Participants
The current study was approved by the Ethics Committee of Nanjing Drum Tower Hospital, and written informed consent was obtained from all patients before entering the study. Twenty-six patients, admitted to the Neurology Department in Drum Tower Hospital of Medical School, Nanjing University, were screened for the current study. Two patients were excluded because of excessive head movement during MRI scanning and two patients were excluded because of loss of imaging data. These participants were composed of MCI and AD patients. The AD, in the presence of AD pathology as supported by cerebrospinal fluid or other imaging biomarker, was diagnosed based on the National Institute of Neurological and Communicative Disorders and Stroke and the AD and Related Disorders Association (NINCDSADRDA) and the Diagnostic and Statistical Manual of Mental Disorders IV criteria (DSM-IV) guidelines[20]. The MCI patients included in this study were diagnosed according to the recommendations of Petersen and described as follows[21]: (1) memory complaint confirmed by the subject and/or an informant; (2) objective cognitive performance documented by an auditory verbal learning test-delayed recall (AVLT-DR) scores below or equal to 1.5 SD of education- and age-adjusted norms; (3) clinical dementia rating (CDR) score = 0.5; (4) the scores for the Mini-Mental State Examination (MMSE) ≥ 24; and (5) not sufficient to dementia according to NINCDS-ADRDA and DSM-IV. Exclusion criteria included brain tumors, epilepsy, Parkinson's disease, serve anxiety and depression, thyroid dysfunction or other neurological or psychiatric disorders which can cause memory loss. Participants were excluded if the MRI scans evidenced significant vascular pathology or micro bleeds, or head motion artefacts that affect T1w3d quality and segmentation. Four patients were excluded for these reasons.
Experiment design and Neuro-navigated rTMS
In the first visit, patients underwent a complete clinical investigation, including medical history and neurological examination, a neuropsychiatric evaluation, brain MRI scanning, and an extensive neuropsychological assessment exploring all cognitive domains. The region, which exhibited significant functional connectivity differences among healthy controls, MCI and AD participants, calculated by our team was located at the left angular gyrus (MNI: -45, -67, 38). The region was calculated by seed-based functional connectivity analysis using the left hippocampus as a seed. All the patients were stimulated the angular gyrus by the Neuro-navigated rTMS for four weeks. rTMS was applied daily at the same 5 times per week. Neuropsychological measurement and brain MRI scanning were performed again after four weeks rTMS treatment. Details of the study design are summarized in Fig. 1.
rTMS was delivered using a commercially available magnetic stimulator (CCY-IV model; YIRUIDE Inc., Wuhan, China) with a 70‐mm figure eight coil and an electromyography device. Each stimulation session consisted of forty circulations of 2 second delivered at 20 Hz spaced-out by 28 s of no stimulation, for a grand total of 1600 stimulations. During the rTMS treatment, the coil was set on the angular gyrus was constantly motored using a navigation system, which was anatomically referred by individual T1-weighed MRI volumes. The treatments lasted about 20 minutes. Intensity of stimulation was set at 100% of the resting motor threshold (RMT), defined as the lowest intensity producing MEPs of > 50 µV in at least five out of 10 trials in the relaxed first dorsal interosseous (FDI) muscle of the right hand. RMT was assessed over the optimal stimulus site to elicit MEPs in the right FDI, which was considered motor spot. For each patient, a source estimation on pre-processed TMS data was run at the beginning of each treatment session to confirm the correct anatomical targeting for rTMS.
MRI scanning
All participants were examined on a Philips 3.0-T scanner (Philips Medical Systems). The examination protocol included the high-resolution T1-weighted turbo gradient echo sequence (repetition time [TR] = 9.8 ms, flip angle [FA] = 8°, echo time [TE] = 4.6 ms, FOV = 250 × 250 mm2, number of slices = 192, acquisition matrix = 256 × 256, thickness = 1.0 mm), the FLAIR sequence (TR = 4.500 ms, TE = 333 ms, time interval [TI] = 1.600 ms, number of slices = 200, voxel size = 0.95 × 0.95 × 0.95 mm3, acquisition matrix = 270 × 260), and the diffusion-weighted imaging sequence (TR = 9.154 ms, TE = 55 ms, acquisition matrix = 112 × 112, FOV = 224 × 224 mm2, thickness = 2.5 mm, voxel size = 2 × 2 × 2.5 mm3, the number of gradient directions = 32 (b = 1000 s/mm2) and one b0 image).
Neuropsychological Measurement
To evaluate the behavioral effects of the rTMS treatment, we employed a standardized neuropsychological test protocol, including global cognitive assessments and multiple cognitive domain examinations. We also completed the Clinical Dementia Rating Scale (CDRS) to assess the degree of cognitive impairment of the participants. Global cognitive function was evaluated by Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment Beijing (MoCA-BJ). The raw test scores were converted to Z-scores, which were used to calculate the compound cognitive index. Episodic memory was calculated as the mean of the Z-scores from Auditory Verbal Learning Test-delayed recall (AVLT-DR) scores and the Wechsler Memory Scale-Visual Reproduction-delayed recall (VR-DR). Information processing speed was calculated as the average Z-scores of the Trail Making Test-A (TMT-A) and the Stroop Color and Word Tests A and B (Stroop A and B). The language function consisted of the Boston Naming Test and Category Verbal Fluency test. Executive function is a compound score of the average Z-scores of the Digit Span Test-backward, Trail Making Test-A (TMT-B) and Stroop Color and Word Tests C (Stroop C). Visuospatial function is a compound score that includes the mean of the Z-scores of the Clock Drawing Test and Visual Reproduction–copy test.
