rTMS modulates early AD progression via synergistic brain network reorganization and peripheral biomarker dynamics

Abstract

Repetitive transcranial magnetic stimulation (rTMS) improves cognition in Alzheimer’s disease (AD), yet its biomarker and neuroplasticity effects remain unclear. Cognitive scale scores, plasma biomarker levels, and resting-state fMRI brain network in AD patients were analyzed before and after a 14-day 20 Hz rTMS intervention. The results demonstrated that rTMS intervention significantly improved cognitive performance (MMSE: Z =  − 2.863, q = 0.017; MoCA: t =  − 6.137, q < 0.001; RAVLT_I: t =  − 3.436, q = 0.011) and reduced neuropsychiatric symptoms (NPI: Z =  − 2.547, q = 0.037; HAMD: Z =  − 3.472, q = 0.009). A 9.4% reduction in neurofilament light chain levels was demonstrated (Z =  − 2.371, P = 0.018), with baseline p-Tau181 levels being inversely correlated to Aβ42 changes (R =  − 0.428, P = 0.033). Enhanced global efficiency (GE: t =  − 1.865, P = 0.081, r = 0.423) and increased connection density (CD: Z =  − 1.823, P = 0.068, r = 0.442) were identified in neural networks. Notably, GE improvements positively correlated with elevated Aβ42/40 (R = 0.596, P = 0.025), while cognitive gains measured by the MoCA were significantly associated with network reorganization metrics (GE: R = 0.486, P = 0.048; CD: R = 0.514, P = 0.035). rTMS demonstrates potential in mitigating neurodegeneration by enhancing brain network integration and modulating Aβ metabolism. This effect was particularly pronounced in early-stage AD patients who exhibit preserved neural integrity. These findings advance therapeutic assessment frameworks and decode TMS neuromodulation mechanisms. The trial was prospectively registered (ChiCTR2400080657, ClinicalTrials.gov; 2024–02-04).

Graphical Abstract

Introduction

As the aging population increases, the incidence of neurodegenerative diseases, particularly Alzheimer’s disease (AD), is on the rise. AD is a degenerative condition of the central nervous system characterized by progressive cognitive impairment, memory loss, personality changes, and behavioral alterations [1]. Its primary pathological features include neuronal loss, neurofibrillary tangles (NFT), and senile plaques [2]. Furthermore, AD is associated with disrupted large-scale connectivity between brain regions, leading to communication breakdowns, prompting scientists to conceptualize Alzheimer’s as a disconnection syndrome [3]. Currently, there are no specific drugs that effectively slow the progression of AD, rendering it one of the diseases that severely impact elderly health and increase family and societal burdens. Consequently, there is an urgent need to identify new clinical intervention strategies. Non-invasive neuromodulation techniques present a promising avenue for the treatment of AD and have garnered significant attention in recent years.

Transcranial magnetic stimulation (TMS), a non-invasive neural modulation technique, has gained considerable attention in recent years for its applications in exploring advanced cognitive functions of the brain and treating major neurological disorders [4]. The guidelines for TMS therapy indicate that repetitive TMS (rTMS) can be effectively employed for the enhancement of cognitive functions in patients with AD [5]. High-frequency rTMS enhances cortical excitability and synaptic plasticity through mechanisms of long-term potentiation, while the function and structure of the dorsolateral prefrontal cortex (DLPFC) are closely associated with memory and executive functions [6]. Consequently, most clinical studies on AD have focused on high-frequency stimulation of the DLPFC. Although current studies indicate that high-frequency rTMS stimulation of the DLPFC can enhance cognitive functions in AD patients, its effectiveness has yet to be validated at the biomarker level.

Currently, the evaluation of rTMS efficacy primarily relies on clinical scales, which are highly subjective and significantly influenced by patients’ educational levels, making them unsuitable for long-term tracking. In contrast, the detection of biomarkers significantly enhances the efficiency, objectivity, and accuracy of AD diagnosis. In 2018, the National Institute on Aging-Alzheimer’s Association (NIA-AA) proposed the AT(N) research framework for diagnosing AD, establishing a precision diagnostic model centered on biomarkers [7]. The AT(N) criteria function as a classification system for AD, enabling the description of the extent and distribution of specific pathological changes in the brain, primarily based on three main pathological features: amyloid plaques (A), tau tangles (T), and neuronal injury (N). In 2024, the NIA-AA released the latest “Revised Criteria for the Diagnosis and Staging of AD (2024)” [8]. The new criteria categorize biomarkers into three broad categories: (1) core biomarkers of neuropathological changes, which are specifically used to identify AD, including Aβ42, p-tau217, p-tau181, and p-tau231; (2) non-core biomarkers during disease progression, which are not characteristic of AD but still play a significant role in the disease’s pathogenesis, such as inflammatory markers and immune activation biomarkers; and (3) common pathological markers shared with non-AD, including neurofilament light (NFL) and GFAP. The revision of the new criteria primarily responds to the rapid development of liquid biomarkers, particularly those found in blood. International organizations have re-emphasized the authoritative detection of AD blood biomarkers; however, the effects of TMS on changes in blood biomarker levels remain unclear. Investigating the impact of TMS on blood biomarkers is of significant scientific value and clinical importance for a deeper understanding of the regulatory effects and mechanisms of rTMS.

