Thalamic volume alterations mediate glymphatic function and cognitive dysfunction in Alzheimer's disease

Abstract

INTRODUCTION

Glymphatic dysfunction is implicated in Alzheimer's disease (AD), but its mechanism remains unclear. This study investigated the associations of the analysis along the perivascular space (ALPS) index, brain reserve, and cognitive outcomes across the AD continuum.

METHODS

This study enrolled two cohorts from the local hospital (n = 95) and the Alzheimer's Disease Neuroimaging Initiative (ADNI; n = 178). We calculated the diffusion tensor image ALPS to assess the whole‐brain glymphatic function and evaluated its associations with cognition and brain structure.

RESULTS

A higher ALPS index was positively associated with better executive, memory, and language abilities (p < 0.05). A lower ALPS index correlated with thalamic atrophy, involving the anterior, lateral, ventral, and midline thalamic nuclei. Thalamic atrophy mediated the link between glymphatic dysfunction and worse cognition via structural association of full or part mediation.

CONCLUSIONS

Glymphatic system deterioration correlates with cognitive decline, potentially mediated by thalamic atrophy. The thalamic nuclei may be critical in this pathway.

Highlights

• A lower analysis along the perivascular space (ALPS) index could indicate poorer ability in the executive, language, and memory domains among patients with Alzheimer's disease or mild cognitive impairment.

• The thalamus is a key locus of vulnerability in glymphatic failure.

• The thalamus is involved in cognitive regulation. The glymphatic system's effect on cognitive outcomes may be via the mediation of gray matter volume of the thalamic nuclei.

1. INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative disorder characterized by extracellular amyloid beta (Aβ) plaques and neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau protein. ¹ The abnormal accumulation of Aβ and tau proteins disrupts neuronal activity and neural circuit function, ultimately resulting in significant cognitive dysfunction, particularly in learning and memory. The glymphatic system is a newly identified pathway of the perivascular fluid transport system that directs the clearance of metabolic waste in the brain. ² Dysfunction of the glymphatic system leads to the accumulation of neurotoxic substances. Emerging evidence suggests that the glymphatic system plays a pivotal role in contributing to the deposition of pathological proteins of AD. ³ , ⁴ Impaired brain clearance mechanisms may be the final common pathway to dementia. ⁵

In 2017, Taoka et al. developed a non‐invasive, image‐based measure called diffusion tensor imaging (DTI) analysis along the perivascular space (ALPS), which assesses perivascular clearance activity in the human brain using diffusion magnetic resonance imaging (MRI). ⁶ The ALPS index has been proposed to be an indirect indicator that may reflect some aspects of glymphatic function, partly based on an early study showing a correlation with glymphatic clearance function which is evaluated by the classical detection method (calculated on MRI after the intrathecal administration of gadolinium). ⁷ The ALPS has been studied in the field of AD and other various neurological disorders. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² A previous study has revealed that the ALPS index was positively associated with cognitive performance. ¹³ Another study found that the ALPS index was negatively associated with Aβ or tau deposition in cerebrospinal fluid (CSF) and positron emission tomography (PET) images. ¹² The ALPS index was also demonstrated to predict clinical progression and cognitive decline in patients with AD. ¹⁴

In addition to cognitive decline, another typical clinical manifestation of AD is cerebral atrophy. ¹⁵ AD is characterized by gray matter (GM) atrophy in the insula, middle frontal gyrus, and medial temporal lobes. ¹⁶ Patterns of cortical atrophy have even been shown to accurately track the disease progression of AD. ¹⁵ While both glymphatic dysfunction and brain atrophy participate in AD progression, their relationship has not been fully elucidated despite advances in relevant studies. This study used the ALPS index to investigate the potential associations between glymphatic activity and the clinical features of AD. Previous studies have shown that a lower ALPS could predict GM atrophy in AD signature regions and even cognitive decline. ¹⁴ Therefore, in this study we hypothesize that a lower ALPS index is associated with decreased cognitive function and lower gray matter volume (GMV), and that the glymphatic system causes cognitive impairment by leading to cerebral atrophy in AD.

2. METHODS

2.1. Participants

The data used in this study were from two independent cohorts, the hospital cohort and the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. All participants (n = 95) in the hospital cohort were recruited from the department of neurology in the China–Japan Friendship Hospital. The patients were diagnosed by two experienced neurologists according to clinical criteria based on comprehensive assessments. AD was clinically diagnosed using the 2011 National Institute on Aging‐Alzheimer's Association (NIA‐AA) criteria. ¹⁷ Mild cognitive impairment (MCI) was defined as the following: (1) concern about a cognitive decline compared to the previous status; (2) decline in at least one cognitive domain after age and education adjustment; (3) maintenance of independent function in daily life activities; and (4) not meeting the diagnostic criteria for dementia. ¹⁸ Cognitively normal (CN) controls showed normal neuropsychological test results without subjective cognitive complaints during the interview. The inclusion criteria for the hospital cohort were as follows: (1) aged 55 to 85 years, (2) right handed, and (3) native Mandarin speakers. Subjects were excluded if they had other nervous system diseases (e.g., Parkinson's disease, epilepsy, or head trauma), severe systemic disorders, psychiatric disorders, and contraindications for MRI. This research protocol was approved by the China–Japan Friendship Hospital ethics committee and institutions (Ethics ID: 2025‐KY‐158), and informed consent was obtained from all participants.

