Transcriptomic decoding of regional cortical vulnerability to drug-resistant epilepsy using 7T MRI Authors: Haixia Mao, Teppei Matsubara, Naoaki Tanaka, Noam Peled, Hana Farzaneh, Diana Melania lon, Nao Suzuki, Robert Mark Richardson, Andrew J. Cole, Xiangming Fang, Steven M. Stufflebeam Published: 2025 ABSTRACT The mechanism by which genetic risk leads to cortical vulnerability in drug-resistant epilepsy (DRE) remains unclear. This study used 7T structural and resting-state functional MRI to investigate cortical neural activity alterations in 105 DRE patients and 105 healthy controls (HCs), and to explore related genetic mechanisms. Vertex-wise analyses of mean amplitude of low-frequency fluctuation (mALFF) and regional homogeneity (ReHo) revealed that DRE patients primarily exhibited decreased mALFF and increased ReHo in the Cingulo-Opercular Network. Using the Allen Human Brain Atlas, we conducted spatial transcriptomic analysis via partial least squares (PLS) and gene enrichment analysis to identify gene categories associated with these functional changes. The results showed that cortical alterations were related to epilepsy-general genes (e.g., TMEM74, KCNN2, RBFOX1) and brain-relevant genes. Genes positively correlated with mALFF alterations enriched in mitochondrial inner membrane, matrix, and carboxylic acid metabolism; negatively in chromatin remodeling, binding, and postsynapse. Genes positively correlated with ReHo alterations enriched in nucleic acid-related catalytic activity, ribonucleoprotein granule, and centrosome; negatively in amyotrophic lateral sclerosis, mitochondrial membrane, and pyrophosphatase activity. These findings link spatial brain activity abnormalities in DRE to specific genetic signatures and biological pathways, suggesting new mechanistic insights and potential therapeutic targets for this difficult-to-treat condition. Introduction Epilepsy affects 0.5–1.5% of the general population1, results in substantial economic burdens2, and contributes to an estimated 0.7% of the global disease burden3. It is a significant neurological disorder that carries a heavy social stigma4, and is often linked with psychiatric conditions5. About one-third of patients continue to have seizures despite medical treatment6. Drug-resistant epilepsy (DRE) is defined as the failure to achieve sustained seizure freedom after trials of two tolerated and appropriately chosen antiseizure drug regimens, whether used as monotherapies or in combination7. Individuals with DRE face significantly higher risks of morbidity, functional impairment, and premature mortality compared to those whose seizures are effectively controlled8. Genes play a crucial role in both the onset and treatment of epilepsy9. Genetic variations can contribute to the development of epilepsy by affecting neuronal excitability10, synaptic transmission11, and network connectivity12, and other related factors in the brain. In addition, genetic factors can influence how individuals respond to antiepileptic drugs, making pharmacogenomic research essential in identifying effective treatments for patients13. However, despite significant progress in understanding the genetic underpinnings of epilepsy, many questions remain unresolved. The complexity of genetic interactions, coupled with the influence of environmental and epigenetic factors, means that pinpointing the exact mechanisms behind epilepsy remains challenging. Furthermore, if genetic causes of DRE are confirmed, surgery may still be an option, but counseling regarding recurrence risk and other factors will be significantly altered14. Resting-state functional MRI (rs-fMRI) is a valuable tool for investigating brain activity in patients with DRE. A wealth of rs-MRI studies has revealed cortical functional abnormalities in DRE patients15–17. These studies commonly employ functional measures such as the amplitude of low-frequency fluctuations (ALFF), which assesses the strength of local spontaneous neural activity by measuring voxel-level activity in the 0.01–0.08 Hz frequency range within local brain regions, and regional homogeneity (ReHo), which evaluates the degree of local functional synchronization by comparing the similarity of a voxel’s time series to those of its neighboring voxels using Kendall’s coefficient of concordance (KCC)18. However, previous studies investigating regional neural activity through ALFF and ReHo in DRE patients have produced inconsistent results, largely due to variations in patient populations, differences in scanning equipment, and other methodological limitations. Ultra-high-field 7T MRI, with its superior signal-to-noise ratio and enhanced sensitivity to susceptibility effects, offers unparalleled lesion visualization19. 7T MRI demonstrates higher stability and reliability in rs-fMRI studies, enabling more precise investigations of brain functional networks20. However, the underlying molecular and cellular mechanisms that predispose specific brain regions to cortical functional activity alterations remain inadequately understood. Large-scale human genome studies have found a range of risk variants, genes, and gene sets in relation to DRE21, yielding an updated framework for our thinking regarding how genetic variants interact with epigenetic and environmental risk factors in the brain to confer risk for DRE. However, most studies on gene expression changes in epileptic brains have been conducted using resected brain tissue samples, aiming to assess average alterations across different types of neurons, glial cells, and non-neuronal cells22. Although such studies provide valuable insights into large-scale gene expression changes in epilepsy, the lack of spatial information makes it difficult to pinpoint the specific brain regions or cells responsible for seizure activity and disease progression. The Allen Human Brain Atlas (AHBA) is a brain-wide gene expression atlas23 that provides high-resolution spatial information, enabling detailed transcriptomic analyses across diverse brain regions and specific cellular subtypes, while bridging the gap between connectomes and transcriptomes24,25. AHBA serves as a crucial reference for understanding brain-wide gene expression patterns and establishing a transcriptional baseline for neurogenetic studies of both normal and pathological brain function26. Moreover, several studies have demonstrated a strong correlation between spatial transcriptional patterns in healthy individuals and the structural and functional alterations observed in neurological and psychiatric disorders27–29. A growing body of neuroimaging genetics research has revealed that genetic variations can shape cortical functional activity, potentially increasing susceptibility to neurological disorders30. Although significant progress has been made, the mechanisms linking genetic risk to region-specific alterations in cortical function in DRE are still poorly understood. With the advent of precision medicine, individualized treatment strategies, such as gene editing, have increasingly garnered attention31. Gaining a deeper understanding of the relationship between cortical functional abnormalities and genetic factors could pave the way for more precise and targeted development of novel treatments of DRE. Our study aimed to investigate the potential associations between gene expression and cortical functional abnormalities in patients with DRE using 7 T MRI. Given the high heritability of cortical surface features32, we hypothesized that alterations in cortical functional activity may reflect a continuum of genetic vulnerability, which manifests as distinct clinical phenotypes and may contribute to treatment resistance in epilepsy. We first examined cortical functional alterations in DRE patients and assessed their associations with clinical variables, such as disease duration, to explore the potential progression-dependent nature of these changes and to reveal possible mechanisms of neuroplasticity or functional compensation. We then conducted spatial transcriptomic analysis to identify candidate genes linked to the observed functional abnormalities. Finally, gene enrichment analysis was