rTMS modulates precuneus-hippocampal subregion circuit in subjective cognitive decline

Background: Hippocampal subregions (HIPsub) and their network connectivities are considered to be abnormal in subjective cognitive decline (SCD). This study aimed to identify whether repetitive transcranial magnetic stimulation (rTMS) could ameliorate the HIPsub network connectivity by modulating one node of HIPsub network in SCD subjects. Methods: In the rst cohort (38 SCD and 55 healthy controls), we analyzed three HIPsub (i.e., hippocampal emotional, cognitive, and perceptual regions: HIPe, HIPc, and HIPp) functional connectivity to identify HIPsub network connectivity alterations that relate to SCD. And we further applied a support vector machine (SVM) approach using the alterations to identify how well this could distinguish SCD from CN. In the second cohort (13 SCD), rTMS with 5-day of once-daily for 2 weeks was used to modulate the altered HIPsub network connectivity in a sham-controlled design. Results: SCD subjects showed differently altered patterns of HIPsub network connectivity compared to controls. The SVM classier showed that the abnormalities had a high power to discriminate SCD from CN, with an ACC of 86.0%, an AUC of 92.9%, 83.8% sensitivity, and 89.1% specicity. Restoration of HIPc connectivity with left parahippocampa gyrus and HIPp connectivity with left middle temporal gyrus corresponded to amelioration of episodic memory in SCD patients after 2 weeks of rTMS. Conclusion: rTMS can restore the posterior hippocampus connectivity by modulating the precuneus in SCD subjects. Correction of breakdown in HIPc and HIPp related to cognitive and perceptual processing can simultaneously ameliorate episodic memory in SCD subjects. Thus, these ndings suggest that rTMS manipulation of precuneus-hippocampal circuit may prevent the disease progression by improving memory as the earliest at-risk state of AD in clinical trials. study provides a novel experimental evidence on correction of breakdown in posterior hippocampus (HIPc and HIPp) related to cognitive and perceptual processing by modulating the precuneus in SCD subjects. Furthermore, rTMS manipulation may the disease progression by improving memory as earliest at-risk state of AD in clinical treatment trials. It further suggests that the precuneus-HIPsub circuit may be used as a useful target circuit for SCD subjects to design rationale for

Neuroimaging evidences have consistently indicated a functional heterogeneity in hippocampus subregions (HIPsub) [13][14][15]. A recent neuroimaging meta-analytic study has de ned the left hippocampus as consisting of the anterior emotional region (HIPe), the middle cognitive region (HIPc), and the posterior perceptual region (HIPp) based on its neurofunctional topography [13]. Furthermore, several studies have indicated that HIPsub have a selective topography of pathological involvement in preclinical AD [12,15,16]. Recently, several studies have reported that SCD present the structural and functional alterations of hippocampus [3,[10][11][12]16]. Converging evidences suggest that SCD may present differently altered patterns of HIPsub network connectivity. Consequently, there is growing interest in answering the question how the dysfunctions in HIPsub network connectivity can be restored.
Neuromodulation techniques, including repetitive transcranial magnetic stimulation (rTMS), afford an opportunity to address this challenge by modulating intrinsic connectivity networks. Recently, rTMS has been broadly applied to study the changes across cortical networks [17][18][19][20]. Local stimulation of an accessible network node can be transmitted across synapses means to remote interconnected nodes with high spatial speci city [18,21,22]. Therefore, this approach can allow to establish a causal link between the applied stimulation and the observed changes in HIPsub network connectivity.
Recently, a neuromodulation study has veri ed that rTMS targeting the parietal cortex can improve hippocampal connectivity networks, and simultaneously improve associative memory performance in healthy individuals [20]. A recent MCI neuromodulation study applies rTMS to promote the improvement of episodic memory by targeting the precuneus [23]. The precuneus is well-known thought as a remote interconnected node of hippocampal intrinsic connectivity networks [20,24] and a critical vulnerability area for the episodic memory de cit observed in early AD [25,26]. Based on above-mentioned evidences, it is reasonable to speculate that the altered HIPsub network connectivity can be causally effected by rTMS modulation upon the precuneus in the HIPsub network in SCD subjects.
In this study, we propose a strategy to empirically discover a HIPsub pathological circuit related to SCD using a pattern classi cation approach (SVM: support vector machine), and subsequently, in a targeted manner, this circuit was experimentally manipulated to assess causal links in a separate cohort. We hypothesized that SCD subjects would display differently altered patterns of HIPsub network connectivity. And we further hypothesized that breakdown in HIPsub circuit related to episodic memory processing could be restored by rTMS modulation upon the precuneus in the HIPsub network in SCD subjects. Figure   S1 shows the data analysis pipeline conducted in this study.