Other data were obtained from the ADNI database (adni.loni.usc.edu) on June 6, 2024. The ADNI cohort participants were selected from the ADNI‐2 and ADNI‐3, comprising 183 subjects (CN, MCI, or AD). The inclusion and exclusion criteria are available at http://www.adni‐info.org. ADNI participants were included if they possessed available DTI, T1‐weighted imaging (T1WI), and complete necessary clinical information (such as age, sex, education, apolipoprotein E [APOE] ε4 genotype, and medical history). Finally, a total of 52 patients with AD, 59 patients with MCI, and 67 CN controls were included in the ADNI cohort study. The ADNI study received ethical approval from the local institutional review boards of all participating centers. Written informed consent was obtained from all participants or their family members.

2.2. MRI data collection and processing

The subjects in the hospital cohort underwent multimodel MRI via a 3.0T MR scanner (GE). High‐resolution T1‐weighted images were acquired following a sagittal 3D magnetization‐prepared rapid gradient echo protocol and covered the entire brain. The parameters of the T1WI were as follows: slice thickness = 1.0 mm, repetition time = 6.7 ms, echo time = Min Full, matrix = 256 mm x 256 mm, field of view = 256 mm x 256 mm, number of excitations = 1, and DTI: maximum b‐value = 1000 s/mm², 64 diff. directions, repetition time = 8084 ms, slice thickness = 2 mm, slice spacing = 0 mm, matrix size = 120 × 120, field of view = 240 mm x 240 mm, axial slices = 65. DTI and T1WI data were acquired for each ADNI participant via a 3.0‐T scanner. The imaging protocol can be accessed on the official ADNI website (https://adni.loni.usc.edu/help‐faqs/adni‐documentation/).

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a systematic literature search on PubMed and Web of Science for traditional articles, meeting abstracts, and presentations. Although there have been several recent publications describing the glymphatic function or relevant imaging indicators in Alzheimer's disease (AD), the mechanism by which decreased glymphatic function impairs cognitive function is poorly understood.

2. Interpretation: The thalamus is a key locus of vulnerability in glymphatic failure. Our findings reveal that damage to the thalamic nucleus is involved in the pathological process of cognitive impairment.

3. Future directions: The study proposes a new hypothesis that enhanced glymphatic function may be an exciting new target for AD treatment. Future studies should focus on investigating the potential of thalamic volume and microstructure as an early biomarker for glymphatic impairment and subsequent cognitive decline in the preclinical AD stage.

All T1WI data were preprocessed with the Computational Anatomy Toolbox (CAT12). Preprocessing was performed with the default settings in CAT12. The T1 images were first segmented into GM, white matter (WM), and CSF. Then all images underwent modulation, spatial normalization, quality check, and homogeneity check. Details are listed in the supporting information. Finally, the normalized images were smoothed with an 8 mm full width at half‐maximum (FWHM) Gaussian kernel with a resulting voxel size of 1.5 mm³. Total intracranial volume (TIV) was obtained using the CAT 12 estimating TIV function. Higher GMV ratios represented larger brain volume.

2.3. ALPS index calculation

The ALPS index was calculated using the FMRIB Software Library (FSL) v6.0. ⁷ After eddy current and head motion correction, preprocessed DTI data were used to generate diffusivity maps and fractional anisotropy (FA) maps. Individual FA maps were co‐registered to the JHU‐ICBM‐FA‐2 mm template. Four 5 mm diameter regions of interest (ROIs) were placed by two neurologists at predetermined Montreal Neurological Institute (MNI) coordinates ([−27, −16, 29], [−39, −16, 29], [27, −16, 30], [39, −16, 30]) with double‐blind verification. ¹⁹ The average of the left and right ALPS indices was used for analyses. A higher value indicates greater diffusivity. Detailed procedures for ALPS index calculation are provided in the supporting information.

2.4. Cognitive assessment

ADNI developed four composite measures representing four specific cognitive domains: memory (ADNI_MEM), language (ADNI_LAN), visuospatial ability (ADNI_VS), and executive function (ADNI_EF). ²⁰ , ²¹ A higher score indicates better cognitive function and a lower score indicates more impaired cognition.

