Transcriptional correlates of structure-function coupling plasticity in trigeminal neuralgia: unveiling the synaptic and metabolic associations Authors: Jian Zhang, Xujing Nie, Shuyue Fu, Yinping Lu, Tiantian Liu, Zhilin Zhang, Jinglong Wu, Xuezhi Tong, Tianyi Yan Published: 2026 ABSTRACT AbstractBackgroundTrigeminal neuralgia (TN) is characterized by severe facial pain, transitioning from peripheral vascular compression to central sensitization. However, the core central pathophysiological mechanisms—particularly how the brain structurally and functionally reorganizes to maintain pain or recover after surgical treatment (primarily microvascular decompression or percutaneous interventions)—remain to be fully elucidated.MethodsWe employed a longitudinal multi-scale design using a graph harmonic model to quantify structure-function (S-F) coupling, a metric reflecting brain network integrity. We analysed multimodal magnetic resonance imaging data from 87 patients with TN and 42 healthy controls (HCs). Post-treatment follow-up data were acquired for 46 patients, of whom 39 had complete longitudinal paired data. We further utilised partial least squares (PLS) regression with rigorous spatial permutation testing (spin-tests) to bridge macroscopic imaging changes with microscopic transcriptomic data from the Allen Human Brain Atlas.ResultsPatients exhibited significantly reduced global S-F coupling, particularly in the somatomotor and dorsal attention networks. Notably, this decoupling was negatively correlated with baseline pain severity and disease duration. Following treatment, global S-F coupling returned to levels statistically indistinguishable from healthy baselines (). Crucially, the magnitude of postoperative reorganization (S-F coupling) significantly correlated with postoperative pain reduction percentages and long-term follow-up NRS scores. This recovery extended beyond focal repair, involving extensive adaptive reorganization in the visual and default mode networks. Molecularly, disease-related decoupling was spatially associated with genes linked to neuronal energy metabolism, cellular ionic homeostasis, and neuroinflammation. Conversely, treatment-induced plasticity strongly correlated with genes modulating chemical synaptic transmission and endogenous opioid signaling.ConclusionsTN pathophysiology is closely linked to ion channel-mediated neuronal metabolism and progressive network decoupling. Effective treatment restores homeostatic brain network coupling primarily by facilitating synaptic plasticity-based adaptive reorganization rather than merely through focal repair. This work offers a new perspective on the neural circuits underlying pain maintenance and provides potential imaging indicators for developing brain network-targeted therapeutic strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s10194-026-02419-7.Keywords: Structural-functional coupling, Graph harmonic model, Brain network, Multimodal MRI, Trigeminal neuralgia, Adaptive reorganization Introduction Trigeminal neuralgia (TN) is a chronic neuropathic pain syndrome characterised by recurrent, transient, electric shock-like paroxysms of pain within the distribution of the trigeminal nerve. The attacks are abrupt, brief and exceptionally severe, profoundly affecting patients’ quality of life. It is widely accepted that its primary peripheral cause is related to vascular compression of the trigeminal nerve root entry zone, leading to the demyelination of nerve fibres. Epidemiological studies indicate that TN incidence is generally higher in women than men, with an average age of onset after 50, and its incidence increases with age [1]. Clinically, first-line treatments are drugs such as carbamazepine, while for patients who do not respond to or cannot tolerate medication, surgical procedures like MVD are a recognized effective treatment [2]. However, the pathophysiology of TN is far more complex than peripheral nerve compression, involving a peripheral-to-central cascade [3]. A growing body of evidence suggests that long-term abnormal pain signalling leads to profound functional and structural remodelling within the central nervous system [4–8]. Persistent nociceptive input sensitises primary neurons in the trigeminal ganglion, and this peripheral sensitization is a critical step towards central sensitization. Within the central nervous system, this triggers a series of maladaptive neuroplastic changes [9], including the over-activation of central nociceptive neurons, neuronal hyperexcitability, and dysfunction of the endogenous pain modulatory system [10]. These central alterations are core factors in the establishment and long-term maintenance of the pain state. At present, how the brain undergoes functional and structural remodelling to adapt to this chronic pain, and the neural circuit principles governing its recovery process following successful peripheral decompression, remain poorly understood. To uncover these central mechanisms, multimodal neuroimaging technologies provide powerful non-invasive research tools. Functional magnetic resonance imaging (fMRI) studies have revealed that patients with TN exhibit abnormal resting-state brain network connectivity, particularly in regions associated with pain processing, emotion and cognition [11]. Recently, several studies utilizing the same TN cohort have provided valuable insights through single-modality approaches. Concurrently, structural analyses based on diffusion tensor imaging (DTI) have shown alterations in the white matter microstructure and aberrant connectome controllability [12–14], while single-omics studies have explored the specific genetic pathogenic factors [15].However, these single-modality approaches possess inherent limitations when deciphering the complex pathophysiology of TN. TN is fundamentally initiated by a focal peripheral structural deficit (vascular compression and demyelination) that ultimately manifests as sudden, paroxysmal functional hyperactivity (central sensitization). Relying solely on structural MRI only captures static white matter microstructural degradation but cannot explain how these physical deficits precipitate the hyper-synchronous functional bursts of pain. Conversely, functional MRI reports aberrant cortical synchrony but lacks the anatomical context to determine whether this hyperactivity is a direct consequence of structural rewiring or a secondary, unconstrained functional adaptation. Therefore, a critical question remains that can only be addressed by a combined structure-function (S-F) coupling framework: how does the brain’s structural architecture constrain, or fail to constrain, maladaptive functional dynamics in chronic TN? Fundamentally, the anatomical white matter backbone severely constrains the repertoire of resting-state functional dynamics [16, 17]. S-F coupling quantifies this precise relationship. In the context of TN, investigating S-F coupling offers a unique opportunity to test whether the brain loses its structural constraints on functional signaling—potentially permitting the unconstrained, “electric shock-like” hyperexcitability characteristic of the disease—and whether surgical decompression can restore this critical structural-functional alignment [18]. Dysregulation of S-F coupling has been demonstrated in various neurological disorders and is considered a key indicator of network dysfunction [19]. To quantify S-F coupling accurately, the choice of a suitable computational model is crucial. Traditional statistical models only consider direct anatomical connections, overlooking the extensive indirect communication pathways in the brain. This study employs an advanced graph harmonic model [20–22], which uses eigendecomposition to break down the brain’s structural network into a series of orthogonal harmonics, capturing the propagation patterns of neuropathological signals across the brain. Compared with traditional models, the graph harmonic model is more sensitive in detecting disease-related pathological changes [23], and its results show a