The progressive deposition of amyloid-beta (Aβ) plaques and hyperphosphorylated tau protein aggregates in the brain is associated with a breakdown of brain connectivity, which in turn may affect cognitive and neuropsychiatric symptoms [9]. The realization of brain functions relies on the efficient coordination and communication among various functional brain systems [9]. Research indicates that many neurological and psychiatric disorders are closely linked to abnormal changes in the characteristics of the human connectome network [10]. Information integration constitutes one of the core principles by which brain networks achieve their functions [11]. “Integration” refers to the facilitation of efficient transmission of long-distance information, based on the realization of local modularity [12]. In patients with AD, the impaired integrative function of brain networks represents a key neural mechanism underlying cognitive dysfunction [13]. Specifically, the global integration capacity of brain networks in AD patients is significantly reduced, as evidenced by an increased average path length and decreased global efficiency [14]. These alterations diminish the efficiency of information transmission between different brain regions, thereby limiting the effective integration of long-distance information and adversely impacting overall cognitive function. By applying TMS to modulate the synchronization or desynchronization patterns of brain oscillations, it is possible to regulate brain activity and its associated functions at the level of brain network interactions [15]. However, the neural network regulatory mechanisms of TMS targeting AD remain unclear. Dickerson et al. [16] proposed that AD may be a disorder characterized by dysfunction across multiple large-scale neural networks rather than being confined to a single brain region. Previous studies on AD have often focused exclusively on connectivity changes within individual networks, such as the default mode network [17], while neglecting the interactions between these networks. Furthermore, these studies primarily relied on correlational data from functional imaging, which limited the ability to infer causality. Granger Causality Weighted Network (GCWN) is a network model constructed based on the analysis of time series causality, and its core advantage lies in the quantification of dynamic relationships and the integration of directional information. Therefore, this study aims to construct a weighted network based on Granger causality (GC) analysis to explore the remodeling effects of TMS on large-scale causal networks and the related mechanisms from the perspective of information integration.

Based on the aforementioned research, we hypothesize that rTMS may enhance the efficiency of information integration within neural networks by reshaping the functional topology of the brain in patients with AD. This, in turn, could improve cognitive function through mechanisms of neuroplasticity and demonstrate dynamic responses to neurodegenerative biomarkers. To validate this hypothesis, the present study investigated the regulatory effects of TMS through assessments of cognitive function scales, analysis of plasma levels of neurodegenerative-related proteins, and exploration of the characteristics of brain functional network reorganization. Additionally, we analyzed the interrelationships among these three factors.

Materials and method

Design

A total of 30 individuals diagnosed with mild to moderate Alzheimer’s disease participated in the research conducted at Huanhu Hospital in Tianjin, China. After their enrollment, every participant underwent an initial cognitive evaluation on day 0, followed by daily rTMS treatments for the subsequent 14 working days. After this 14-day intervention phase, patients were scheduled for a cognitive assessment the day following the end of treatment on day 15. Furthermore, blood samples were obtained from 25 participants on days 0 and 15 to analyze the alterations in blood biomarkers related to AD pre- and post-treatment. Furthermore, the fMRI data was collected from 17 AD patients to examine alterations in neural network properties (Fig. 1). Throughout the trial, the collected data were appropriately labeled. In this study, only the trial arrangers with direct contact with the patients had access to the patients’ information. Other researchers, including cognitive assessors and data analysts, remained blinded throughout the trial. This study received approval from the Institutional Review Boards of Hebei University of Technology, Tianjin, China (No: HEBUThMEC2024009), and written informed consent was obtained from all participants prior to enrollment.

Fig. 1.

Experimental design. Note: Enrolled patients received three weeks of daily sessions (5-day per week, for a total of 14 days). At baseline and after three weeks of treatment, patients underwent cognitive assessment. Additionally, blood samples and fMRI were also collected from patients

Participants

Diagnoses and enrollment decisions were conducted by neurologists with senior professional credentials. Participants in the study were patients attending the cognitive impairment clinic at Huanhu Hospital in Tianjin, China. All individuals took part after providing written informed consent and signing the necessary forms, in line with the Declaration of Helsinki. The criteria for including subjects were as follows: (1) individuals aged between 50 and 85 years who fulfill the AD diagnostic criteria set forth by the NIA-AA (2018); (2) patients with a Mini-Mental State Examination (MMSE) score of 10 to 27; (3) individuals diagnosed with mild to moderate AD, characterized by a Clinical Dementia Rating (CDR) of no more than 2 points; (4) both patients and their family members were required to have the legal capacity to understand the information and sign the consent form. The exclusion criteria for individuals with AD were as follows: (1) any history of epilepsy or other neurological conditions; (2) significant depression, as indicated by a score exceeding 20 on the 24-item Hamilton Depression Scale (HAMD); (3) sensory impairments, whether visual or auditory; (4) a background of severe chronic systemic illnesses, brain injuries, cranial surgeries, or substance use disorders (alcohol or drugs); and (5) any contraindications to TMS. Baseline patient information is presented in Table 1.

