Introduction: Anti-N-methyl-D-aspartate receptor (NMDAR) encephalitis appears an autoimmune disease characterized by neuropsychiatric symptoms, but the molecular mechanisms bridging brain structural alterations and gene expression remain unclear. This study aimed to explore the associations between Morphometric Inverse Divergence (MIND) abnormalities and transcriptional profiles in anti-NMDAR encephalitis, integrating multimodal analyses of neuroimaging, transcriptomics, neurotransmitter systems, and cell-type specificity, with a particular focus on identifying structural and molecular signatures underlying cognitive dysfunction.

Methods: Thirty - seven healthy controls and 37 anti-NMDAR encephalitis patients were enrolled, and analyzed for MIND region values between groups. Using transcriptomic data from the Allen Human Brain Atlas, Partial Least Squares (PLS) regression was applied to analyze spatial correlations between regional MIND changes and gene expression. Functional enrichment, Protein-Protein Interaction (PPI) networks, neurotransmitter receptor/transporter analyses, and cell-type assignment were further performed to explore molecular pathways potentially linked to cognitive impairment.

Results: Significant global and regional MIND abnormalities were presented in anti-NMDAR encephalitis patients, with increased values in regions such as the left paracentral cortex, insula, and cingulate cortex, involving multiple Yeo 7 functional networks (default mode and limbic networks). PLS regression identified genes associated with MIND changes, enriched in metabolic pathways, synaptic processes, and signaling pathways. Neurotransmitter systems (glutamate, gamma-aminobutyric acid) and cell types (astrocytes, excitatory/inhibitory neurons) were implicated in these associations, supporting a model of excitation-inhibition imbalance as a driver of both network disruption and cognitive impairment.

Conclusion: This study reveals multidimensional mechanisms linking macroscale brain structural abnormalities, microscale gene expression, and neurotransmitter dysfunction in anti-NMDAR encephalitis, providing insights for developing imaging-based biomarkers of cognitive decline and molecularly targeted strategies for cognitive restoration in affected patients.

• Anti-NMDAR Encephalitis; Morphometric Inverse Divergence; Transcriptomics; Neuroimaging; Neurotransmitter Systems; Cell-type specificity; Cognitive Impairment

Qian R, Guo R, Cai N, Gao C, Dai W, et al. (2026) Cognitive Impairment Related Neurochemical and Molecular Signatures of Altered Mor phometric Inverse Divergence in Anti-NMDAR Encephalitis. J Mem Disord Rehabil 4(1): 1008.

Anti-N-methyl-D-aspartate encephalitis which occurs receptor as an (NMDAR) autoimmune encephalopathy facilitated by anti-NMDAR antibodies, is recognized by the multisystem symptoms including psychobehavioral abnormalities, seizures, and cognitive impairment. Among these, cognitive impairment — including deficits in memory, attention, executive function, and information processing speed — represents a core and often persistent clinical feature that profoundly affects patients’ quality of life and long-term functional outcomes. Unlike psychotic symptoms or seizures, which may respond favorably to immunotherapy, cognitive deficits frequently persist even after resolution of acute illness. In-depth elucidation of its pathophysiological mechanisms remains a key challenge in current research [1,2]. Previous neuroimaging research has proven that 33%-55% of anti NMDAR encephalitis patients exhibit definite cerebral lesions, such as gray matter volume changes and metabolic abnormalities in the medial temporal lobe, frontal, and parietal cortices [3]. Notably, brain function is not achieved through the independent operation of isolated regions but via dynamic interregional coordination, with structural and functional integration relying on complex network systems. Anti-NMDAR encephalitis involves not only impaired interregional connectivity within individual neural circuits but also affects multiple distributed brain systems, suggesting that the disease is inherently a global network dysfunction disorder [4]. Currently, numerous studies using functional MRI and diffusion tensor imaging have identified variations in the structure and function of various brain districts of patients, intimately tying to cognitive impairment and additional clinical symptoms [5]. To systematically dissect the multidimensional characteristics of such network disruptions, Morphometric Similarity Network (MSN) has shown potential as a method to construct individual connectomes by integrating multimodal structural MRI features. However, Morphometric Inverse Divergence (MIND), an emerging approach that estimates intracortical regional similarity via multivariate distribution divergence, exhibits significant advantages over MSN: studies have demonstrated that MIND networks possess greater stability and stronger associations with gene co expression in cortical regions [6,7]. Currently, research on MIND network alterations in patients with anti-NMDAR encephalitis remains unexplored, which greatly limits a comprehensive understanding of the network-level pathological characteristics of the disease.