stronger correlation with biological metrics such as myelin distribution and macroscopic gradients, providing a more biologically plausible framework for understanding S-F coupling. This study aims to employ a longitudinal design and an advanced graph harmonic model to precisely quantify the dynamic changes in S-F coupling in patients with TN before and after treatment, thereby exploring the core recovery mechanisms at both macroscopic network and microscopic molecular levels. At the macroscopic level, we hypothesise that the normalization of S-F coupling in patients with TN following treatment is not a simple process of pathological reversal (i.e., the direct repair of affected brain regions). Instead, it is more likely a complex neural reorganization process dominated by widespread, adaptive reorganization. To move beyond these descriptive neuroimaging observations, we integrated transcriptomic data from the Allen Human Brain Atlas (AHBA) to investigate the molecular underpinnings of regional vulnerability and plasticity in TN. While S-F coupling analysis identifies macroscopic network disruptions, it cannot fully explain why specific cortical regions are more susceptible to recurrent paroxysmal pain. Given that TN is characterized by abnormal neuronal hyperexcitability [24] and maladaptive synaptic plasticity [25], we hypothesized that brain regions exhibiting altered S-F coupling would show distinct expression patterns of pain-related genes, including voltage-gated sodium channels (SCN9A), calcium channels (CACNA1H), and synaptic remodeling markers (BDNF, TNF). By elucidating these intertwined processes, this multimodal approach bridges S-F reorganization with underlying molecular mechanisms, offering new insights into the neural circuits of paroxysmal pain and providing potential targets for developing brain network-directed therapeutic strategies. Materials and methods Participants This study constitutes a secondary analysis of a longitudinal neuroimaging dataset. All raw MRI and clinical data were retrieved from the OpenNeuro repository (Accession Number: ds005713). To ensure strict transparency regarding dataset versioning, the data utilization was precisely stratified: All raw multimodal MRI data were originally acquired and processed from Version 1.0.1 to establish the neuroimaging cohort and compute all S-F coupling metrics. Subsequently, the repository underwent a major metadata upgrade (Version 2.0.1, DOI: 10.18112/openneuro.ds005713.v2.0.1), which introduced critical quantitative clinical endpoints that were absent in the initial release (e.g., quantitative baseline Pain_severity, Postsurgery_pain_reduction percentages, Second_follow-up_pain_NRS, and precise follow-up intervals in days). To leverage these enriched clinical endpoints without altering the established imaging cohort, we performed a rigorous, subject-by-subject matching between the updated v2.0.1 clinical metadata and the v1.0.1 imaging cohort using unique BIDS participant identifiers. Consequently, all reported imaging results and group-level brain maps mathematically reflect the v1.0.1 imaging cohort, while all clinical correlation analyses were executed utilizing the comprehensive clinical variables from v2.0.1. All data underwent a rigorous quality control process (Fig. 1). Subjects were excluded for significant image artefacts or excessive head motion (maximal framewise displacement > 3 mm or translation > 3 mm). Furthermore, a one-way analysis of variance (ANOVA) confirmed that there were no significant differences in mean framewise displacement (mean FD) among the three groups (F = 2.55, P = 0.081). Detailed mean FD distributions (mean ± SD) were reported to ensure data transparency: healthy controls (0.25 ± 0.12 mm, range: 0.09 – 0.69 mm), the total pre-surgical TN cohort (0.30 ± 0.16 mm, range: 0.10 – 0.71 mm), and the overall post-surgical TN cohort (0.32 ± 0.14 mm, range: 0.12 – 0.75 mm). Additionally, within the strictly paired longitudinal cohort (N = 39), the pre-surgical mean FD was 0.28 ± 0.12 mm (range: 0.10 – 0.56 mm) and the post-surgical mean FD was 0.33 ± 0.15 mm (range: 0.12 – 0.75 mm). Flowchart of participant selection. The flowchart illustrates the participant inclusion and exclusion process for the study. The initial dataset consisted of 217 scans from a cohort including HCs, TNpre, and TNpost. During the screening process, a total of 11 scans were excluded due to missing data (3 HCs, 3 TNpre, 5 TNpost), and another 31 scans were excluded due to excessive head motion (3 HCs, 21 TNpre, 7 TNpost). The final sample included 175 scans for analysis, comprising 42 HCs, 87 TNpre, and 46 TNpost. Of these, 39 patients had complete, high-quality pre- and post-treatment paired data available for the longitudinal analysis After quality control, the final sample for cross-sectional analysis included 129 individuals: 87 patients with pre-treatment TN (TNpre), 42 HCs. Within the TNpre group , 62 patients (71.3%) were scheduled for microvascular decompression (MVD), 19 (21.8%) for percutaneous interventions (PI), and 6 (6.9%) were outpatient cases. Regarding pain lateralization in this sample, 31 patients reported left-sided pain, 54 reported right-sided pain, and 2 reported bilateral pain. For the longitudinal analysis, 39 patients had complete pre- and post-treatment paired data. This paired cohort comprised 29 patients who underwent MVD and 10 who received PI. In this longitudinal sample, 17 patients reported left-sided pain, 20 reported right-sided pain, and 2 reported bilateral pain. MRI acquisition The MRI acquisition protocols have been comprehensively documented in the primary data descriptor [26]. In brief, all imaging data were acquired using a 3.0 T Philips Ingenia scanner equipped with a multi-channel head coil. The dataset includes high-resolution T1-weighted structural images, DTI (64 directions, b = 1500 s/mm²), and resting-state fMRI (TR = 3000 ms). As specified in the OpenNeuro repository, the structural scans were de-faced for anonymity and pre-organized according to the Brain Imaging Data Structure standard by the original data contributors prior to our retrieval. fMRI preprocessing and FC construction fMRI data were preprocessed using fMRIPrep 23.1.4, a pipeline that integrates tools from FSL, ANTs, and FreeSurfer [27]. The fMRIPrep workflow included slice-timing correction, head motion correction, co-registration to the subject’s T1w image, and non-linear normalization to MNI standard space. The T1w images were simultaneously processed for bias field correction, skull-stripping, and brain tissue segmentation. Following fMRIPrep, the preprocessed BOLD time series were extracted and further denoised. This post-processing step, performed using nilearn, included linear detrending, band-pass filtering (0.01 – 0.08 Hz), and regressing out nuisance signals (six head motion parameters, and the mean signals from white matter and cerebrospinal fluid). Network nodes were defined using the Schaefer 400-parcel, seven-network atlas [28]. The mean denoised BOLD time series was extracted from each of the 400 cortical regions. Functional connectivity (FC) matrices were then constructed for each subject by calculating the pairwise Pearson’s correlation coefficients between all regional time series. The resulting correlation values were Fisher’s r-to-z transformed to improve normality, yielding a 400 × 400 symmetric FC matrix for each subject. DTI preprocessing and structural connectivity construction DTI data were preprocessed using the PANDA toolbox, which included steps for denoising, B1 field correction, and eddy current correction. Subsequently, each subject’s preprocessed diffusion image was co-registered to their native T1-weighted structural image. To account for crossing fibers within each voxel, diffusion modelling was performed using BedPostX, and probabilistic tractography was conducted using ProbTrackX [29]. Tractography was executed in each subject’s native diffusion space, initiated from voxels with a fractional anisotropy value > 0.2 