Tai Chi modulating multimodal connectivity patterns and cognitive function in cerebral small vessel disease

Abstract

Cerebral small vessel disease (CSVD) significantly contributes to cognitive decline and poses potential risk factors for both dementia and acute cerebrovascular events. Tai Chi has been recognized as an effective mind–body intervention for enhancing cognitive function and promoting neural plasticity. However, the neurobiological mechanisms that underlie how Tai Chi exercises affect cognitive improvement in patients with CSVD remain unclear. We recruited 51 patients with CSVD, randomly assigning them to either a 24-week Tai Chi exercise group (n = 26) or a health education control group (n = 25). We collected and analyzed data on neuropsychological assessments, plasma homocysteine levels, and brain structural and functional connectivity (SC and FC). Additionally, we employed network-based statistics along with correlational and mediation effect analyses to analyze the intervention process. Patients in the intervention group demonstrated a significant improvement in cognitive function. Furthermore, they demonstrated an increased FC pattern in the frontal–parietal–occipital regions, along with a significant rise of structural–functional coupling within the frontal network and a reduction in the occipital network. The enhanced structural–functional coupling in the frontal network partially mediated the reduction in homocysteine levels and the improvement in cognitive function.

Introduction

Cerebral small vessel disease (CSVD) is a key factor in cognitive dysfunction and vascular dementia, encompassing various clinical, imaging, and pathological syndromes due to impairment to white matter and grey matter [1]. In China, CSVD is responsible for about half of acute strokes, which is a significantly higher rate compared to Western countries [2]. As the population ages and the incidence of CSVD rises, CSVD with mild cognitive impairment (CSVD-MCI) has become the most common cognitive disorder, influencing over 60% of patients with CSVD [3]. Moreover, vascular cognitive impairment because of CSVD has become one of the leading causes of dementia [4].

Considering the importance of addressing early cognitive decline in patients with CSVD, especially at the MCI stage, there is an urgent requirement to utilize effective intervention strategies. However, pharmacological treatments for CSVD-MCI remain controversial, as patients with CSVD might face an increased risk of long-term cerebral hemorrhage following thrombolytic therapy [5]. Thus, the current treatment plan primarily focuses on preventive and non-pharmacological strategies. Among these methods, physical exercise is particularly beneficial for neurogenesis and vascular health, contributing to the delay of cerebrovascular disease and cognitive impairment [6, 7].

In exploring non-pharmacological physical interventions, traditional Chinese practices such as Tai Chi offer a balanced and effective option to mitigate the risks associated with high-intensity interval training while retaining the benefits of mind–body exercises [8, 9]. Tai Chi combines gentle movements with flexible breath control, making it an ideal exercise for older adults with CSVD [10]. Furthermore, existing evidence supports that Tai Chi can enhance overall cognitive ability, help regulate blood pressure, and improve quality of life among older adults and those with neurodegenerative disorders [11–13]. Despite these advantages, it remains unclear whether this intervention can specifically enhance cognitive ability related to CSVD-MCI, and the neurobiological mechanisms linking potential cognitive improvements to Tai Chi exercises are still not well understood.

Understanding the neurobiological mechanisms is essential for elucidating how Tai Chi may influence cognitive ability in patients with CSVD. Emerging evidences have supported that CSVD pathology might influence the white matter microstructure, causing disruptions in brain structural connectivity (SC) networks, by investigating through diffusion tensor imaging (DTI) [14, 15]. Previous studies have indicated that the patients with CSVD-MCI suffer connection disruptions in global structural networks [16–18]. Moreover, resting-state functional Magnetic Resonance Imaging (rs-fMRI) has become another widely used approach for constructing functional networks associated with multidimensional cognitive functions [19–21]. Previous studies have found that large-scale functional connectivity (FC) patterns such as frontal–parietal networks have also been impaired in patients with CSVD, relating to the general cognitive deficits including attention, execution, and emotional dysregulation [22–24]. Additionally, recent studies have verified the impaired SC–FC network coupling as another effective measurement to evaluate the burden of CSVD pathology and cognitive impairment, as it could reflect the interplay between structural and functional abnormalities related to cognition [25, 26]. However, despite numerous studies exploring neuroimaging biomarkers associated with cognitive deficits in CSVD patients, it remains unclear how effective mind–body interventions like Tai Chi could influence large-scale brain SC–FC networks in relation to varying cognitive burdens.

In this study, we recruited 106 patients with CSVD and randomly assigned them to either a Tai Chi training group or a health education control group. The aim is to analyze the effects of Tai Chi intervention on cognitive abilities in patients with CSVD-MCI and its underlying brain SC-FC network alteration. We hypothesize that patients with CSVD-MCI who experienced a 24-week intervention training might show significant alterations in the SC-FC coupling networks leading to favorable cognitive improvements.