Multimodal magnetic resonance image preprocessing
In recent years, more and more neuroimaging studies suggested that white matter alterations may be an important pathophysiological feature and a potential target of AD[22]. However, whether patterns of white matter change in different fiber tracts are different and what happens to the white matter after the intervention are still largely unknown[13]. We decided to use AFQ, applying deterministic tractography approach, to reconstruct whole-brain white matter and analyze point-wise diffusion parameters in specific fiber tracts. AFQ can not only trace the fiber tracts associated with the thalamus, but also analyze the important tracts in the brain, such as corticospinal tract, cingulate fasciculus, uncinate fasciculus, and arch fasciculus and so on to provide a comprehensive detection for whole-brain[23]. However, the changes in the microstructure of the white matter tracts are not necessarily consistent with alterations in the brain's complex networks[24] and it is better to combine multi-modality data to detect complex network changes in AD patients then a single modality[25, 26]. It is effective to quantify the complex brain network topology by graph theory using rs-fMRI to build functional network[27, 28]. As a result, we combined DTI and rs-fMRI to better explore thalamus and related network alteration after treatment.
For diffusion images, the data preprocessing was carried out by FSL 5.0.9 software (Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, University of Oxford; https://www.fmrib.ox.ac.uk/fsl/). The preprocessing included the following steps: DICOM-to NIfTI format conversion, registering DWI images (b = 1000 s/mm2) to the non-DWI image (B0), eddy current and head motion correction, and then nonbrain tissue exclusion. After preprocessing, using DTIFIT command of FSL to obtain the whole brain images of diffusion metrics, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (DA) and radial diffusivity (RD), Specific calculation indexes are as follows:
FA = \(\sqrt{3{(\left({\lambda }_{1}-\stackrel{-}{\lambda }\right)}^{2}+{\left({\lambda }_{2}-\stackrel{-}{\lambda }\right)}^{2}+{\left({\lambda }_{3}-\stackrel{-}{\lambda }\right)}^{2})}/\sqrt{2\left({\lambda }_{1}^{2}+{\lambda }_{2}^{2}+{\lambda }_{3}^{2}\right)}\),MD = \(\left({\lambda }_{1}+{\lambda }_{2}+{\lambda }_{3}\right)/3\), DA = \({\lambda }_{1}\), RD = \(({\lambda }_{2}+{\lambda }_{3})/2\), λ1, λ2 ,λ3 reflect the three dispersion directions of water molecules[13]. λ1 travels in or against the direction of the fiber bundle of the voxel, which called axial direction and λ2, λ3, corresponds to the direction perpendicular to the axis, which called radial direction.
For rs-fMRI images, the data were preprocessed by GRETNA, a graph theoretical network analysis toolbox for imaging connectomes[29]. During the preprocessing, the first 10 volumes for signal were removed to reach a steady state, leaving 220 functional volumes for each participant. The remaining functional volumes were corrected for acquisition time delay between slices (slice timing) and head motion between volumes (realignment). Then, these functional data were normalized to the T1 segmentation individually and spatially smoothed with a Gaussian kernel (full width at half-maximum of 4 mm). We regressing out covariates (white matter, cerebral spinal fluid, global signals, and head-motion profiles) by multiple regression analysis to avoid noise signals. Other steps in preprocessing consisting of temporally linear detrending, temporal band-pass filtering (0.01–0.1 Hz), and scrubbing to reduce the effects of head motion on rs-fMRI data. The network construction was based on a voxel or region of interest approach. The Human Brainnetome Atlas was used to parcellate the brain into 246 regions. All network analyses were performed using GRETNA.
Automated fiber quantification procedure
We identified 20 major tracts in whole brain and further quantified the diffusion metrics along the tract trajectory by applying the AFQ package. This is a description of AFQ steps in this result:
(1) Fiber Tract Identification
First, 3D T1-weighted images were co-registered into the b0 image for each participant based on FSL, and poorly aligned images were excluded by visual evaluation. Second, using deterministic tractography and a fourth-order Runge–Kutta path integration method[30] to perform whole-brain tractography with thresholds of turning angle < 30° and FA > 0.2. The tracking procedure generates a database of candidate fibers in the whole-brain, which can be broken down into anatomically defined bundles; Third, based on the waypoint ROI procedure described in Wakana et al[31], fiber tract is segmented. In this procedure, if they pass through two waypoints defined by ROI of AFQ, fibers are assigned to a specific fiber group. Fourth, by comparing each candidate fiber to fiber tract probability maps, the fiber tract refinement is accomplished. Each fiber conforms to the shape of the tracts defined by the fiber tract probability maps
(2) Fiber Tract Cleaning.