Meanwhile, the neurobiological and genetic mechanisms underlying brain functional alteration in anti-NMDAR encephalitis remain to be elucidated. The consolidation of neuroimaging and transcriptomics provides a new method to bridging this gap, by combining brain imaging data with whole-brain gene expression profiles, the molecular underpinnings of functional and structural abnormalities in brain disease can be revealed [8,9]. Numerous studies have clarified the patterns of functional brain and hereditary basis of structure/function alterations in neurological diseases through spatial correlation analyses of neuroimaging measurements and spatial transcriptomic data [10,11]. Whole-brain gene expression profiling acquiring from the Allen Human Brain Atlas (AHBA) database provides a novel avenue to investigate latent linkages between macrostructural abnormalities and specific patterns of transcriptional expression of diverse psychiatric conditions, promising to facilitating our comprehension of the underlying biological mechanisms of variations in neuroimaging among anti NMDAR encephalitis [12]. Neurotransmitter systems play a pivotal part in proper brain function and disease states. Neurotransmitters closely associated with cognitive and psychiatric symptoms (e.g., glutamate, γ-aminobutyric acid, and dopamine) have been demonstrated to play critical roles in the pathogenesis and progression of anti-NMDAR encephalitis [13-15], and the therapeutic efficacy of clinical drugs is also linked to changes in these neurotransmitter pathways [16-18]. Based on neurotransmitter density maps constructed from Positron Emission Tomography (PET) data of more than 1,200 healthy participants [19], we carried out functional enriched analysis with Protein Protein Interaction (PPI) networks constructed to identify hub genes and signaling pathways that play key roles in disease processes [20,21].

The major aim in this study was to investigate the latent relationship among molecular machinery and macroscopic structural variations in anti-NMDAR encephalitis patients by correlating MIND abnormalities associated with the disease with transcriptional profiles. We calculated regional MIND values and compared patients with healthy controls to identify characteristic patterns of MIND abnormalities. Then we correlated these MIND alterations with cognitive performance (MMSE scores) to determine whether specific network disruptions serve as structural substrates of cognitive dysfunction. Subsequently, integrating transcriptomic data deprived from the Allen Human Brain Atlas (AHBA), we employed methods such as Partial Least Squares (PLS) regression to analyze the spatial associations among regional MIND variation and gene expression patterns. We further analyzed functional enrichment and structured PPI networks to elucidate the enrichment characteristics of relevant genes in neurotransmitter signaling pathways (particularly glutamatergic and GABAergic systems) and the role of core hub genes, thereby investigating the coordinated regulatory mechanisms linking gene expression, neurotransmitter system function, and MIND abnormalities. Finally, we profiled particular cell types to estimate their contributions to transcriptomic relationships linked to MIND changes in anti-NMDAR encephalitis patients. Through this comprehensive framework, we aim to unravel the molecular pathological pathological mechanisms of the disease from three dimensions: macroscale brain structural networks, microscale gene expression, and neurotransmitter systems. This will provide a theoretical basis for enhancing our understanding of pathogenic mechanisms of cognitive decline and developing molecularly targeted therapeutic strategies for cognitive restoration.

Participants and clinical assessments

Thirty-seven individuals fulfilled diagnostic guidelines for anti-NMDAR encephalitis, were collected from a tertiary general hospital in Hefei, China, which included characteristic clinical manifestations and a positive cerebrospinal fluid assay for immunoglobulin First-line G anti-NMDAR antibodies, as previously described [1]. immunotherapeutic interventions, including glucocorticoids, intravenous immunoglobulin, plasmapheresis, or combinations were administered to all patients according to standard thereof, clinical guidelines. Concurrently, 37 healthy control subjects without a history of psychiatric or neurological disorders were enrolled, and were carefully matched to the patient cohort for sex, age, and educational level, and all tested negative for anti-NMDAR encephalitis antibodies via validated serological methods. No participant had contraindications for MRI. The study protocol was approved by the Ethics Committee of Anhui Medical University, and written informed consent was obtained from each participant prior to inclusion, in accordance with the Declaration of Helsinki principles. The Hamilton Anxiety Scale (HAMA) and the Hamilton Depression Rating Scale (HAMD) and to assess mental health and emotional status [22-24]. Cognitive function tests were evaluated using the Mini-Mental State Examination (MMSE), a global cognitive screening tool assessing orientation, attention, calculation, registration, recall, and language [25].

Imaging acquisition and preprocessing

A 3.0 T Siemens MRI scanner captured all the images, at the University of Science and Technology of China.Three-dimensional fast low-angle shot sequences yielded T1-weighted (T1w) anatomical pictures with high spatial resolution. Acquisition settings were described below: 188 axial slices, voxel size = 1 × 1 × 1 mm3, flip angle = 12?, field of view = 256 × 256mm2, slice thickness = 1 mm, time echo = 3.18 ms and time repetition = 8,160 ms. Afterwards, the T1w data with FreeSurfer 6.0 (http://surfer.nmr. mgh.harvard.edu/) were pre-processed on surface-based space [26]. Furthermore, Total Intracranial Volume (TIV) was computed for all participants. Cortical preprocessing consisted of tissue segmentation, hemibrain and subcortical structure segmentation, skull stripping, and the creation of cortical surfaces and gray-white interfaces [27].