Materials and methods

Study design and subjects

The whole study was approved by the Medical Ethics Committee in the Affiliated People’s Hospital of Fujian University of Traditional Chinese (IRB number: 2019–058-04). The study population consisted of patients diagnosed with CSVD from local community residents who received medical treatment at this hospital. The study involved two parallel groups (an intervention group with Tai Chi exercises and a control group with health education). During the study design phase, we performed a priori sample size estimation using G*Power 3.1.9.7. Based on the primary outcome measure and two-sample independent t-test, the parameters were set as follows: Effect size (Cohen’s d): 0.5, α: 0.05, Power (1-β): 0.8, Allocation ratio N2/N1: 1. The required total sample size was calculated to be 52 participants with 26 in each group [27]. Patients were all Han Chinese enrolled between November 2020 and October 2023. Each subject has signed an informed consent before the whole experiment. Notably, this is a sub-study of the project, and our current analysis focuses solely on samples that have neuroimaging data.

The enrolled subjects should satisfy the criteria: (1) meet the diagnostic criteria for patients with CSVD-MCI [28]; (2) age ≥ 50; (3) no regular physical exercise habits (such as over 20 min per session with over three times a week); (4) Montreal Cognitive Assessment (MoCA) scores between 19 and 25 [29]; The exclusion criteria were shown as follows: (1) cognitive impairment from other diseases, such as Alzheimer’s disease or Parkinson’s disease; (2) cortical or mixed lesions, subcortical non-lacunar infarcts, and hemorrhagic lesions; (3) white matter lesions from diseases such as multiple sclerosis; (4) severe medical conditions contraindicating exercise; (5) prior cognitive impairment documented in medical records; (6) history of alcohol or drug abuse; (7) dementia and inability to perform self-care or severe mental illnesses; (8) MRI contraindications; (9) participation in other clinical experiments affecting study results.

Each participant was evaluated by two experienced radiologists, who independently assigned MRI-based CSVD scores. Any discrepancies were resolved through discussion to reach a consensus. We assessed global CSVD burden using an ordinal scale ranging from zero to six [30]. To focus on those with a mild to moderate global CSVD burden, only patients with scores between one and three points were included in the current study.

Intervention

Subjects in the intervention group underwent 24 weeks’ training in eight-style Tai Chi, with three 30-min sessions each week. Each session included a 5-min warm-up, 20 min’ Tai Chi practice, and a 5 min’ cool-down. Additionally, participants attended a 30-min health education session every 8 weeks, which provided information on cognitive impairment and primary prevention strategies for cerebrovascular disease. To minimize commuting burdens, participants were assigned to 2–3 community centers for their Tai Chi training. The Tai Chi exercise routine was developed in accordance with the General Administration of Sport in China [31].

Subjects in the control group received the same 30-min health education sessions on cerebrovascular disease and cognitive impairment as those in the intervention group, occurring every 8 weeks, without any accompanying exercise intervention. To monitor the potential confounding effects of daily physical activity on outcomes, all subjects were required to document their routine physical activities in a daily activity log throughout the intervention period. To enhance participant adherence to the study protocol, individuals were organized into small groups during exercise sessions, allowing for supervision and social interaction among group members, which facilitated the completion of training with high quality. Additional supportive strategies, including free health consultations and rewards, were implemented for subjects with full adherence.

Primary outcome

Global cognitive function was assessed as the primary outcome using the Chinese version of the Montreal Cognitive Assessment (MoCA) at the 24-week mark [32]. This brief screening tool is designed to detect cognitive impairment in approximately 10 min. The assessment evaluates eight cognitive domains, which include visuospatial and executive functioning, naming, delayed recall, attention, language, abstraction, and orientation. Higher scores indicate better overall cognitive status. To account for educational background, subjects with six years or fewer of education receive two additional points, those with more than six but up to twelve years of education receive one additional point, while no additional points are given to subjects with more than twelve years of education [33]. The scale demonstrates strong validity, reliability, and sensitivity, making it suitable as the primary outcome measure for assessing global cognitive function [34, 35].

Secondary outcomes

Plasma homocysteine (HCY) levels were measured using an HCY assay kit (24–0603, Beijing Strong Biotechnologies, Inc., Beijing, China) on the Abbott Architect ci6200 chemistry analyzer, following the enzyme cycling method. Blood samples were collected in the morning after a 12-h fasting, both within three days prior to the intervention at baseline and within three days following the completion of the intervention. Moreover, we examined structural–functional coupling patterns underlying the intervention effects using multimodal neuroimaging data. Collectively, HCY level, structural–functional coupling, and cognitive subdomains are designated as secondary outcomes in the current analyses.

