Associations of neurodegenerative proteins with brain iron deposition and cognition in cerebral small vessel disease: a quantitative susceptibility mapping and plasma biomarker study

Abstract

INTRODUCTION

Cerebral small vessel disease (CSVD) is a common neurological disorder with limited pathology on conventional magnetic resonance imaging. This study uses quantitative susceptibility mapping (QSM) to investigate links among brain iron, plasma neurodegenerative proteins, and cognition in CSVD.

METHODS

This study enrolled 319 CSVD patients, grouped into CSVD‐M and CSVD‐S. Plasma proteins were measured in 178 participants, with 80 being followed up after 2 years. QSM‐based voxel‐wise analysis assessed brain iron, CSVD severity, and protein correlations. A cross‐lagged panel model was used to analyze the temporal association between plasma protein levels and brain iron levels.

RESULTS

In CSVD‐S, elevated QSM values in the right Rolandic operculum/superior temporal gyrus negatively correlated with plasma Aβ42 and executive function. Aβ42 also negatively correlated with QSM in cortical regions, tied to episodic memory decline. Higher baseline Aβ40 predicted increased QSM in the left putamen at follow‐up.

DISCUSSION

Plasma Aβ42 and Aβ40 may drive brain iron deposition and cognitive impairment in CSVD, serving as potential early biomarkers for disease progression.

Highlights

• QSM reveals brain iron links to Aβ42, cognition in CSVD.

• Plasma Aβ42 correlates with iron in motor and frontal areas.

• High Aβ40 predicts putamen iron increase in CSVD follow‐up.

• Iron deposition is tied to executive, memory deficits in CSVD.

1. BACKGROUND

Cerebral small vessel disease (CSVD) is a common neurological disorder affecting small brain vessels (< 1 mm), contributing to 25% of strokes and 45% of dementias. ¹ , ² , ³ The diagnosis of CSVD relies on magnetic resonance imaging (MRI) features, including lacunes, white matter hyperintensities (WMHs), cerebral microbleeds (CMBs), and perivascular spaces. ⁴ However, the lesions detected by conventional MRI represent only a fraction of the underlying pathological changes in CSVD, necessitating deeper investigation.

Iron is essential for various neurobiological processes, including DNA synthesis, oxygen transport, myelin formation, mitochondrial function, and neurotransmitter synthesis and metabolism. ⁵ Excessive iron deposition causes oxidative stress, contributing to neurodegenerative diseases and cognitive impairments. ⁵ , ⁶ Brain iron, stored as ferritin and hemosiderin, distorts local fields. Local magnetic field distortions can be detected by typical susceptibility‐weighted imaging (SWI) techniques, including those measuring transverse relaxation rates (R2, R2*, T2, and T2*) and quantitative susceptibility mapping (QSM). ⁷ Unlike other SWI techniques, QSM precisely quantifies tissue magnetic susceptibility at the voxel level by solving the inverse problem – specifically, the deconvolution process from the measured magnetic field to the underlying susceptibility sources. ⁸ , ⁹ This enables an in‐depth investigation of susceptibility changes caused by iron distribution, metabolic oxygen consumption, calcification, demyelination, and other pathophysiological alterations. ⁹ A strong linear correlation exists between the susceptibility values of gray matter and chemically measured iron concentrations. This enables susceptibility values to serve as a non‐invasive biomarker for quantifying pathological iron accumulation within brain tissue in patients with neurodegenerative diseases. Consequently, QSM provides crucial technical support for research into the etiology and treatment of neurodegenerative disorders. ¹⁰

Amyloid beta (Aβ) is a 36‐ to 43‐amino‐acid peptide derived from Aβ precursor protein (APP), expressed mainly in brain neurons and non‐neuronal cells. Aβ40 is the most abundant, whereas Aβ42 is widely regarded as the primary protein constituent of amyloid plaques in Alzheimer's disease (AD). ¹¹ , ¹² Tau, encoded by the microtubule‐associated protein tau (MAPT) gene located on chromosome 17q21, is a microtubule‐associated protein critical for the assembly and stabilization of microtubules, as well as DNA/RNA protection. ¹² , ¹³ Tau is predominantly expressed in neurons, with lesser expression in astrocytes and oligodendrocytes. In pathological conditions, soluble, unfolded tau undergoes changes influenced by genetic mutations, conformational alterations, and post‐translational modifications (PTMs), rendering it insoluble and leading to its aggregation into paired helical filaments and neurofibrillary tangles (NFTs). ¹³ Tau phosphorylation has been the most extensively studied and is considered a pivotal event in the pathological aggregation of tau protein.

Plasma Aβ42/40 and phosphorylated tau (pTau) exhibit a high degree of consistency with cerebrospinal fluid (CSF) levels, accurately reflecting intracerebral Aβ plaques and NFTs accumulation. ¹⁴ These biomarkers show high specificity in distinguishing between abnormal and normal Aβ‐positron emission tomography (PET) statuses, which are established indicators for detecting AD. ¹⁵ , ¹⁶ Although amyloid plaques and NFTs are pathological hallmarks of AD, their presence has also been reported in other conditions, such as Huntington's disease, ¹⁷ , ¹⁸ Parkinson's disease, ¹⁷ , ¹⁹ traumatic brain injury, ²⁰ and cerebral amyloid angiopathy (CAA). ²¹ It suggests that pathological changes and mechanisms associated with Aβ and tau proteins play significant roles in non‐AD populations as well. Prior research identified correlations between plasma/CSF levels of Aβ and tau proteins and various imaging markers of CSVD, such as WMHs, CMBs, and lacunes. Furthermore, these protein levels have been closely linked to cognitive function, highlighting their broader relevance in CSVD. ¹⁴ , ²² , ²³ , ²⁴

Research on CSVD has largely ignored interactions between neurodegenerative proteins and brain iron levels. This study hypothesizes that, in CSVD, the abnormal aggregation of neurodegenerative proteins may disrupt local iron transport or activate microglia responsible for clearance functions, causing aberrant iron deposition. This may trigger neuronal damage via oxidative stress responses, ultimately mediating cognitive impairment in CSVD patients. ²⁵ , ²⁶ The present study addresses this hypothesis through the following approaches: (1) voxel‐wise QSM analysis compared QSM values between mild CSVD and severe CSVD groups; (2) voxel‐wise correlation between plasma protein concentrations and QSM values; and (3) longitudinal voxel‐wise QSM changes in a follow‐up cohort, using a cross‐lagged panel model to assess plasma protein‐QSM relationships.