Table 1.

Principal demographic and clinical characteristics of AD patients

Age (years)	68.7 ± 8.2
Sex (% female)	53
Education (years)	7.2 ± 3.0
MMSE (> 27 points)	19.97 ± 5.77
MoCA (> 26 points)	15.77 ± 6.14
Aβ1-42 (> 5.782 pg/ml)	5.47 ± 1.87
Aβ1-42/40 (> 0.0543 pg/ml)	0.04 ± 0.02
p-tau181 (< 1.692 pg/ml)	3.28 ± 2.03

The table shows the mean ± SD. The normal range for each item is shown in parentheses.

rTMS implementation

The rTMS procedure was carried out utilizing a Magstim Rapid2 transcranial magnetic stimulator, which is manufactured by Magstim in Whitland, UK. This device was paired with a specific coil known as the Magstim Double 70 mm Cooled Coil. For the course of treatment, the patient underwent a total of 14 sessions of rTMS, with each session lasting 20 min and conducted once daily, five days a week. The targeted stimulation was applied to the left DLPFC, and the precise location for this stimulation was determined by adjusting the position 5 cm anterior to the motor cortex, in accordance with the natural curvature of the patient’s scalp [18]. rTMS was applied over the DLPFC with the coil placed tangentially to the skull, with a handle pointing in the posterolateral direction. The junction point of the figure of eight was placed directly over the localized region. The stimulation intensity was set at 90% of the resting motor threshold (RMT), with a stimulation frequency of 20 Hz. Each series of continuous stimulation lasted for 2 s, followed by an interval of 28 s. In total, there were 40 stimulation series administered during each session. The total number of pulses per session was 1,600, and the parameter selection strictly adheres to the standards of international rTMS safety guidelines [19, 20]. RMT was defined as the minimum output stimulation intensity that evoked a small response (> 50 µV) in more than five of ten consecutive trials in the right first dorsal interosseous muscle [21].

Clinical scale assessment

Patients were evaluated using a standardized scale at the beginning of the study and again after completing a total of 14 treatment sessions. These scale scores were derived from a clinical neuropsychological assessment carried out by a physician who was not part of the TMS intervention, ensuring the objectivity of the evaluation. The assessment covered various domains critical to understanding the patients’ overall cognitive health and functionality. The evaluation comprised several specific assessment areas. Firstly, the overall cognitive function of patients was primarily gauged using the MMSE and the Montreal Cognitive Assessment (MoCA), both of which are established tools for assessing cognitive impairments. Secondly, to measure patients’ abilities in daily living, the Activities of Daily Living (ADL) scale was implemented, providing insights into their functional independence and everyday skills. Additionally, emotional and psychiatric symptoms were assessed using the HAMD for depressive symptoms, alongside the Neuropsychiatric Inventory (NPI), which provides a broader evaluation of psychiatric symptoms experienced by the patients. Cognitive functions such as memory and attention were examined through the Rey Auditory Verbal Learning Test (RAVLT), which assesses memory through components that capture immediate recall (RAVLT-I), delayed recall (RAVLT-D), and recognition (RAVLT-R). The attention levels of patients were measured using the Digit Span Test (DST), offering a comprehensive overview of their attentional capabilities. Furthermore, language function was scrutinized through the Boston Naming Test (BNT) and the Verbal Fluency Test (VFT), both of which are instrumental in evaluating expressive language skills. Lastly, executive function, which is critical for higher-level cognitive processes, was assessed through the Stroop Test and the Trail Making Test (TMT). The Stroop Test differentiates between various types of word-color interactions, including color point recognition, different color word recognition, and interference word color recognition, highlighting the complexities of cognitive control. The TMT is divided into two parts, A and B, where part A assesses visuospatial abilities while part B specifically measures executive functioning, providing a nuanced understanding of the patients’ cognitive capabilities.

Blood biomarker testing

A stringent inclusion protocol was implemented for blood biomarker analysis, selecting AD patients exhibiting a Hachinski Ischemic Score (HIS) ≤ 4 through validated vascular comorbidity screening. Longitudinal venous sampling was completed in 25 participants with paired pre-/post-stimulation phlebotomy procedures. Blood was collected through venipuncture into collection tubes, mixed by inversion, and allowed to rest at room temperature for 10 min. Subsequently, the samples were centrifuged at 1800 × g for 10 min at room temperature to achieve plasma separation. The upper plasma layer was carefully extracted and divided into 1.5 ml centrifuge tubes, with each tube containing a volume of 100–300 µl, and labeled accordingly. The aliquoted plasma is then systematically stored in a freezing box before being transferred to − 80 °C for long-term storage. To avoid multiple freeze–thaw cycles, we utilize individually aliquoted samples for testing. The processed samples are forwarded to the Ultra-sensitive Simoa platform for detection and analysis. Both pre- and post-stimulation levels of p-tau181, NFL, GFAP, Aβ42, Aβ40, and the Aβ42/Aβ40 ratio in the blood were recorded.