MRI Data preprocessing

The rs-fMRI data were acquired using a 3.0 T Siemens MRI system (Germany), with scanning parameters aligned with those used in our previous protocol [10]. Specifically, the rs-fMRI data were performed using an echo-planar imaging (EPI) sequence (33 slices, field of view (FOV) = 224 × 224 mm², slice thickness = 3.5 mm, TR = 2000 ms, TE = 30 ms, FA = 90°, and voxel size = 3.5 × 3.5 × 3.5 mm). The DTI preprocessing utilized the diffusion toolbox from the FSL, correcting for eddy current distortions and head movements, followed by skull stripping [36]. The diffusion tensor matrix was computed to estimate fractional anisotropy (FA) maps, with fiber tract reconstruction performed using the DiffusionKit toolbox and whole-brain tractography via the FACT algorithm [37, 38].

For rs-fMRI, data were preprocessed with SPM12 and DPABI version 8.2 [39, 40]. The pipeline comprised removal of the first 6 volumes, slice-timing correction, and realignment. For each participant, we computed translation, angular motion, framewise displacement (FD). Volumes were flagged as outliers if FD > 0.3 mm, maximum translation > 2 mm, angular motion > 2° [41]. Groups were comparable in head motion: mean FD (Tai Chi = 0.12 ± 0.05 mm; control = 0.13 ± 0.06 mm) and number of outlier volumes (Tai Chi = 8.2 ± 6.1; control = 7.9 ± 5.8); no between-group differences were observed (all p > 0.05). The functional images were co-registered with the structural images and transformed into Montreal Neurological Institute (MNI) space, with a resampled voxel size of 3 × 3 × 3 mm³. Spatial smoothing was applied using a 6-mm full-width at half maximum Gaussian kernel, followed by linear detrending and band-pass filtering in the range of 0.01 to 0.08 Hz. Nuisance regression was conducted, incorporating the Friston-24 motion parameters along with signals from the global brain, white matter, and cerebrospinal fluid [42].

Brain connectivity network and SC-FC coupling pattern

We calculated the SC and FC maps based on the automatic anatomical labeling (AAL) parcellation [43]. SC was determined based on white matter projections measured by the averaged fractional anisotropy (FA). Two regions were considered structurally connected if at least two fibers, each longer than 10 mm, passed through them. The weight of SC was defined as the averaged FA value along all fibers linking the two regions. Meanwhile, we computed Pearson correlations on the mean fMRI time series for all pairs of regions, resulting in a whole-brain connectivity matrix. FC was defined as the Z-transformed correlation coefficient to ensure normality.

To compute the SC–FC coupling at the regional level, we constructed an adjacency matrix for the individualized SC matrix, where each row represented the strength of connections from a specific brain region to all other regions. Next, we extracted the regional connectivity feature from each row of this adjacency matrix, indicating the connection strength from a particular region to the other regions of the brain. We then isolated the non-zero structural connection weights and rescaled them into a normal distribution to facilitate further analysis [44, 45]. Corresponding functional connectivity values were extracted to ensure they aligned with the structural data. Eventually, we computed the non-parametric Spearman correlation between the structural and functional networks, getting 90 regional SC–FC coupling values.

Statistical analysis

Demographic characteristics were investigated using Statistical Product and Service Solutions (SPSS, version 22.0). Categorical features, such as gender differences and education duration, were evaluated using χ² tests. For continuous variables, such as age, HCY levels, MoCA scores (including subdomains), and brain imaging metrics, we conducted between-group comparisons (e.g., differences between pre- and post-intervention, or at either pre- or post-intervention) using independent t-tests or Mann–Whitney tests, depending on the normality of the data. For within-group comparisons between different time points, we utilized paired t-tests or Wilcoxon signed-rank tests, based on the normality of the data.

To investigate the intervention impact of Tai Chi exercise on brain networks, we employed Network-Based Statistics (NBS), which utilizes the cluster-based thresholding principles to control the familywise error rate across all connections in the graph [46]. For the between-group comparisons (Tai Chi vs. Control) on post-intervention and Δ networks, the primary component-forming threshold was set at an uncorrected p < 0.001, and t > 3.1. Connections surviving this threshold were grouped into components based on their extent. The significance of each component was then assessed using a non-parametric permutation test with 5,000 repetitions, which generated an empirical null distribution of the maximum component size. The FWER-corrected p-value for a component was defined as the proportion of permutations yielding a component of equal or larger size. To evaluate robustness, we conducted sensitivity analyses using alternative primary thresholds of p < 0.001 (t > 2.8) and p < 0.001 (t > 3.4). The Δ (change) values for both FC and SC were calculated from a repeated-measures ANOVA model that included the factors Group (Tai Chi, Control) and Time (pre, post) to explicitly model the group × time interaction. When testing for post-intervention differences, the corresponding baseline (pre-intervention) network metric was included as a covariate in the model. For each significant network identified, the cluster-level effect size was computed as the mean t-statistic of all edges within the component. We then investigated the significant connections identified between the two groups by masking the network connectivity matrix and conducted an independent two-sample t-test to compare averaged connectivity values for these connections.