and a curvature angle threshold of 45°. Network nodes were defined using the same Schaefer 400-parcel, seven-network atlas as in the functional analysis. This atlas was transformed from MNI standard space into each subject’s native diffusion space. For probabilistic tractography, 5,000 streamlines were sampled per voxel within each seed region. The structural connection strength (edge weight) between any two regions was quantified by the probability of connectivity, defined as the number of streamlines reaching the target region divided by the total number of streamlines sampled in the source region. To ensure undirected graph properties, the resulting matrix was subsequently symmetrized [30], yielding a 400 × 400 weighted, symmetric, and normalized structural connectivity (SC) matrix for each subject. Structural-functional coupling: the graph harmonic model We employed a graph harmonic model to quantify S-F coupling by assessing the alignment of FC with the intrinsic eigenmodes of the structural connectome (Fig. 2). First, the SC matrix was represented as a weighted, undirected adjacency matrix (A), and the structural Laplacian (L) was computed as: Overview of the multi-scale analysis pipeline. (A) Construction of S-F networks. The top panel shows the structural pipeline: DTI data are processed to generate a SC matrix, and subsequent eigendecomposition of the graph Laplacian yields a set of low-frequency structural harmonics. The bottom panel shows the functional pipeline: fMRI data are preprocessed to generate a FC matrix. represents the functional connectivity profile of a single brain region i. (B) Quantification of S-F coupling. For each brain region i, a multiple linear regression model is constructed, using a basis set of low-frequency structural harmonics (e.g., ) as predictors to explain the region’s . The model’s goodness-of-fit (coefficient of determination, R²) is defined as the S-F coupling strength for that region and is mapped back onto the cortical surface. (C) Transcriptomic association analysis. The group difference map of S-F coupling and the whole-brain gene expression matrix serve as inputs for a PLS regression. The PLS analysis identifies the primary pattern of gene expression (PLS1) that is most strongly associated with the imaging changes, with statistical significance assessed via rigorous spatial permutation testing (spin-tests). The top-weighted genes are then subjected to a GO enrichment analysis to uncover their underlying biological functions where is the diagonal degree matrix. To mitigate effects of network density, L was normalized by its largest eigenvalue: . Eigendecomposition of yields a set of orthonormal eigenmodes and their corresponding eigenvalues. These eigenmodes, or graph harmonics, were sorted by their increasing eigenvalues. Eigenmodes with low eigenvalues (near zero) capture global, persistent patterns, while those with high eigenvalues capture spatially complex processes. Following previous work, we modelled the FC profile of each brain region i (the i-th column of the FC matrix, ) as a linear combination of the 2nd through 14th structural harmonics. For each region, an ordinary least squares regression was fit, excluding the region’s self-connection (i.e., the i-th element): where is the k eigenmode (harmonic), is the intercept, are the estimated parameters, and is the error term. This model uses eigenmodes through , as the first eigenmode represents a trivial constant pattern captured by the intercept . Prior to fitting, the predictor eigenmodes ( through ) were z-scored. The S-F coupling strength for region i was quantified as the model’s coefficient of determination (R²). A higher R² indicates that a region’s FC is more strongly constrained by the underlying structural architecture. This process was repeated for all 400 regions, yielding a 400 × 1 S-F coupling vector for each subject. Gene expression analysis To investigate the biological mechanisms underlying S-F coupling changes, whole-brain gene expression data were obtained from the AHBA [31]. Data were processed using the Abagen toolbox to map normalized expression levels of protein-coding genes onto the same Schaefer 400-parcel atlas used for network construction. This process yielded a 400 × 15,633 (region × gene) expression matrix [32]. We then employed partial least squares (PLS) regression analysis to identify multivariate gene expression patterns that were spatially correlated with our S-F coupling findings [33]. Two separate PLS analyses were conducted: the gene expression matrix served as the independent variable (X) to predict the dependent variable (Y), which was (1) the disease effect t-map (HCs vs. TNpre) or (2) the treatment effect t-map (paired TNpost vs. TNpre). For significant PLS components, genes with weights in the top 5% (based on z-score) were extracted for functional enrichment analysis using Metascape [34]. To specifically test whether the S-F coupling alterations were associated with known disease-related genes, we compiled a list of high-risk candidate genes for TN based on a search using the keyword Trigeminal Neuralgia in the GeneCards database. This resulted in a final list of 31 candidate genes with a relevance score > 10. Some genes (e.g., SCN10A, SCN11A) were excluded due to the unavailability of expression data in the AHBA, leaving 25 candidate genes for the final analysis. We calculated the spatial Pearson correlation between the expression map of each of these 25 genes and both the disease and treatment effect t-maps. The two resulting sets of p-values were separately corrected for multiple comparisons using the False Discovery Rate (FDR). Statistical analysis All statistical analyses were performed in Python. We assessed group differences in S-F coupling using independent-samples t-tests for cross-sectional comparisons (HCs vs. TNpre; HCs vs. TNpost) and a paired-samples t-test for longitudinal changes (TNpre vs. TNpost). To accurately link the imaging phenotypes with clinical outcomes, group comparisons and clinical relevance analyses were performed at three spatial scales: global (whole-brain mean), network (seven functional networks), and regional (400 parcels). For macroscopic analyses (global and network levels), S-F coupling values were averaged across respective parcels. Because these metrics aggregate data across both hemispheres, they are mathematically invariant to pain laterality. However, given the highly lateralized nature of TN pathology, we strictly aligned the neuroimaging data to the side of pain for precise regional analyses to prevent lateralized effects from canceling out. Specifically, for patients with right-sided TN, the functional and structural arrays of the two hemispheres were symmetrically flipped, ensuring that the first 200 regions of the Schaefer atlas uniformly represented the ipsilateral (pain-affected) hemisphere. Clinical relevance was examined using partial Spearman correlation analyses (reported as partial Spearman’s ). For the baseline analysis, correlations were calculated between S-F coupling, baseline pain severity, and disease duration, strictly adjusting for age, sex, mean FD, and surgery type as covariates. For the longitudinal analysis, we correlated the change in S-F coupling (Δ) with the postoperative pain reduction percentage and follow-up NRS scores. To isolate the effects of recovery, the specific interval of follow-up time was additionally included as a covariate in the longitudinal model for the second follow-up NRS scores. Missing clinical data across all aforementioned models were handled using listwise deletion. Comprehensive details regarding the exact reasons for missingness, model-specific covariates, and final effective sample sizes (N) for both unaligned and aligned pipelines are transparently documented in Supplementary Table S16. To ensure statistical robustness and appropriately account for these nuisance covariates, empirical significance for all partial correlations was assessed using non-parametric permutation testing (1,000 iterations) based