This is a retrospective study to study the neuroplasticity mechanisms associated with VS in Chinese PLA General Hospital from January 2019 to April 2022. It was approved by the institutional review board and the independent scientific advisory committee at the Chinese PLA General Hospital. The data are anonymous, and the requirement for informed consent was, therefore, waived. The data for post hoc retrospective analysis can be divided into two parts: (1) We reviewed all patients with VS who underwent unilateral resection via retrosigmoid approach from January 2019 to April 2022, and 110 cases with preoperative and postoperative resting-state functional MRI (rs-fMRI) imaging were screened out. The tumor diagnosis of these patients was based on WHO criteria and was confirmed by postoperative pathological examination. The main exclusion criteria included any other abnormality or disease, a history of alcohol or illegal substance abuse, and a history of other brain surgery or neuromodulation interventions. In the end, the data of a total of 32 patients were finally included in the present study. The data of patients with VS were divided into a pre-operative group (VS_pre) and a post-operative group (VS_post). Moreover, the data of the patients were also divided into a group with tinnitus (VS_+tinnitus) and a group without tinnitus (VS_−tinnitus) according to tinnitus symptoms. (2) The data of healthy subjects. All the health controls accepted health examinations in the Chinese PLA General Hospital during the same period. The inclusion criteria included being older than 18 years and having functional imaging data. The main exclusion criteria include any other abnormality or disease, a history of alcohol or illegal substance abuse, MRI contraindications or intolerance, and any history of surgery or invasive intervention. Thirty-two of them matching the sex, age, and education of patients with VS were selected as healthy controls, and the image data of another 32 healthy subjects were used as an unrelated healthy dataset for gradients’ alignment.

MRI data were acquired by a 3T MRI scanner (Discovery 750, GE Healthcare, USA). Participants were asked to keep their eyes closed, relax, and wear earplugs to reduce noise, and foam pads were placed around the head to minimize head motion during MRI acquisition. The rs-fMRI data were collected based on the echo-planar imaging sequence: TE = 30 msec, TR = 2,000 msec, FA = 90°, FOV = 240 mm × 240 mm, matrix = 64 × 64, slice thickness = 3.5 mm, slice gap = 0.5 mm, and volume = 180. Two senior radiologists independently reviewed the sequences and checked the quality of the images.

The rs-fMRI data were preprocessed using Graph Theoretical Network Analysis (GRETNA) toolbox (Wang et al., 2015). Briefly, the preprocess consisted of the following steps: (1) the Dicom images were converted to NIfTI format; (2) the top five time points were removed; (3) the images underwent slice timing and correction for head motion; (4) all images were spatially normalized with the standard Montreal Neurological Institute (MNI) space; (5) spatial smooth was performed using a 4 mm × 4 mm × 4 mm full with a half-maximum Gaussian kernel; and (6) linear trend removal and nuisance covariate regression was performed with nuisance variables including Friston 24 parameter correction, white matter signal, and cerebrospinal fluid signal. Bandpass filtering was 0.01–0.1 Hz.

In order to explore whole-brain gradients alteration, we merged multiple intrinsic functional connectivity-based atlases, including cortical parcellations (Schaefer et al., 2018), cerebellar parcellations (Buckner et al., 2011), striatal parcellations (Choi et al., 2012), and thalamic parcellations (Horn and Kühn, 2015). There were three subcortical regions of interest (ROIs) that were discarded during 3 mm × 3 mm × 3 mm down-sampling due to small sizes. The final remaining 1,039 ROIs have corresponding functional community annotation, including Visual (Vis), Somatomotor (SM), Dorsal Attention (DA), Ventral Attention (VA), Limbic (Lim), Frontoparietal (FP), and Default Mode Network (DMN) (Yeo et al., 2011). Pearson correlation coefficients were computed for each pair of brain regions as the functional connectome. Detailed information about this brain atlas could be found in Supplementary Table 1 or https://github.com/louxin-lab.