Each hospital subject underwent a battery of neuropsychological assessments, including the Rey Auditory Verbal Learning Test immediate recall (RAVLT‐I), RAVLT delayed recall (RAVLT‐D), Digit Span Test (DST) Backward, DST Forward, Trail Making Test Part A (TMTA), Trail Making Test Part B (TMTB), Stroop Color and Word Test (SCWT), Symbol Digit Modalities Test (SDMT), Boston Naming Test (BNT), Verbal Fluency Test (VFT), Clock Drawing Test (CDT), and Rey–Osterrieth Complex Figure Test (ROCF). The cognitive test scores were time matched to the closest MRI examination with a 6‐month maximum difference cutoff.

2.5. Confirmatory factor analysis

We referred to the method of ADNI data to establish the score of each cognitive domain. ²⁰ , ²¹ As recommended by Laura E. Gibbons, we constructed a model using confirmatory factor analysis (CFA) for the hospital cohort (Figure 1), with raw data transformed into ordered categorical scores. ²² Methods for reordered scores can be found in Table S1 in supporting information. The CFA model included four latent factors: executive function (EF), language function (LAN), visuospatial ability (VS), and memory function (MEM). The executive function domain was measured by the SDMT, SCWT, TMTA, TMTB, and DST Backward. The language domain comprised category‐based VFT and the BNT. Visuospatial ability was assessed using the ROCF and CDT. The memory domain was measured by each RAVLT score. Each latent factor yields a composite value that represents its cognitive domain score. Maximum‐likelihood estimation was used for model estimation via the R package lavaan. ²³ We computed the confirmatory fit index (CFI), the Tucker–Lewis Index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR) to assess the fit quality of a model. We considered a model an acceptable fit when it respected the following thresholds: CFI ≥ 0.90; TLI ≥ 0.90; RMSEA ≤0.10; SRMR ≤0.08. ²⁴ , ²⁵

FIGURE 1.

Model of cognitive composite score for each domain. Numbers on the single‐headed arrows from each domain to each test are standardized factor loadings, indicating the strength of association between the latent domain and the observed test. Left column numbers are residuals of each test, representing the proportion of variance in each test not explained by the corresponding latent domain. Numbers on the double‐headed arrows between domains are correlations among latent variables. BNT, Boston Naming Test; CDT, Clock Drawing Test; DSTback, Digit Span Test Backward; N1N3, the total score of round 1 to 3 in the learning tasks of the Rey Auditory Verbal Learning Test; N4, Rey Auditory Verbal Learning Test immediate recall; N5, Rey Auditory Verbal Learning Test delayed recall (30 minutes); ROCFcopy, Rey–Osterrieth Complex Figure Test copy; SCWT‐C, Stroop Color and Word Test (part C); SDMT, Symbol Digit Modalities Test; TMTA, Trail Making Test Part A; TMTB, Trail Making Test Part B; VFT, Verbal Fluency Test.

2.6. Statistical analysis

Given all continuous variables were non‐normal (tested by Q‐Q plots), they were reported as median (interquartile range [IQR]); categorical variables were summarized as frequencies (%). Group comparisons used Wilcoxon rank‐sum or Kruskal‐Wallis tests. Multivariate linear regression analysis was used to evaluate the correlation between ALPS and cognitive function, adjusting for confounding variables such as age, sex, education, and APOE ε4. Statistical significance was set at p < 0.05 after false discovery rate (FDR) correction. All analyses were performed using R software version 4.2.1.

Voxel‐based morphometry (VBM) was conducted using the data processing and analysis for brain imaging (DPABI) ²⁶ software in MATLAB R2017b. The correlation of ALPS with GMV was assessed after controlling for age, sex, education year, and TIV. When analyzing the ADNI data, we included the site number as the “fixed effect” to remove the unwanted variation introduced by different sites. FDR correction was performed for multiple comparison correction with a threshold of voxel p < 0.05. The regional volumes were quantified according to the automated anatomical atlas 3 (AAL3), which adds 15 subdivision parts of the thalamus. ²⁷

Mediation analyses were carried out to explore whether injured GM mediates the relationship between ALPS and cognitive performance. In this study we hypothesized that ALPS decreases before brain atrophy and then causes cognitive dysfunction. First, we extracted the average GM value of each ALPS‐related region in the VBM analysis. We subsequently constructed simple mediation models to specify the mediator role of each region. ALPS was included as an independent variable, and the composite score of each cognitive domain was included as a dependent variable. Possible confounders, such as age, sex, education, and APOE ε4 genotype, were set as covariates. A 95% bootstrap confidence interval based on 5000 bootstrap replicates was used to estimate significance. The FDR correction method was used for multiple comparison correction. ²⁸ All mediation analyses were performed using the R mediation package.