on the rigorous Freedman–Lane procedure. In each permutation, residuals of the clinical variable after regressing out covariates were randomly shuffled and added back to the fitted values. Furthermore, a univariate analysis of covariance (ANCOVA) was employed to test the effects of specific clinical subtypes (pain laterality, paroxysmal vs. persistent pain, and Sindou severity grading) on baseline S-F coupling, controlling for age, sex, and mean FD. For group comparisons (t-tests), P-values were corrected for multiple comparisons using the FDR [35], with a < 0.05 considered significant. However, given the conservative penalty of correcting across the entire cortex for clinical correlations, network- and ROI-level clinical partial correlations were treated as exploratory analyses. Crucially, to avoid arbitrary selection and focus on pathologically relevant areas, an empirical ROI mask was predefined based on our group-level findings. Specifically, we extracted all regions that exhibited significant group-level S-F coupling alterations derived from the aforementioned pairwise comparisons. This data-driven procedure yielded restricted feature spaces of 171 ROIs for unaligned analyses and 169 ROIs for pain-side aligned analyses (excluding two patients with bilateral pain). Exploratory clinical associations within these predefined regions and the seven functional networks are reported at an uncorrected permutation . Results Demographic and clinical characteristics Based on the secondary analysis of the OpenNeuro dataset (ds005713), the TNpre (N = 87) and HCs (N = 42) groups were successfully matched, showing no significant differences in age (TNpre: 59.15 ± 12.20 years; HCs: 56.31 ± 7.32 years; t(120.79) = -1.65, P = 0.102) or sex distribution (TNpre: 52 F/35 M; HCs: 25 F/17M; χ²(1) = 0, P = 1.000). As expected, the groups differed significantly in the Sindou grade of neurovascular compression (Mann-Whitney U test, W = 810, P < 0.001). The majority of HCs (88.1%) had no compression (Sindou grade 0), while 66.7% of patients exhibited compression (grades 1–3). Primary pre-scan medications included carbamazepine, gabapentin, and pregabalin. Detailed demographic and clinical data are provided in Table 1, while comprehensive clinical characteristics, including baseline pain severity and disease duration, are detailed in Supplementary Table 15. Table 1Demographic and clinical characteristics of the participantsHealthy Control (N = 42)Trigeminal Neuralgia (N = 87)P-valueStatisticAge (years)56.31 (7.32)59.15 (12.20)0.102t = 1.65Sex (Female/Male)25 / 1752 / 351χ² = 0Sindou Grade (N)< 0.001W = 810Grade 03729Grade 1011Grade 2420Grade 3127Disease Duration(years)N/A7.8 ± 5.2N/AN/AData are presented as mean ± standard deviation for continuous variables and as count (N) for categorical variables. P-values for group comparisons were calculated using an independent samples t-test for age, a Chi-squared (χ²) test for sex, and a Mann-Whitney U test for Sindou grade. A P-value < 0.05 was considered statistically significant Demographic and clinical characteristics of the participants Data are presented as mean ± standard deviation for continuous variables and as count (N) for categorical variables. P-values for group comparisons were calculated using an independent samples t-test for age, a Chi-squared (χ²) test for sex, and a Mann-Whitney U test for Sindou grade. A P-value < 0.05 was considered statistically significant Spatial distribution and network hierarchy of S-F coupling Whole-brain spatial maps, generated via the graph harmonic model, revealed a clear functional hierarchy in S-F coupling across all groups (Fig. 3A). Consistent with established brain organization, coupling strength was highest in unimodal cortices (e.g., visual and sensorimotor networks) and progressively decreased towards transmodal association cortices (e.g., limbic and default mode networks). Spatial distribution and network hierarchical pattern of coupling. (A) Average spatial maps of S-F coupling strength on the cortical surface for the HCs, TNpre, and TNpost groups. A consistent macroscale gradient pattern is observed across all groups, with higher coupling in unimodal cortices (e.g., visual and somatomotor) and lower coupling (e.g., limbic and default mode). (B) Boxplots illustrating the distribution of mean S-F coupling within each of the seven functional networks. The left plot displays the disease effect (HCs vs. TNpre), while the right plot displays the treatment effect (paired TNpre vs. TNpost). (C) T-statistic heatmaps visualizing the intrinsic hierarchical pattern of S-F coupling across networks within each group. Cells represent T-statistics for pairwise network comparisons, highlighting the abnormally widened coupling gradient in the TNpre group and its subsequent normalization post-treatment. Asterisks indicate FDR-corrected significance (*p < 0.05, **p < 0.01, ***p < 0.001) Analysis at the seven-network level significantly quantified this spatial distribution and highlighted specific group-level alterations (Fig. 3B). For the disease effect (HCs vs. TNpre), TNpre exhibited visibly lower S-F coupling compared to HCs. As comprehensively detailed in Supplementary Table S2, this decoupling was most prominent in the somatomotor (, Cohen’s , ), dorsal attention (, Cohen’s , ), and ventral attention/salience (, Cohen’s , ) networks. Conversely, the treatment effect (paired TNpre vs. TNpost) was characterized by a widespread and significant increase in S-F coupling that extended across all seven functional networks (all ). The most substantial recovery was observed in the visual (, Cohen’s , ) and limbic (, Cohen’s , ) networks. All corresponding t-statistics, effect sizes, and FDR-corrected p-values for network-level S-F coupling alterations are reported in Supplementary Table S2. Beyond mean coupling changes, the intrinsic network hierarchy itself differed between groups, as evaluated by within-group pairwise network comparisons (paired t -tests, FDR corrected) (Fig. 3C). HCs demonstrated a stable hierarchy where unimodal networks (visual, sensorimotor) had significantly higher S-F coupling than transmodal networks (default mode, limbic). In TNpre, this hierarchy gradient was abnormally widened, showing a more pronounced differentiation between networks (indicated by higher t-values) and suggesting a progressive loss of structural constraints on FC along the cortical hierarchy. Following TNpost, this aberrant differentiation was significantly attenuated, with the overall hierarchical distribution closely resembling that of HCs. Significant group differences in S-F coupling At the global level (Fig. 4A), the TNpre group had significantly lower whole-brain mean S-F coupling than the HCs group (t = 2.49, Cohen’s d = 0.45, PFDR= 0.022). Following treatment, S-F coupling in patients increased significantly (t = -4.04, Cohen’s d = 0.65, PFDR< 0.001). Importantly, an equivalence-style comparison between HCs and TNpost revealed a near-zero effect size (t = -0.002, Cohen’s d = 0.004, PFDR= 0.999), indicating that global S-F coupling was successfully restored and was no longer significantly different from HCs (Supplementary Table S1). Dynamic reorganization of S-F coupling and its association with longitudinal pain relief. (A) Group comparisons of whole-brain mean S-F coupling. The left boxplot shows the disease effect, where TNpre have significantly lower S-F coupling than HCs. The right boxplot shows the treatment effect, where patients exhibit a significant increase in S-F coupling in TNpost state compared to their TNpre baseline. (B) Radar plots of network-level S-F coupling differences. The left plot illustrates the distribution of the disease effect across the seven functional networks, while the right plot demonstrates the widespread nature of the treatment effect across all networks. (C) Examples of dynamic change patterns in representative ROIs. This panel illustrates four primary patterns of S-F reorganization: Normalization, where S-F coupling was significantly lower than