performed to investigate the biological processes associated with these candidate genes, providing insights into the molecular mechanisms potentially underlying drug resistance in DRE. Results Demographic and clinical data A total of 105 patients with DRE were included, consisting of 48 males and 57 females. They had a mean age of 30.0 ± 12.9 years and a median age of 26 years. For the 105 healthy controls (HCs), there were 50 males and 55 females, with a mean age of 27.7 ± 2.4 years and a median age of 28 years. No statistically significant differences were observed in gender or age between the two groups (p-values = 0.89 and 0.271, respectively). The detailed demographic and clinical data for both groups are summarized in Table 1.Table 1Clinical characteristics of the study cohortHCs (n = 105)DRE (n = 105)pSex0.89a Male5048 Female5557Age(years)0.271b Mean ± SD27.7 ± 2.430.0 ± 12.9 Median [IQR]28(28,28)26(21,40)Age at onset (years) Mean ± SD-16.8 ± 12.4 Median [IQR]-16 (7, 22)Duration of disease (years) Mean ± SD-13.2 ± 10.6 Median [IQR]-10 (5,18.75)Total treatment duration (years) Mean ± SD-11.5 ± 9.4 Median [IQR]-9 (5, 15.5)Localization Temporal Right-19 Left-24 Both-9 Frontal Right-15 Left-8 Both-6 Parietal Right-3 Left-3 Both-0 Occipital Right-1 Left-0 Both-0 Multiple Lobes Right-1 Left-9 Both-7HCs healthy controls, DRE drug-resistant epilepsy, SD standard deviation, IQR interquartile range.aChi-squared test (χ² = 0.019; p = 0.89).bSkewed distribution, Mann–Whitney U test (Z = −1.101; p = 0.271). Table 1Clinical characteristics of the study cohortHCs (n = 105)DRE (n = 105)pSex0.89a Male5048 Female5557Age(years)0.271b Mean ± SD27.7 ± 2.430.0 ± 12.9 Median [IQR]28(28,28)26(21,40)Age at onset (years) Mean ± SD-16.8 ± 12.4 Median [IQR]-16 (7, 22)Duration of disease (years) Mean ± SD-13.2 ± 10.6 Median [IQR]-10 (5,18.75)Total treatment duration (years) Mean ± SD-11.5 ± 9.4 Median [IQR]-9 (5, 15.5)Localization Temporal Right-19 Left-24 Both-9 Frontal Right-15 Left-8 Both-6 Parietal Right-3 Left-3 Both-0 Occipital Right-1 Left-0 Both-0 Multiple Lobes Right-1 Left-9 Both-7HCs healthy controls, DRE drug-resistant epilepsy, SD standard deviation, IQR interquartile range.aChi-squared test (χ² = 0.019; p = 0.89).bSkewed distribution, Mann–Whitney U test (Z = −1.101; p = 0.271). Clinical characteristics of the study cohort HCs healthy controls, DRE drug-resistant epilepsy, SD standard deviation, IQR interquartile range. aChi-squared test (χ² = 0.019; p = 0.89). bSkewed distribution, Mann–Whitney U test (Z = −1.101; p = 0.271). Changes in cortical indicators The mALFF and ReHo analysis revealed significant differences between the DRE and HCs groups (Figs. 1A and 2A). For mALFF mapping, the DRE group exhibited increased mALFF primarily in lower-order unimodal primary networks, such as the Visual Network (VN). Additionally, localized increases were observed in certain transmodal networks, including the Dorsal Attention Network (DAN) and Ventral Attention Network (VAN). In contrast, a general pattern of decreased mALFF was identified across higher-order transmodal associative networks, including the Cingulo-Opercular Network (CON) and Frontoparietal Network (FPN). Detailed information on brain regions with significant mALFF differences between the DRE and HC groups, including the number of peaks and their coordinates, is presented in Table 2 (random field theory [RFT] correction, p < 0.001, cluster size > 100 vertices).Fig. 1Alterations in mALFF in patients with DRE and its association with the PLS1 gene.A Compared to HCs, DRE patients exhibited significant increases and decreases in mALFF (p < 0.001, cluster-level RFT corrected). Warm colors indicate regions with increased imaging metrics, while cool colors indicate decreased metrics. B Standardized and ranked PLS1 gene scores associated with mALFF alterations. C A scatterplot showing the relationship between regional PLS1 scores (a weighted sum of 15,632 gene expression scores) and regional mALFF changes in the left hemisphere.Fig. 2Alterations in ReHo in patients with DRE and its association with the PLS1 gene.A Compared to HCs, DRE patients demonstrated significant increases and decreases in ReHo (p < 0.001, cluster-level RFT corrected). Warm colors represent regions with increased synchronization, while cool colors represent decreased synchronization. B Standardized and ranked PLS1 gene scores related to ReHo alterations. C A scatterplot illustrating the association between regional PLS1 scores (a weighted sum of 15,632 gene expression scores) and regional ReHo changes in the left hemisphere.Table 2Brain regions with differences in mALFF between the DRE and HCs groupsCluster IDSizeT-valuep-valuePeak coordinatesAtlas region(vertices)Peak XPeak YPeak Z11934.218058<0.0001−15.7464−76.672632.94602L_Visual_7a21594.79728<0.000214.006−39.0984−4.02625R_Visual_22a31384.815004<0.0003−12.2122−43.2703−2.22558L_Visual_6a41544.370208<0.0004−41.233832.33434−1.34212L_VentralAttn_8a51323.787136<0.000521.91819−66.691954.35797R_DorsalAttn_27a63009−5.5491<0.000647.55512−27.987922.53287R_CinguloOperc_30b7665−4.15512<0.000762.84552−39.9219−6.97217R_FrontoParietal_12baIndicates a higher value in the DRE group than in the HCs group.bIndicates a lower value in the DRE group than in the HCs group. Alterations in mALFF in patients with DRE and its association with the PLS1 gene. A Compared to HCs, DRE patients exhibited significant increases and decreases in mALFF (p < 0.001, cluster-level RFT corrected). Warm colors indicate regions with increased imaging metrics, while cool colors indicate decreased metrics. B Standardized and ranked PLS1 gene scores associated with mALFF alterations. C A scatterplot showing the relationship between regional PLS1 scores (a weighted sum of 15,632 gene expression scores) and regional mALFF changes in the left hemisphere. Alterations in ReHo in patients with DRE and its association with the PLS1 gene. A Compared to HCs, DRE patients demonstrated significant increases and decreases in ReHo (p < 0.001, cluster-level RFT corrected). Warm colors represent regions with increased synchronization, while cool colors represent decreased synchronization. B Standardized and ranked PLS1 gene scores related to ReHo alterations. C A scatterplot illustrating the association between regional PLS1 scores (a weighted sum of 15,632 gene expression scores) and regional ReHo changes in the left hemisphere. Table 2Brain regions with differences in mALFF between the DRE and HCs groupsCluster IDSizeT-valuep-valuePeak coordinatesAtlas region(vertices)Peak XPeak YPeak Z11934.218058<0.0001−15.7464−76.672632.94602L_Visual_7a21594.79728<0.000214.006−39.0984−4.02625R_Visual_22a31384.815004<0.0003−12.2122−43.2703−2.22558L_Visual_6a41544.370208<0.0004−41.233832.33434−1.34212L_VentralAttn_8a51323.787136<0.000521.91819−66.691954.35797R_DorsalAttn_27a63009−5.5491<0.000647.55512−27.987922.53287R_CinguloOperc_30b7665−4.15512<0.000762.84552−39.9219−6.97217R_FrontoParietal_12baIndicates a higher value in the DRE group than in the HCs group.bIndicates a lower value in the DRE group than in the HCs group. Brain regions with differences in mALFF between the DRE and HCs groups aIndicates a higher value in the DRE group than in the HCs group. bIndicates a lower value in the DRE group than in the HCs group. For ReHo mapping, the DRE group showed increased ReHo in lower-order unimodal primary regions within Somatomotor Network (SMN), VN, and auditory network (AN), as well as in the transmodal network CON and Default Mode Network (DMN). Conversely, decreased ReHo was observed in transmodal networks, including VAN, DMN, DAN, CON, and FPN. Locations of significant ReHo differences between the DRE and HC groups, including peak number and coordinates, are detailed in Table 3 (RFT correction, p < 0.001, cluster size > 100 vertices).Table 3Brain regions with differences in ReHo between the DRE and HCs groupsCluster IDSizeT-valuep-valuePeak coordinatesAtlas region(vertices)Peak XPeak YPeak