3. RESULTS

3.1. Participant characteristics

Table 1 presents the demographic and clinical characteristics of all participants from the ADNI cohort and the hospital cohort. No significant difference was observed in age (p = 0.264), sex distribution (p = 0.161), or APOE ε4 status (p = 0.814) between two groups. In the hospital cohort, there were more patients with AD (p = 0.001), and more subjects with fewer years of education (p < 0.001). Therefore, it is reasonable that hospital patients have lower Mini‐Mental State Examination scores (p < 0.001).

TABLE 1.

Baseline characteristics for the ADNI and hospital cohorts.

	Cohort
Characteristic	ADNI, N = 178 1	Hospital, N = 95 1	p value
Age			0.264 2
Median (IQR)	72 (68, 77)	72 (66, 77)
Sex			0.161 3
Female	76 (42.7%)	49 (51.6%)
Male	102 (57.3%)	46 (48.4%)
Education year			<0.001 2
Median (IQR)	16.0 (14.0, 18.0)	12.0 (9.0, 16.0)
APOE ε4			0.814 3
0	100 (56.2%)	56 (60.2%)
1	65 (36.5%)	31 (33.3%)
2	13 (7.3%)	6 (6.5%)
Cognitive status			0.001 3
CN	67 (37.6%)	16 (16.8%)
MCI	59 (33.1%)	38 (40.0%)
AD	52 (29.2%)	41 (43.2%)
BMI			<0.001 2
Median (IQR)	27.0 (24.6, 29.8)	23.5 (21.3, 24.9)
MMSE			<0.001 2
Median (IQR)	28.0 (25.0, 30.0)	24.0 (19.0, 26.5)
Smoke	70 (41.2%)	20 (21.1%)	<0.001 3
Drink	9 (5.3%)	45 (47.4%)	<0.001 3
Heart attack	126 (72.4%)	0 (0.0%)	<0.001 3
Diabetes	98 (55.4%)	37 (39.8%)	0.015 3
ALPS			0.480 2
Median (IQR)	1.25 (1.14, 1.39)	1.29 (1.18, 1.38)

Abbreviations: AD, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; ALPS, diffusion tensor image analysis along the perivascular space; APOE, apolipoprotein E; BMI, body mass index; IQR, interquartile range; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; CN, cognitively normal.

¹ n (%).

² Wilcoxon rank sum test.

³ Pearson chi‐squared test.

3.2. ALPS and cognition

We further explored the relationships between ALPS and cognitive function. Table 2 summarizes the influence of ALPS on each cognitive domain. ALPS was significantly associated with lower executive function (β = 1.20, p = 0.003), language (β = 0.83, p = 0.044), and memory (β = 0.85, p = 0.048) abilities in the ADNI population. We proceeded to identify the influence of ALPS on cognition in the hospital cohort. The performance of each cognitive domain was converted into a composite score using the CFA. The final model for the composite score (Figure 1) demonstrated acceptable fit (CFI = 0.931, RMSEA = 0.091, TLI = 0.912, SRMR = 0.072). Table S2 in the supporting information presents the factor loadings and standard errors of each item in the CFA model. ALPS was still significantly associated with lower cognition in the executive function (β = 3.36, p = 0.048), memory (β = 4.17, p = 0.037), and language (β = 4.83, p = 0.028) domains. In addition, higher ALPS was associated with better visuospatial ability in the hospital cohort (β = 2.90, p = 0.042). At different disease stages (CN, MCI, and AD), no correlation was found between the ALPS index and each cognitive domain.

TABLE 2.

Multiple linear regression showing the relationship between the ALPS index and cognition.

ADNI					Hospital
	β	SE	t	p		β	SE	t	p
ADNI					Hospital
ADNI_EF	1.20	0.40	3.00	0.003	Executive function	3.59	1.65	2.18	0.032
ADNI_LAN	0.83	0.41	2.03	0.044	Language	5.44	2.15	2.53	0.013
ADNI_MEM	0.85	0.43	1.99	0.048	Memory	4.17	1.97	2.12	0.037
ADNI_VS	0.38	0.33	1.15	0.253	Visuospatial	2.90	1.41	2.06	0.042
ADNI‐CN					Hospital‐CN
ADNI_EF	0.28	0.47	0.60	0.552	Executive function	−2.22	6.49	−0.34	0.739
ADNI_LAN	0.16	0.51	0.32	0.753	Language	−3.76	10.23	−0.37	0.720
ADNI_MEM	−0.51	0.44	−1.18	0.242	Memory	−2.46	8.76	−0.28	0.785
ADNI_VS	0.18	0.43	0.42	0.674	Visuospatial	−9.31	5.78	−1.61	0.138
ADNI‐MCI					Hospital‐MCI
ADNI_EF	0.15	0.42	0.35	0.727	Executive function	3.42	2.54	1.35	0.187
ADNI_LAN	−0.60	0.44	−1.36	0.181	Language	3.28	3.26	1.01	0.323
ADNI_MEM	−0.46	0.43	−1.08	0.284	Memory	5.13	3.68	1.39	0.174
ADNI_VS	−0.14	0.54	−0.26	0.800	Visuospatial	1.92	1.89	1.02	0.318
ADNI‐AD					Hospital‐AD
ADNI_EF	1.46	0.71	2.06	0.045	Executive function	1.67	1.88	0.89	0.381
ADNI_LAN	1.45	0.78	1.85	0.071	Language	3.11	2.39	1.30	0.203
ADNI_MEM	0.99	0.50	1.97	0.056	Memory	0.91	2.00	0.46	0.652
ADNI_VS	0.43	0.71	0.61	0.548	Visuospatial	2.56	2.21	1.16	0.255