fMRI data acquisition and processing

Data acquisition

Resting-state functional magnetic resonance imaging (fMRI) data were prospectively acquired from AD patients before and after magnetic stimulation. All imaging data of the subjects were obtained from Tianjin Huanhu Hospital using a Siemens Prisma scanner. The subjects were instructed to close their eyes, relax, avoid falling asleep, and remain as still as possible. During the scanning process, structural three-dimensional T1-weighted images were acquired using a turbo field echo sequence with the following parameters: repetition time (TR) = 2300 ms, echo time (TE) = 2.98 ms, flip angle (FA) = 9°, field of view (FOV) = 220 × 220 mm, matrix resolution = 256 × 256, number of slices = 170, voxel size = 0.87 × 0.87 × 0.87 mm, no gap, and a total number of volumes = 240. The sagittal resting-state fMRI dataset of the entire brain was acquired using an echo-planar imaging sequence within 6 min, with the following parameters: TR = 735 ms, TE = 39 ms, number of slices = 64, FA = 52°, FOV = 88 × 88 mm, matrix size = 88 × 88, voxel size = 2.4 × 2.4 × 2.4 mm, and no gap.

Data preprocessing

All fMRI data were preprocessed using DPARSF version 5.4 (http://rfmri.org/DPARSF) [22]. The specific steps are as follows: (1) format conversion, converting the data from DICOM format to NIFTI format; (2) removal of the first 10 time points to reduce signal fluctuations caused by magnetic field instability at the beginning of the scan and the subjects’ initial adaptation to the environment; (3) slice timing correction, which ensures that images from the same time period are acquired simultaneously; (4) motion correction, excluding subjects with translations exceeding 2 mm or rotations exceeding 2°; (5) spatial normalization, aligning the local spatial data of different subjects to the MNI standard space for statistical analysis, with the normalized voxel size being 3 mm × 3 mm × 3 mm; (6) covariate removal, using linear regression to eliminate the effects of motion parameters, white matter, and cerebrospinal fluid signals; and (7) filtering, applying a band-pass filter to the data within the range of 0.01 to 0.1 Hz to reduce the influence of low-frequency drift and high-frequency noise. Smoothing was performed using a 6 mm × 6 mm × 6 mm full width at half maximum Gaussian kernel for spatial smoothing to enhance the signal-to-noise ratio.

Construction of Granger causality networks

This study utilized the Schaefer2018_100Parcels_17Networks atlas to obtain the gray matter volume (cm³) of 100 cortical parcels [23]. These parcels were categorized into 17 distinct brain functional networks (Fig. 2). The cortical volume for each brain network is calculated as the sum of the gray matter volumes of the corresponding brain regions. A large-scale Granger-weighted causal network was constructed both before and after stimulation, based on the 100 cortical parcels from the Schaefer template. The differences in their graph-theoretical properties were analyzed to explore the regulatory effects of rTMS on the connectivity of large-scale brain networks. Specifically, within each region of interest (ROI), the BOLD signals of all voxels were extracted and averaged. The Augmented Dickey-Fuller (ADF) test was used to test for unit roots in the time series of multiple ROIs to determine whether the data were stationary. If the test results showed the existence of a unit root, the data were processed by first-order differencing, and the ADF test was performed again on the differenced series to ensure that it met the stationarity requirement. A directed network connectivity among multiple ROIs was established based on GC. By calculating the intensity between paired signals and applying Bonferroni correction to compute significant GC values (P < 0.05, corrected P = 0.05/9900), invalid connections were removed, yielding a causality-weighted connectivity matrix (GC^d). The impact of rTMS on the integration of brain networks in AD patients was then examined based on connection density and global efficiency. Subsequently, the causal connectivity strength matrix was allocated to 17 cognitive systems: two visual, two somatomotor, two dorsal attention, two salience/ventral attention, one limbic, three frontoparietal, three default mode networks, and one temporoparietal system. This allocation facilitates further analysis of the functional connectivity characteristics among the 17 subnetworks within the large-scale brain networks, assessment of the functional states and interaction strengths of different subnetworks, and exploration of the impact of rTMS on the functional interactions among different subnetworks in AD patients. The distribution of the 17 subnetworks is illustrated (Fig. 2).

Fig. 2.