HCs at baseline but was restored to a level indistinguishable from HCs post-treatment; Adaptation, where coupling was not different from HCs at baseline but showed a significant increase post-treatment; and No Response, where coupling was lower than HCs at baseline and showed no significant change after treatment; and Pain-Side Aligned, highlighting severe ipsilateral decoupling (e.g., Ipsi_Vis_2) unmasked after systematically aligning the data to the affected side. (D) Cortical surface maps of significant group differences. The top panel (“Disease effect”) displays the brain regions where S-F coupling was significantly lower in the TNpre group compared to HCs. The bottom panel (Treatment effect) displays the regions where S-F coupling was significantly higher in the TNpost group compared to the TNpre group. (E) Raw-scale scatter plots illustrating the correlations between the raw change in whole-brain S-F coupling (Δ S-F coupling) and raw clinical outcomes (postsurgery pain reduction percentage and long-term follow-up NRS scores). The overlaid statistical values represent the partial Spearman’s correlation coefficient () and permutation P-values, adjusting for age, sex, mean FD, surgery type, and follow-up intervals. (F) Bar charts detailing the partial Spearman’s between network-level S-F coupling and long-term follow-up NRS / pain reduction. Asterisks (*) indicate significant associations based on rigorous non-parametric permutation testing (P < 0.05) At the network level (Fig. 4B), consistent with the spatial patterns described above, the disease-related decrease was most evident in the somatomotor, dorsal attention networks and ventral attention/salience networks. In contrast, the treatment-induced increase was widespread, with all seven functional networks showing a significant enhancement (all ). The most prominent increases occurred in the visual, limbic, and dorsal attention networks. For complete network-level statistics, including exact t-statistics and effect sizes, please refer to Supplementary Table S2. At the regional level, mapping the significant S-F coupling alterations onto the cortical surface revealed that disease-related decreases were concentrated in the bilateral sensorimotor cortex and dorsal attention network, while treatment-induced increases occurred in similar but vastly more extensive regions with exceptionally large effect sizes (e.g., ) (Fig. 4D).Rather than a uniform global response, we identified four representative patterns of regional change tracking longitudinal trajectories (Fig. 4C): (i) Restoration: Several regions (e.g., LH_Vis_20, LH_SomMot_1) exhibited S-F coupling that was significantly lower than in HCs at baseline which subsequently returned to levels statistically indistinguishable from HCs following treatment. (ii) Adaptation: Certain regions in the visual network (e.g., LH_Vis_1) were not different from HCs at baseline but exhibited a significant increase above HCs levels following treatment. (iii) No Response: Some regions that were abnormal at baseline (e.g., in the dorsal attention network) did not exhibit significant longitudinal changes following treatment. (iv) Laterality-Masked Alterations (Pain-Side Aligned): While the aforementioned three patterns were observed in standard whole-brain analyses, the heterogeneous lateralization of TN masked highly specific pathological changes. By systematically aligning the data to the affected side, we unveiled severe ipsilateral decoupling (e.g., in the ipsilateral limbic network) that was previously obscured by mixed lateralities (Supplementary Table S3, S4, S5). Comparing aligned and unaligned outcomes confirmed the necessity of this procedure. Alignment eliminated numerous artefactual, diffusely distributed regions (e.g., removing 60 false-positive regions in the TNpost vs. TNpre comparison) driven by mixed lateralities. Conversely, it unveiled highly specific, masked alterations localized strictly to the ipsilateral hemisphere. For example, baseline disease effects showed severe S-F decoupling peaking in the ipsilateral limbic network (e.g., LH_Limbic_OFC_3, Cohen’s d = 0.857) and control network (LH_Cont_Par_6, Cohen’s d = 0.908). Furthermore, the network distribution of these significant regions of interest (ROIs) was overwhelmingly asymmetric (Supplementary Fig. S1), confirming that the pathological S-F decoupling is tightly lateralized to the affected side. Association between S-F coupling and clinical outcomes We assessed whether the S-F coupling was influenced by baseline clinical subgroup characteristics. ANCOVA controlling for age, sex, and mean FD revealed that pain laterality (left / right), pain phenotype (paroxysmal / persistent), or the severity of neurovascular compression (Sindou grade) had no significant impact on baseline S-F coupling at both the global and network levels (all ). Furthermore, these categorical clinical variables explained a negligible portion of the brain network variance partial . Consequently, it was not necessary to include them as confounding factors in subsequent analyses. To ensure statistical robustness, clinical relevance was examined using partial Spearman correlation analyses with 1,000 permutations. At the macroscopic network level, the baseline S-F coupling of the dorsal attention network showed an exploratory negative correlation with disease duration (, , ). At the regional level, utilizing the lateralization-aligned dataset, exploratory associations were observed between baseline S-F coupling and baseline pain severity in several core regions, including the ipsilateral SalVentAttn_FrOperIns_5 (, ) and contralateral SomMot_16 (, ). In the longitudinal dimension, the change in S-F coupling (S-F coupling) exhibited strong clinical relevance to postoperative recovery. Crucially, without the need for multiple testing correction at the single global metric level, the whole-brain S-F coupling was significantly and positively correlated with the postoperative pain reduction rate (, , ), and showed a significant negative correlation with long-term follow-up NRS scores (, , ; Fig. 4E). At the network level, given the stringent penalty of multiple comparisons, findings were treated as exploratory. Unadjusted associations indicated that S-F coupling in the limbic (, ) and frontoparietal control (, ) networks correlated with pain reduction. Furthermore, robust exploratory associations with long-term pain relief (follow-up NRS) were observed across the somatomotor (, ), dorsal attention (, ), limbic (, ), frontoparietal control (, ), and ventral attention / salience networks (, ; Fig. 4F). Consistently, regional S-F coupling within the bilateral somatomotor networks (e.g., Ipsi_SomMot_31, , ) was also closely associated with postoperative pain relief. This multi-scale evidence suggests that a greater magnitude of postoperative S-F coupling reorganization corresponds to better long-term pain control. Transcriptomic association analysis: exploring spatial transcriptomic correlates of S-F coupling changes To explore the spatial transcriptomic profiles co-localized with the observed changes in S-F coupling, we performed a spatial gene expression association analysis using the human brain transcriptome atlas. We separately identified gene expression patterns that spatially correlated with the disease effect (difference between HCs and TNpre) and the treatment effect (difference between TNpost and TNpre). Spatial transcriptomic associations with the disease effect We first employed a PLS regression analysis to identify the gene expression pattern that was spatially correlated with the S-F coupling difference map for the disease effect. Crucially, to control for spatial autocorrelation, we assessed statistical significance using a rigorous spatial permutation procedure (1,000 spin-tests). We then screened the genes in the top 5% of PLS1 weights for subsequent gene enrichment. PLS1 explained 17.16% of the spatial variance in gene expression (Fig. 5A), and its expression pattern showed a significant positive spatial