu ammonis/Dentate gyrus/Subiculum/Amygdala R	24	0	−43.5	6317	5.54
Superior/middle temporal gyrus, posterior division L	−54	−36	1.5	399	2.68
Middle temporal gyrus, temporooccipital part R	52.5	−57	7.5	319	2.87
Subcallosal cortex/Paracingulate gyrus LR	0	31.5	−10.5	265	3.05
Frontal operculum cortex R	43.5	22.5	3	258	2.78
Central opercular cortex L	−49.5	−19.5	13.5	237	3.05
Intracalcarine cortex L	−22.5	−66	6	234	3.48
Inferior frontal gyrus, pars opercularis R	40.5	16.5	27	183	2.75
Lateral occipital cortex, inferior division L	−45	−72	3	151	2.39
Increase
Lateral occipital cortex, superior division R	28.5	−66	37.5	1073	3.15
Caudate R	18	25.5	6	585	3.18
Crus I R	15	−87	−19.5	440	2.57
Callosal body LR	0	19.5	7.5	273	5.69
Crus II L	−9	−84	−40.5	204	2.51
Caudate L	−10.5	4.5	7.5	181	3.09
Cingulate gyrus, posterior division LR	0	−51	13.5	145	2.53
Caudate R	18	−7.5	22.5	142	2.47
VI L	−31.5	−36	−34.5	130	2.42
Cuneal cortex L	−12	−85.5	31.5	129	2.27
Supramarginal gyrus, posterior division	−54	−42	54	106	3.01

Brain region labels were determined using the Harvard–Oxford cortical structural atlas, Jülich histological atlas,¹⁴ and cerebellar atlas in FSLeyes (FSL version 6.0.7.10).¹⁵ Clusters with >100 voxels and Z-scores >1.96 (corresponding to P < 0.05) were reported. Bootstrap estimation (1000 iterations) demonstrated the reliability of the voxel weights (|inverse coefficient of variation| > 1.96). AGD: argyrophilic grain disease; L: left; MNI: Montreal Neurological Institute; R: right.

Discussion

We explored a gray matter volume network related to pathologically confirmed AGD. The network, derived from a small cohort of a single center, was characterized by relative reductions in the ambient gyrus, entorhinal cortex, hippocampus, amygdala, and thalamus. These findings are largely consistent with previous studies on region-level structural changes in AGD^3,4 and provide a network-level perspective on gray matter volume alterations in this disease. Methodologically, our results suggest that the SSM is useful for uncovering disease-related volume patterns that extend beyond conventional regional analyses. Validation in larger, independent cohorts is warranted to determine its clinical relevance.

The identified AGD-related network highlights reductions in regions most affected by tau pathology in AGD, including the ambient gyrus, entorhinal cortex, and hippocampus. The ambient gyrus, in particular, has been identified as the initial and most severely affected site of pathology.² Previous neuroimaging and neuropathology studies have demonstrated prominent atrophy and tau deposition in these limbic structures.^2–4 The amygdala, also significantly affected, reflects AGD's impact on regions involved in memory and emotional processing.¹⁶ These findings corroborate the notion that AGD pathology predominantly targets the limbic system.

In addition, thalamic involvement, identified as part of the AGD-related network, aligns with a recent report of reduced gray matter volume in the bilateral limbic lobes and thalami in AGD patients compared with healthy controls using voxel-based morphometry.⁴ However, significant neuropathological alterations in the thalamus have not been documented in AGD.² These findings warrant cautious interpretation. Whether these changes reflect secondary disruptions or specific structural involvement in AGD requires further investigation, such as longitudinal studies or multimodal imaging to explore their progression and functional implications.

Comparison of the present AGD-related network with a recently reported AD-related network underscores both commonalities and differences between the two conditions. The AD-related network⁶ is characterized by reductions in the amygdala and hippocampus as well as the inferior parietal lobule and precuneus. The involvement of these areas aligns with the hallmark cognitive deficits of AD, particularly in memory and visuospatial domains. In contrast, the AGD-related network's focus on limbic regions, with minimal parietal involvement, suggests a more localized pattern of structural degeneration.

Despite these advances, this study has several limitations. First, the sample size was small, and no independent validation cohort was included, which may limit the generalizability of the results and underscores the need for replication in larger, external cohorts. Second, although the inclusion of pathologically confirmed cases strengthens diagnostic accuracy, it may introduce selection bias. For example, the mean MMSE scores of 15.5 in AGD and 9.8 in AD were relatively low compared with those for potential eligibility for anti-amyloid-β therapies. This may reduce the generalizability of the present findings to early-stage patients. Third, the exclusive focus on gray matter volume limits the ability to detect other AGD-related changes, such as functional or molecular alterations.