Whole-brain regional Δ SC-FC coupling was analyzed between groups using independent two-sample t-tests or Mann–Whitney tests, based on their data normality.

To control for multiple comparisons across nodes (n = 90), we applied Benjamini–Hochberg false discovery rate (FDR) correction with a q < 0.05 threshold. Furthermore, we prespecified the frontal network, which is defined according to the anatomical structure in the AAL template. Network-level ΔSC–FC coupling was computed as the unweighted average of the nodewise coupling values across all regions in this network. Partial correlation analyses were conducted to examine associations between network-level ΔSC–FC coupling and Δ cognitive function (MoCA) and Δ HCY levels, adjusting for age, sex, and education duration. Mediation analyses were performed using the PROCESS macro 4.0 to test whether network-level ΔSC–FC coupling mediated the group effect on Δ MoCA and ΔHCY, with 10,000 bootstrap samples to derive bias-corrected 95% confidence intervals (CI). The conditional indirect effect was considered significant when the 95% CI did not encompass zero. To improve the specificity of the model, the analyses were adjusted for relevant baseline covariates, including age, sex, and education level.

Results

Enrollment

Of the 302 subjects initially recruited, 196 were excluded during baseline screening due to ineligibility, scheduling conflicts, or poor adherence. 106 subjects successfully passed the baseline screening, with 53 assigned to each group (Tai Chi and Control). During the follow-up period, 21 subjects dropped out due to loss of contact or interest, scheduling conflicts, or relocation. Additionally, 34 subjects were excluded because of lack of MRI data. Ultimately, 51 subjects were included in our analysis pipeline, with 26 of the intervention group and 25 of the control group (Fig. 1).

Fig. 1.

Flowchart of the enrollment procedures for patients with CSVD-MCI. CSVD-MCI, Cerebral small vessel disease with mild cognitive impairment

Baseline characteristics of subjects

Subjects’ demographic characteristics and assessments were presented in Table 1. No significant statistical differences were observed in baseline age, sex, education duration, and neuropathology severity score, HCY level, or general cognitive status between subjects in the Tai Chi and Control groups. Additionally, there were no significant between-group differences in MRI head motion parameters, as well as in FC/SC network patterns, and SC-FC coupling at both regional and network levels (p > 0.05).

Table 1.

Baseline demographic and clinical assessments

	Tai Chi group (N = 26)	Control group (N = 25)	Statistics (t/Z/χ2)	P value
Demographics and CSVD-related characteristics
Age a	62.58 ± 7.94	63.24 ± 6.29	0.33	0.74
Sex (Male/Female) b	12/14	12/13	0.02	0.89
BMI a	24.32 ± 3.26	23.60 ± 2.85	0.84	0.40
WHR c	0.87 ± 0.06	0.90 ± 0.05	1.55	0.13
Educational duration b (1: ≤ 6 years, 2: > 6 and ≤ 12 years, 3: > 12 years), n
Level 1	3	2	0.30	0.86
Level 2	18	17
Level 3	5	6
Tatal CSVD score b, n
Score 1	17	17	0.07	0.97
Score 2	6	5
Score 3	3	3
Participants with metabolic disorder, (with/without)
Hypertension b	14/12	13/12	0.02	0.89
Hyperlipidemia b	11/15	9/16	0.21	0.64
Diabetes b	3/23	5/20	0.69	0.41
Lifestyle factors, (Never/occasionally/often)
Smoking b	17/6/3	16/5/4	0.24	0.88
Drinking b	6/16/4	5/15/5	0.21	0.90
Exercise b	9/17/0	7/18/0	0.26	0.61
General Cognitive function, mean ± standard deviation
MoCA c	21.35 ± 1.94	21.72 ± 1.51	0.77	0.45
CSVD-related Biomarker, mean ± standard deviation
HCY a, umol/L a	9.88 ± 2.77	9.12 ± 2.56	1.00	0.32
SBP a, mm Hg c	127.31 ± 11.64	129.56 ± 10.56	0.72	0.47
DBP a, mm Hg c	81.42 ± 5.49	83.44 ± 7.58	1.09	0.28
TC c, mmol/L a	5.02 ± 1.36	5.01 ± 1.00	0.03	0.98
TG c, mmol/L a	1.55 ± 1.07	1.31 ± 0.59	0.96	0.34
LDL-C, mmol/L a	3.06 ± 1.12	2.96 ± 1.05	0.32	0.75
FBG, mmol/L c	6.55 ± 2.46	6.50 ± 1.79	0.07	0.94