2. MATERIALS AND METHODS

2.1. Participants

This community‐based cohort study enrolled 319 participants with CSVD from communities surrounding Shandong Provincial Hospital between September 2019 and February 2025. The inclusion criteria for participants were based on the updated Standards for Reporting Vascular Changes on Neuroimaging (STRIVE) ²⁷ : (1) age ≥ 40 years; (2) MRI compatibility (no ferromagnetic implants, claustrophobia, or motion artifacts > 2 mm); (3) psychotropic medication‐free; (4) preserved sensory function.

The exclusion criteria were as follows: (1) major neuropsychiatric disorders (stroke, traumatic brain injury), neurodegenerative diseases (AD, Parkinson's disease, Huntington's disease), or intracranial lesions; (2) significant organ dysfunction including cardiac, renal, or hepatic impairment; (3) confirmed non‐vascular white matter lesions, include multiple sclerosis, adult leukodystrophy, or metabolic encephalopathy.

Clinical data, including age, sex, body mass index (BMI), education, and vascular risk factors (hypertension, diabetes, hyperlipidemia, smoking), were systematically recorded. The presence or absence of each vascular risk factor was determined based on participants' self‐reported medical histories, with each factor assigned a binary value (0 or 1). These individual risk factors were subsequently combined to form a composite vascular risk factor score, ranging from 0 to 4.

2.2. Plasma neurodegenerative protein measurement

Plasma collection for neurodegeneration biomarker assays began in January 2021, involving a subsample of 178 participants. Eighty participants completed a follow‐up assessment 2 years later, with the same protein types and measurement methods applied as at baseline.

RESEARCH IN CONTEXT

1. Systematic review: We systematically reviewed all PubMed peer‐reviewed articles on CSVD and plasma neurodegenerative protein biomarkers. This is the first study using QSM to examine cerebral iron deposition – plasma biomarker relationships in CSVD. Nevertheless, several recent studies have explored associations between CSVD and plasma biomarkers. All referenced literature is appropriately cited.

2. Interpretation: This study revealed that in cross‐sectional investigations of CSVD, iron deposition levels in multiple brain regions demonstrated significant correlations with concentrations of plasma neurodegenerative protein biomarkers and were further associated with domain‐specific cognitive functions. Longitudinal analyses additionally indicated that baseline plasma Aβ40 levels could effectively predict future iron deposition changes in the left putamen.

3. Future directions: Future studies should extend this research by (1) incorporating novel plasma biomarker panels or establishing more precise quantification methodologies and (2) combining white matter tractography with QSM to elucidate white matter integrity – plasma biomarker pathological relationships.

Blood samples (5 mL) were collected via venipuncture into EDTA tubes using a standardized protocol. Samples were centrifuged at 3000 rpm for 15 min. Plasma was separated, aliquoted, and stored at −20°C for short‐term preservation, followed by long‐term storage at −70°C. Plasma levels of Aβ1‐42 and Aβ1‐40 were measured using human‐specific enzyme‐linked immunosorbent assay (ELISA) kits (Thermo Fisher/Invitrogen, USA). All samples were processed with a single reagent lot according to the manufacturer's protocol to ensure consistency. Total tau, pTau 181, pTau 231, and pTau 217 were analyzed using the same ELISA platform under identical conditions. All tau measurements were performed in the same batch as amyloid biomarkers to maintain analytical uniformity.

2.3. Neuropsychological assessment

All participants underwent comprehensive cognitive function assessments. The Montreal Cognitive Assessment (MoCA) was used to evaluate global cognitive function. The MoCA is a 30‐point test that covers eight cognitive domains, with higher scores indicating better cognitive performance. ²⁸ To ensure applicability to the Chinese population, the Beijing version of the MoCA was used in this study.

The cognitive subdomain assessments included the Auditory Verbal Learning Test (AVLT), Stroop Color and Word Test (SCWT), Trail Making Test (TMT), and Symbol Digit Modalities Test (SDMT). The AVLT is a neuropsychological assessment designed to evaluate episodic memory function. Through auditory‐verbal recall tasks, it comprehensively measures an individual's episodic memory abilities. The Chinese version of the SCWT reflects information processing speed and the ability to resist cognitive interference. ²⁹ Performance was assessed based on the accuracy and timing of verbal responses to colors or words, with higher scores indicating lower cognitive flexibility. TMT evaluates information processing speed, sequencing ability, cognitive flexibility, and visual–motor coordination. ³⁰ Scores were based on the accuracy and timing of the sorting tasks, with higher scores reflecting slower information processing speed and poorer cognitive flexibility. The SDMT primarily measures information processing speed. ³¹ Participants were asked to verbally provide the number corresponding to a symbol within 90 s, with the total score reflecting the number of correct responses.