Z110434.229772<0.000158.47347−44.009525.50527R_CinguloOperc_29a24255.132438<0.000113.8397−39.678−3.38536R_Visual_22a34215.586371<0.00016.69815942.6284−14.2164R_Default_31a41884.568738<0.0001−18.8471−48.5807−2.41176L_Visual_6a51824.433801<0.0001-51.7051-55.490124.80135L_Default_3a61544.20757<0.000138.81203−22.37591.438535R_Auditory_23a71424.143284<0.0001−21.8254−99.9141−2.81463L_Visual_17a81264.485048<0.000118.59371−31.566367.95574R_SMhand_20a91224.066265<0.000150.08604−13.949954.40916R_SMmouth_8a10367−4.4794<0.0001−44.53320.941938.085825L_VentralAttn_5b11331−4.26639<0.000110.3470952.18534−2.31659R_Default_31b12316−4.22863<0.0001−33.16526.87281829.32133L_DorsalAttn_16b13315−4.18993<0.0001−57.6557−4.96656−5.29216L_Default_12b14242−4.96728<0.000134.750144.24220813.06329R_CinguloOperc_33b15232−4.83555<0.000155.0915620.8519112.94751R_VentralAttn_18b16223−4.387<0.0001-13.2863-49.932135.88238L_Default_1b17183−4.26433<0.0001-48.127238.70552−7.57028L_VentralAttn_6b18168−4.61513<0.0001−42.9658−58.4131−17.1717L_DorsalAttn_14b19157−4.56227<0.0001−28.673128.778310.557251L_CinguloOperc_13b20148−4.09124<0.000142.7593329.2816522.74445R_FrontoParietal_18baIndicates a higher value in the DRE group than in the HCs group.bIndicates a lower value in the DRE group than in the HCs group. Table 3Brain regions with differences in ReHo between the DRE and HCs groupsCluster IDSizeT-valuep-valuePeak coordinatesAtlas region(vertices)Peak XPeak YPeak Z110434.229772<0.000158.47347−44.009525.50527R_CinguloOperc_29a24255.132438<0.000113.8397−39.678−3.38536R_Visual_22a34215.586371<0.00016.69815942.6284−14.2164R_Default_31a41884.568738<0.0001−18.8471−48.5807−2.41176L_Visual_6a51824.433801<0.0001-51.7051-55.490124.80135L_Default_3a61544.20757<0.000138.81203−22.37591.438535R_Auditory_23a71424.143284<0.0001−21.8254−99.9141−2.81463L_Visual_17a81264.485048<0.000118.59371−31.566367.95574R_SMhand_20a91224.066265<0.000150.08604−13.949954.40916R_SMmouth_8a10367−4.4794<0.0001−44.53320.941938.085825L_VentralAttn_5b11331−4.26639<0.000110.3470952.18534−2.31659R_Default_31b12316−4.22863<0.0001−33.16526.87281829.32133L_DorsalAttn_16b13315−4.18993<0.0001−57.6557−4.96656−5.29216L_Default_12b14242−4.96728<0.000134.750144.24220813.06329R_CinguloOperc_33b15232−4.83555<0.000155.0915620.8519112.94751R_VentralAttn_18b16223−4.387<0.0001-13.2863-49.932135.88238L_Default_1b17183−4.26433<0.0001-48.127238.70552−7.57028L_VentralAttn_6b18168−4.61513<0.0001−42.9658−58.4131−17.1717L_DorsalAttn_14b19157−4.56227<0.0001−28.673128.778310.557251L_CinguloOperc_13b20148−4.09124<0.000142.7593329.2816522.74445R_FrontoParietal_18baIndicates a higher value in the DRE group than in the HCs group.bIndicates a lower value in the DRE group than in the HCs group. Brain regions with differences in ReHo between the DRE and HCs groups aIndicates a higher value in the DRE group than in the HCs group. bIndicates a lower value in the DRE group than in the HCs group. Comparison of cluster sizes in the altered networks revealed that patients with DRE predominantly exhibited decreased mALFF and increased ReHo in the CON. Correlation analysis In the DRE group, several negative correlations were observed between the average mALFF/ReHo values of individually identified regions (which exhibited statistical differences) and clinical variables. Specifically, the mALFF in the right fronto-parietal region 12 (R_FrontoParietal_12) was negatively correlated with age (r = −0.24, p = 0.014). Meanwhile, ReHo in the right sensorimotor hand region 20 (R_SMhand_20) showed a negative correlation with epilepsy duration (r = −0.21, p = 0.036), and ReHo in the left visual region 17 (L_Visual_17) was negatively correlated with epilepsy onset age (r = −0.23, p = 0.018). Notably, none of these correlations remained statistically significant after false discovery rate (FDR) correction for multiple comparisons (Fig. 3).Fig. 3Correlations between mALFF/ReHo and age, onset age, and duration of epilepsy.A mALFF in the right fronto-parietal region 12 (R_FrontoParietal_12) was negatively correlated with age (r = −0.24, p = 0.014). B ReHo in the right sensorimotor hand region 20 (R_SMhand_20) showed a negative correlation with epilepsy duration (r = −0.21, p = 0.036). C ReHo in the left visual region 17 (L_Visual_17) was negatively correlated with epilepsy onset age (r = −0.23, p = 0.018). Notably, no correlation remained statistically significant after FDR correction for multiple comparisons. Correlations between mALFF/ReHo and age, onset age, and duration of epilepsy. A mALFF in the right fronto-parietal region 12 (R_FrontoParietal_12) was negatively correlated with age (r = −0.24, p = 0.014). B ReHo in the right sensorimotor hand region 20 (R_SMhand_20) showed a negative correlation with epilepsy duration (r = −0.21, p = 0.036). C ReHo in the left visual region 17 (L_Visual_17) was negatively correlated with epilepsy onset age (r = −0.23, p = 0.018). Notably, no correlation remained statistically significant after FDR correction for multiple comparisons. Cortical gene expression related to regional changes in mALFF and ReHo Using standardized protocols33, we constructed a gene expression matrix of 15,633 genes across 161 regions in the left hemisphere (161 regions × 15,633 genes) based on the Gordon 333 Atlas, using high-resolution spatial data from the AHBA database. Partial least squares (PLS) analysis identified the first component (PLS1), which represented the linear combination of gene expression values most strongly correlated with regional changes in mALFF and ReHo maps, was defined as the spatial map accounting for the largest portion of gene expression variance across cortical regions34. PLS1 explained 21.3% and 23.0% of the variance in mALFF and ReHo mappings, respectively (permutation test: 5000 iterations, p < 0.05). We ranked the normalized weights of PLS1 based on univariate one-sample Z tests (Figs. 1B and 2B). For mALFF, we found 906 PLS1+ (Z > 3) genes (Supplementary Data 1) and 945 PLS1− (Z < -3) genes (Supplementary Data 2) (all FDR-corrected p < 0.05) as positively (or negatively) weighted gene expressions34. For ReHo, we found 1120 PLS1+ (Z > 3) genes (Supplementary Data 3) and 1603 PLS1− (Z < −3) genes (Supplementary Data 4) (all FDR-corrected p < 0.05) as positively (or negatively) weighted gene expressions. Both mALFF and ReHo mappings (parcellated using the Gordon 333 template) demonstrated strong correspondence with the brain transcriptomic co-expression network, with linear regression results of r = 0.479, p < 0.0001 for mALFF (Fig. 1C) and r = 0.462, p < 0.0001 for ReHo (Fig. 2C). Gene Integration of epilepsy-related GWAS and transcriptomics of DRE To further explore the relationships between prior epilepsy-related gene expressions and regional changes in mALFF and ReHo, we identified a total of nine overlapping genes from the intersection of PLS1 background genes derived from mALFF and ReHo with 29 candidate epilepsy-associated genes, based on genome-wide significant loci reported by the ILAE Consortium on Complex Epilepsies11. Specifically, seven overlapping genes were identified for mALFF (two PLS1+ and five in PLS1−), while two overlapping genes were identified for ReHo (both in PLS1−) (Fig. 4).Fig. 4Overlapping genes from the intersection of PLS1 background genes derived from mALFF and ReHo with 29 candidate epilepsy-associated genes.A For mALFF, seven overlapping genes were identified, namely TMEM74, KCNN2, RBFOX1, RORB, RYR2, BCL11A, and RSPO2. Among these, TMEM74 and KCNN2 exhibited positive correlations (with regional mALFF changes), while RBFOX1, RORB, RYR2, BCL11A, and RSPO2 showed negative correlations. B For ReHo, two overlapping genes were identified: PTPRK and SCN8A. Both genes were negatively correlated with regional ReHo changes. Overlapping genes from the intersection of PLS1 background genes derived from mALFF and ReHo with 29 candidate epilepsy-associated genes. A For mALFF, seven overlapping genes were identified, namely TMEM74, KCNN2, RBFOX1, RORB, RYR2, BCL11A, and RSPO2. Among these, TMEM74 and KCNN2 exhibited positive correlations (with regional mALFF changes), while RBFOX1, RORB, RYR2, BCL11A, and RSPO2 showed negative correlations. B For ReHo, two overlapping genes were identified: PTPRK and SCN8A. Both genes were negatively correlated with regional ReHo changes. Among