Schaefer’s 100 cortical regions. Note: The parcels are assigned to 17 resting-state networks, which are further grouped into 8 major networks: default mode (Default A/B/C), executive control (Cont A/B/C), somatomotor (SomMot A/B), visual (Viscent/Visperi), limbic (Limbic A/B), salience/ventral attention (Sal/VentAttn A/B), dorsal attention (DorsAttn A/B), and temporoparietal (TempPar)

Statistical analysis

A statistical analysis was conducted to examine the changes in cognitive function scale scores, neurodegeneration-associated peripheral blood biomarker levels, and brain network integration characteristics before and after rTMS intervention. Prior to comparison, perform normality tests on the data using the Shapiro–Wilk test and assess homogeneity of variance using Levene’s test [24]. For data that meet the criteria of normal distribution and homogeneity of variance, use paired t-tests, with statistical results denoted by t. For variables that do not meet the normal distribution criteria, employ non-parametric Wilcoxon matched-pairs tests for comparison, with statistical results denoted by Z, and r representing the effect size. In addition, to control for the risk of Type I error due to multiple comparisons, this study used a stratified correction strategy, correcting for statistical results of scales using the Benjamini–Hochberg method (FDR correction). The significance threshold for the preset primary outcome metrics was set at q < 0.05. During the construction of the weighted causal network, weak connections were removed by Bonferroni correction. When analyzing the correlations between changes in scale scores before and after stimulation, alterations in blood biomarker levels, and changes in brain network integration properties, Pearson correlation analysis is used for data that conform to a normal distribution, while Spearman correlation analysis is applied for data that do not conform to a normal distribution.

Results

Effects of rTMS on cognitive function

The regulatory effects of rTMS were observed in terms of overall cognitive function, daily living abilities, and neuropsychiatric symptoms (Fig. 3). After 14 days of stimulation, the scores for the MMSE (9.0% change, Z =  − 2.863^b, P = 0.004, q = 0.017 r = 0.523), MoCA (21.1% change, t =  − 6.137, P = 0.000, q = 0.000, r = 0.752), RAVLT-I (16.4% change, t =  − 3.436, P = 0.002, q = 0.011, r = 0.545) demonstrated significant improvement compared to baseline measurements. Additionally, the scores for the NPI (49.7% change, Z =  − 2.547^c, P = 0.011, q = 0.037, r = 0.465) and HAMD (53.7% change, Z =  − 3.472^c, P = 0.001, q = 0.009, r = 0.634) showed a significant reduction from baseline, indicating that the overall cognitive function and depressive symptoms of AD patients improved after rTMS. No statistically significant differences were observed in performance on the ADL, BNT, VFT, Stroop task, and TMT between pre- and post-intervention assessments (q > 0.05 for all comparisons).

Fig. 3.

Longitudinal changes in neuropsychological assessment outcomes. Abbreviation: MMSE = Mini-Mental State Examination, MoCA = Montreal Cognitive Assessment, ADL = Activities of Daily Living, NPI = Neuropsychiatric Inventory, HAMD = Hamilton Depression Scale, RAVIT = Rey Auditory Verbal Learning Test, VFT = Verbal Fluency Test, BNT = Boston Naming Test, DS = Digit Span Test, Stroop = Stroop Test, TMT = Trail Making Test. *P < 0.05, **P < 0.01

Effect of TMS on degenerative peripheral blood markers

The deviation between each subject’s biomarker content and the normal reference value was calculated, and the correlation between this difference, cognitive function scores, and the patient’s age was analyzed. P-tau181 and GFAP levels demonstrated significant inverse correlations with global cognitive function (R =  − 0.36, P < 0.05; R =  − 0.35, P < 0.05, respectively), whereas Aβ42 exhibited a positive association with MoCA scores (R = 0.29, P < 0.05) (Fig. 4a). Peripheral blood neurodegeneration-related biomarker levels were normalized before and after stimulation. Compared to pre-stimulation levels, NFL content in AD patients decreased significantly (Z =  − 2.371, P = 0.018, r = 0.474) (Fig. 4b), indicating a reduction of 9.39% prior to the stimulus (Fig. 4c). Degeneration-related peripheral blood markers other than NFL did not show significant pre- and post-stimulation differences (P > 0.05), Furthermore, a significant negative correlation was observed between baseline p-tau181 content and the change rate of Aβ42 content before and after stimulation (R =  − 0.428, P = 0.033, n = 25) (Fig. 4d).

Fig. 4.

Biomarker content with several key analyses. a The correlation analysis between baseline biomarker levels and cognitive function scores is presented, where the lower left corner of the heat map indicates the correlation values, and the upper right corner denotes the significance of these correlations. The color bar represents the correlation coefficients, while an asterisk (*) indicates the significance of the correlation. b Changes in biomarker content before and after stimulation are depicted, showcasing the normalized results of biomarker levels pre- and post-stimulation. c The rate of change in biomarker content after stimulation, in comparison to baseline levels, is presented. d A correlation analysis between baseline p-tau181 content and the change in Aβ42 content before and after stimulation is shown, with significance indicated by * P < 0.05