2.4. MRI acquisition

MRI examinations were conducted using a Siemens MAGNETOM Skyra 3.0T system (Erlangen, Germany) with a 32‐channel head receiver coil. To reduce scanner noise, participants were given foam earplugs. Head motion was minimized using a head restraint system, which included foam padding around the participant's head. The imaging protocol included 3D T1‐magnetization‐prepared rapid gradient echo (MPRAGE) sequence for structural analysis (repetition time [TR]/echo time [TE]/inversion time [TI] = 7.3/2.4/900 ms; flip angle 9°; 1 mm³ isotropic resolution); 3D multi‐echo gradient echo (mGRE) sequence for QSM reconstruction (TR = 50 ms; TE/ΔTE = 6.8/4.1 ms; 10 echoes; 15° flip angle; 1×1×2 mm³ resolution); T2‐weighted turbo spin echo (T2W), T2‐weighted fluid‐attenuated inversion recovery (FLAIR), diffusion‐weighted imaging (DWI), and susceptibility‐weighted imaging (SWI) sequences were also performed to detect brain abnormalities.

2.5. Neuroimaging biomarker evaluation

The identification and severity evaluation of imaging biomarkers of CSVD were performed following the updated Standards for Reporting Vascular Changes on Neuroimaging 1(STRIVE‐1). ²⁷

A composite CSVD burden score (0 to 4 points) was calculated to assess overall disease severity, as validated for cognitive impairment prediction. ³² One point was allocated for each of the following: presence of ≥1 lacune, presence of ≥1 CMB, confluent deep WMH, or periventricular hyperintensities extending into deep white matter, and moderate‐to‐severe enlarged perivascular spaces (PVS) (>10).

Based on the total CSVD burden score, participants were stratified into two distinct groups: the CSVD‐M group (scores 0 to 1), representing no/mild small vessel disease burden, and the CSVD‐S group (scores 2 to 4), indicative of moderate–severe cerebrovascular pathology.

2.6. QSM reconstruction

QSM reconstruction was performed using the morphology‐enabled dipole inversion with automatic uniform cerebrospinal fluid zero reference algorithm (MEDI+0) framework on multi‐echo GRE data. ³³ In summary, non‐linear fitting of the multi‐echo data was first conducted to estimate the total magnetic field. The total field was spatially unwrapped using a quality‐guided region‐growing algorithm. ³⁴ This was followed by background field removal using the projection onto dipole fields (PDF) algorithm to compute the local field. ³⁵ The local field was then inverted to produce the final susceptibility map. To enhance QSM accuracy and provide an automatic susceptibility reference, structural priors (edges) from the magnitude image, along with a regularization term to enforce uniform CSF susceptibility in the lateral ventricles, were incorporated into the numerical inversion. The CSF mask was generated by thresholding the R2* map derived from the mGRE magnitude data, with voxel connectivity constraints. ³³

We initiated our analysis by segmenting the T1‐weighted anatomical images using Statistical Parametric Mapping (SPM12) to obtain gray matter volume images. Concurrently, the QSM images were resampled to match the isotropic resolution (1 × 1 × 1 mm³) of the gray matter volume images. Subsequently, a study‐specific brain template was created utilizing the Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) toolbox. ³⁶ The gray matter volume and QSM maps were then normalized into Montreal Neurological Institute (MNI) space and subjected to smoothing with a 3‐mm full‐width at half‐maximum Gaussian kernel, following established protocols. ³⁷

3. STATISTICAL ANALYSES

3.1. Demographic and clinical variable comparisons

Statistical analysis was performed using SPSS (version 27.0) and R (version 4.2.3). Normality assumptions were verified through Shapiro–Wilk testing. Between‐group comparisons (CSVD‐M vs CSVD‐S) employed parametric (independent t‐test) or non‐parametric (Mann–Whitney U test) approaches for continuous variables based on distribution characteristics, while categorical variables were analyzed using χ² or Fisher's exact tests as appropriate. A two‐tailed α level of 0.05 defined statistical significance.

3.2. Voxel‐wise QSM analysis

Using Statistical Parametric Mapping (SPM12) in the MATLAB R2022b environment, a voxel‐wise two‐sample t‐test was conducted within a mask created by the Anatomical Automatic Labeling (AAL) atlas with the cerebellum removed. This analysis compared voxel‐wise QSM values between the CSVD‐M and CSVD‐S groups, with sex, age, and vascular risk factor scores included as covariates.

Additionally, multiple linear regression was employed to investigate the correlation between voxel‐wise QSM values and concentrations of plasma neurodegenerative proteins (Aβ40, Aβ42, pTau231, pTau217, and pTau181), with sex, age, and vascular risk factor scores included as covariates. Analyses were repeated in CSVD‐M and CSVD‐S subgroups. The significance thresholds were set at a voxel‐wise p < 0.001 and a cluster‐wise family‐wise error (FWE)‐corrected p < 0.05, with only clusters exceeding 100 voxels considered significant. Significant clusters were defined as regions of interest (ROIs), and their mean QSM values were extracted for subsequent analyses.

3.3. ROI‐cognitive score association analysis

ROI_diff was defined as a significant cluster identified through voxel‐wise two‐sample t‐test. Spearman rank correlation analysis assessed associations between plasma proteins and QSM values of ROI_diff. Variables with uncorrected p < 0.1 were entered into multivariate stepwise linear regression models, which were adjusted for age, sex, vascular risk factor score, and education years. Stratified subgroup analyses were rigorously performed within the CSVD‐M and CSVD‐S cohorts.

Spearman rank correlation analysis was conducted between QSM values of ROI_diff and cognitive scores across all participants. Variables with uncorrected p < 0.1 were incorporated into multivariate stepwise linear regression models, with age, sex, and education years as covariates. This analysis was repeated within the CSVD‐M and CSVD‐S groups.

Spearman rank correlation analysis was conducted between QSM values of plasma protein‐associated clusters and cognitive scores across all participants. Variables with uncorrected p < 0.1 were incorporated into multivariate stepwise linear regression models, which were adjusted for age, sex, and education years. Stepwise regression was employed, and multicollinearity was assessed using variance inflation factors (VIF < 5). Stratified subgroup analyses were systematically performed within the CSVD‐M and CSVD‐S cohorts.