The functional connectome was then z-transformed and the top 10% thresholded, and the cosine similarity matrix was calculated to capture similarity in connectivity profiles (Vos de Wael et al., 2020). Principal component analysis (PCA), the most reproducible dimensionality reduction algorithm for gradient framework, was applied to identify primary gradient components for the majority of connectome variance (Hong et al., 2020). A group-level gradient component template was generated from an average connectivity matrix based on unrelated health datasets as mentioned earlier, and we performed Procrustes rotation to align components to the template (Vos de Wael et al., 2020). As with most gradient studies, we mainly focused on the primary two components (Gradient-1 and Gradient-2) as they explained the majority of the total variance. These components, initially defined in connectivity space, were then mapped back onto the ROIs to visualize macroscale transitions in overall connectivity patterns. To analyze the functional distance alteration between ROIs, we used the primary two gradients to calculate the Euclidean distance in the functional hierarchical architecture. As for statistical analysis, independent and paired t-tests were applied for VS vs. healthy controls (HCs) comparison or VS_pre vs. VS_post comparison, respectively, with False Discovery Rate (FDR) correction.

During the rs-fMRI follow-up, we mainly used tinnitus handicap inventory (THI) and the visual analog scale (VAS) to evaluate the ipsilateral tinnitus of patients with VS. The THI is the most commonly used questionnaire of a 25-item to measure tinnitus-produced handicap. Patients can answer “no,” “sometimes,” or “yes” to each item (corresponding to scores 0, 2, and 4) (Newman et al., 1996). The global score is used as an index of the severity of tinnitus, with a grading included between 0 and 100. According to the THI score, it could be divided into different THI levels, including Level-1 (very mild, 1–16 points), Level-2 (mild, 18–36 points), Level-3 (moderate, 38–56 points), Level-4 (severe, 58–76 points), and Level-5 (catastrophic, 78–100 points). The change in the patient’s tinnitus status (improve or worsen) is mainly determined by the alteration of the THI score (increase or decrease). The VAS of tinnitus intensity was used with the left and right extremes labeled “very faint” and “very loud,” respectively (Raj-Koziak et al., 2018). Patients were required to give a first estimation of the intensity of their tinnitus at the beginning of the testing session before any sound was presented, and then be asked to use this VAS only when partial residual inhibition was obtained in a condition.

We also tried to apply gradient features for tinnitus measurements (VAS rating, THI score, and THI level) using stepwise linear models with leave-one-out cross-validation (LOOSWR). The gradient features were extracted from the brain regions with the most significant alteration and were fit into linear regression with adjustment of the subject’s gender, age, and constant. At each iteration, the predictive performance for the tinnitus measurements was assessed and only P < 0.05 was considered statistically significant when making predictions. Using both forward and backward iterations, over 98% of iterations (gradient predictors were significant and entered a model) were reported as reliable features. The Pearson correlation coefficient and the mean absolute error (MAE) between the predicted value and the actual measurements were used to evaluate the prediction effect.

The Allen Human Brain Atlas (AHBA) was introduced to study neuropathophysiological features underlying gradient framework alteration subjected to tinnitus. AHBA comprises available transcriptomic dataset post-mortem samples from six adult male donors (Markello et al., 2021). Detailed descriptions of the methods of whole genome microarray analysis could be found in Allen Institute for Brain Science technical white paper. In the present study, 20,737 genes of 1,039 regions were extracted as a 20,737 × 1,039 matrix. The framework alteration was established through a t-test between the pre- and post-operative global gradient features. We then used partial least squares regression (PLSR) to investigate the fundamental relationships between the framework alteration and gene expression with 10,000 permutation tests. The PLS1 is defined as the most strongly correlated spatial signature during the linear analysis of the gene expression weights. The gene ranked list according to the PLS1 performance was fitted into the WebGestalt to identify Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment by Gene Set Enrichment Analysis (GSEA) (Liao et al., 2019).

A total of 32 patients with unilateral VS were finally included in the present study. The mean age of all patients was 46.44 ± 12.13 years, and 12 of these patients were male. Among the patients, 65.63% of patients had preoperative hearing impairment of varying degrees with an average tumor size of 2.44 ± 1.21 cm. As for tinnitus symptoms, there were 56.25% of preoperative patients and 65.63% of postoperative patients suffered from subjective tinnitus. The preoperative THI scores, THI level, and VAS rating of patients with tinnitus were 22.11 ± 18.38, 1.83 ± 1.02, and 4.33 ± 2.14, respectively, while the postoperative THI scores, THI level, and VAS rating were 20.95 ± 17.54, 1.71 ± 0.91, and 4.10 ± 1.90, respectively. The detailed cohort info could be found in Table 1.