Restructuring effects of rTMS on resting-state fMRI network connectivity

The impact of rTMS on functional network connectivity in AD patients

Based on the “Construction of Granger causality networks” section, the Granger causality matrices before and after stimulation (Fig. 5a and b), as well as the Granger-weighted causality matrices (Fig. 5c and d), were constructed and visualized (Fig. 5e and f). The results indicated that, compared to the pre-stimulation phase, rTMS modulation resulted in intensified information transmission between brain regions and an expanded range of information transmission. The connection density (Fig. 6a) and global efficiency (Fig. 6b) were further calculated to assess the integration ability of the brains of AD patients before and after stimulation. It was observed that, compared to the pre-stimulation measurements, both connection density (t =  − 1.865, P = 0.081, r = 0.423) and global efficiency (Z =  − 1.823, P = 0.068, r = 0.442) exhibited an increasing trend following stimulation. This finding indicates that rTMS enhanced the information integration ability of AD patients. Furthermore, the increase in global efficiency was positively correlated with the rise in Aβ42/40 levels (R = 0.596, P = 0.025, n = 14) (Fig. 6c), suggesting that changes in global integration ability are related to the pathological alterations in AD patients. Additionally, the increases in global efficiency (R = 0.486, P = 0.048, n = 17) (Fig. 6d) and connection density (R = 0.514, P = 0.035, n = 17) (Fig. 6e) were both associated with baseline MoCA scores, indicating that the improvement in information integration ability among AD patients is linked to their baseline cognitive function.

Fig. 5.

The construction of the average Granger causal network both before and after stimulation. a The original Granger network prior to stimulation. b The original Granger network following stimulation. c The Granger-weighted causal network before stimulation. By calculating the GC intensity between paired signals and applying Bonferroni correction to determine significant GC values (P < 0.05, corrected P < 0.05/9900), invalid connections were eliminated, resulting in a causality-weighted connectivity matrix (GC^d). d The Granger-weighted causal network after stimulation. e The visualization of the Granger-weighted causal network before stimulation. f The visualization of the Granger-weighted causal network after stimulation

Fig. 6.

The changes in brain attributes before and after stimulation, along with their correlations to cognitive functions. a The comparison of brain connection density prior to and following stimulation. b The comparison of the global efficiency of the brain before and after stimulation. c A correlation analysis is conducted between the change in global efficiency and the change in Aβ42/40 levels. d The correlation between the change in connection density and the baseline MoCA score is examined. e The correlation between the change in global efficiency and the baseline MoCA score

Reconstruction of the directional functional connectivity within the network

Furthermore, 100 regions were divided into 17 functional networks, and the effective connections within each network were visualized. It was found that rTMS enhanced the inter-hemispheric information flow within the network. Further calculations of the inter-hemispheric index within different functional networks, conducted before and after stimulation, revealed that, compared to the pre-stimulation phase, the inter-hemispheric information index of Visual-Central (VisCent) (Z =  − 2.014, P = 0.044, r = 0.489), the limbic system (Z =  − 2.000, P = 0.046, r = 0.555), and DefaultB (Z =  − 2.121, P = 0.034, r = 0.515) significantly increased following rTMS regulation (Fig. 7). These results indicate that rTMS promotes inter-hemispheric information flow in the VisCent, limbic system, and DefaultB. This may represent one of the mechanisms underlying the enhancement of information integration ability in patients with AD.

Fig. 7.

The analysis of effective connectivity within the network and interhemispheric indices both before and after stimulation. Notably, the LimbicB network exhibited no effective connections detected. The functional connection patterns of the remaining 16 networks are comprehensively depicted. The lower right of this figure highlights the dynamic changes in the cross-hemispheric index under stimulation conditions. This index is defined as the ratio of intra-hemispheric connection density to inter-hemispheric connection density (intra-hemispheric density/inter-hemispheric density)

The reconstruction of the directed functional connectivity between networks

To explore how TMS affects the information flow between networks, we first visualized the information flow before and after stimulation (Fig. 8a and b). It was observed that the connection density between networks increased following stimulation. Further analysis of the changes in information flow for each network revealed that, compared to pre-stimulation levels, the information inflow of the temporal-parietal network (TempPar) was significantly enhanced (Z =  − 2.223, P = 0.026, r = 0.539) (Fig. 8c). Additionally, the information outflow of the VisCent (t =  − 2.522, P = 0.023, r = 0.533) and the Dorsal Attention A (DorsAttnA) network (Z =  − 2.722, P = 0.006, r = 0.660) was also significantly increased (Fig. 8d). Notably, the Visual-Peripheral (VisPeri) network transitioned from information inflow to information outflow (t = 2.286, P = 0.036, r = 0.496), while the TempPar network shifted from information outflow to information inflow (t = 2.129, P = 0.490, r = 0.470) (Fig. 8e). Furthermore, the betweenness centrality of the DefaultB was significantly enhanced compared to pre-stimulation measurements (Z =  − 2.489, P = 0.013, r = 0.604) (Fig. 8f). Furthermore, it is proposed that the dorsolateral attention network, the temporal-parietal network, and the default network exhibit significant changes in information flow before and after stimulation. An exploration of the alterations in information flow among these three networks reveals that, compared to the pre-stimulation phase, the information flow observed is as follows: DorsAttnA → TempPar → DefaultC (Fig. 8g, h, and i).

Fig. 8.