the seven genes overlapping with PLS1 in mALFF, all seven showed expression levels associated with regional mALFF alterations (FDR-corrected p < 0.05). Specifically, two genes (TMEM74 and KCNN2) exhibited positive correlations, while five genes (RBFOX1, RORB, RYR2, BCL11A, and RSPO2) showed negative correlations (Fig. 4A). Additionally, two epilepsy-related genes from genome-wide association study (GWAS) (PTPRK and SCN8A) overlapped with PLS1 genes in ReHo, both of which exhibited significant negative correlations with regional changes in ReHo (both FDR-corrected p < 0.05, Fig. 4B). Enrichment analysis The PLS1– and PLS1+ gene enrichment analysis results of mALFF and ReHo were obtained after correcting for enrichment terms (FDR-corrected p < 0.05) and discarding discrete enrichment clusters. For mALFF mapping, the top 20 clusters of PLS1+ gene enrichment spanned eight Gene Ontology (GO) biological process categories, seven GO cellular component category, two GO molecular function category, and three Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. The top three ontology terms identified were “mitochondrial inner membrane” and “mitochondrial matrix” (from the GO cellular component category) and “carboxylic acid metabolic process” (from the GO biological process category) (Fig. 5A). Meanwhile, the top 20 clusters of mALFF-related PLS1− gene enrichment encompassed eleven GO biological process categories, three GO cellular component categories, five GO molecular function categories, and one KEGG pathways. The top terms from the three GO categories—biological process, cellular component, and molecular function—were “chromatin remodeling”, “postsynapse”, and “chromatin binding”, respectively (Fig. 5B).Fig. 5Transcription-neuroimaging spatial correlation for PLS1 genes associated with mALFF.A Network visualization of enriched terms for PLS1+ genes associated with mALFF, where nodes represent terms grouped by cluster ID to indicate functional similarity. Node colors reflect p-values, with larger gene-involved terms generally having smaller p values. B Bubble chart of the top 10 enriched term clusters for PLS1+ genes associated with mALFF. Bubble color represents the multi-test adjusted p-value (log10), and size indicates the number of genes involved. C Network visualization for PLS1− genes, following the same conventions as (A). D Bubble chart of the top 10 enriched term clusters for PLS1− genes, with bubble color and size reflecting p-values (log10) and gene count, respectively. Transcription-neuroimaging spatial correlation for PLS1 genes associated with mALFF. A Network visualization of enriched terms for PLS1+ genes associated with mALFF, where nodes represent terms grouped by cluster ID to indicate functional similarity. Node colors reflect p-values, with larger gene-involved terms generally having smaller p values. B Bubble chart of the top 10 enriched term clusters for PLS1+ genes associated with mALFF. Bubble color represents the multi-test adjusted p-value (log10), and size indicates the number of genes involved. C Network visualization for PLS1− genes, following the same conventions as (A). D Bubble chart of the top 10 enriched term clusters for PLS1− genes, with bubble color and size reflecting p-values (log10) and gene count, respectively. For ReHo mapping, the top 20 clusters of PLS1+ gene enrichment included twelve GO biological process categories, four GO cellular component categories, and four GO molecular function categories. The top three ontology terms identified were “catalytic activity, acting on a nucleic acid” (molecular function), “ribonucleoprotein granule” (cellular component), and “centrosome” (cellular component) (Fig. 6A). Additionally, the top 20 clusters of ReHo-related PLS1− gene enrichment encompassed ten GO biological process categories, six GO cellular component categories, two GO molecular function categories, and two KEGG pathways. The top three ontology descriptions were “amyotrophic lateral sclerosis” (KEGG Pathway), “mitochondrial membrane” (GO Cellular Component), and “pyrophosphatase activity” (GO Molecular Function) (Fig. 6B).Fig. 6Transcription-neuroimaging spatial correlation for PLS1 genes associated with ReHo.A Network visualization of enriched terms for PLS1+ genes associated with ReHo, where nodes represent terms grouped by cluster ID to indicate functional relatedness. Node colors reflect p-values, with larger gene-involved terms generally having smaller p-values. B Bubble chart of the top 10 enriched term clusters for PLS1+ genes associated with ReHo. Bubble color represents the multi-test adjusted p-value (log10), and size indicates the number of genes involved. C Network visualization for PLS1− genes, following the same conventions as (A). D Bubble chart of the top 10 enriched term clusters for PLS1− genes, with bubble color and size reflecting p-values (log10) and gene count, respectively. Transcription-neuroimaging spatial correlation for PLS1 genes associated with ReHo. A Network visualization of enriched terms for PLS1+ genes associated with ReHo, where nodes represent terms grouped by cluster ID to indicate functional relatedness. Node colors reflect p-values, with larger gene-involved terms generally having smaller p-values. B Bubble chart of the top 10 enriched term clusters for PLS1+ genes associated with ReHo. Bubble color represents the multi-test adjusted p-value (log10), and size indicates the number of genes involved. C Network visualization for PLS1− genes, following the same conventions as (A). D Bubble chart of the top 10 enriched term clusters for PLS1− genes, with bubble color and size reflecting p-values (log10) and gene count, respectively. Discussion This study used 7T structural and functional MRI to examine cortical neural activity in DRE patients and its associations with clinical variables. Building on these findings, we conducted transcriptomic and enrichment analyses to investigate whether the affected regions are linked to specific gene expression patterns and biological pathways that may underlie the observed functional alterations and contribute to disease intractability. The results indicated that, compared to the HC group, patients with DRE exhibited significant alterations in mALFF and ReHo, which were correlated with clinical epilepsy-related variables. Functionally, based on cluster size, patients with DRE primarily exhibited decreased mALFF and increased ReHo within the CON. Further transcription-neuroimaging spatial correlation analysis revealed that mALFF and ReHo changes in DRE were strongly associated with genes enriched in biological functions broadly related to DRE. The results may provide insights into the underlying mechanisms of disease pathogenesis in DRE, although further studies including drug-responsive patients are needed to better understand mechanisms specific to drug resistance. This study used high-resolution 7T MRI to examine mALFF and ReHo alterations in DRE patients, reflecting spontaneous neural activity. Compared to HCs, DRE patients showed increased activity in lower-order unimodal regions and decreased activity in higher-order transmodal areas, except in the CON, DMN, and localized VAN and DAN regions. Additionally, mALFF in the FPN was negatively correlated with age. Meanwhile, ReHo in the SMN showed a negative correlation with epilepsy duration, and ReHo in the VN was negatively correlated with age at epilepsy onset. Transmodal regions, such as the DMN, DAN, FPN, and CON—key for attention, memory, internal thought, and information integration35–37—exhibit decreased mALFF and ReHo, indicating reduced spontaneous neural activity38 and impaired local synchronization of neural activity39. This likely reflects pathological inhibition or degeneration in the neural network. Conversely, increased mALFF and ReHo in unimodal regions (SMN, VN, AN) suggest enhanced neural activity and synchronization. Elevated mALFF may indicate compensatory mechanisms or network reorganization to preserve cognitive function under disease conditions. Additionally, they might represent epileptogenic focus in epilepsy. Similarly, areas with increased ReHo may reflect