The change in the patient’s tinnitus status (improve or worsen) is mainly determined by the alteration of THI score (Figure 1). After VS resection, there was only one (5.56%) that was totally resolved, five (27.78%) improved, and 13 (50.00%) worse. It was interesting to note that 28.57% of the patients with no preoperative tinnitus developed postoperative tinnitus. To analyze the factors affecting postoperative tinnitus, we divided patients with tinnitus into two groups: Resolved + Improved (33.34%) and Unchanged + Worse (66.66%). We compared the differences between the two groups in terms of demographics info, preoperative hearing, tumor features (tumor size, occupation of internal auditory canal, tumor nature, and cerebellum brain stem extrusion), and surgery approach (cochlear nerve and degree of resection). The results showed that none of these factors had a significant effect on postoperative tinnitus alteration (Table 2).

The primary Gradient-1 and Gradient-2 accounted for a total variance of 69% in VS_pre (Figure 2A). No statistical difference was identified for variance explanation of global functional connectome in the comparison of VS_pre vs. HCs or VS_pre vs. VS_post (Figure 2B). We then investigated regional alteration. Independent t-test revealed that Gradient-1 features of the No. 501 ROI of the visual cortex on the right hemisphere (known as RH_Vis_1 in Schaefer 1,000 Parcels 7 Networks atlas, MNI: x = 33, y = −36, z = −23, VS_pre vs. HCs: t-test P_FDR = 0.004) and Gradient-2 features of the No. 24 ROI of the visual cortex on the left hemisphere (known as LH_Vis_24 in Schaefer 1,000 Parcels 7 Networks atlas, MNI: x = −34, y = −92, z = −9, VS_pre vs. HCs: t-test P_FDR = 0.042) were found abnormal in VS_pre compared with matched HCs (Figures 2C, D). Both altered features of RH_Vis_1 and LH_Vis_24 were found rescued after tumor resection (VS_post vs. HCs: both t-test P_FDR > 0.05) (Figures 2C, D). No gradient features were found significantly altered in any of the subcortical structures (cerebellum, thalamus, and striatum). Moreover, it was found Gradient-2 features of the No. 708 ROI of the posterior cortex of dorsal attention on the right hemisphere (known as RH_DorsAttn_Post_24 in Schaefer 1,000 Parcels 7 Networks atlas, MNI: x = 29, y = −64, z = 52) were significantly decreased both preoperatively and postoperatively (VS_pre vs. HCs: t-test P_FDR > 0.05; VS_post vs. HCs: t-test P_FDR = 0.016) (Refer to Result section “3.3. Postcentral gradient features in patients with VS reflected tinnitus symptoms”; Figure 3A). These results were robust and virtually identical when controlling for confounding features of rs-fMRI processing, including connectivity matrix thresholding (10–20%; van Wijk et al., 2010; Supplementary Figure 1). Moreover, we obtained similar results with the exclusion of the patients with preserved cochlear nerves (Supplementary Figure 1).

According to functional hierarchy theory, we analyzed the gradient distances related to altered regions. With the threshold of t-test P_FDR < 0.05, the gradient distances from RH_Vis_1 to the other 668 nodes were significantly altered in the VS_pre compared with HCs, which mainly were restored in VS_post (58 nodes, Figure 2E). As for LH_Vis_24, there were 375 altered gradient distances in the VS_pre compared with HCs, and still 312 altered gradient distances in the VS_post (Figure 2E). Gradient distances from LH_Vis_24 to DMN regions were significantly increased after VS resection (23.2% in VS_pre vs. 48.1% in VS_post).