3.4. Longitudinal voxel‐wise QSM analysis and its association with plasma neurodegenerative protein concentrations

To investigate the longitudinal association between QSM values and plasma neurodegenerative proteins, this study enrolled 80 CSVD patients, who completed full follow‐up between September 2019 and February 2025, with an average follow‐up duration of 19.95 ± 8.80 months. There were 75 CSVD‐M participants and 5 CSVD‐S participants among the 80 CSVD patients. Given the small number of participants in the CSVD‐S group, we only conducted a longitudinal analysis on the subjects from the CSVD‐M group.

Using SPM12 (in the MATLAB R2022b environment), voxel‐wise paired t‐tests were performed on baseline and follow‐up QSM values (corrected at the cluster level with FWE correction, with thresholds set at a voxel‐wise p < 0.001 and cluster size ≥ 100 contiguous voxels) to identify brain regions with significant QSM changes during the follow‐up period. A cross‐lagged panel model was employed to examine the dynamic associations between the mean QSM values of the cluster and plasma neurodegenerative protein concentrations, adjusting for covariates including age, sex, vascular risk factors, and follow‐up interval (in months).

4. RESULTS

4.1. Missing data analysis

Among the total cohort of 319 CSVD participants, 178 individuals with measured plasma biomarker levels were included in the plasma biomarker analyses. A comparison between participants with plasma biomarker data (n = 178) and those without (n = 141) revealed significant differences in several baseline characteristics: those with biomarker data were younger, had higher educational levels and a lower prevalence of hypertension, and demonstrated better global cognitive function (Table S1). Furthermore, within the subset of 178 participants with plasma biomarker measurements, 80 completed the follow‐up assessment at 2 years. Participants who returned for follow‐up exhibited higher educational levels and better performance in specific cognitive domains compared to those lost to follow‐up (Table S2).

4.2. Characteristics of study participants

This study enrolled a total of 319 participants with CSVD, comprising 226 CSVD‐M participants and 93 CSVD‐S participants. Among these, 178 participants provided plasma protein samples, with 137 from the CSVD‐M group and 41 from the CSVD‐S group.

No significant differences were observed between the CSVD‐M and CSVD‐S groups in terms of BMI, hyperlipidemia, and smoking status. However, compared to the CSVD‐M group, the CSVD‐S group was older, had a higher proportion of males, lower educational attainment, and a greater prevalence of hypertension and diabetes (p < 0.05). Significant differences were also found between the two groups in the AVLT, MoCA, SCWT, and TMT measurements (Table 1).

TABLE 1.

Demographic and clinical data.

Variables	CSVD‐M (226)	CSVD‐S (93)	T/χ2/z value	p
Age	57.78 ± 8.54	65.71 ± 6.89	−7.95	<0.001 a
Sex
1	105 (46.5)	56 (60.2)	4.45	0.035 b
2	121 (53.5)	37 (39.8)
BMI	24.35 ± 4.22	24.92 ± 2.77	−1.19	0.236 a
Education years	13.11 ± 3.45	11.48 ± 3.12	3.93	<0.001 a
Hypertension
0	141 (62.4)	24 (26.4)	32.29	<0.001 b
1	85 (37.6)	67 (73.6)
Diabetes
0	155 (68.9)	46 (51.1)	8.05	0.005 b
1	70 (31.1)	44 (48.9)
Hyperlipidaemia
0	119 (58.0)	43 (47.3)	2.55	0.111 b
1	86 (42.0)	48 (52.7)
Smoking
0	176 (78.2)	64 (68.8)	2.66	0.103 b
1	49 (21.8)	29 (31.2)
Vascular risk factor score
0	65 (28.8)	8 (8.6)	27.7	<0.001 b
1	72 (31.9)	26 (28.0)
2	56 (24.8)	27 (29.0)
3	26 (11.5)	20 (21.5)
4	7 (3.1)	12 (12.9)
AVLT	60.00 (51.00, 69.00)	54.00 (46.00, 63.00)	−3.15	0.002 c
MoCA	27.00 (25.00, 28.00)	25.00 (22.00, 27.00)	−5.56	<0.001 c
Stroop1	23.00 (20.00, 27.00)	27.00 (24.00, 33.00)	5.54	<0.001 c
Stroop2	35.00 (30.00, 40.00)	41.00 (35.00, 51.00)	5.67	<0.001 c
Stroop3	66.00 (57.00, 80.00)	86.00 (67.25, 108.00)	6.17	<0.001 c
SCWT	120.00 (107.00, 146.00)	150.00 (130.25, 188.00)	6.37	<0.001 c
TMTA	50.00 (38.00, 71.00)	67.50 (51.25, 95.25)	5.02	<0.001 c
TMTB	140.50 (100.00, 201.77)	207.00 (153.25, 294.50)	6.38	<0.001 c
TMT (A+B)	190.50 (140.00, 275.00)	263.50 (208.50, 380.25)	6.12	<0.001 c
TMT (B−A)	85.00 (58.00, 137.25)	139.50 (89.25, 212.50)	5.85	<0.001 c

Abbreviations: AVLT, Auditory Verbal Learning Test; CSVD‐M, cerebral small vessel disease participants with no/mild small vessel disease burden (total CSVD burden score < 2); CSVD‐S, cerebral small vessel disease participants with moderate/severe small vessel disease burden (total CSVD burden score ≥ 2); MoCA, Montreal Cognitive Assessment; SCWT, Stroop Color and Word Test; SDMT, Symbol Digit Modalities Test; TMT, Trail Making Test.

^a Two‐sample t‐tests.

^b Chi‐squared test.

^c Mann–Whitney U test.

5. RESULTS OF VOXEL‐WISE QSM ANALYSIS

5.1. Voxel‐wise QSM differences in CSVD cohort

In the CSVD cohort, two‐sample t‐tests adjusted for age, sex, and vascular risk factor scores demonstrated significantly higher QSM values in the CSVD‐S group compared to CSVD‐M within the right Rolandic operculum and right superior temporal gyrus (peak MNI: 43, −23, 16; t = 5.06, FWE‐corrected p = < 0.001) (Table 2, Figure 1). No regions exhibited lower QSM values in CSVD‐S relative to CSVD‐M.