Data were presented as the range of mean ± standard deviation

^a independent t-test for continuous data with normal distribution

^b chi-square tests for categorical variables

^c Mann‒Whitney U test for continuous data with skewed distribution

CSVD: Cerebral small vessel disease, BMI: Body mass index, WHR: Waist-to-Hip Ratio, SBP: Systolic blood pressure, DBP: Diastolic blood pressure, TC: Total cholesterol, TG: Triglyceride, HDL-C: High density lipoprotein cholesterol, LDL-C: Low density lipoprotein cholesterol, FBG: Fasting blood glucose, MoCA: Montreal Cognitive Assessment, HCY: homocysteine

Primary outcomes

Independent samples t-tests revealed that the difference in Δ MoCA scores between the two groups was statistically significant (t(49) = 4.8145, p < 0.001, d = 0.85, CI: [1.61, 3.93]). Additionally, the paired samples t-tests indicated that the Tai Chi intervention group experienced a significant increase in MoCA scores after 24 weeks of exercise (t(25) = 5.3906, p < 0.001, d = 1.44, confidential interval (CI): [1.99, 3.55], Fig. 2A). In contrast, the health education group showed no significant change (t(24) = 0, p > 0.05, Fig. 2B).

Fig. 2.

Comparison of MoCA Scores and HCY Levels in Tai Chi Intervention and Control Groups Pre- and Post–Intervention. (A-B) Comparison of MoCA scores between the Tai Chi intervention and control groups between pre- and post–intervention. (C-D) Comparison of HCY levels between the Tai Chi intervention and control groups between pre- and post–intervention

Secondary outcomes

Intervention effects on cognitive subdomains and HCY

Cognitive subscores at baseline and after 24 weeks in Tai Chi and control groups were shown in Table 2. Independent samples t-tests revealed that the difference in several Δ subdomains of MoCA scores and HCY between the two groups was statistically significant (visuospatial and executive functioning, t(49) = 3.2592, p = 0.002089, d = 0.527, CI: [0.415, 1.754], survived after FDR correction; HCY, t(49) = −4.2548, p < 0.001, d = −0.870, CI: [−4.939, −1.770]). The paired samples t-tests indicated that the Tai Chi intervention group experienced a significant increase in several subdomains of MoCA scores and HCY after 24 weeks of exercise (visuospatial and executive functioning, t(25) = 4.2133, p < 0.001, d = 0.826, CI: [0.45, 1.317], naming, t(25) = 3.638, p = 0.001247, d = 0.713, CI: [0.250, 0..904], abstraction, t(25) = 3.3477, p = 0.002582, d = 0.657, CI: [0.192, 0..808], survived after FDR correction; HCY, t(25) = −3.6731, p = 0.001141, d = −0.720, CI: [−3.269, −0.920], Fig. 2C). In contrast, the health education group showed no significant change (all cognitive subdomains and HCY, p > 0.05, Fig. 2D).

Table 2.

Cognitive Subscores at Baseline and After 24 Weeks in Tai Chi and Control Groups

Cognitive Subscores	Visuospatial and executive	Naming	Attention	Language	Abstraction	Delayed recall	Orientation
Baseline
Tai Chi group	3.69 ± 1.12	2.46 ± 0.51	4.73 ± 0.67	1.88 ± 0.77	0.96 ± 0.66	0.88 ± 0.91	5.77 ± 0.43
Control group	3.8 ± 0.96	2.76 ± 0.60	4.64 ± 0.70	2.04 ± 0.89	1.04 ± 0.61	1.12 ± 0.88	5.48 ± 0.65
24 Weeks
Tai Chi group	4.57 ± 0.64	3.04 ± 0.60	4.85 ± 0.78	1.92 ± 0.69	1.46 ± 0.58	1.46 ± 0.86	5.85 ± 0.37
Control group	3.6 ± 1.15	2.76 ± 0.66	4.56 ± 0.92	1.76 ± 0.83	1.24 ± 0.60	1.56 ± 0.82	5.4 ± 0.87

Data were presented as the range of mean ± standard deviation

Intervention effects on connectivity network

We first analyzed between-group differences in post-intervention SC and ΔSC using NBS. No significant connected components were found at the primary threshold (p < 0.001, FWER-corrected) or in the sensitivity analyses (thresholds of p < 0.005 and p < 0.0005). In contrast, for functional connectivity, NBS revealed a significant network for the between-group difference in post-intervention FC (FWER-corrected p = 0.020, component size: 16 edges). This result was robust to sensitivity analyses. The cluster-level effect size (mean t-statistic) for this network was 3.45 (Cohen’s d = 0.49) (Fig. 3A-B). Similarly, for the ΔFC analysis (testing the group × time interaction), a significant network was identified (FWER-corrected p = 0.005, component size: 16 edges), which was also stable across sensitivity thresholds. The cluster-level effect size for this ΔFC network was 3.80 (Cohen’s d = 0.67) (Fig. 3D-E), mainly featuring the bilateral inferior triangular frontal gyrus, right inferior opercular frontal gyrus, right middle frontal orbital gyrus, bilateral superior occipital gyrus, left inferior occipital gyrus, and bilateral inferior parietal gyrus. The Tai Chi group also showed significant increases in Δ FC patterns (Fig. 3F, t(49) = 3.1262, p = 0.00298, d = 0.666, CI: [0.0402, 0.185]).