The VS occurrence initialed the gradient features alteration of RH_DorsAtten_post_24, and these changes were further exacerbated after tumor resection (VS_pre vs. HCs: t-test P_FDR > 0.05; VS_post vs. HC : t-test P_FDR = 0.016; Figure 3A). To explore the potential link between gradient features and tinnitus symptoms, the VS data were regrouped into two clusters according to tinnitus status: (1) VS_+tinnitus group: 39 cases with tinnitus including 18 preoperative and 21 postoperative data and (2) VS_−tinnitus group: 25 cases without including 14 preoperative and 11 postoperative data. As for global gradient features, no statistical difference was identified for functional connectome variance in the comparison of VS_+tinnitus vs. VS_−tinnitus or VS_+tinnitus vs. HCs. As for regional alteration analysis, it was found that Gradient-2 features of RH_DorsAtten_post_24 were also significantly decreased in VS_+tinnitus (VS_−tinnitus vs. HCs: t-test P_FDR > 0.05; VS_+tinnitus vs. HCs: t-test P_FDR = 0.022; Figure 3B).

We also analyzed the gradient distance alteration of RH_DorsAtten_post_24 after tumor resection. Although there were only 39 nodes with significantly altered gradient distances from RH_DorsAtten_post_24 in the VS_pre compared with HCs, there was a dramatical increase of 261 regions in VS_post with the threshold of t-test P_FDR < 0.05 (Figure 3C). Based on the Yeo subnetwork theory, it can be found that the altered gradient distance of RH_DorsAtten_post_24 mainly connected to the brain regions within the DMN subnetwork (64.1% in VS_pre and 35.2% in VS_post). Moreover, we compared the specific gradient distance features of the RH_DorsAtten_post_24 in VS_+tinnitus or VS_−tinnitus compared with HCs. It could be found that gradient distances were only significantly increased from RH_DorsAtten_post_24 with RH_Cont_PFCI_10/15/22 in VS_+tinnitus vs. HCs (t-test P_FDR = 0.047/0.040/0.047, respectively; Figure 3D). All these gradient distances were found significantly correlated with tinnitus status evaluation (Table 3).

Gradient-2 features of RH_DorsAtten_post_24 were used for spearman correlation with tinnitus status evaluation. It was found that Gradient-2 features of RH_DorsAtten_post_24 were significantly correlated with THI score (r = −0.30, P = 0.013), THI level (r = −0.31, P = 0.010), and VAS rating (r = −0.31, P = 0.0093). Considering the tinnitus symptoms and the RH_DorsAtten_post_24 features were both relatively independent of tumor resection, Gradient-2 features of RH_DorsAtten_post_24 were selected as independent variables in LOOSWR models for tinnitus symptoms prediction (Figures 3D, E). With iterative repetition of 100% and predictive significance <0.05, it could be used to predict postoperative improvement of VAS rating (MAE = 1.94 ± 1.21, r = −0.23 and P = 0.028). We also tried to use gradient features to predict THI score or THI level improvement but found no significant results.

We introduced AHBA to investigate featured genes underlying gradient, framework alteration related to tinnitus (VS_+tinnitus vs. HCs). The primary PLS component of featured genes accounted for 10.13% and 11.61% of the total variance significantly more than a random chance for Gradient-1 and Gradient-2 alteration features, respectively (P < 0.001). KEGG pathway analysis revealed that Gradient-1 featured genes were most significantly enriched in terms related to hsa03010 Ribosome (normalized enrichment score (NES) = 2.615, P_FDR < 0.001) with GO0070972 protein localization to the endoplasmic reticulum (NES = 2.379, P_FDR < 0.001), GO0006413 translational initiation (NES = 2.288, P_FDR < 0.001), GO0005840 ribosome (NES = 2.275, P_FDR < 0.001), and other GO annotations related to Ribosome (Figure 4A). Moreover, KEGG pathway analysis revealed that Gradient-2 featured genes were most significantly enriched in terms related to hsa00190 oxidative phosphorylation (NES = 2.361, P_FDR < 0.001) with GO0033108 mitochondrial respiratory chain complex assembly (NES = 2.270, P_FDR < 0.001), GO0044455 mitochondrial membrane part (NES = 2.400, P_FDR < 0.001), GO0070469 respiratory chain (NES = 2.256, P_FDR < 0.001), and other GO annotations related to oxidative phosphorylation (Figure 4B). Detailed gene enrichment results could be found in Supplementary Tables 2–9.