Abbreviations: ADNI, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; EF, executive function; LAN, language; MCI, mild cognitive impairment; MEM, memory; CN, cognitively normal; SE, standard error; VS, visuospatial function.

3.3. ALPS‐related GM clusters

We explored GM structures associated with the ALPS index in the ADNI and hospital group separately, adjusting for age, sex, education year, APOE ε4 status, and TIV (Figure 2). Then we overlapped the GM maps of the ADNI population and the hospital population. The topography revealed that both figures shared overlapping clusters in thalamus, Rolandic operculum, supramarginal gyrus, Heschl gyrus, superior temporal gyrus, and middle temporal gyrus. Among these regions, the thalamus has the largest cluster. The results of the thalamic clusters were situated in the anterior (Thal_AV), lateral (Thal_LP), ventral (Thal_VL), and midline (Thal_MDm, Thal_MDl) thalamic nuclei, not including the posterior group nucleus. The correlation with ALPS was stronger with right thalamic volume than with left. We first examined whether this lateralization could be explained by side differences in ALPS itself by performing separate correlation analyses of left or right ALPS with bilateral thalamic volumes. However, no consistent lateralization pattern was observed in either cohort (Figure S1 in supporting information). We further considered intrinsic volumetric differences of thalamic subnuclei as a potential explanation. GMV comparisons of left and right thalamic nuclei revealed that the lateral posterior thalamic nucleus (Thal_LP) and anteroventral thalamic nucleu (Thal_AV) were significantly larger on the right side than the left in both cohorts (Figures S2 and S3 in supporting information).

FIGURE 2.

A, Gray matter regions correlated with the ALPS index in the ADNI patients (pink) p < 0.05 correlated with multiple comparisons using FDR. B, Gray matter regions correlated with the ALPS index in the hospital patients with a statistical threshold of pfdr < 0.05 and a cluster size >100 voxels (blue). C, Regions showing overlap in (A) and (B). The color bar represents the t statistic. D, A ROI illustration of the thalamic nuclei showing their spatial distribution. ADNI, Alzheimer's Disease Neuroimaging Initiative; ALPS, diffusion tensor image analysis along the perivascular space; FDR, false discovery rate; ltAV, left anteroventral thalamic nucleus; ltLP, left lateral posterior thalamic nucleus; ltMDl, left mediodorsal lateral parvocellular thalamic nucleus; ltMDm, left mediodorsal medial magnocellular thalamic nucleus; ltVL, left ventral lateral thalamic nucleus; ROI, region of interest; rtAV, right anteroventral thalamic nucleus; rtLP, right lateral posterior thalamic nucleus; rtMDl, right mediodorsal lateral parvocellular thalamic nucleus; rtMDm, right mediodorsal medial magnocellular thalamic nucleus; rtVL, right ventral lateral thalamic nucleus.

3.4. Significant mediation of thalamic nucleus in the ALPS index and cognition

After establishing significant correlations between the ALPS index and cognitive outcomes, we further examined whether these relationships were mediated by ALPS‐related clusters. Given that the thalamus constituted the largest significant cluster, mediation analysis was conducted with the ALPS index as the predictor, the GMV of each thalamic nucleus as the mediator, and domain‐specific cognitive scores as the outcome. Table 3 presents mediating effects of each thalamic nucleus reported by cohort and cognitive domain. The results from the ADNI cohort showed that executive ability was affected by the ALPS index via the structural association of full or partial mediation of predefined GM clusters. (This study uses a cross‐sectional design, and the mediation effect only reflects the structural association between variables and does not represent the confirmation of a causal pathway.) Most thalamic nuclei, such as the right lateral posterior nucleus (rtLP), left anteroventral nucleus (rtAV), bilateral ventral lateral nucleus (ltVL, rtVL), and left mediodorsal lateral parvocellular nucleus (ltMDI), partially mediated the association between ALPS and executive function (ADNI_EF; proportion of mediation:18.7% to 35.4%). Other nuclei, such as the bilateral mediodorsal medial magnocellular (rtMDm, ltMDm) and right mediodorsal lateral parvocellular (rtMDI) nuclei significantly mediated the path from ALPS to executive function (Figure 3A). In addition, some thalamic nuclei significantly mediated the relationships between ALPS and language ability or memory ability (Figures 3B–3C). However, only a small number of thalamic nuclei played a mediating role in the effect of ALPS on cognition, limited to language and memory domains (Figure 3D). We also extracted the volumes of other significant brain regions related to ALPS (Rolandic operculum, supramarginal gyrus, Heschl gyrus, superior temporal gyrus, and middle temporal gyrus) for mediation analysis, but no significant result was obtained.