TABLE 2.

Regions with QSM value differences between every two groups (voxel‐wise p < 0.001 and cluster‐wise FWE‐corrected p < 0.05).

Clusters	Cluster size (total voxels)	t‐value	p FWE value	Peak MINI coordinates (mm)			Brain regions (cluster size)
				X	Y	Z
CSVD‐M < CSVD‐S
1	789	5.06	<0.001	43	−23	16	Rolandic_Oper_R (517), Temporal_Sup_R (154)

Abbreviations: CSVD‐S, cerebral small vessel disease with moderate–severe cerebrovascular pathology; FWE, family‐wise error; MNI, Montreal Neurological Institute; QSM, Quantitative Susceptibility Mapping; Rolandic_Oper_R, right Rolandic Operculum; Temporal_Sup_R, right superior temporal gyrus.

FIGURE 1.

Regions with QSM value differences between the two groups (voxel‐wise p < 0.001 and cluster‐wise FWE‐corrected p < 0.05). (A) Depicts the cluster with QSM value differences between CSVD‐M and CSVD‐S in sagittal, coronal, and axial planes; Maps are displayed at the p < 0.05 level with the T‐values displayed in the color bar. Correction for multiple comparisons was applied with FWE correction. (B) The distribution of the QSM values in the cluster with statistical differences in two‐sample t test.

5.2. Voxel‐wise QSM and plasma protein concentration correlations

Voxel‐wise correlation analyses adjusted for age, sex, and vascular risk factor scores revealed widespread negative associations between mean QSM values and plasma Aβ42 levels across bilateral supplementary motor areas/paracentral lobules/superior frontal gyri/postcentral gyri, and right precentral (voxel‐wise p < 0.001, cluster‐level FWE‐corrected p < 0.05) (Table 3, Figure S1).

TABLE 3.

Regions with QSM value significantly associated with plasma neurodegenerative protein in CSVD patients (voxel‐wise p < 0.001 and cluster‐wise FWE‐corrected p < 0.05).

Groups	Variable	Clusters	Cluster size	t value	p FWE value	Peak MINI coordinates (mm)			Brain regions (cluster size)
						X	Y	Z
All	Aβ42‐negative correlation	1	14746	5.481	<0.001	12	−33	72	Supp_Motor_Area_R (5308), Supp_Motor_Area_L (2135), Paracentral_Lobule_R (1501), Paracentral_Lobule_L (1042), Frontal_Sup_Medial_R (999), Frontal_Sup_R (971), Frontal_Sup_L (865), Postcentral_R (733), Precentral_R (315), Postcentral_L (912)
		2	1330	5.076	<0.001	−19	−36	76	Postcentral_L (912)
CSVD‐M	Aβ42‐negative correlation	1	1758	4.421	<0.001	10	7	64	Supp_Motor_Area_R (1388), Frontal_Sup_R (244)
		2	1074	5.276	0.001	13	−33	72	Paracentral_Lobule_R (501), Postcentral_R (453)
		3	1047	4.628	0.002	8	−17	68	Supp_Motor_Area_R (842)
		4	947	4.121	0.003	−3	−11	75	Supp_Motor_Area_L (674)
		5	670	4.305	0.022	−19	−36	76	Postcentral_L (536)
	Tau‐positive correlation	1	647	4.4	0.025	−2	52	−5	Frontal_Med_Orb_L (479)
	pTau181‐positive correlation	1	739	4.379	0.013	3	−10	73	Paracentral_Lobule_L (452)
CSVD‐S	Aβ42‐negative correlation	1	440	4.722	0.036	−46	12	0	Frontal_Inf_Oper_L (242)
		2	610	5.026	0.006	35	−49	44	Parietal_Inf_R (444)
	pTau217‐positive correlation	1	757	5.165	0.002	46	11	−1	Insula_R (444), Rolandic_Oper_R (200)

Abbreviations: CSVD, cerebral small vessel disease; Inf_Oper, inferior opercular frontal cortex; Frontal_Med_Orb, medial orbital frontal cortex; Frontal_Sup_Medial, superior medial frontal cortex; Frontal_Sup_R, superior frontal gyrus; FWE, family‐wise error; Insula, insular cortex; L, left hemisphere; MNI, Montreal Neurological Institute; Paracentral_Lobule, paracentral lobule; Parietal_Inf, inferior parietal lobule; Postcentral, postcentral gyrus; Precentral, precentral gyrus; QSM, quantitative susceptibility mapping; R, right hemisphere; Rolandic_Oper, Rolandic operculum; Supp_Motor_Area, supplementary motor area.

In the CSVD‐M subgroup, after adjusting age, sex, and vascular risk factor scores, voxel‐wise correlation analyses revealed (1) negative associations between mean QSM values and plasma Aβ42 levels across right supplementary motor areas/superior frontal gyri/paracentral lobule/postcentral gyri/supplementary motor areas, left supplementary motor areas/postcentral gyri; (2) positive associations between mean QSM values and plasma tau protein levels in the left medial orbital frontal gyrus; (3) positive associations between mean QSM values and plasma pTau181 protein levels in the left paracentral lobule (Table 3, Figure S1).

In the CSVD‐S subgroup, after adjusting for age, sex, and vascular risk factor scores, voxel‐wise correlation analyses revealed (1) negative associations between mean QSM values and plasma Aβ42 levels across the left inferior frontal operculum/right inferior parietal lobule and (2) positive associations between mean QSM values and plasma PTau217 protein levels in the right insula/right Rolandic operculum. Details of the voxel‐wise correlation analyses between QSM and plasma protein levels are shown in Table 3 and Figure S1.