In conclusion, this exploratory study found a gray matter volume network related to AGD, offering novel insights into its neuropathological basis. By leveraging network-based imaging approaches, these findings enhance our understanding of AGD and provide a foundation for future research aimed at improving clinical diagnosis and developing therapeutic strategies to mitigate the disease's impact.

Footnotes

Acknowledgements

ChatGPT 4o, a language model developed by OpenAI (), was used for language refinement of this manuscript.

ORCID iDs

Koji Fujita

Tomoyasu Matsubara

Masafumi Harada

Ethical considerations

This study was approved by the Ethics Committees of Fukushimura Hospital and Tokushima University Hospital.

Consent to participate

Informed consent was obtained using the opt-out method because of the retrospective and noninvasive nature of the study.

Consent for publication

Not applicable

Author contribution(s)

Koji Fujita: Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Software; Validation; Visualization; Writing – original draft; Writing – review & editing.

Daita Kaneda: Data curation; Formal analysis; Investigation; Resources; Validation; Writing – review & editing.

Keita Sakurai: Conceptualization; Data curation; Formal analysis; Investigation; Project administration; Resources; Software; Validation; Writing – original draft; Writing – review & editing.

Tomoyasu Matsubara: Investigation; Project administration; Writing – review & editing.

Masafumi Harada: Funding acquisition; Investigation; Resources; Writing – review & editing.

Yuishin Izumi: Funding acquisition; Investigation; Supervision; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Japan Society for the Promotion of Science (JSPS) Grant-in-Aid for Scientific Research (C) Grant Number JP24K15767 and Japan Agency for Medical Research and Development (AMED) under Grant Number JP18he1402002 and JP24dk0207068.

Declaration of conflicting interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Keita Sakurai is an Editorial Board Member of this journal but was not involved in the peer-review process of this article nor had access to any information regarding its peer-review.

The remaining authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data supporting the findings of this study are available on reasonable request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

References

Ferrer

Santpere

van Leeuwen

. Argyrophilic grain disease. Brain 2008; 131: 1416–1432.

Saito

Ruberu

Sawabe

, et al. Staging of argyrophilic grains: an age-associated tauopathy. J Neuropathol Exp Neurol 2004; 63: 911–918.

Sakurai

Tokumaru

Ikeda

, et al. Characteristic asymmetric limbic and anterior temporal atrophy in demented patients with pathologically confirmed argyrophilic grain disease. Neuroradiology 2019; 61: 1239–1249.

Sakurai

Kaneda

Morimoto

, et al. Voxel-based and surface-based morphometry analysis in patients with pathologically confirmed argyrophilic grain disease and Alzheimer's disease. J Alzheimers Dis 2023; 93: 379–387.

Spetsieris

Eidelberg

. Scaled subprofile modeling of resting state imaging data in Parkinson's disease: methodological issues. Neuroimage 2011; 54: 2899–2914.

Chuluunbat

Matsuda

Fujita

, et al. Identification and validation of a gray matter volume network in Alzheimer's disease. J Neurol Sci 2022; 440: 120344.

Montine

Phelps

Beach

, et al. National Institute on Aging-Alzheimer's Association guidelines for the neuropathologic assessment of Alzheimer's disease: a practical approach. Acta Neuropathol 2012; 123: 1–11.

Mirra

Heyman

McKeel

, et al. The Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Part II. Standardization of the neuropathologic assessment of Alzheimer's disease. Neurology 1991; 41: 479–486.

Sakurai

Kaneda

Morimoto

, et al. Diverse limbic comorbidities cause limbic and temporal atrophy in Lewy body disease. Parkinsonism Relat Disord 2022; 105: 52–57.

10.

Gaser

Dahnke

Thompson

, et al. CAT: a computational anatomy toolbox for the analysis of structural MRI data. Gigascience 2024; 13: giae049.

11.

Ridgway

Omar

Ourselin

, et al. Issues with threshold masking in voxel-based morphometry of atrophied brains. Neuroimage 2009; 44: 99–111.

12.

Habeck

Foster

Perneczky

, et al. Multivariate and univariate neuroimaging biomarkers of Alzheimer's disease. Neuroimage 2008; 40: 1503–1515.

13.

Habeck

Stern

. Alzheimer’s Disease Neuroimaging Initiative. Multivariate data analysis for neuroimaging data: overview and application to Alzheimer's disease. Cell Biochem Biophys 2010; 58: 53–67.

14.

Amunts

Mohlberg

Bludau

, et al. Julich-Brain: a 3D probabilistic atlas of the human brain's cytoarchitecture. Science 2020; 369: 988–992.

15.

Jenkinson

Beckmann

Behrens

, et al. FSL. Neuroimage 2012; 62: 782–790.

16.

Tolnay

Clavaguera

. Argyrophilic grain disease: a late-onset dementia with distinctive features among tauopathies. Neuropathol Appl Neurobiol 2004; 30: 185–196.