TABLE 3.

Statistical results of the mediation effect for each thalamic nucleus by cohort and cognitive domain.

			Hospital cohort					ADNI cohort
	Nuclear	Indirect effect	Direct effect	Prop_ Mediated	p value	Nuclear	Indirect effect	Direct effect	Prop_ Mediated	p value
Executive function	ltLP	0.733	2.834	0.206	0.048	ltLP	0.240	0.902	0.210	0.004
	rtLP	0.899	2.668	0.252	0.024	rtLP	0.404	0.738	0.354	0.002
	ltAV	0.788	2.779	0.221	0.032	ltAV	0.268	0.874	0.235	0.006
	rtAV	0.765	2.803	0.214	0.030	rtAV	0.275	0.868	0.241	0.004
	ltVL	0.427	3.141	0.120	0.024	ltVL	0.269	0.874	0.235	0.006
	rtVL	1.023	2.544	0.287	0.038	rtVL	0.214	0.928	0.187	0.002
	ltMDm	0.664	2.904	0.186	0.038	ltMDm	0.320	0.823	0.280	0.004
	rtMDm	0.966	2.602	0.271	0.048	rtMDm	0.436	0.706	0.382	0.004
	ltMDl	0.517	3.051	0.145	0.026	ltMDl	0.290	0.853	0.254	0.006
	rtMDl	0.796	2.772	0.223	0.036	rtMDl	0.385	0.757	0.337	0.002
Language ability	ltLP	1.022	4.363	0.190	0.018	ltLP	0.374	0.378	0.497	0.046
	rtLP	1.463	3.922	0.272	0.016	rtLP	0.477	0.276	0.634	0.026
	ltAV	1.255	4.130	0.233	0.010	ltAV	0.380	0.372	0.505	0.030
	rtAV	0.870	4.516	0.161	0.010	rtAV	0.318	0.435	0.422	0.046
	ltVL	0.697	4.688	0.129	0.000	ltVL	0.436	0.317	0.579	0.066
	rtVL	1.274	4.112	0.236	0.012	rtVL	0.303	0.449	0.403	0.044
	ltMDm	1.103	4.283	0.205	0.010	ltMDm	0.480	0.272	0.639	0.048
	rtMDm	1.357	4.029	0.252	0.016	rtMDm	0.514	0.238	0.683	0.050
	ltMDl	0.710	4.675	0.132	0.014	ltMDl	0.437	0.315	0.581	0.042
	rtMDl	0.905	4.480	0.168	0.004	rtMDl	0.508	0.244	0.675	0.030
Memory ability	ltLP	1.290	2.851	0.312	0.020	ltLP	0.253	0.550	0.315	0.062
	rtLP	1.190	2.952	0.287	0.030	rtLP	0.365	0.438	0.455	0.064
	ltAV	1.252	2.889	0.302	0.042	ltAV	0.371	0.433	0.461	0.050
	rtAV	0.737	3.404	0.178	0.036	rtAV	0.366	0.438	0.455	0.050
	ltVL	0.679	3.462	0.164	0.048	ltVL	0.227	0.577	0.282	0.044
	rtVL	1.347	2.794	0.325	0.018	rtVL	0.146	0.657	0.182	0.048
	ltMDm	1.035	3.106	0.250	0.042	ltMDm	0.379	0.424	0.472	0.056
	rtMDm	1.246	2.895	0.301	0.034	rtMDm	0.405	0.399	0.504	0.048
	ltMDl	0.747	3.394	0.180	0.040	ltMDl	0.277	0.526	0.345	0.066
	rtMDl	1.043	3.099	0.252	0.026	rtMDl	0.282	0.521	0.351	0.068
Visuospatial ability	ltLP	0.189	2.718	0.065	0.032	ltLP	0.114	0.214	0.347	0.372
	rtLP	0.431	2.476	0.148	0.038	rtLP	0.221	0.107	0.674	0.354
	ltAV	0.058	2.849	0.020	0.022	ltAV	0.095	0.233	0.289	0.336
	rtAV	−0.030	2.937	−0.010	0.034	rtAV	0.158	0.170	0.481	0.324
	ltVL	0.125	2.782	0.043	0.030	ltVL	0.086	0.242	0.261	0.354
	rtVL	0.307	2.600	0.106	0.040	rtVL	0.071	0.257	0.217	0.304
	ltMDm	0.121	2.786	0.042	0.032	ltMDm	0.104	0.224	0.317	0.362
	rtMDm	0.357	2.550	0.123	0.026	rtMDm	0.196	0.132	0.597	0.346
	ltMDl	0.163	2.744	0.056	0.044	ltMDl	0.099	0.229	0.301	0.322
	rtMDl	0.393	2.515	0.135	0.026	rtMDl	0.159	0.169	0.485	0.352