6. RESULTS OF ROI–COGNITIVE SCORE ASSOCIATION ANALYSIS

6.1. Relationship between ROI_diff QSM values and plasma protein levels

In the CSVD cohort, Spearman correlation analysis revealed that in the CSVD‐S group, the ROI_diff mean QSM values were negatively correlated with plasma Aβ42 protein concentrations (r = −0.455, p = 0.004). After adjusting for age, sex, vascular risk factor scores, and education years, multiple linear regression confirmed this negative correlation (β [95% confidence interval [CI] = −0.140 (−0.220,−0.059), p = 0.001) (Table S3, Figure S2). However, no association was observed between the ROI_diff mean QSM values and plasma protein concentrations in either the overall population or the CSVD‐M group.

6.2. Relationship between ROI_diff QSM values and cognitive performance

After adjusting for age, sex, and education years, multiple linear regression analyses revealed the following associations:

All CSVD participants: The ROI_diff mean QSM values (CSVD‐M vs CSVD‐S) were positively correlated with SCWT (Stroop3: β [95% CI] = 0.372 [0.119, 0.624], p = 0.004) and TMT Part A (TMT‐A: β [95% CI] = 0.323 (0.063, 0.583), p = 0.015) (Table 4, Figure 2);

TABLE 4.

Associations between regions with QSM value differences and cognitive function among the two groups.

Cognition	Variables	β (95% CI)	t‐value	p	R 2 of model	p value of model
ALL
Stroop3	Age	0.7 (0.363,1.037)	4.088	<0.001	0.281	<0.001
	Education	−2.526 (−3.39, −1.661)	−5.746	<0.001
	Sex	−11.204 (−16.597, −5.811)	−4.088	<0.001
	ROI_diff	0.372 (0.119, 0.624)	2.899	0.004
TMT‐A	Education	−3.436 (−4.323, −2.549)	−7.621	<0.001	0.43	<0.001
	Age	1.239 (0.892, 1.587)	7.022	<0.001
	Sex	8.486 (2.935, 14.038)	3.008	0.003
	ROI_diff	0.323 (0.063, 0.583)	2.443	0.015
CSVD‐M
Stroop3	Education	−2.318 (−3.26, −1.376)	−4.848	<0.001	0.242	<0.001
	Age	0.607 (0.234, 0.981)	3.207	0.002
	Sex	−9.658 (−15.58, −3.735)	−3.214	0.002
	ROI_diff	0.301 (0.006, 0.595)	2.011	0.046

Abbreviations: CI, confidence interval; CSVD‐M, cerebral small vessel disease with no/mild small vessel disease burden (total CSVD burden score < 2); CSVD‐S, cerebral small vessel disease with moderate/severe small vessel disease burden (total CSVD burden score ≥ 2); QSM, quantitative susceptibility mapping; ROI_diff, regions with QSM value differences between each two groups; Stroop3, Stroop Color and Word Test Part 3; TMT, Trail Making Test.

FIGURE 2.

Associations between regions with QSM value differences and cognitive function among the two groups.

CSVD‐M group: The mean QSM values of ROI_diff was positively correlated with SCWT (Stroop3: β [95% CI] = 0.301 (0.006, 0.595), p = 0.046) (Table 4, Figure 2).

6.3. Relationship between QSM values in plasma protein concentration‐related clusters and cognitive performance in plasma protein cohort

In the plasma protein cohort, after adjusting for age, sex, and education years, multiple linear regression analyses revealed the following associations:

All CSVD participants: Clusters with mean QSM values negatively correlated with plasma Aβ42 levels exhibited inverse associations with Auditory Verbal Learning Test delayed recall scores (AVLT: β [95% CI] = −0.394 (−0.625, −0.163), p = 0.001) (Table 5);

TABLE 5.

Associations between protein‐associated regions and cognitive function among the two groups.

Cognition	Variables	β (95% CI)	t‐value	p	R2 of model	p value of model
ALL
AVLT	Sex	4.718 (0.966,8.47)	2.482	0.014	0.352	<0.001
	Age	−0.442 (−0.662, −0.222)	−3.966	<0.001
	Education	1.347 (0.807, 1.887)	4.924	<0.001
	Aβ42_Negative_Cluster1	−0.394 (−0.625, −0.163)	−3.366	0.001
CSVD‐M
AVLT	Sex	5.858 (1.611, 10.104)	2.729	0.007	0.399	<0.001
	Age	−0.416 (−0.671, −0.161)	−3.231	0.002
	Education	1.603 (0.999, 2.207)	5.252	<0.001
	Aβ42_Negative_Cluster1	−0.43 (−0.664, −0.197)	−3.65	<0.001
CSVD‐S
AVLT	Sex	4.511 (−2.724, 11.747)	1.266	0.214	0.353	0.004
	Age	−0.764 (−1.269, −0.26)	−3.074	0.004
	Education	0.582 (−0.577, 1.742)	1.02	0.315
	Aβ42_Negative_Cluster2	−0.506 (−0.834, −0.178)	−3.136	0.003

Abbreviations: Aβ42_Negative_Cluster, clusters with QSM values correlated to plasma Aβ42 protein; AVLT, auditory verbal learning test; CI, confidence interval; CSVD‐M, cerebral small vessel disease with no/mild small vessel disease burden; CSVD‐S, cerebral small vessel disease with moderate/severe small vessel disease burden.

CSVD‐M group: Clusters with mean QSM values negatively correlated with plasma Aβ42 levels exhibited inverse associations with Auditory Verbal Learning Test delayed recall scores (AVLT: β [95% CI] = −0.403 (−0.664, −0.197), p < 0.001) (Table 5);

CSVD‐S group: Clusters with mean QSM values negatively correlated with plasma Aβ42 levels were positively correlated to AVLT score (AVLT: β (95% CI) = −0.506 (−0.834, −0.178), p = 0.003) (Table 5).