pathological synchrony or localized compensatory adjustments aimed at preserving visual, auditory, and sensorimotor functions. SMN plays a crucial role in the subcortical-cortical circuits of epilepsy patients40. Previous studies have found that epilepsy patients exhibited increased connectivity within the AN and SMN, along with decreased connectivity within the DAN41, which is consistent with the results of this study. VN plays a key role in epilepsy, with visual auras, hallucinations, and transient vision loss occurring across seizure stages. Additionally, certain antiseizure medications (ASMs) may contribute to specific visual disturbances, thereby affecting therapeutic efficiency for patients with epilepsy42. DMN was reported to have closely associated with the generation and propagation of epileptic activity43. A previous study found significant DMN and DAN connectivity changes in DRE, with increased activity and abnormal interactions potentially contributing to frequent seizures44. However, other studies45 have reported decreased connectivity within the DMN in children with medically refractory epilepsy. These findings align with this study, which showed region-specific ReHo increases and decreases in DMN. Increased ReHo may reflect pathological inhibition, neuronal loss, or network dysfunction, while decreased ReHo may indicate compensatory synchronization or local functional reorganization. The DMN’s complex structure and function may underlie these heterogeneous alterations. As a widely distributed resting-state network, the DMN comprises three subsystems, including the core hub with the anterior medial prefrontal cortex (amPFC) and posterior cingulate cortex (PCC), the dorsal medial prefrontal cortex (dmPFC) subsystem, and the medial temporal lobe (MTL) subsystem46. Several studies have reported regionally divergent changes within the DMN during disease progression. For example, a long-term follow-up study in temporal lobe epilepsy (TLE) patients revealed abnormal activity in the left mPFC and disrupted connectivity in the left inferior parietal gyrus within the DMN47. Similarly, the CON, also referred to as action-mode network, is anatomically complex but functionally anticorrelated with the DMN48. In the present study, based on cluster size, patients with DRE primarily exhibited decreased mALFF and increased ReHo within the CON. Decreased mALFF in certain CON regions may indicate reduced activation or deactivation, whereas increased ReHo within the CON suggests enhanced neural synchronization. Strongly connected to the ascending reticular activating system, the CON is associated with more frequent consciousness-impairing seizures49. A hypothesis has been proposed to explain interictal neocortical dysfunction in TLE, suggesting that the spread of limbic seizure activity to subcortical activating structures leads to inhibition of the CON49,50. Animal studies confirm that limbic seizure spread to subcortical structures inhibits and deactivates the CON51, which is consistent with the findings of the present study. Epilepsy is a network disorder, where localized abnormal activity disrupts widespread brain networks52, as supported by lesion network mapping research53. In DRE, correlations between mALFF/ReHo and clinical factors reveal the following: mALFF within the FPN is negatively correlated with age, indicating that older patients are more prone to connectivity disruption or activity inhibition in the FPN. Additionally, ReHo in SMN regions correlates with age at epilepsy onset, suggesting that early-onset epilepsy may impair motor network integration, which could reflect compensatory mechanisms or developmental adaptations. Furthermore, ReHo in the VN is negatively correlated with epilepsy duration, implying that the VN is vulnerable to the long-term effects of epilepsy, which may contribute to cognitive and behavioral impairments. These findings highlight the distinct roles of the FPN, SMN, and VN in DRE, with mALFF and ReHo serving as dynamic markers of disease progression and treatment history in this patient population. Epigenetic processes in the brain translate transient cellular signals and changes in neuronal activity into enduring modifications of gene expression. Advancements in gene sequencing technology support the hypothesis that genetics plays a key role in epilepsy54. Various studies indicate that alterations in specific genes and their expressions, along with distinct metabolic dysfunctions, are associated with resistance to anti-seizure medications and the pathophysiology of DRE55. Applying the multivariate PLS method, we identified cortical patterns of weighted gene expression that were significantly co-localized with alterations in mALFF and ReHo. Additionally, we highlighted significantly weighted genes in the first PLS component that could play a role in the pathogenesis of DRE. Transcriptome analysis of altered cortical functional activity in DRE identified 1851 risk genes (PLS1) associated with mALFF changes, and 2723 risk genes (PLS1) linked to ReHo changes. PLS1+ genes, positively correlated with mALFF/ReHo, represent the upregulated genes associated with mALFF/ReHo changes, while PLS1− genes, negatively correlated with mALFF/ReHo, represent the downregulated genes linked to mALFF/ReHo changes34. The prioritized candidate genes overlap with those identified in the recent multi-ancestry genome-wide association study on epilepsy to date11. Analysis of DRE-related genes revealed a significant association between the overlapping genes and alterations in mALFF/ReHo. This study revealed that TMEM74 and KCNN2 are positively correlated with mALFF alteration. Tmem74 has been identified as a novel gene associated with epilepsy11; however, the physiological basis of its role in epilepsy remains unclear. Research indicates that Tmem74 is linked to increased excitability of pyramidal neurons in the prelimbic cortex56 and has been shown to induce autophagy57, but its direct involvement in seizure mechanisms has yet to be confirmed. KCNN2 encodes the small conductance calcium-activated potassium channel 2 (SK2)58. SK ion channels regulate neuronal firing rates by mediating the after-hyperpolarization current, making them critical therapeutic targets for conditions involving dysfunctional neuronal firing rates, such as epilepsy59. KCNN2 and related pathways, which exhibit a dichotomous expression pattern—being upregulated in one mutant line and downregulated in the other—may function as “switches” to modulate the directionality of neural homeostasis, a mechanism that is relevant to pathologies such as stress-related disorders and epilepsy60. Additionally, RBFOX1, RORB, RYR2, BCL11A, and RSPO2 showed negative correlations with mALFF alteration. RORB expression significantly affected axonal morphology, exhibiting a hypomorphic axonal effect, and has been associated with susceptibility to idiopathic generalized epilepsy61. Animal study has shown that RBFOX1 stimulates Vamp1 expression in inhibitory neurons, partially by blocking microRNA-9. Abnormal expression of RBFOX1 leads to reduced inhibitory synaptic transmission and an excitation/inhibition imbalance, which has been linked to epilepsy62. RYR2 encodes the ryanodine receptor 2 (RYR-2) protein, which regulates intracellular calcium concentrations, and strong evidence from the ClinGen Clinical Validity Framework supports an association between RYR2 variants and epilepsy63. Studies have shown that changes in BCL11A expression lead to abnormal neuronal development, which causes abnormal neuronal discharge and thus results in seizures, and the protein encoded by BCL11A interacts with the targets of approved antiepileptic drugs64. As an agonist of the WNT signaling pathway, RSPO2 regulates the activity of the WNT/β-catenin pathway by binding to its receptors LGR4 and LGR5, and emerging evidence suggests a potential link between aberrant WNT signaling and epilepsy pathogenesis65. Furthermore, PTPRK and SCN8A showed significant negative correlations with regional alterations in ReHo. The human receptor-type PTPRK gene, highly expressed in the brain, has been linked to an increased