correlation with the disease effect map () (Fig. 5B). The specific gene list is provided in Supplementary Table S6. Molecular basis of the disease effect on S-F coupling. (A) Explained variance and spatial permutation testing of PLS components. The left plot shows the percentage of spatial variance in gene expression explained by the first 10 PLS components, with PLS1 explaining the most variance. The right plot displays the null distribution generated from spatial permutation tests (spin-tests), confirming the statistical significance of the PLS1 spatial correlation (Pspin = 0.0380) (B) Spatial correlation between PLS1 and the disease effect map. The scatterplot demonstrates a significant positive correlation (r = 0.414, P < 0.001) between the gene expression scores of PLS1 (X-axis) and the “Disease Effect” S-F coupling T-map (Y-axis). This indicates that the reduction in S-F coupling is spatially coupled with a specific multivariate gene expression pattern. (C) GO enrichment analysis. This panel displays the biological processes enriched in the top-weighted genes (top 5%) of PLS1, highlighting key pathways related to energy metabolism (e.g., generation of precursor metabolites and energy) and cellular ionic homeostasis. (D) Association between key candidate genes and the S-F coupling difference. The scatter plots show the results of the hypothesis-driven analysis, where the expression maps of several pain-related candidate genes (e.g., SCN3A, KNG1, and SCN8A) were significantly correlated with the “Disease Effect” T-map. All displayed correlations survived correction for multiple comparisons ( < 0.001) Gene Ontology (GO) enrichment analysis of the top-weighted (top 5%) genes in PLS1 indicated that regions with pronounced disease-related decrease in S-F coupling were spatially co-localized with the expression of genes primarily enriched in pathways related to energy metabolism and cellular ionic homeostasis (Fig. 5C). These included “generation of precursor metabolites and energy” (GO:0006091), “glycolysis and gluconeogenesis” (WP534), and “mitochondrion organization” (GO:0007005). Functions classically implicated in neuronal excitability and central sensitization, such as “monoatomic cation transmembrane transport” (GO:0098655) and “gamma-aminobutyric acid transport” (GO:0015812), were also significantly enriched (Supplementary Table S7). In the exploratory hypothesis-driven analysis, we evaluated a curated list of classical pain-related candidate genes (GeneCards relevance score > 10). Of the 25 candidate genes available in the AHBA dataset, the spatial expression of eight genes was significantly correlated with the disease effect map. For instance, S-F coupling abnormalities were spatially co-distributed with the expression of several sodium ion channel subunit genes, including SCN3A (), SCN8A (), and SCN1A (). Furthermore, genes implicated in for neuroinflammation and synaptic plasticity, such as KNG1 (), IL1B (), and BDNF (), also exhibited robust spatial associations with the disease effect (Fig. 5D, Supplementary Table S5, Supplementary Table S8). Spatial transcriptomic associations with the treatment effect A corresponding PLS analysis for the treatment effect t-map (TNpost − TNpre) identified a PLS1 that explained 27.71% of the spatial variance in gene expression (Fig. 6A) and screened the genes in the top 5% of PLS1 weights for subsequent gene enrichment. The specific gene list is provided in Supplementary Table S9. This component was strongly positively correlated with the spatial distribution of treatment-induced S-F coupling changes (r = 0.508, ; Fig. 6B). Molecular basis of the treatment effect on S-F coupling. (A) Explained variance and spatial permutation testing of PLS components. The left plot shows the percentage of spatial variance in gene expression explained by the first 10 PLS components. The right plot displays the null distribution generated from spatial permutation tests (spin-tests), confirming the statistical significance of the PLS1 spatial correlation(Pspin = 0.0130). (B) Spatial correlation between PLS1 and the treatment effect map. The scatterplot demonstrates a strong and significant positive correlation (r = 0.526, P < 0.001) between the gene expression scores of PLS1 (X-axis) and the “Treatment Effect” S-F coupling T-map (Y-axis). (C) GO enrichment analysis. This panel displays the biological processes enriched in the top-weighted genes (top 5%) of PLS1, highlighting key pathways related to neuroplasticity and synaptic signaling (e.g., modulation of chemical synaptic transmission and neuroactive ligand signaling). (D) Association between key candidate genes and the S-F coupling difference. The scatterplots show the results of the hypothesis-driven analysis, where the expression maps of several plasticity- and pain-modulation-related candidate genes(e.g., CACNA1H, OPRM1, and TNF) were significantly correlated with the “Treatment Effect” T-map. All displayed correlations survived correction for multiple comparisons ( < 0.001) GO and pathway enrichment analysis of the top-weighted genes (top 5%) in PLS1 revealed significant enrichment for biological processes related to neuroplasticity and synaptic signaling (Fig. 6C). Notably, the most significant pathway identified was “Metallothioneins bind metals” (R-HSA-5661231), a pathway functionally associated with regulating ionic homeostasis and oxidative stress. This was closely followed by pathways implicated in functional and structural remodeling, including “modulation of chemical synaptic transmission” (GO:0050804), “cholesterol biosynthesis pathway” (WP197), “neuroactive ligand signaling” (hsa04082), and “regulation of secretion by cell” (GO:1903530) (Supplementary Table S10). At the candidate gene level, 16 of the 25 genes showed expression patterns significantly spatially correlated with the treatment effect. Representative examples included genes related to calcium channel function (CACNA1H; ), opioid signaling (OPRM1; ), neuroinflammation (TNF;) and sodium channel function (SCN9A;) (Fig. 6D, Supplementary Table S8). Discussion By integrating multi-scale neuroimaging with transcriptomic association analysis, this study systematically provides insights into the pathophysiological correlates of TN and the neuroplastic associated with changes induced by surgical treatment. Resonating with emerging multi-omics studies on TN and migraine published in this journal [36, 37], our framework bridges the gap between macroscopic imaging phenotypes and microscopic biological underpinnings drivers. We found that: (i) compared with HCs, patients with TN exhibit a widespread reduction in S-F coupling, which is significantly correlated with disease duration, and their brain’s S-F coupling hierarchy presents an abnormally widened gradient; (ii) treatment is accompanied by a global normalization of this abnormality, restoring S-F coupling to a level statistically indistinguishable from healthy baselines. However, tracking regional trajectories reveals that this recovery is not a simple repair process but rather a complex reorganization reflecting both the targeted restoration of affected ipsilateral regions and widespread adaptive plasticity; and (iii) these macroscopic imaging changes are closely associated with the expression patterns of known pain-related and neuroinflammatory genes (e.g., CACNA1H, TNF, and SCN9A) and biological pathways related to neuroplasticity and synaptic signaling [38, 39]. The disease effect: a progressive cerebral footprint of chronic pain A core finding of this study is the significant global reduction of S-F coupling in patients with TN. S-F coupling reflects the constraining relationship between the brain’s functional activity and its underlying structural connectivity [40]. A reduction in coupling may represent a degree of decoupling between function and structure, implying a disruption in the structural constraints on functional dynamics [41]. We interpret this phenomenon, particularly the significant reductions in the sensorimotor and attention