6.4. Longitudinal voxel‐wise QSM analysis and its association with plasma neurodegenerative protein concentrations

Longitudinal studies indicate that, over approximately 2 years of follow‐up, 75 CSVD‐M patients exhibit a significant increase in QSM values in the left putamen (voxel‐wise p < 0.001, cluster‐level FWE‐corrected p < 0.05) (Table S4, Figure 3). Baseline plasma Aβ40 concentrations show a positive correlation with QSM values in this region at follow‐up (p = 0.026) (Figure 3).

FIGURE 3.

Longitudinal voxel‐based QSM analysis and its association with plasma neurodegenerative protein concentrations. (A) QSM voxel‐wise paired t‐tests were performed on baseline and follow‐up participants. (B) The cross‐lagged panel models assess associations between the longitudinal cluster QSM and the plasma protein; Time 1, baseline; Time 2, follow‐up. (C) Wilcoxon signed rank test assesses the differences of QSM values between the baseline and follow‐up cluster.

7. DISCUSSION

This study is the first to employ QSM voxel‐based analysis to investigate the relationships among plasma neurodegenerative proteins, brain iron, and cognitive function in CSVD, aiming to offer new perspectives for future diagnostic and therapeutic strategies.

Iron deposits are linked to dysfunctional iron regulatory mechanisms within the brain. The core mechanism underlying CSVD‐related brain injury is diffuse cerebral vascular endothelial dysfunction. ³⁸ CSVD‐related cerebrovascular endothelial dysfunction increases blood–brain barrier (BBB) permeability, promoting the accumulation of harmful substances such as iron and hemoglobin, thereby triggering neuroinflammation. ³⁹ , ⁴⁰

This study found that elevated QSM values in the right Rolandic operculum and right superior temporal gyrus were negatively correlated with executive function decline in CSVD‐S patients, suggesting that increased iron load may contribute to cognitive impairment, particularly in interference suppression.

The Rolandic operculum, also known as the subcentral gyrus or central/basal operculum, is a large, highly interconnected structure adjacent to the insula, playing a critical role in sensorimotor function and language processing. ⁴¹ Although current research has not specifically focused on the relationship between CSVD burden and iron deposition in the Rolandic operculum, the stability of this region's function tends to decline as CSVD burden increases. A recent study employing voxel‐based morphometry and amplitude of low‐frequency fluctuations (ALFF) analysis found that the Rolandic operculum in CSVD patients not only underwent volumetric atrophy but also exhibits a compensatory increase in ALFF. ⁴² Consequently, the observed increase in iron deposition in the Rolandic operculum may serve as a marker of disease progression in CSVD patients. Future studies are needed to elucidate the mechanisms by which iron accumulation in this region impacts cognitive function in CSVD and to explore its potential as a target for therapeutic intervention.

In the CSVD‐S population, plasma Aβ42 protein levels were negatively correlated with QSM values in the right Rolandic operculum/superior temporal gyrus, suggesting that plasma Aβ42 levels may decrease as CSVD burden increases. Aβ42 aggregates more rapidly than Aβ40, potentially rendering it more toxic and making it a primary component of amyloid plaques. ¹² Previous studies demonstrated a strong consistency between plasma Aβ42/40 concentrations and CSF levels. Earlier research also showed that reduced CSF Aβ42 levels were associated with increased CSVD burden and more severe neurodegeneration. ⁴³ To date, no studies have explored the relationship between brain iron levels in CSVD and plasma neurodegenerative proteins.

To investigate the relationship between plasma levels of neurodegenerative proteins (Aβ40, Aβ42, pTau231, pTau217, and pTau181) and brain iron in CSVD patients, this study employed voxel‐wise multiple linear regression analysis. The results revealed a negative correlation between plasma Aβ42 levels and QSM values in specific regions of the frontal and parietal lobes. Furthermore, across the entire study population and the two subgroup, QSM values in these clusters were negatively correlated with AVLT scores, indicating that brain iron deposition in these regions was associated with impaired episodic memory function in CSVD subjects.

Disruption of brain iron homeostasis is a key factor in Aβ protein deposition. Elevated intracellular iron levels upregulate APP expression by enhancing iron regulatory protein/iron regulatory element interactions. During iron overload, impaired furin suppresses α‐secretase activity while activating β‐secretase, cleaving the upregulated APP into Aβ40/42 and accelerating its deposition. ⁴⁴ APP stabilizes ferroportin‐1 (FPN1) to promote iron efflux, but under iron overload, excessive consumption of this mechanism impairs iron export, exacerbating deposition. ⁴⁵ Soluble Aβ binds to Fe³⁺ to remove excess iron but is difficult to dissociate; it also promotes the reduction of Fe³⁺ to Fe²⁺, releasing reactive oxygen species (ROS) and further accelerating Aβ deposition. ⁴⁶ Consequently, in CSVD patients, increased iron accumulation may trigger Aβ42 deposition, manifesting as reduced Aβ42 levels in plasma and CSF.

Local iron deposition may accelerate neuronal death by disrupting protein synthesis or increasing oxidative stress, manifesting as gray matter volume reduction. ⁴⁷ , ⁴⁸ Studies have previously identified atrophy in the frontal and parietal gray matter in CSVD patients with mild cognitive impairment. ⁴⁹ Additionally, while univariate correlation analysis indicated that QSM values of plasma Aβ42 protein‐related clusters were associated with multiple cognitive domains, these associations were attenuated after adjusting for age, sex, and education. Only a negative correlation between QSM values and the AVLT survived, suggesting that brain iron deposition in this region was linked to episodic memory impairment. Prior research noted that, compared to other cognitive domains, amyloid burden exhibits a stronger association with episodic memory. ⁵⁰ Notably, voxel‐wise multivariate linear regression revealed no associations between QSM values in temporal/hippocampal regions – areas of earliest AD involvement – and plasma Aβ42 levels. This spatial dissociation suggests distinct patterns of iron‐associated Aβ42 aggregation in CSVD compared to classic AD pathology.