Materials And Methods
Participants Data used in this study were obtained from our in-home database: Nanjing Brain Hospital-Alzheimer's Disease Spectrum Neuroimaging Project (NBH-ADsnp) (Nanjing, China), which is constantly being updated. Relevant information of NBH-ADsnp is summarized in SI Methods S.2.
Network discovery of altered HIPsub related to SCD A total of 99 elderly individuals participated in this study. Of the subjects, we excluded 2 healthy controls (CN) and 4 SCD subjects due to excessive head movement (see quality assurance section below), and incomplete or missing MRI data. The nal analyses included 55 CN and 38 SCD eligible subjects. SI Methods S.2 shows detailed inclusion and exclusion criteria.
Network validation of altered HIPsub related to SCD with rTMS A total of 20 SCD subjects participated in the clinical trial (No. ChiCTR2000034533) in this study from the NBH-ADsnp database.
We used rTMS (or sham) with 5-day of once-daily to modulate the precuneus of SCD subjects for 2 weeks in a sham-controlled design. Clinical measures, neuropsychological assessments, and MRI data were collected at baseline (pre-rTMS or sham intervention) and at the end of 2 weeks of rTMS or sham. A total of 20 SCD subjects were enrolled in the study, of which 16 SCD subjects were randomly divided into real rTMS (8 SCD) or sham (8 SCD), and 13 SCD subjects (8 SCD for real rTMS, 5 SCD for sham rTMS) completed the trial of 2 weeks of rTMS.

Neuropsychological assessment
Neuropsychological assessments are summarized in SI MethodsS.3. This study performed a standardized clinical interview and comprehensive neuropsychological assessment to evaluate general cognitive function, executive function, information processing speed, episodic memory, and visuo-spatial function.

MRI data acquisition
Detailed MRI data acquisition parameters in NBH-ADsnp are summarized in SI MethodsS.4. fMRI data preprocessing In this study, MATLAB2015b and DPABI software [27] were used to preprocess all fMRI data. The image processing procedure was performed as described in a previous study [28] and is summarized inSI Methods S.5.
Quality assurance (QA) Brain atrophy effect Given that signi cant hippocampal GM atrophies in SCD subjects have been reported [12,29], the anatomical differences between groups may affect these differences on the FCs of HIPsub. To clarify this issue, we computed global intracranial volumes (ITV) based on native GM, WM, and CSF in CN and SCD subjects by using in-home MATLAB codes. Furthermore, we used ITV as an additional covariate when general linear model (GLM) analysis is used to investigate the differences on the network connectivity of HIPsub between CN and SCD subjects.

Head motion effect
In this study, we used three approaches to control the head motion effect both at the individual and at group levels. Firstly, we excluded SCD subjects with excessive head motion (cumulative translation or rotation > 3.0 mm or 3.0°). Then we used a Friston 24-parameter model to regress out head motion effects from the realigned data [30]. Secondly, we performed a 'scrubbing' procedure to scrub frames (volumes) with an excessively high whole-brain root mean square (RMS) signal change over time in the preprocessed fMRI data for each individual [31][32][33]. Furthermore, we regressed out all volumes with a framewise displacement (FD) greater than 0.2 mm as nuisance covariates, and discarded any scan with 50% of volumes removed as described in a previous study [34]. Overall, we excluded 1 CN because of excessive head movement. No signi cant differences were observed in the head motion parameters between quali ed CN and SCD subjects ( Table 1).

Strict multiple comparison correction strategy
To ensure the reproducibility, test-retest reliability, and replicability on the fMRI metrics, we performed a strict multiple comparison correction [35], that is, statistical maps were thresholded using the permutation test with Threshold-Free Cluster Enhancement (TFCE) [36] and the false discovery rate (FDR), as implemented in DPABI [27]. For cluster-extent permutation tests, voxel thresholds of two-tailed p < 0.02 (Z > 2.3) were set. Finally, we set a two-tailed p < 0.05 threshold (1,000 permutations in FDR evaluation).