The regulatory effect of TMS on the information flow between networks in AD patients. a The connections between networks prior to stimulation; b the connections between networks following stimulation; c the amount of information inflow for the network before and after stimulation; d the amount of information outflow for each network before and after stimulation; e the changes in information flow for each network before and after stimulation; f the betweenness centrality of each network before and after stimulation; g the information flow among the three networks prior to stimulation; these networks include the dorsal attention network, the temporoparietal network, and the default network. h The information flow among the networks following stimulation; i the changes in information flow among the networks before and after stimulation

Discussion

To investigate the clinical efficacy of rTMS in patients with AD, this study provides comprehensive evidence that high-frequency rTMS targeting the DLPFC exerted multifaceted therapeutic effects in AD, including cognitive enhancement, alleviation of neuropsychiatric symptoms, and attenuation of neurodegenerative biomarkers. Specifically, the results showed that 20 Hz rTMS significantly enhanced both overall cognitive function and neuropsychiatric symptoms in AD patients. Notably, however, no significant post-intervention changes were observed in certain cognitive domains (ADL, BNT, VFT, Stroop, and TMT). This finding is consistent with a recent meta-analysis published this year [25]. ADL reflects patients’ ability to perform daily activities. Since the enrolled cohort predominantly consisted of mild AD patients with less severe baseline impairment in daily functioning, this may have constrained observable intervention improvements due to a ceiling effect. Furthermore, ADL serves as a composite indicator integrating multidimensional cognitive functions. Although rTMS enhances brain network information integration, our findings demonstrate that its neuromodulatory effects primarily target the dorsolateral attention network—temporoparietal network—default mode network pathway—a circuit more directly associated with improvements in memory and neuropsychiatric symptoms in AD. Consequently, rTMS-induced cognitive improvements exhibit domain specificity, resulting in limited enhancement of comprehensive ADL scores. Regarding language assessments (BNT and VFT), most enrolled AD patients (except rare cases) presented without severe language dysfunction at baseline. The ceiling effect likely contributed significantly to the non-significant pre-/post-intervention score differences. In addition, the Stroop and TMT were primarily used to assess executive function. The lack of significant improvement in executive function may be attributed to insufficient intervention duration, which resulted in inadequate remodeling of brain networks induced by rTMS to yield substantial improvement. Moreover, to ensure the rigor of the results, we conducted strict FDR correction on all scale scores, which may explain why some scores did not reach statistical significance despite showing a trend of improvement. Alongside these cognitive improvements, a 9.4% reduction in NFL levels was observed. NFL, a critical component of neuronal cytoskeletons [26], is predominantly released upon neuronal damage or death [27], establishing its levels as robust indicators of brain neuronal injury [28]. Collectively, these findings demonstrate that rTMS slowed the progression of neuronal damage, providing neurobiological-level validation of its therapeutic efficacy in AD at the synaptic level.

The neuroprotective effects observed at the synaptic level were subsequently investigated for their relationship with global network integration mechanisms. Our findings establish a connection between synaptic dysfunction and large-scale network reorganization. GC analysis demonstrated that rTMS improved global network integration, as indicated by increased connection density and global efficiency. AD patients exhibit reduced functional connectivity and degenerative changes within large-scale brain networks. The findings of this study indicate that rTMS can enhance global integration capabilities in AD patients. Notably, the correlation observed between the gains in global efficiency and the elevation of the Aβ42/40 ratio suggests a synergistic relationship between network remodeling and the metabolism of Aβ. The latest advancements in biomarker research have revealed a significant correlation between abnormal plasma Aβ levels in the AD population and Aβ pathological changes measured by cerebrospinal fluid testing and Aβ-PET imaging [29]. These findings not only provide crucial evidence for plasma Aβ42/Aβ40 as a biomarker for early Aβ pathology but also offer a molecular-level explanatory framework for the innovative discoveries in this study. This experiment observed a notable correlation between changes in the integration properties of neural networks following repetitive transcranial magnetic stimulation intervention and variations in the Aβ42/40 ratio. This association not only validates the effectiveness of rTMS in improving AD pathology at the neural network level but also reveals that TMS may exert its therapeutic effects through a dual mechanism of “network remodeling and molecular metabolism.”

We further investigated the information flow within and between 17 networks to explore how rTMS affects information integration in patients with AD. Our findings indicate that rTMS enhances interhemispheric information flow predominantly within networks associated with the visual network, limbic system, and DMN. Furthermore, rTMS significantly increases the betweenness centrality of the DMN in AD patients. The DMN, which is implicated in self-referential cognition and memory consolidation, has been extensively documented as a core network disrupted by the pathology of AD [30]. Patients with Mild Cognitive Impairment (MCI) and early AD exhibit reduced functional connectivity within the DMN, with the loss of connectivity associated with deficits in episodic memory [31]. Longitudinal studies indicate that the disintegration of the DMN parallels the accumulation of Aβ as disease severity increases [32]. In contrast, our study demonstrates that intervention can counteract this pathological trajectory, suggesting that this remodeling of the DMN may underlie improvements in episodic memory and psychiatric symptoms. We further investigated how targeted rTMS of the DLPFC reorganizes the DMN by examining the changes in information flow among the dorsal attention network, the temporoparietal network, and the DMN. Our analysis revealed that, following stimulation, the information flow among these three networks transitioned from the dorsal attention network to the temporo-parietal network and subsequently to the default mode network. The effects of rTMS involve the selective enhancement of the “DorsAttenA- TempPar- DefaultC” cascade, a tri-network axis that is crucial for integrating attentional control, multisensory processing, and self-referential cognition. This result not only verifies the regulatory potential of rTMS on local network nodes but also confirms its enhancing effect on cross-network information integration, thereby providing direct evidence for the regulatory mechanism of rTMS in the treatment of AD.