networks, as being closely associated with persistent disruption and damage to core pain-processing circuits by chronic, intense nociceptive input. This aligns with findings of network dysfunction observed in other chronic pain conditions [42]. We also discovered that the S-F coupling network hierarchy within patients with TN exhibits an abnormally widened gradient compared with HCs. We propose that this may reflect a form of maladaptive plasticity [43], at the core of which is a progressive loss of structural constraints along the cortical hierarchy [44]. To cope with the continuous and disruptive pain signals, the brain may be forced to strengthen the segregation between different functional systems, reducing flexible crosstalk between networks to maintain essential functions [45]. However, this short-term coping strategy is detrimental in the long run, potentially predisposing the brain to a loss of healthy flexibility in the brain’s functional organization. Our disease duration analysis provides strong longitudinal evidence for this hypothesis: the longer the disease duration, the lower the S-F coupling. This strongly suggests that the impact of TN on the brain may reflect a progressive and cumulative process of damage, with the brain becoming increasingly entrenched in this inefficient and rigid functional mode over time. A new framework for neural recovery: adaptation-led, limited repair In our longitudinal analysis, threshold-based significance testing initially suggested that direct restoration to normal patterns was highly focal. However, effect-size analyses revealed a markedly different picture. At the global level, surgical treatment was associated with a robust restoration of S-F coupling to levels indistinguishable from HCs, indicating a return toward global homeostasis (). This global normalization was supported by substantial regional reorganization, with paired-samples t-tests identifying 144 cortical parcels exhibiting large recovery effect sizes (e.g., LH_Vis_10, ). Rather than a simple focal repair of initially affected regions, recovery appeared to involve widespread, network-specific reconfiguration. Although surgical treatment successfully relieved peripheral neurovascular compression, the prolonged abnormal nociceptive and sensorimotor inputs characteristic of TN may leave a persistent imprint on core sensory-attentional systems [46]. Indeed, evidence suggests that the neural basis of TN chronification can shift over time from primary sensory processing circuits to brain networks associated with emotion and memory [47], reflecting a form of maladaptive plasticity that is not readily undone. Because this profound imprint is difficult to reverse directly, our effect-size analyses suggest that the homeostatic balance of the entire brain system likely relies on a synergistic combination of targeted S-F restoration and broader adaptive reorganization. Within this broader network-wide reorganization, the widespread adaptive reorganization of the visual network was particularly prominent, exhibiting a massive recovery effect size (, PFDR < 0.001). Although the exact underlying process is unclear, one possible explanation is sensory re-weighting: when somatosensory input (especially from the face) is chronically unreliable or corrupted by paroxysmal pain, the brain may systematically increase its reliance on other stable sensory channels, such as vision, to maintain an accurate perception of the body and the external environment [48]. Meanwhile, the substantial enhancement of higher-order association cortices, such as the default mode network and limbic networks(e.g., limbic network , default mode network ), may reflect a form of top-down cognitive modulation [42]. These networks are central to self-referential thought, episodic memory, and emotion regulation. The involvement of the limbic network observed in our study is consistent with recent high-resolution MRI evidence suggesting structural alterations in the hippocampus and amygdala of TN patients [49]. Consequently, their enhancement may be involved in suppressing residual abnormal sensory signals, regulating negative emotions associated with TN, or remodelling the self-perception of a new post-treatment physical state, thereby facilitating functional recovery at a higher cognitive level [50]. This broad, effect-size-supported reorganization is a core feature of the treatment-induced brain plasticity observed in our study. It suggests that the brain’s recovery is not a simple linear repair process but an active, adaptive functional reconstruction involving multiple systems and levels. Theoretical implications of altered S-F coupling in TN From a network neuroscience perspective, structural connectivity forms a stable anatomical scaffold that shapes and constrains the dynamic repertoire of functional interactions. Our finding of regional S-F decoupling in TN suggests a pathological divergence where functional network synchrony operates independently of its underlying white matter constraints. In TN, the constant barrage of ectopic nociceptive inputs from the demyelinated trigeminal nerve may drive excessive, unconstrained functional plasticity in the cortex. This S-F decoupling could represent a macroscopic network analogue to the microscopic “hyperexcitability” seen in TN, where functional pain circuits fire aberrantly due to a loss of structural regulation. Furthermore, the postoperative re-coupling observed in our study indicates that microvascular decompression not only removes the peripheral trigger but fundamentally restores the centralized structural constraints over functional dynamics, guiding the brain back toward a healthy, structurally-constrained functional state. Spatial transcriptomic associations: bridging macroscopic imaging and microscopic molecular architectures This study links the macroscopic S-F coupling changes in TN to spatial transcriptomic correlates [33], highlighting the distinct biological pathways that may spatially co-localize with the disease state and treatment-induced recovery. Our data-driven PLS analysis found that the gene set spatially associated with the disease effect was primarily enriched in pathways related to energy metabolism (e.g., glycolysis and mitochondrion organization) and cellular ionic homeostasis (e.g., monoatomic cation and gamma-aminobutyric acid [GABA] transport). In contrast, the genes whose spatial expression patterns co-aligned with the treatment effect were significantly enriched in pathways like neuroplasticity and synaptic signaling (e.g., modulation of chemical synaptic transmission and neuroactive ligand signaling) [51]. This spatial shift in transcriptomic correlates from metabolic exhaustion to active synaptic signaling strongly suggests that the brain’s recovery following treatment is not a simple functional quiescence but an active biological process involving structural remodelling and connectional plasticity. We further focused these findings on specific candidate genes [52]. The results demonstrated that the S-F coupling abnormality map was significantly spatially correlated with the normative expression patterns of several key sodium ion channel genes (e.g., SCN3A, SCN8A, SCN1A) and genes critical for neuroinflammation and synaptic plasticity [53] (e.g., IL1B, KNG1, and BDNF). This spatial co-localization is consistent with the hypothesis that TN is fundamentally a channelopathy complicated by neuroinflammatory processes [54]. Notably, this finding is highly consistent with clinical pharmacology: the first-line drugs for TN, such as carbamazepine and oxcarbazepine, act primarily by blocking voltage-gated sodium channels to stabilise hyperexcitable neuronal membranes and suppress ectopic discharges [55]. Our S-F coupling map may therefore provide a macroscopic visualization of the spatial distribution of these clinical drug targets. More interestingly, when analysing the treatment effect, we found that brain regions with higher SCN9A expression also showed a greater improvement in S-F coupling