De nition of hippocampal subregions
Our de nition of HIPsub referred to recent studies from Robinson et al. [13] and Bai et al. [37], who used coactivation-based parcellation to reveal a subspecialization in the hippocampus by a data-driven method. We only selected the left HIPsub as regions of interest (ROI) (Fig. S2) based on the recent study published by Bai and colleagues [37]. The left hippocampus was de ned as three subregions (HIPe, HIPc, and HIPp).

Functional connectivity analyses
First, we extracted the average time courses for all voxels within each HIPsub as the reference time course. Second, we performed voxelwise cross-correlation analysis between the averaged time courses of all voxels within the seed HIPsub region and each voxel in the remainder of the whole brain within the group-speci c GM mask. Finally, we performed a Fisher's z-transform analysis to enhance the normality of the correlation coe cients. rTMS protocol rTMS was used to stimulate the precuneus of all aMCI participants using a Magstim Rapid2 magnetic stimulator with a 70-mm gure-8-shaped coil. We used the Pz site of the 10-20 electroencephalogram system to locate the precuneus, and the tip of the intersection of the two coil loops was placed at the Pz site to stimulate the precuneus [23]. rTMS was applied, using trains of 1000 stimuli at a frequency of 20 Hz and at an intensity of 100% of the motor threshold (MT). We de ned the MT as the lowest intensity producing motor evoked potentials of greater than 50 μV in at least 5 out of 10 trials in the relaxed rst dorsal interosseous (FDI) muscle of the contralateral (right) hand [38]. Participants received 25 sessions of either rTMS or sham stimulation over the precuneus. Each daily stimulation session consisted of a stimulation of 42 s duration with an interval of 28 s. The entire session lasted approximately 30 minutes each daily. We performed the sham rTMS blocks with the coil held close to the precuneus, but angled away.

TMS protocol adverse events
The participants did not report any adverse events during the rTMS trial.

Statistical Analysis Demographics and neuropsychological data
We performed two-sample t-test and chi-square tests to assess differences in demographic data, clinical, cognitive performance, ITV, and head rotation parameters between SCD and CN subjects (p < 0.05).

Network discovery of altered HIPsub related to SCD
To characterize the HIPsub network FC patterns at a group level, we performed a random-effect analysis using one-sample t-tests in the spatial maps of FC in CN and SCD subjects with a stringently threshold of p < 0.001 using the permutation test with TFCE and the family-wise error (FWE) correction together with a cluster extent k > 100 voxels (2700 mm 3 ).
We performed GLM analysis to investigate the differences in the FCs of HIPsub between SCD subjects and CN before rTMS treatment after controlling for age, sex, education, ITV, and mean FD (TFCE-FDR-corrected p < 0.05 and cluster size > 405 mm 3 ). Then we made masks based on brain regions showing differences in the FCs of HIPsub in SCD compared to CN. These masks were used for the analysis of prev.s. post-rTMS (pre-sham-v.s. post-sham-rTMS) fMRI data from study #2 (i.e., network validation of altered HIPsub related to SCD). These ndings identi ed network connectivity of altered HIPsub related to SCD, which can explain the changes that are related to SCD during rTMS treatment.
Pattern classi cation based on the altered HIPsub GM and FC To further identify GM and network connectivity of altered HIPsub as closely related to SCD patients, we applied a support vector machine (SVM) approach using the alterations in the identi ed ROIs as a biomarker to test how well this could distinguish SCD patients from CN subjects. A leave-one-out crossvalidation (LOOCV) strategy was used to assess the generalization of this SVM classi er and to assess its accuracy, sensitivity, and speci city. These ndings identi ed GM and network connectivity of altered HIPsub related to SCD patients, which can explain the changes that are related to SCD during rTMS treatment.

Network validation of altered HIPsub related to SCD with rTMS
To empirically validate altered HIPsub network connectivity related to SCD, we used paired t-tests to calculate the changes in network FC of HIPsub pre-v.s. post-rTMS (or pre-sham-v.s. post-sham-rTMS) in SCD subjects after controlling for age, sex, education, and GM.
Sham v.s. real rTMS comparison Of the 13 SCD subjects with full clinical assessments, usable sMRI, and fMRI scan data at baseline and 2 weeks of post-rTMS (or sham), 8 subjects had been randomized to real rTMS and 5 subjects received the sham rTMS. We performed a two-sample t-test to investigate the differences in the changes in FC of HIPsub between pre-post real rTMS and pre-post sham rTMS. Pre-real-rTMS (or sham-rTMS) maps were subtracted from post-real-rTMS (or sham-rTMS) maps to generate maps of FC changes for each subject.