Fig. 3.

Changes in FC Post-Intervention. (A–B) Comparison of post-intervention FC patterns between the Tai Chi and Control groups. (C) Average connection values for the significant clusters identified in (A–B). (D–E) Δ FC patterns between the two groups. (F) Average connection values for the significant clusters identified in (D–E). *p < 0.05, **p < 0.01, Δ FC, functional connectivity changes from pre-intervention to post-intervention

Intervention effects on structural–functional coupling change

In comparisons between two groups, we generated a statistical difference map of the regional Δ SC-FC (Fig. 4A) and identified 10 regions with significant differences related to the frontal, occipital, parietal, and limbic networks (abs (Cohen’s d) > 0.8, corrected p < 0.05) (Fig. 4B–C). Specifically, as shown in Fig. 4D–E, the Tai Chi group demonstrated significantly higher Δ SC-FC coupling in six regions of the frontal network (bilateral superior orbital frontal gyrus, bilateral middle orbital frontal gyrus, left inferior opercular frontal gyrus, right inferior orbital frontal gyrus), and one region in the parietal network (right inferior parietal gyrus). Additionally, there were statistically significant decreases in Δ SC-FC coupling in two regions of the occipital network (left inferior occipital gyrus, right superior occipital gyrus) and the right hippocampus. Furthermore, we compared the averaged network-level Δ SC-FC coupling between the two groups and found significant increases in both the frontal and occipital networks (Fig. 4F) (frontal: t(49) = 3.344, p = 0.001667, d = 0.595, CI: [0.0646, 0.260]; occipital: t(49) = −2.91, p = 0.0057, d = −0.458, CI: [−0.245, −0.044], survived after FDR correction).

Fig. 4.

Statistical Differences in Regional Δ SC-FC Coupling. (A) Brain map showing statistical differences in regional Δ SC-FC between the two groups. (B) Brain map highlighting significant statistical differences in regional Δ SC-FC between the two groups. (C) Identification of 10 regions from (B) (Cohen’s d > 0.8, corrected p < 0.05). (D) Significantly higher Δ SC-FC coupling observed in six regions of the frontal cortices. (E) Notable changes in Δ SC-FC coupling in the occipital, parietal, and limbic cortices. (F) Average network-level Δ SC-FC coupling. *p < 0.05, **p < 0.01, ***p < 0.001, Δ SC-FC, structural–functional coupling changes from pre-intervention to post-intervention

Intervention effect on the cognitive functions, HCY and structural–functional coupling

Further correlational analyses revealed that the averaged Δ SC-FC coupling in the frontal network was significantly negatively associated with the Δ HCY level (p < 0.001, r = −0.488) and positively correlated with the Δ MoCA (p < 0.001, r = 0.480) (Fig. 5A–B). Moreover, mediating effects analysis demonstrated that Δ SC-FC coupling of the frontal network partially mediated the group effect on the Δ MoCA (Indirect effect: 0.9306, bootstrap CI [0.1111, 2.0789]) and Δ HCY levels (Indirect effect: −0.6217, bootstrap CI [−1.5116, −0.0360]) (Fig. 5C–D).

Fig. 5.

Correlation and Mediation Analyses of Frontal Network SC-FC Coupling with HCY Levels and MoCA Scores. (A) Correlation analysis between averaged Δ SC-FC coupling in the frontal network and Δ HCY levels. (B) Correlation analysis between averaged Δ SC-FC coupling in the frontal network and Δ MoCA scores. (C–D) Mediation effect analysis demonstrating the role of Δ SC-FC coupling in the frontal network on Δ MoCA and Δ HCY levels. **p < 0.01, n.s, p > 0.05. Δ HCY/MoCA/SC-FC, Plasma homocysteine/Montreal Cognitive Assessment scores/structural–functional coupling changes from pre-intervention to post-intervention