post-treatment. This was further corroborated by spatial associations with genes related to calcium channels (CACNA1H) and the endogenous opioid system (OPRM1) [56], and neuroinflammation (TNF). The seemingly paradoxical dual spatial mapping of SCN9A in both disease and recovery may point to a profound spatial organizing principle: brain regions that are more specialised at the molecular level (i.e., have high normative SCN9A expression) for processing nociceptive signals may also be the ones with the greatest potential for plasticity and recovery once the peripheral pathology is resolved. This suggests that the spatial transcriptomic landscapes co-localizing with disease pathogenesis and treatment-induced recovery may be highly overlapping. In summary, our findings indicate that the S-F coupling map has the potential to serve as a multi-dimensional imaging biomarker that reflects brain network integrity, underlying spatial transcriptomic architectures, and its plastic potential. Clinical implications and future directions Our results suggest that S-F coupling patterns reflect the brain’s plastic potential. Pre-treatment S-F coupling maps, particularly the baseline integrity of adaptive networks (e.g., default mode, visual), could potentially serve as indicators for MVD outcomes. Future longitudinal validation of this link would enable clinicians to identify optimal surgical candidates and personalize treatment decisions. Indeed, the strategy of utilizing neuroimaging-derived spatiotemporal markers to evaluate therapeutic efficacy and guide clinical decisions has been successfully demonstrated in other neurological fields, such as Parkinson’s disease rehabilitation [57]. Our findings also highlight a potential adjunctive therapeutic strategy: moving from merely suppressing pain signals to actively enhancing the brain’s adaptive capacity. Non-invasive brain stimulation could theoretically target the specific adaptive regions we identified (such as the visual or default mode networks) to facilitate or optimize recovery. This strategy may offer a novel therapeutic option for patients refractory to traditional medication or with suboptimal surgical outcomes. Finally, our exploratory transcriptomic analysis points to potential new molecular targets. While current drugs for TN primarily target ion channels, our findings reveal that successful recovery is spatially linked to gene pathways for governing neuroplasticity (e.g., modulation of chemical synaptic transmission and neuroactive ligand signaling). This suggests a promising avenue for drug development: novel agents that modulate adaptive synaptic plasticity. Such agents could potentially alleviate not only chronic pain but also its psychiatric comorbidities by helping the global brain network regain homeostasis. Limitations of the study Several limitations should be noted. First, our transcriptomic associations were derived from the publicly available, AHBA. While this validated approach captures spatial correlations between canonical gene expression profiles and macroscopic imaging phenotypes, it does not reflect individual genetic variations in patient’s dynamic disease-induced transcriptomic alterations, or imply any direct causal relationships. Second, the modest sample size for the longitudinal analysis (N = 39) limited our statistical power to differentiate the potential effects of different treatment modalities (MVD vs. PI) or medication use. Finally, although we established the clinical relevance of S-F coupling by demonstrating its significant associations with pain severity, disease duration, and postoperative relief, these findings remain fundamentally associative. Therefore, while we observed a robust network-level reorganization associated with surgical intervention, future studies with larger, independent cohorts and patient-specific multi-omics data are necessary to validate these findings and translate them into clinical practice. Conclusions Through a longitudinal, multi-scale study design, this work reveals that the progressive impairment of brain S-F coupling in patients with TN closely tracks clinical pain severity and disease chronification. Following surgical treatment, the return to global homeostasis is driven not merely by localized repair, but by an extensive, network-specific adaptive reorganization that dictates long-term pain relief. Furthermore, we link these macroscopic imaging changes to specific transcriptomic profiles associated with neuronal metabolism and synaptic plasticity, providing a new perspective for understanding the central pathophysiology of the disease. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Acknowledgements We thank the research team at the Federal Neurosurgical Centre in Novosibirsk, Russia, for acquiring and publicly sharing the dataset used in this study. We are also grateful to all the participants of the original study. This work was made possible by the support of several open-source analysis tools, and we extend our sincere gratitude to the developers of software including fMRIPrep, PANDA, Abagen, and Nilearn for their invaluable contributions to the neuroimaging community. Abbreviations Trigeminal Neuralgia Healthy Controls Pre-treatment Trigeminal Neuralgia Post-treatment Trigeminal Neuralgia Structure-Function coupling Functional Magnetic Resonance Imaging Diffusion Tensor Imaging Functional Connectivity Structural Connectivity Partial Least Squares Allen Human Brain Atlas Microvascular Decompression Percutaneous Interventions Numeric Rating Scale Framewise Displacement Analysis of Variance Analysis of Covariance False Discovery Rate Regions of Interest Gene Ontology Visual Network Somatomotor Network Dorsal Attention Network Ventral Attention Network Limbic Network Frontoparietal Network Default Mode Network Author contributions Jian Zhang, Xujing Nie, and Shuyue Fu contributed to the conception, design, acquisition, analysis, and interpretation of the work and drafted the manuscript. Yinping Lu, Tiantian Liu, and Zhilin Zhang contributed to the acquisition, analysis, interpretation, and drafted the manuscript. Jinglong Wu, Xuezhi Tong, and Tianyi Yan contributed to the conception, design, and critically revised the manuscript. All authors gave their final approval and agree to be accountable for all aspects of the work. Funding This work was supported by the National High Level Hospital Clinical Research Funding, Elite Medical Professionals Project of China-Japan Friendship Hospital(NO.ZRJY2023-QM19), the National Natural Science Foundation of China (grant numbers 62306035, 62373056, 62336002, 62406025), Key-Area Research and Development Program of Guangdong Province (2023B0303030002), STI2030-Major Projects (2022ZD0208500), Shenzhen Basic Research Program (JCYJ20241202124804007), Beijing Institute of Technology Young Faculty Academic Launch Program(xsqd-6120230052). Data availability The raw neuroimaging data used in this study were obtained from the publicly available OpenNeuro database (accession number: ds005713, version: 2.0.1) and can be accessed at https://openneuro.org/datasets/ds005713/versions/2.0.1. The custom code, parameter files, and computational environment details used for the graph harmonic modeling and PLS analysis in this study are available on GitHub at https://github.com/starz0275/TN_SF_Code. Declarations Ethics approval and consent to participate The study protocol was approved by the local ethics committee. All participants provided written informed consent. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Citations Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The raw neuroimaging data used in this study were obtained from the publicly available OpenNeuro database (accession number: ds005713, version: 2.0.1) and can be accessed at https://openneuro.org/datasets/ds005713/versions/2.0.1. The custom code, parameter files, and computational environment details used for the graph harmonic modeling and PLS analysis in this study are available on GitHub at https://github.com/starz0275/TN_SF_Code.