Non-parametric statistics
To improve the statistical power with a low sample size, we performed a re-sampling method of stationary 10,000 bootstrap samplings to obtain signi cance in demographic data, clinical characteristics, cognitive performance, and FC of HIPsub between baseline assessment and at 2 weeks of post-rTMS (sham rTMS) for all statistical analyses (i.e., chi-square test, two-sample t-test, Pearson correlation, and paired-sample t-test).

Study approval
This study was approved by the responsible Human Participants Ethics Committee of the A liated Brain Hospital of Nanjing Medical University (No. 2018-KY010-01 and No. 2020-KY010-02) (Nanjing, China).
Written informed consent was received from participants prior to inclusion in the study.

Demographics, and clinical and cognitive function characteristics
As shown in Table 1, compared with CN, SCD subjects did not show no any signi cant differences in age, gender, education, general cognitive function (i.e., MMSE, MoCA, MDRS scores), and multimodal cognitive function (i.e., episodic memory, information processing speed, executive function, visuospatial function) (all p > 0.05, 10000 bootstraps) when showed higher HAMD and SCD-Q (all p < 0.05, 10000 bootstraps).
Network discovery of altered HIPsub related to SCD As shown in Figure 1-3 and Figure S3, SCD subjects displayed distinctly differently altered patterns of HIPsub network connectivity compared to controls.
As shown in Figure 4A and Table S3, in the HIPe network, compared with CN, SCD subjects displayed reduced FC in right CERpos, increased FC in FusG.L, Insula.L and PHG.L (P TFCE-FDR < 0.05, cluster size > 405 mm 3 ). As shown in Figure 4B and Table S3, in the HIPc network, compared with CN, SCD displayed reduced FC in IFGorb.R, increased FC in PHG.L (P TFCE-FDR < 0.05, cluster size > 405 mm 3 ).As shown in Figure 4C and Table S3,  Classi cation of aMCI patients based on the altered HIPsub GM volumes and functional connectivities in SCD patients The SVM classi er's classi cation accuracy was 86.0%. As shown in Figure 5, the SVM classi er's receiver operating characteristic (ROC) curve shows a high power to discriminate SCD patients from CN on an individual subject basis, with an AUC of 92.9%, 83.8% sensitivity, and 89.1% speci city.

Network validation of altered HIPsub related to SCD with rTMS
Changes of HIPsub FC pre-v.s. post-rTMS (or sham rTMS) As shown in Figure 6A, in the HIPc network, compared with pre-rTMS, SCD subjects showed signi cantly reduced FC in PHG.L at 2 weeks of post-rTMS (p < 0.05, 10000 bootstraps). As shown in Figure 6B, in the HIPp network, compared with pre-rTMS, SCD subjects showed signi cantly reduced FC FC in MTG.L at 2 weeks of post-rTMS (p < 0.05, 10000 bootstraps). No differences were found in the HIPe, HIPc and HIPp connectivity pre-v.s. post-sham rTMS.
Changes of episodic memory pre-v.s. post-rTMS (or sham rTMS) As shown in Figure 7, SCD subjects showed improvement of episodic memory (AVLT) after 2 weeks of real rTMS treatment (p < 0.001, 10000 bootstraps) while no differences were observed after 2 weeks of sham TMS treatment (p > 0.05, 10000 bootstraps).