Sensitivity analyses

Post-hoc sensitivity analyses were conducted for the primary and key secondary endpoints. The analysis of the primary and secondary endpoints revealed significant findings. For the change in MoCA scores (Δ MoCA), the effect size was Cohen’s d = 0.85, with a 95% confidence interval of [1.61, 3.93], and the achieved power for this analysis was approximately 0.99. The change in HCY levels showed an effect size of Cohen’s d = −0.87, with a 95% confidence interval of [−4.939, −1.770], yielding a power of about 0.99 as well. In terms of specific subdomains of MoCA, the visuospatial and executive functioning subdomain demonstrated an effect size of Cohen’s d = 0.53, with a confidence interval of [0.415, 1.754], and the achieved power was approximately 0.76. Additionally, the Tai Chi intervention group exhibited notable increases in several subdomains: for visuospatial and executive functioning, Cohen’s d = 0.83 (CI: [0.452, 1.317]) with a power of about 0.99; for naming, Cohen’s d = 0.71 (CI: [0.250, 0.904]) with a power of approximately 0.94; and for abstraction, Cohen’s d = 0.66 (CI: [0.192, 0.808]) with a power of around 0.91. The effect size for HCY in this group was Cohen’s d = −0.72, with a confidence interval of [−3.269, −0.920], and the power was approximately 0.95. Furthermore, when examining the frontal SC-FC coupling, the effect size was Cohen’s d = 0.60 (CI: [0.0646, 0.260]), with an achieved power of about 0.85, while the occipital SC-FC coupling showed an effect size of Cohen’s d = −0.46 (CI: [−0.245, −0.044]), with a lower power of approximately 0.63.

To evaluate the robustness of our neuroimaging findings, we re-ran all group comparisons using various preprocessing and modeling configurations, including analyses with and without global signal regression (GSR) and different temporal band-pass filters (0.01–0.03 Hz and 0.01–0.10 Hz compared to the primary 0.01–0.08 Hz). Additionally, spatial smoothing was applied prior to ROI time-series extraction using 4 mm and 8 mm full-width at half maximum kernels, alongside the primary 6 mm kernel. The results from NBS for the pre-to-post change in ΔFC network demonstrated consistent robustness across these specifications, with values of 0.008 without GSR, 0.007 for the 0.008–0.09 Hz filter, 0.009 for the 0.01–0.10 Hz filter, 0.005 for the 4 mm kernel, 0.011 for the 8 mm kernel, and 0.008 for the 6-parameter model. The change-network topology consistently favored the Tai Chi group in group-level statistics, which showed t(49) values ranging from 2.87 to 3.32, p-values from 0.0020 to 0.0060, and effect sizes (d) from 0.61 to 0.7. Similarly, the Δ SC–FC results remained stable across different specifications, with a persistent set of 10 significant regions indicating increased coupling in frontal and right inferior parietal areas, and decreased coupling in occipital regions and the right hippocampus, all with absolute Cohen’s d values exceeding 0.8 and corrected p-values below 0.05. The direction of regional effects remained consistent across alternative pipelines, with per-region effect sizes varying within ± 0.05 of the primary values. At the network level, group differences in averaged Δ SC–FC coupling were significant after FDR correction for both the frontal and occipital networks, with primary statistics showing t(49) = 3.344, p = 0.001667, d = 0.595 for the frontal network, and t(49) =  − 2.91, p = 0.0057, d =  − 0.458 for the occipital network, with robustness ranges for these statistics indicating strong consistency across analyses.

Discussion

In this study, we aimed to investigate the effects of a 24-week Tai Chi intervention on patients with CSVD-MCI and underlying neurobiological mechanisms. By comparing the Tai Chi group with the Control group, we found not only statistically significant improvements in cognitive function but also notable alterations in FC patterns within the frontal, occipital, and parietal networks in the Tai Chi group. Additionally, enhanced SC-FC coupling in the frontal network, along with decreased SC-FC coupling, was statistically linked to the intervention effects. Further correlational and mediation analyses indicated that SC-FC coupling in the frontal network mediates improvements in overall cognitive function and HCY levels, suggesting potential compensatory mechanisms in the brain that help regulate cognitive performance.

By comparing between-group network differences, we found a significant increase in connection clusters in patients who participated in Tai Chi exercises, primarily involving brain areas in the frontal, occipital, and parietal cortices. The frontal–parietal network is recognized to play an essential role in higher-order executive functions necessary for cognitive tasks, and decreased activation in this network may contribute to working memory and attention deficits observed in patients with CSVD [22, 47]. The enhanced connectivity in the frontal–parietal network suggests that Tai Chi exercises could help reshape the impaired brain hierarchy, thereby improving cognitive flexibility [48–50]. Furthermore, the occipital network is involved in spatial–visual preprocessing and played a role in emotional regulation [51, 52]. The interaction between the frontal and occipital networks may indicate a coordinated effort to manage visual information alongside cognitive tasks [53, 54]. Previous studies have identified structural disruptions between the frontal and occipital areas in patients with CSVD, highlighting the importance of coordinated patterns among the frontal–parietal–occipital networks in understanding the pathophysiology of CSVD-MCI [22, 55, 56]. Our analyses emphasize that Tai Chi exercises can modulate and enhance functional connectivity between these critical networks, potentially contributing to cognitive improvements.