risk of neuropsychiatric disorders and cancer through single nucleotide polymorphisms (SNPs)66, although the exact mechanism of its association with epilepsy remains unidentified. SCN8A is an established epileptic encephalopathy gene that encodes the voltage-gated Na+ channel Nav1.6, which is targeted by commonly used sodium channel-blocking drugs, the most effective anti-seizure medications for individuals with monogenic SCN8A-related epilepsies67. Since this study did not include drug-sensitive patients, whether the aforementioned genes are definitively associated with epilepsy drug resistance requires further investigation. An important direction for future research is to determine to what extent genes influencing mALFF and ReHo in DRE patients are related to DRE drug resistance. Markers of disease progression, such as cortical functional activity changes, may relate to drug responsiveness in epilepsy. Transcriptomics connects brain function to molecular mechanisms, identifying potential DRE biomarkers55. This study showed that genes positively correlated with mALFF are associated with the mitochondrial inner membrane, mitochondrial matrix, and carboxylic acid metabolic processes, whereas genes negatively correlated with mALFF are linked to chromatin remodeling, postsynaptic structures, and chromatin binding. Mitochondria are the primary source of reactive oxygen species (ROS) in cells, and oxidative damage can lead to mitochondrial dysfunction and the transmission of cell death signals, thereby triggering a variety of pathological conditions, including epilepsy68. Loss of the inner mitochondrial membrane protein prohibitin 1 alters the permeability of the inner mitochondrial membrane, which in turn leads to the release of mitochondrial DNA into the cytoplasm and the subsequent activation of the innate immune response, and this mechanism may contribute to the pathogenesis of epilepsy through neuroinflammation or oxidative stress69. Additionally, increased neuron-specific mitochondrial ROS is sufficient to drive complex epileptic phenotypes, and the loss of superoxide dismutase 2 exacerbates epileptic seizures through a cell-nonautonomous mechanism70. Disorders in carboxylic acid metabolism can induce excessive neuronal excitability by compromising energy supply and disrupting ion gradients. Research has demonstrated that rapid neuronal firing during epileptic seizures accelerates glycolysis, leading to the accumulation of carboxylic acid metabolites (e.g., lactic acid), and these metabolic alterations may modulate neuronal excitability via the hydroxycarboxylic acid receptor 1 (HCA1R), thereby exerting an impact on epileptic activity71. Furthermore, abnormalities in critical nodes of citric acid metabolism—specifically, defects in the sodium-dependent citrate transporter—are directly linked to the pathogenesis of epilepsy in pediatric populations72. Additionally, carboxylic acid metabolism can also influence the pathophysiological processes of epilepsy by regulating neurotransmitter balance73, and future studies are needed to further clarify the spatiotemporal dynamics of specific carboxylic acid metabolites (e.g., lactic acid, ketone bodies) as well as their causal relationship with epileptic seizures74,75. Overexpression of genes related to mitochondrial inner membrane, mitochondrial matrix, and carboxylic acid metabolic processes in regions with increased mALFF may play a role in DRE pathology. Chromatin remodeling and chromatin binding play crucial roles in the pathogenesis of epilepsy. Chromatin remodeling alters chromatin structure via ATP-dependent complexes, thereby influencing the regulation of gene transcription. For instance, BORIS recruits the SRCAP complex to replace histone H2A with H2A.Z, which relaxes chromatin around CTCF-binding sites and facilitates the recruitment of transcription factors76,77. In epilepsy, aberrant chromatin remodeling may cause neuronal network dysfunction. For example, SETD2 gene loss increases chromatin occupancy of chromatin-associated proteins (e.g., RNA polymerase II regulatory proteins) while reducing other proteins’ binding, thereby impairing transcriptional regulation78. Molecular abnormalities in postsynaptic structures, cytoskeletal disorders, imbalances in inhibitory/excitatory signaling, and interactions with glial cells constitute key processes in the development and progression of epilepsy79. Seizures can rewire neural circuits80, causing brain injury and localized functional reductions, reflected in decreased mALFF. The abnormal expression of synaptic-related genes in these regions is likely linked to the disruption of neural circuits. This study suggests that upregulated genes linked to mALFF, chromatin remodeling, and binding, postsynaptic structures not only play a key role in epilepsy pathogenesis, but are also associated, to some extent, with treatment response. Further studies incorporating drug-sensitive epilepsy patients are required to definitively establish this relationship. For ReHo mapping, genes positively correlated with ReHo are mainly involved in catalytic activity on nucleic acids, ribonucleoprotein granule, and centrosome, while negatively correlated genes are linked to amyotrophic lateral sclerosis, mitochondrial membrane, and pyrophosphatase activity. Catalytic activities on nucleic acids play a crucial role in the development and progression of epilepsy by regulating processes such as RNA metabolism, protein translation, and the dynamic balance of stress granules. In particular, the phosphorylation modification of RNA-binding proteins provides a novel perspective for understanding the molecular mechanisms of epilepsy81. Additionally, studies have demonstrated that dysfunction of ribonucleoprotein granules may contribute to epileptic pathogenesis through mechanisms including dysregulated mRNA metabolism, aberrant stress responses, and unregulated phase separation82,83, which may explain their positive correlation with ReHo. Centrosomes are crucial for cell division, intracellular trafficking, and neuronal polarity, playing a key role in axonal and dendritic development that affects connectivity and synchronization. Studies have found that centrosomal dysfunction can lead to subcortical heterotopia, which in turn can cause associated epilepsy84. Epilepsy involves multiple epileptogenic processes, including neuroinflammation, neurodegeneration, and abnormal neurogenesis85. This study suggests that low ReHo regions are linked to neurodegenerative pathways, leading to structural and functional impairments and reduced local synchronization. Studies have revealed that genes associated with amyotrophic lateral sclerosis also overlap with those linked to other neurodegenerative diseases. This overlap sheds light on the phenotypic spectrum of neurodegenerative diseases. However, there is currently a lack of direct evidence regarding the relationship between amyotrophic lateral sclerosis and epilepsy. Future studies may focus on whether common pathways, such as neuroinflammation, intersect in these two diseases86. Similar to changes in mALFF, PLS1 + genes associated with ReHo alterations are also enriched in pathways related to the mitochondrial membrane. Pyrophosphatase is a key enzyme involved in DNA/RNA biosynthesis, bone formation, and lipid/carbohydrate metabolism; it can affect metabolic disorders related to neural function87. Pyrophosphatases may be involved in the development and progression of epilepsy, through processes such as regulating neuronal metabolism and energy homeostasis87, influencing pyroptosis-related neuronal damage88, participating in neuroinflammatory processes89, and affecting neuronal excitability via inositol pyrophosphate signaling pathways90, thereby impacting neural activity and ReHo patterns. This may explain the negative correlation between pyrophosphatase activity and ReHo observed in this study. In summary, lower ReHo may indicate neurodegeneration, mitochondrial membrane, and pyrophosphatase activity, contributing to reduced local neural coherence. Further exploration of these pathways is crucial to understanding ReHo alterations and their