Discussion
To our knowledge, this study was the rst to answer questions that the dysfunctions in posterior hippocampus (i.e., HIPc and HIPp) network connectivity could be causally restored by rTMS modulation upon the precuneus in SCD subjects. It further suggests that the precuneus-HIPsub circuit may be a potential target circuit to prevent the clinical progression of SCD to MCI/AD in therapeutic trials.
This study showed that SCD subjects presented differently altered patterns of HIPsub network connectivity compared to controls, which supports the notion that HIPsub involves in functional heterogeneity [13][14][15]. The SVM classi er showed that the abnormalities in brain memory networks had a high power to discriminate SCD from CN on an individual subject basis, with an ACC of 86.0%, an AUC of 92.9%, 83.8% sensitivity, and 89.1% speci city, which suggests that this study identi es a HIPsub pathological circuit related to SCD. The HIPe network connectivity in SCD subjects is predominantly abnormal in the brain regions involved in many aspects of emotional processing, including the cerebellum [39,40], anterior insula [39,41], parahippocampa gyrus [39,42], and fusiform gyrus [43]. Indeed, SCD subjects are characterized by self-reported concerns regarding their memory decline [44], which signi cantly increases the risk of AD [5]. Based on above-mentioned evidence, it suggests that SCD subjects exist the abnormal network connectivity associated with emotional processing. In particular, SCD subjects showed altered HIPc and HIPp network connectivity in these brain regions involved in memory processing: the input and integration of sensory perception spatial information, visual object recognition memory, and the formation of episodic memory [15,[45][46][47]. Based on the processing theory of memory system, memory formation requires two neruopathways: occipito-temporal visual object processing pathway (the "what" stream) [47, 48] and parieto-temporal visuospatial pathway (the "where" stream) [49,50]. Therefore, our results suggest that although SCD self-reported memory declines within the normal range, there are abnormalities in the networks associated with memory processing. That is, abnormalities in brain memory networks may precede the onset of clinical symptoms in SCD subjects.
More especially, the most fascinating nding was that rTMS modulation upon the precuneus for 2 weeks could restore HIPc connectivity with left parahippocampa gyrus and HIPp connectivity with left middle temporal gyrus accompanied by improvement of episodic memory. Indeed, numerous studies have consistently indicated that the precuneus is a node belonging to hippocampal intrinsic connectivity networks [15,20,24], a key node for episodic memory de cits observed in early AD [25,26], and a vulnerable region for the progress of MCI to AD [23]. Local rTMS stimulating accessible network nodes can be transmitted across synapses to distant but highly spatially speci c interconnected nodes [18,21,22]. Therefore, it is reasonable to speculate that the effects of rTMS may propagate in the hippocampus through synaptic transmission in the precuneus-HIPsub pathway. Furthermore, this study only showed that rTMS modulation could restore the dysfunctions in posterior hippocampus (i.e., HIPc and HIPp) network connectivity related to memory processing but did not the dysfunctions in anterior hippocampus (i.e., HIPe) network connectivity related to emotional processing. It suggests that the precuneus-HIPsub pathway is an ideal target circuit for tailored rTMS intervention to improve episodic memory decline in SCD.

Limitations
First, a relatively small sample size was used to validate the effects of rTMS on the HIPsub pathological circuit related to SCD in this study. According to the general rTMS clinical trials, a single clinical symptom indicator was used as the clinical e cacy criteria. In this case, our sample size is indeed relatively small. However, this study combined clinical response with functional connectivity as the e cacy criteria, so it can be speculated that our conclusions are still reliable. Second, our ndings suggested a potential target circuit the restoring the dysfunctions in HIPc and HIPp network connectivity, but this study did not nd a target circuit for the restoration of HIPe network connectivity. In the future, a large sample size is needed to explore the therapeutic pathways of anterior hippocampus network connectivity.

Conclusions
This study provides a novel experimental evidence on correction of breakdown in posterior hippocampus (HIPc and HIPp) related to cognitive and perceptual processing by modulating the precuneus in SCD subjects. Furthermore, rTMS manipulation may prevent the disease progression by improving memory as the earliest at-risk state of AD in clinical treatment trials. It further suggests that the precuneus-HIPsub circuit may be used as a useful target circuit for SCD subjects to design rationale strategies for therapeutic trials.   scores. This study used the composite Z scores to indicate the level of each cognitive domain. Note, to improve the statistical power, this study used a re-sampling method of stationary bootstrap (10,000 bootstrap samplings) to obtain significance. Figure 1 Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPe seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.

Figure 1
Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPe seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.

Figure 2
Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPc seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.

Figure 2
Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPc seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.

Figure 3
Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPp seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.

Figure 3
Schematic polar plot and bar chart depicting distinct functional connectivity patterns of HIPp seeds with target regions of interest (ROI) distributed across the whole brain among CN and SCD subjects. The concentric circles depict parameter estimates representing the connectivity strength. Note that the functional connectivity data are only extracted from the brain regions corresponding to Fig. S3.