Abbreviations: ltAV, left anteroventral thalamic nucleus; ltLP, left lateral posterior thalamic nucleus; ltMDl, left mediodorsal lateral parvocellular thalamic nucleus; ltMDm, left mediodorsal medial magnocellular thalamic nucleus; ltVL, left ventral lateral thalamic nucleus; MEM, memory; rtAV, right anteroventral thalamic nucleus; rtLP, right lateral posterior thalamic nucleus; rtMDl, right mediodorsal lateral parvocellular thalamic nucleus; rtMDm, right mediodorsal medial magnocellular thalamic nucleus; rtVL, right ventral lateral thalamic nucleus; Prop_Mediated: proportion of the mediation effect to the total effect.

FIGURE 3.

Simple mediation diagram for the ADNI subjects (A–C) and hospital subjects (D). ADNI, Alzheimer's Disease Neuroimaging Initiative; ALPS, diffusion tensor image analysis along the perivascular space; CI, confidence interval; DE, direct effect (average direct effect); EF, executive function; IE, indirect effect (average causal mediation effects); LAN, language; ltAV, left anteroventral thalamic nucleus; ltLP, left lateral posterior thalamic nucleus; ltMDl, left mediodorsal lateral parvocellular thalamic nucleus; ltMDm, left mediodorsal medial magnocellular thalamic nucleus; ltVL, left ventral lateral thalamic nucleus; MEM, memory; rtAV, right anteroventral thalamic nucleus; rtLP, right lateral posterior thalamic nucleus; rtMDl, right mediodorsal lateral parvocellular thalamic nucleus; rtMDm, right mediodorsal medial magnocellular thalamic nucleus; rtVL, right ventral lateral thalamic nucleus.

4. DISCUSSION

This study used the non‐invasive ALPS index to serve as a surrogate indicator for estimating the glymphatic function in patients with MCI and AD. The ALPS index is related to cognitive abilities across multiple cognitive domains. The observed lower ALPS index could indicate poorer ability in the executive, language, or memory domains. The thalamus cluster was the largest overlapping cluster in both the ADNI and hospital cohorts. The GMV of the thalamus might be vulnerable to glymphatic dysfunction. This research aligns with our previous hypothesis that the clinical significance of the glymphatic system on cognitive outcomes may be via the mediation of ALPS‐related GM regions, which are thalamic nuclei in this study.

4.1. Association between thalamic volume and glymphatic function

Our findings provide additional evidence about the relationship between the ALPS index and the GM of subcortical nuclei. A previous study demonstrated that a decreased ALPS index was partially associated with thalamus atrophy in patients with relapsing‐remitting multiple sclerosis. ²⁹ This study demonstrated that ALPS‐related thalamic damage remained in patients with AD. Located between WM and CSF, the thalamus is vulnerable to diverse pathological alterations. ³⁰ , ³¹ Impairment of glymphatic function may lead to inefficient clearance of pathological proteins, metabolic waste, or inflammatory cells. ³² The accumulation of above substances triggers neuroinflammation and tissue damage. ³⁰ The subsequent demyelination, oligodendrocyte reduction, and axonal loss cause brain atrophy. ²⁹

In this study, the anterior, ventral lateral, and mediodorsal thalamic nucleus were susceptible to glymphatic dysfunction, while no significant result was found in the posterior nuclei. This is closely related to the anatomical and functional characteristics of the thalamus. Glymphatic flux is significantly higher in the ventral and anterior part of the thalamus, which are adjacent to the third ventricle and large‐caliber penetrating arteries with well‐developed perivascular spaces. ³³ , ³⁴ In the thalamus, tau pathology initially accumulates in the anterior and ventral lateral nucleus, which are directly connected to the hippocampus and limbic cortices. ³⁵ In addition, NFTs and Aβ plaques are also deposited particularly in the anterior and midline thalamic nuclei. ³⁶ , ³⁷ Additionally, these susceptible nuclei are all high‐connectivity hub nodes within large‐scale functional networks. The anterior and mediodorsal thalamic nuclei are core components of the default mode network (DMN). ³⁸ Similarly, the ventral lateral nucleus serves as the central relay of the cortico–striatal–thalamic executive/motor circuit. These hub regions display elevated baseline neuronal activity and high metabolic clearance demand, which rely on proper glymphatic function.