Although no association between voxel‐wise QSM values and plasma pTau levels has been observed in the general population, subgroup analyses provide valuable insights. In the CSVD‐M group, the QSM value of the left paracentral lobule positively correlates with plasma pTau181 levels. In the CSVD‐S group, the QSM values of the right insula and right Rolandic operculum positively correlate with plasma pTau217 levels. Previous studies found that increased plasma levels of pTau181 and pTau217 correlated with a greater CSVD burden, including increased WMH volume, as well as a higher number of cerebral microbleeds and lacunes. ⁵¹ , ⁵² In addition, plasma pTau181 and pTau217 levels reflect underlying tau pathology and amyloid plaque deposition. ⁵³ , ⁵⁴ When local tauopathy and amyloid accumulation occur, misfolded proteins induce neuroinflammation characterized by microglial clustering, iron deposition, and elevated ROS generation. ⁵⁵ Subsequent ROS overproduction compromises cerebrovascular endothelial function through diminished nitric oxide (NO) bioavailability alongside enhanced platelet aggregation, leukocyte adhesion, and cellular apoptosis. ⁵⁶ The progressive deterioration and dissemination of endothelial dysfunction consequently exacerbates CSVD severity.

Longitudinal studies indicate that, over approximately 2 years of follow‐up, CSVD patients exhibit a significant increase in QSM values in the left putamen. Baseline plasma Aβ40 concentrations show a positive correlation with QSM values in this region at follow‐up, suggesting that higher Aβ40 levels may predict future iron accumulation in the brain. Although cross‐sectional studies have not identified a direct association, longitudinal data support a potential link between Aβ40 and iron deposition. Mechanistically, elevated Aβ40 concentrations may reflect more severe amyloid pathology, potentially leading to microglial activation, neuroinflammation, and iron deposition, which could ultimately result in oligodendrocyte damage and vascular endothelial cell injury. ⁵⁵ , ⁵⁷ These findings align with previous research, which showed that higher plasma Aβ40 levels were not only associated with more severe WMH volume, as well as increased numbers of lacunes and microbleeds in the brain, but also with the future progression of these lesions. ²³ , ⁵⁸

The primary strengths of this study include the inclusion of a large sample of community‐dwelling CSVD individuals, ensuring a robust sample size; a comprehensive evaluation of neuroimaging, plasma biomarkers, and cognitive function, providing rich multidimensional data; and the adoption of a longitudinal follow‐up design, which enabled the capture of temporal trends in disease progression. Additionally, the neuroimaging and plasma biomarker assessments were blinded to clinical data, minimizing data bias. Most notably, this study is the first to utilize QSM to assess regional iron deposition in CSVD patients, integrating plasma biomarkers and clinical cognitive function to uncover a potential link between iron accumulation and neurodegenerative protein metabolism. This finding enhances our understanding of the pathophysiological mechanisms underlying the onset and progression of CSVD and offers potential targets for future intervention strategies.

However, this study has several limitations. It did not include a healthy control group, instead categorizing participants into mild and severe CSVD groups based on total small vessel disease burden scores. This approach may underestimate the impact of CSVD on brain iron and plasma proteins due to potentially subtle differences in disease severity between groups. In addition, in the cross‐sectional analysis of plasma biomarkers, the CSVD‐S subgroup (n = 41) had a substantially smaller sample size than the CSVD‐M subgroup (n = 137). This imbalance may reduce statistical sensitivity. As a single‐center study conducted in a hospital‐adjacent community population, sample bias may exist, potentially reducing statistical power. The long‐term follow‐up design resulted in some participants being unable to complete the entire study, which could further compromise the statistical power of the follow‐up analysis. This study did not systematically collect records of participant medication use, including antidiabetic, antihypertensive, antimicrobial, and anti‐inflammatory drugs. These medications may modulate the relationships observed between QSM values and biomarkers of neurodegeneration by counteracting free radical generation, neuroinflammation, Aβ aggregation, and tau hyperphosphorylation. ⁵⁹ While the current study employed rigorously controlled ELISA protocols to measure plasma biomarkers, this method may have limited sensitivity for detecting subtle concentration changes in low‐abundance targets such as Aβ isoforms and pTau variants (pTau 181/231/217). Emerging platforms (e.g., SIMOA, immunoprecipitation‐mass spectrometry, single‐molecule array and Meso Scale Discovery immunoassay platforms) demonstrate superior performance in quantifying such biomarkers. ⁶⁰ , ⁶¹ This study could not adjust for participants' APOE genotype status. Given that APOE status is strongly associated with increased amyloid deposition and impaired clearance, future studies incorporating complete APOE genetic data would facilitate further interpretation of results. The longitudinal analyses in this study primarily reflect iron deposition dynamics in the CSVD‐M subgroup. Definitive conclusions regarding the high‐severity cohort (CSVD‐S) remain limited by its underrepresentation in the follow‐up sample (n = 5). Future investigations should prioritize enlarging the high‐severity subgroup cohort or establishing multicenter collaborations to achieve statistically robust analyses. In addition, recent evidence identifies blood neurofilament light chain (NfL) as a potential biomarker for CSVD burden and progression, mechanistically associated with acute/chronic neuroaxonal injury. ⁶² Future studies should incorporate NfL measurements to comprehensively elucidate multidimensional pathological processes in CSVD.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All study procedures were approved by the Ethical Committee of the Institutional Review Board of Shandong Institute of Medical Imaging (2019‐002). Informed consent was obtained from all participants, and the consent forms were securely stored.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Chen Y, Li M, Li J, et al. Associations of neurodegenerative proteins with brain iron deposition and cognition in cerebral small vessel disease: a quantitative susceptibility mapping and plasma biomarker study. Alzheimer's Dement. 2025;21:e70710. 10.1002/alz.70710