role in neurological disorders. Using 7T MRI, we identified widespread alterations in mALFF and ReHo in DRE. Patients exhibited a distinct pattern of heightened activity in lower-order unimodal primary regions and reduced activity in higher-order transmodal associative areas. Transcription-neuroimaging analysis revealed associations between these changes and epilepsy-related as well as brain-specific genes. By integrating neuroimaging with molecular markers, our findings provide deeper insights into cortical vulnerability and the underlying mechanisms of drug resistance in epilepsy. Despite these contributions, the study has several limitations. First, although high-resolution 7T rs-fMRI was utilized, the relatively small sample size (105 DRE, 105 HCs) may limit the generalisability of the findings. Incorporating larger, multi-center datasets would enhance external validity, increase statistical power, and improve the robustness of the results. Second, while the study controlled for age, sex, and site effects, it is important to note that the 7T scans were primarily used for pre-surgical assessments of DRE patients, resulting in a limited sample size of well-controlled epilepsy patients. Consequently, we were unable to use well-controlled epilepsy patients as a more appropriate control group. As a result, the potential influence of other confounding factors, such as seizure frequency, medication effects, and comorbidities, was not fully accounted for. While our findings provide some insight into the alterations in brain activity and related gene expression in DRE, they do not conclusively demonstrate that these changes are specifically linked to drug resistance. Future studies should incorporate more detailed clinical stratification to refine the analysis. Third, there was a lack of direct gene expression validation. The spatial transcriptomic analysis relied on data from the AHBA rather than direct patient-specific gene expression profiling. While the AHBA integrates structural, functional, and gene expression data to advance both basic and clinical research on the human brain in health and disease, enabling the mapping of gene expression patterns to neuroimaging changes, validation with individual-level transcriptomic data would enhance the robustness of the findings and improve their biological interpretability. Although the AHBA contains over 20,000 genes, there are still regions with sparse sampling. Despite our use of the “centroids” method to address this issue during data processing, it may still impact the accuracy of the results—and this remains a limitation in transcriptomic analysis. Fourth, it is worth noting that in this study, unlike the HCs, DRE patients were not instructed to fixate on a cross during the rs-fMRI scan, which may have contributed to the observed differences in the VN, despite the inclusion of site effects as covariates in the analysis. This potential confound should be taken into careful consideration when interpreting the results and should ideally be addressed in future studies through the use of standardized visual fixation protocols during data acquisition. Finally, this study focused on overall group comparisons. Because of the small sample size, we did not consider differences between seizure locations in different lobes. Future research with larger, more balanced samples is needed for detailed lobe-specific analysis. Future research addressing these limitations will help refine our understanding of the neurobiological mechanisms underlying DRE and enhance the clinical applicability of advanced neuroimaging techniques. Methods Participants A total of 105 patients with diabetic retinopathy with macular edema (DRE) (50 males, 55 females; mean age, 27.7 ± 2.4 years) and 105 gender- and age-matched HCs (48 males, 57 females; mean age, 30.0 ± 12.9 years) were enrolled. The DRE patients were recruited from a single center, whereas data from the HCs were obtained from the Young Adult Human Connectome Project (HCP). Patients with DRE who underwent comprehensive presurgical evaluation and for whom the decision to proceed with surgery and/or the optimal surgical strategy remained uncertain were referred for 7T MRI. This retrospective study was approved by the Mass General Brigham Institutional Review Board (Protocol Number: 2010P001169), and the requirement for informed consent was waived. All ethical regulations relevant to human research participants were followed. A total of 105 individuals with DRE were recruited between 2020 and 2024. The inclusion criteria were as follows: (1) a clinical diagnosis of epilepsy with ongoing seizures within 12 months prior to MRI acquisition despite adequate trials of at least two appropriate and well-tolerated ASMs7; (2) completion of inpatient continuous video electroencephalography (EEG) monitoring, including both ictal and interictal scalp EEG recordings; (3) additional evaluation with magnetoencephalography (MEG); and (4) absence of detectable lesions on prior 3T MRI. Exclusion criteria included the inability to tolerate 7T MRI due to claustrophobia or other contraindications; a history of neurosurgical intervention or the presence of implanted intracranial stimulation devices (e.g., vagus nerve stimulation or deep brain stimulation); poor image quality; and incomplete clinical or imaging data; and identification of tumors, cortical dysplasia, cavernomas, or other focal lesions on 7T MRI. 7T MRI data from 105 gender- and age-matched HCs were obtained from HCP (Washington U. in Saint Louis, U. of Minnesota), which has ethical approval, and all patients consented for use of clinical data and MRI in research. For DRE patients, clinical characteristics include onset age, duration of epilepsy, total treatment duration, and brain region classification. The data inclusion flowchart is shown in Supplementary Fig. 1. Image acquisition and quality control For DRE patients, MRI scans were obtained using a 7.0-Tesla MR system (Terra, Siemens Medical Solutions, Erlangen, Germany) with a 32-channel head coil. Foam padding was used to minimize head motion. All patients were instructed to relax, remain still, keep their eyes open with natural gaze (without a fixation cross), refrain from thinking, and avoid falling asleep during the scan. Each participant’s 3D-T1WI and rs-fMRI were acquired. 3D T1-weighted structural images (3D-T1WI) were acquired with TR = 2000 ms, TE = 2.49 ms, FOV = 256 × 170 mm, matrix = 342 × 227, isotropic voxel size = 0.75 mm, FA = 9°, iPAT = 2, and 192 sagittal slices. Resting-state fMRI (rs-fMRI) was obtained using a gradient-echo EPI sequence with simultaneous multislice acquisition (Multiband Acceleration Factor = 5), iPAT = 2, TR = 802 ms, TE = 21 ms, FOV = 144 × 144 mm, matrix = 96 × 96, isotropic voxel size = 1.5 mm, FA = 33°, and 80 transverse slices. For HCs, 3D-T1WI data were collected at 3T MRI, and rs-fMRI data were collected at 7T MRI. Available protocols are listed in HCP Protocols (https://www.humanconnectome.org/hcp-protocols). Image preprocessing All images underwent visual inspection to exclude those with visible artifacts, excessive head motion, segmentation inaccuracies, structural-functional registration discrepancies, or incomplete brain coverage. Before performing the preprocessing workflow, the 7T data underwent bias field correction using in-house software, originally developed by Lankinen K et al.91, and implemented in Matlab 2019b (The MathWorks, Inc., Natick, Massachusetts, USA). The fMRI data were preprocessed using fMRIPrep 24.0.1 workflow92, followed by post-processing with XCP-D93,94. A detailed description of the preprocessing and post-processing procedures can be found in the supplementary materials. Cortical MRI measurements and statistical analysis This study used 7T rs-fMRI data to calculate the vertex-wise mean ALFF (mALFF) and ReHo for each participant, followed by statistical comparison analysis. Detailed procedures can be found in the Supplementary Materials. A fixed-effect model was applied within the Spatial Linear Modeling (SLM) framework to analyze group differences in cortical mALFF and ReHo between DRE and HCs. To control