Notably, the correlation between ALPS and thalamic volume exhibited right‐sided lateralization. No consistent lateralization was observed when analyzing ALPS indices separately for each hemisphere, suggesting that the right‐dominant phenomenon could not be fully explained by side differences in ALPS itself. In the comparison of bilateral thalamic volume, Thal_LP and Thal_AV showed rightward volume dominance. We speculate that the observed lateralization of ALPS–thalamus correlation may partly arise from inherent interhemispheric volume differences in specific thalamic subnuclei, as the thalamic lateralization patterns found in our elderly sample differ from those of healthy young individuals in the prior study. ³⁹ The mechanism underlying this asymmetric ALPS–thalamus interaction warrants further investigation.

4.2. Thalamus and cognitive regulation

The contribution of the thalamus to cognition has been somewhat overlooked. Indeed, the thalamus integrates inputs from the cortex, basal ganglia, limbic systems, brainstem nuclei, and cerebellum. ³⁶ It regulates functional connectivity within and between cortical regions and is crucial for supporting complex cognitive processes. ⁴⁰ , ⁴¹ Our study expands on traditional opinions about the thalamus, as it reveals nucleus‐specific pathophysiological links between glymphatic clearance, thalamic integrity, and specific cognitive impairment. As a core node in the Papez circuit, the anterior thalamic nucleus supports memory encoding, consolidation, and maintenance via hippocampal–cingulate connections. ⁴² , ⁴³ , ⁴⁴ For the mediodorsal thalamic nucleus, its extensive reciprocal connections with the prefrontal cortex enable it to regulate executive functions such as memory retrieval, decision making, and attentional control. ⁴⁵ Meanwhile, the ventral lateral nucleus is integral to the cortico–striatal–thalamic circuit, integrating information about executive control and task coordination. ⁴⁶ Together, these nucleus‐specific effects reveal selective cognitive dysfunction rather than global cognitive decline.

4.3. Clinical translational implications

This work has promising clinical translational implications. The combination of the ALPS index and thalamic nuclear volumes holds potential as a non‐invasive imaging biomarker for early risk stratification and subtype differentiation in AD. Although subgroup analyses showed no significant association between the ALPS index and cognitive performance, possibly due to the limited sample size and non‐linear changes of glymphatic function along disease progression, the ALPS index still correlated with multiple cognitive domains. Our results also provide a neuroanatomical basis for AD‐targeted interventions. ALPS‐related thalamic nuclei could serve as imaging endpoints to evaluate the efficacy of glymphatic enhancing therapies. Moreover, neuromodulation targeting the thalamocortical circuit could be combined with glymphatic improvement strategies to produce synergistic effects.

4.4. Limitations

This study has several inherent limitations that should be acknowledged. First, the ALPS method is theoretically deductive, and the relationship between ALPS indices and human glymphatic function has yet to be substantially and rigorously validated through pathophysiological studies. A recent study revealed no statistically significant association between the ALPS index and cortical glymphatic function measures. ⁴⁷ Consequently, caution is needed when using ALPS to interpret whole‐brain glymphatic clearance rates. ⁴⁸ The cerebral glymphatic system should be comprehensively assessed with multiple indicators. Second, we found no significant relationship between the ALPS index and cognitive function across CN, MCI, and AD subgroups, so we did not further validate their association among ALPS, the thalamus, and cognition across different disease stages. The cross‐sectional study design also limits the causal inferences drawn from the mediation results. Furthermore, residual confounding (total neurodegeneration burden, vascular pathology, and sleep quality) may inflate the mediation effect estimates and introduce bias. In light of these limitations, the mediation findings in this study should be interpreted as exploratory and associative.

5. CONCLUSION

Cognitive function across the executive, language, and memory domains was associated with the ALPS index. Thalamic nuclear damage is linked to the interplay between glymphatic dysfunction and cognitive impairment. Targeting the glymphatic system emerges as a promising potential therapeutic direction for AD.

AUTHOR CONTRIBUTIONS

Xiao Zhou, Yu Sun, and Dantao Peng provided the human specimens and collected the samples. Leian Chen developed the idea, performed the analysis, and wrote the manuscript. Bin Zhang, and Ying Hou participated in the study design and revised the manuscript. All authors read and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

ETHICAL APPROVAL

This research protocol was approved by the China–Japan Friendship Hospital ethics committee and institutions (Ethics ID: 2025‐KY‐158). Consent forms were obtained from all participants. The research was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). The ADNI study received ethical approval from the local institutional review boards of all participating centers. Written informed consent was obtained from all participants or their study partner.

ADNI CONSORTIA INFORMATION

The data used in this study were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. As such, the investigators within the ADNI contributed to the design and implementation of the ADNI and/or provided data but did not participate in the analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wpcontent/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf

Supporting information

Supporting Information

DATA AVAILABILITY STATEMENT

The code of this article is available in the “open science framework” repository (https://osf.io/wgtn7/). The datasets analysed in this study are available from the corresponding author on reasonable request.