Our exploratory findings further found that SC–FC coupling patterns change significantly in the Tai Chi group, especially with increased frontal coupling and decreased occipital coupling. These phenomena can be explained with the posterior–anterior shift in aging (PASA) mechanisms, which reflect a shift in the neural resource allocation from posterior brain regions (such as occipital cortices) to anterior regions (such as frontal cortices) [57–59]. The significant increase in SC–FC coupling in the frontal cortices suggested that engagement in Tai Chi may boost the brain’s capacity to allocate resources effectively, due to the cognitive or physical demands of the practice. Conversely, the observed decreased occipital coupling might reflect a reallocation of neural resources, which could be part of the brain’s adaptive strategy to prioritize cognitive functions that are more heavily reliant on frontal cortices. These changes in SC–FC coupling patterns suggested that Tai Chi may enhance the brain’s ability to implement PASA as a compensatory strategy [60–62]. By constructing a more robust coupling pattern in the frontal network while simultaneously modulating the occipital network, Tai Chi appears to promote a more efficient neural network organization, allowing patients to better manage cognitive demands despite underlying structural impairment.

Another notable finding is that the increase in frontal SC-FC coupling was associated with the effects of Tai Chi on improvements in general cognitive function and reductions in HCY levels. Previous studies have indicated that patients with cerebrovascular diseases often experience cognitive decline alongside elevated HCY levels [63, 64], and HCY levels have been linked to neural abnormalities in degenerative disorders [65, 66]. Our findings build upon this foundation by suggesting that enhanced SC-FC coupling within the frontal network is correlated with improvements in cognitive abilities and reductions in HCY levels. One potential explanation for this pattern of results is that Tai Chi exercise, by promoting physical activity, may enhance neuroplasticity, which could influence SC-FC coupling. Additionally, the mindfulness component of Tai Chi may help reduce stress, which is a known factor in HCY metabolism [67, 68]. Therefore, based on the observed associations, increased SC-FC coupling in the frontal cortex may play an important role in the relationship between Tai Chi practice and improvements in cognitive and metabolic outcomes. While we acknowledge that the mediation analysis aligns with the hypothesis of a causal pathway, it is important to note that it does not provide definitive evidence for causation. Alternative explanations, including the potential influence of unmeasured confounding variables, should be considered.

Several limitations should be considered for the current analyses. First, while the results have achieved the statistical power required for the initial RCT design, a larger sample size could further validate and enhance our findings. Second, the analysis is based on follow-up data from 24 weeks of Tai Chi practice; longer follow-up periods would allow us to observe the sustained effects of the intervention. Third, our analysis focused solely on general cognitive ability without examining specific cognitive dimensions. Additionally, a limitation of this study is that both subjects and coaches could not be blinded, making it challenging to implement blinding in non-pharmacological trials. Consequently, this may introduce some performance bias during the experiment.

Conclusion

In conclusion, our study demonstrated that 24 weeks of Tai Chi exercises had significant intervention effects on patients with CSVD-MCI. These effects included altered SC-FC coupling patterns, reduced HCY metabolism levels, and improved cognitive abilities. Specifically, patients engaging in Tai Chi exhibited a posterior–anterior shift characterized by increased SC-FC coupling in the frontal network and decreased coupling in the occipital network. Additionally, the enhanced coupling in the frontal network partially mediated the reduction in HCY levels and the improvement in general cognitive function. Overall, our findings suggest that Tai Chi exercises may serve as a beneficial intervention strategy for patients with CSVD-MCI, helping to enhance their cognitive control ability and promote neurobiological rehabilitation.

Author contribution

Acquisition, analysis, or interpretation of data: X. Z., W. L., B. C.; Drafting of the manuscript: Z. D., L.C.; Critical review of the manuscript for important intellectual content: Z. D., X. Z., W. L., B. C., S. Z.; Statistical analysis: Z. D., B. C., H. Z.; Supervision: W. L., B. C.

Funding

This study was supported by the Natural Science Foundation of Fujian Province (2023J01148, 2023J01846), Young and Middle-aged Core Personnel Training Project of Fujian Province Health Department (2022GGA046), Special Research Project of the National Clinical Research Base of Traditional Chinese Medicine (JDZX201935), and Special Research Project of the National Natural Science Foundation of China's Basic Improvement Program (JCZX202302, JCZX202307, JCZX202413), Joint Funds for the Innovation of Science and Technology, Fujian Province (No. 2024Y9490).

Data availability

The data are not publicly available due to privacy or ethical restrictions. The data and codes that support the findings of this study are available on request from the corresponding author.

Declarations

Ethics approval

The whole study was approved by the Medical Ethics Committee in the Affiliated People’s Hospital of Fujian University of Traditional Chinese (IRB number: 2019–058-04). The whole trial was conducted according to the Declaration of Helsinki.

Conflict of interest

The authors have no conflict of interest to declare.