false positives, RFT correction95 was applied, with a cluster-forming threshold of 0.001 and requiring a cluster size greater than 100 vertices to detect significant clusters. Given that the 105 patients with DRE and 105 healthy controls were recruited from two different sites, and that two slightly different scanning protocols were used, we included site as a covariate in the SLM to account for potential confounding effects related to site differences. The SLM model was then fitted to the data to identify spatially significant group differences while accounting for spatial dependencies in the data, with site, gender, and age included as covariates. Cortical gene expression data processing Normalized microarray expression data from six donated human brains, encompassing over 20,000 genes and 3702 brain tissue samples, is available through the AHBA (http://human.brain-map.org)27. The preprocessing of the AHBA data was performed using the abagen toolbox96(https://www.github.com/netneurolab/abagen). First, the gene expression and anatomical data were loaded and spatially registered, with the gene expression data mapped onto the cortical surface of the brain using the FreeSurfer-based surface model. Low-quality gene expression data and samples with high variance were removed to ensure reliability. The probes were then mapped to gene annotations using Re-Notator, where probe IDs were updated to the latest gene annotations. A normalization step follows, using a robust sigmoidal normalization method, which scaled the expression data while accounting for outliers and minimizing their impact. Only probes with signal intensities significantly higher than the background noise threshold, detected in more than 50% of the samples, were retained. For quality control, data were visually inspected for consistency, and further filtering steps were be applied to remove any residual noise or inconsistencies. Finally, the data were organized into a gene expression matrix, where each gene’s expression profile was spatially mapped to corresponding brain regions for further analysis. After preprocessing, a total of 15,633 genes were retained, ensuring that the AHBA data were standardized, cleaned, and ready for downstream analyses. To ensure accuracy, we further focused our analyses on the tissue samples within the 333-node Gordon cortical atlas (Gordon 333 Atlas)97, resulting in a 333 × 15,633 gene expression matrix. Due to the AHBA dataset containing two right hemisphere data points exclusively, our analysis was restricted to the left hemisphere34. Therefore, the matrix of transcriptional level values (161 regions × 15,663 gene expression levels) was obtained by calculating the mean expression of all samples within each region. Gene expression–neuroimaging spatial correlation analyses The genes most correlated with mALFF and ReHo changes in DRE were identified by spatially correlating AHBA gene expression data with mALFF and ReHo maps in the left hemisphere. PLS regression98 was used to analyze the relationship between the transcriptional activity of 15,633 genes and regional changes in mALFF and ReHo. In the PLS regression, gene expression data served as predictor variables for regional changes in mALFF and ReHo maps. To prevent overfitting, the variability of PLS was evaluated by assessing the statistical significance of the components through a permutation test with 5000 iterations. To test the null hypothesis that PLS1 did not explain more covariance between the mALFF and ReHo map and whole-genome expression than expected by chance, permutation testing (performed 10,000 times) with spherical rotations of the mALFF and ReHo map was conducted to account for spatial autocorrelation99. And to evaluate the stability of PLS1 for each gene, bootstrap resampling was performed (5000 iterations). The Z scores were then calculated by dividing the weight of each gene by its bootstrap standard error. Finally, genes were ranked according to their normalized weight and Z scores, reflecting their contribution to PLS1100. The gene list associated with regional changes in mALFF and ReHo maps includes genes with Z scores > 3 or <−3, all with FDR-corrected p < 5%, classified as either positive (PLS1+) or negative (PLS1−)34. Genes with Z scores > 3 or <−3 and a FDR-corrected p < 5% were classified as either positive (PLS1+) or negative (PLS1−). Further methodological details, including permutation testing, bootstrap resampling, ranking procedures, and spatial autocorrelation are provided in the Supplementary Materials. Analysis of epilepsy-related genes from GWAS study and PLS ± genes in the present study We selected prior defined epilepsy-related genes from recent GWAS11, including 29,944 people with epilepsy and 52,538 controls, which implicated 29 likely causal genes underlying these 26 loci(see Supplementary Materials). These prioritized candidate genes, which overlapped with monogenic epilepsy genes and targets of current antiseizure medications, were known to be involved in physiological pathways that play a role in epilepsy. To investigate the contribution of epilepsy-related genes in the PLS analysis, we first identified the overlapping genes between the 29 epilepsy-related gene list and the PLS1 background genes (p < 0.05, with FDR correction). Then we estimated the associations between overlapped gene expressions and mALFF and ReHo changes in the left hemisphere of cases and controls. Significance was set at p < 0.05 with FDR correction for multiple comparisons. Enrichment analysis Metascape is a web-based platform (https://metascape.org/gp/index.html#/main/step1) that offers comprehensive gene list annotation and analysis by integrating over 40 knowledgebases, combining functional enrichment, interactome analysis, gene annotation, and membership search101. The PLS1+ and PLS1− genes were uploaded to the Metascape platform, where enriched pathways were identified and evaluated for significance at a 5% FDR-adjusted threshold. The GO enrichment analysis, encompassing biological processes, cellular components, and molecular functions, along with KEGG pathways, was performed for the biological interpretation of DRE. All genes in the genome were used as the background for enrichment analysis. Terms with a p-value < 0.05, a minimum count of three, and an enrichment factor > 1.5 (the ratio of observed counts to expected counts by chance) were selected and clustered based on their membership similarities. The most statistically significant term within a cluster was chosen to represent the cluster. To further capture the relationships between the terms, a subset of enriched terms with a similarity > 0.3 was selected and rendered as a network plot. The relationship between the Allen brain atlas, cortical gene expression, mALFF, and ReHo alteration maps, and epilepsy-related genes is illustrated in a flowchart (Supplementary Fig. 2). Statistics and reproducibility Statistical analyses were conducted using SPSS 26.0. To better illustrate the distribution characteristics of the data, continuous variables were presented as both mean ± standard deviation (SD) and median with interquartile range (IQR). The Chi-squared (χ2) test was used for categorical variables. Independent t-tests were performed for comparisons between groups with normally distributed continuous variables, whereas the Mann–Whitney U test was applied for non-normally distributed data. The correlations between the average mALFF/ReHo values of each individually identified region showing statistical differences and clinical variables, including onset age, duration of epilepsy, and total treatment duration, were analyzed using Spearman correlation. A significance level of p < 0.05 (two-tailed test) was used, with FDR correction for multiple comparisons. Other statistical methods, correction methods, p-value thresholds, and the number of replicates were all detailed in the corresponding methods section. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary materials Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Reporting Summary Supplementary materials Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Reporting Summary