The therapeutic response was modulated by the baseline pathophysiological status. Baseline p-Tau181 levels inversely associated with Aβ42 increases post-rTMS, which might reflect preserved neuroplasticity in early-stage AD patients. This hypothesis aligns with evidence that early tau pathology correlates with synaptic dysfunction [33]. Furthermore, higher baseline MoCA scores predicted greater gains in network efficiency, consistent with the cognitive reserve theory, the brain’s ability to optimize network resource allocation despite pathology [34]. This triad of biomarker dynamics, neurophysiological remodeling, and cognitive stratification suggests that patients with milder cognitive impairment exhibit greater neuroplastic potential, facilitating more robust network reorganization in response to rTMS. These findings support the implementation of precision neuromodulation strategies that are tailored to the disease stage and the residual capacity of neural networks.

In summary, rTMS may slow the process of neuronal damage while synergistically enhancing the integration capabilities of brain networks and the metabolism of amyloid proteins, which could contribute to slowing AD progression. The effects of rTMS involve a selective enhancement of the dorsal attention network and the temporoparietal region, as well as the DMN cascade. This is accompanied by increased interhemispheric information flow and betweenness centrality in the DMN, which may underlie improvements in episodic memory and attention control. Additionally, the response to rTMS treatment is influenced by baseline pathophysiological states. This investigation establishes an innovative methodological framework that systematically combines standardized neurocognitive assessments, blood biomarker profiling, and brain functional network analysis to elucidate the neurophysiological mechanisms underlying rTMS-mediated therapeutic effects in AD pathology.

Although the current results are promising, several limitations must be considered. Because a sham rTMS stimulation group was not set up, it is difficult to completely exclude the practice effect from interfering with the results in multiple cognitive assessments. Multiple repetitions of practice on the same cognitive test task may result in higher test scores [35]. Although it has been shown that cognitive scores improved less in the control group (sham group) than in the stimulation group, and even that there was no significant difference in pre- and post-stimulation changes [36], this practice effect may still confound the true effects of rTMS interventions with changes resulting from repeated practice, thus increasing the difficulty of accurately assessing the true effects of rTMS on cognitive functioning. Of note, although sham rTMS stimulation may result in fluctuations in cognitive test scores due to the practice effect, the intervention did not substantially affect functional brain connectivity patterns or changes in blood marker levels in AD patients. This conclusion suggests that, despite the interference of the practice effect in assessing changes in cognitive function, the observation of rTMS on the remodeling of brain functional networks and the modulation of pathological metabolic indexes in AD patients in the present study not only provides a new empirical basis for the intervention of rTMS in AD, but also helps to reveal the potential mechanisms underlying the association between neurological and metabolic changes in the pathological process of AD. Furthermore, the absence of long-term follow-up data may limit comprehensive assessment of rTMS treatment durability and its potential for long-term therapeutic application. Notably, existing studies have demonstrated that rTMS exerts treatment effects on cognitive function in AD persisting beyond two months [37]. Future research should implement multicenter randomized controlled trials featuring extended follow-up periods and integrated longitudinal biomarker monitoring to elucidate rTMS’s sustained efficacy and mechanisms of action in neurodegenerative diseases.

Author contribution

PW, MMG, and GZX designed and administrated this study, and also reviewed the writing. All authors participated in data collection. TW made the analysis of data and drafted the paper. SL, YJ, and HW were responsible for keeping statistics on cognitive scores and biomarker content. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the National Key R&D Program of China (No. 2022YFC2402200), the Natural Science Foundation of Hebei Province, China (No. F2024202085), the Funds for International Cooperation and Exchange of the National Natural Science Foundation of China (No. 52320105008), Tianjin Health Science and Technology Program of China (TJWJ2022MS032), the Natural Science Foundation of Hunan Province (No. 2025JJ70517), and the Cross-disciplinary graduate program of HEBUT (No. XKJC-2024009).

Data availability

The datasets used during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study received approval from the Institutional Review Boards of Hebei University of Technology, Tianjin, China (No: HEBUThMEC2024009) and written informed consent was obtained from all participants before enrollment.

Consent for publication

All participants provided written informed consent for aggregate, de-identified study results to be published in peer-reviewed journals and to be presented at conferences or other professional forums.

Competing interests

The authors declare no competing interests.