Due to the noise in the data, areas with complex fiber orientation and ambiguous stopping criteria, a few fibers may differ from the rest of the fiber group. The fibers were resampled to 100 equidistant nodes firstly and the fiber tract core is calculated as the mean of each fibers x, y, z coordinates at each node. The spread of fibers in 3-dimensional space is calculated by computing the covariance between each fiber’s x, y, z coordinates at 100 nodes. Thus, each node on the tract is represented as a mean coordinate, m, and a 3 by 3 covariance matrix, S. Then we can calculate its Mahalanobis distance Dm(x). The specific formula is as follows: Dm(x)=\(\sqrt{{\left(x-\mu \right)}^{T}{s}^{-1}{\left(x-\mu \right)}^{T}}\)
Dm(x) corresponds to the probability that a given point belongs to the distribution. Abnormal fibers are removed if fibers deviate substantially from the average position.
(3) Fiber Tract Quantification
The fiber group is clipped to the central portion that spans between the two defining ROIs and each fiber was resampled to 100 equally spaced nodes. The properties such as DA, FA, MD and RD at each fiber node are summarized by a weighted average of the diffusion properties. This probability is calculated based on the fiber’s Mahalanobis distance from the fiber tract core.
The identified 20 WM tracts in the whole brain are listed in supplementary Table 1. It did not succeed in identifying 20 white matter tracts per participant because of the strict criteria applied by AFQ in the identification of white matter tracts. We excluded 3 fiber tracts, right arcuate fasciculus (AF) and the bilateral cingulum hippocampus (CH) which largely unidentified. Only the remaining 17 fiber tracts would be analyzed in further study. The thresholds were set at a p < 0.05.
Network parameter analysis
Graph theoretical analysis was performed on the interregional connectivity matrix by using GRETNA. The weight network properties were calculated under the threshold set by network sparsity with a range of 0.05–0.5 step size of 0.05. GRTNA was used to calculate the global network metrics including global efficiency, global clustering coefficient (Cp), characteristic path length (Lp), and nodal network metrics including node degree centrality (DC), nodal global efficiency (Ne) and nodal shortest path (Nlp). The calculating formula and descriptions of these topological properties for a network G with N nodes and V edges are as follows[32]:
-
Characteristic Lp at the level of network is an indicator of overall network connectedness and quantifies the parallel information propagation ability, which can be calculated as Lp(G) = \(\frac{1}{1∕N\left(N-1\right){\sum }_{i=1}^{N}{\sum }_{j\ne 1}^{N}1∕{L}_{ij}}\) Lij is the characteristic Lp between nodes i and j.
-
Eg is defined as the inverse of the harmonic mean of shortest path between each pair of nodes within the network, which effectively measures the information communication capacity of the whole network and is calculated as Eg (G) = \(\frac{1}{N\left(N-1\right)}\sum _{i\ne jϵG}\frac{1}{{d}_{ij}}\), dij is the shortest Lp between node i and j in the network.
-
Cp at the network level represents the degree of local cliquishness or interconnectedness within the network. It can be calculated as Cp(G) =
-
Node degree centrality is number of links connected to a node. It can be represented as DC =\(\sum _{j\in N}{d}_{{ij}_{}}\)
-
Nodal Lp (Nlp) quantifies the mean distance or routing efficiency between one node and all the other nodes in the network, which calculated as: Lp (i) = \(\frac{1}{N-1}\sum _{i\ne j\notin G}dij\), dij is the shortest Lp between node i and j in the network.
-
Nodal efficiency means the efficiency of parallel information transfer of one node in the network. It can be calculated as: Ne (i) = \(\frac{1}{N-1}\sum _{i\ne j\notin G}\frac{1}{dij}\), dij is the shortest Lp between node i and j in the network.
Statistical analyses
To examine the point-wise difference of white matter tracts between baseline and post-treatment, we sorted DTI metrics (FA, MD, DA, and RD) of 100 nodes along each white matter tract calculated by AFQ in all patients. Then, paired T test was used to detect the differences in the DTI metrics of each fiber tract. Paired T tests were performed in Gretna's nodal metric comparison toolkit and false discovery rate (FDR) was applied to determine the significance for p-values (p < 0.05). Within each fiber, we only chose more than or equal to three adjacent nodes corrected by FDR to further analyze [26]. Differences of whole-network and nodal properties in functional network between pretherapy and post-treatment was performed in Gretna toolbox using paired samples T test and FDR corrected p < 0.05.
Baseline and post-treatment cognitive assessment were compared using paired-samples T test in SPSS software (Version 22). We divided the participants into the AD group and MCI group. We would conduct data analysis from the perspective of the whole participant, AD group and MCI group respectively. In order to investigate possible relationships between alteration of white matter fiber and cognition change, we tested correlations using the Spearmen coefficient (two-tailed) between the altered diffusion metrics and cognitive change (calculated by using post-treatment data minus the baseline data). The thresholds were set at a p < 0.05.