Construction of MIND

Cortical surface was segmented into 500 spherically contiguous areas via the Schaefer atlas. We adopted a backtracking algorithm in order to minimize changes of plot sizes to ensure that each area is roughly equal in area. The segmentation outcomes of the Schaefer atlas were separately plotted onto the cortical surface of each subject to achieve customized surface parsing. The conversion contributes to five structural traits in each brain district, ranging from cortical thickness, mean curvature, gray matter volume, sulcus depth and surface area [7]. We then standardized the structural properties at distinct scales and structured a 500 × 500 MIND matrix per sample. Figure S1 offers a detailed illustration of the MIND calculation. Regional MIND values were averaged over all connections in 500 cortical regions.

Case-control analysis of regional MIND

We employed general linear modeling (GLM) using age, sex, and TIV as codependent variables for measuring variation of MIND regions in patients with NMDAR-resistant encephalitis and HC, with two-sided t-tests. The Yeo 7 functional network was utilized to ascertain alterations in MIND by exploiting diverse sub-networks between anti-NMDAR encephalitis and HC [28]. Within the GLM framework with age, sex, and TIV as covariates. Correcting for multiple comparisons, a False Discovery Rate (FDR) with a threshold of p < 0.05 was utilized.

Multi-model diagnosis of anti-NMDAR encephalitis patients

Brain region MIND values from the whole brain were used to train and validate machine learning models. To evaluate the identification of MIND values for patients with anti-NMDAR encephalitis, we used two machine learning models, namely, gradient boosting (XGBoost) and support vector machines (SVM). Sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC) were computed the capacity to recognize anti-NMDAR encephalitis. We conducted a comprehensive analysis by segmenting the entire brain into the seven canonical networks as defined by Yeo et al. Subsequently, we employed SVM to evaluate the comparative contributions of every network feature to the predictive model. The discriminative power of each network feature was quantified based on the size of its average weight across all cross-validation partitions. A detailed description of the model parameters and methodology can be found in the Supplementary Material, particularly in the Methods section.

Gene expressional profiles obtainment and preprocessing

Gene expression profiles were extracted from six postmortem brain tissues, comprising 3702 spatially distinct locations, obtained from the AHBA database (http://human.brain-map.org) [29]. Given that only two brains in AHBA datasets provided samples of the right hemisphere, our subsequent analysis focused on the transcriptional profiles of the left hemisphere. A python toolkit named “abagen” (https://github.com/rmarkello/ abagen is available) was used to preprocess the AHBA dataset and map the gene expression data to 250 brain regions on the respective left hemisphere [30]. Collectively, the procedures were concluded as per standard protocols: (i) upgrading probe-to-symbol transformation; (ii) filtering low-intensity prodders; (iii) filtering high homogeneity probes; (iv) management of lost data; (v) allocation of samples to brain zones; (vi) sample and gene normalization, and (vii) recognition of credible genes. Therefore, a matrix of transcript expression was yielded, consisting of two hundred and fifty brain areas and the respective 15,631 genes.

Analysis of correlation among MIND transcripts and regional variations

We applied PLS regression according to the expression levels of all 15,631 genes (independent variable) and case control t-value variances of 250 MIND regions (dependent variable) to model the association between gene expression and regional alteration of the MIND among anti-NMDAR encephalitis patients [31]. PLS composites stand for linear portfolios of gene expression values, and the first or second component (PLS1 or PLS2) usually is ascertained to be the best low-dimensional explanation of the covariance of the high-dimensional data matrices. We estimated whether the explained covariance between MIND t-statistic profiles and transcriptome scores in the PLS component was beyond what would be expected by chance with spatial autocorrelation analyses and 10,000 iterations of the permutation test. In accordance, the variability for each gene in the PLS component was determine using bootstrapping (10,000 bootstrap samples) [32]. For each region weight, the Z-value was computed by dividing the expression of each region weight based on its bootstrap standard error, with ranking all genes on the basis of their weights in the PLS component, subsequently [33]. In the end, with the Spearman methods we analyzed the spatial correlation between the PLS scores and the MIND t-statistic maps.

Functional enrichment of PLS-weighted genes

We picked PLS1 as the optimal component yielding the highest variance explained. The important genes were classified into two vastly divergent groups depending on the PLS1 weights, called PLS1+ genes (Z >3.40, FDR-corrected p < 0.005) and PLS2- genes (Z < − 3.40, FDR-corrected p < 0.005) [33]. The determined genes should have particular representation across the brain if transcriptome neuroimaging correlation analysis is justified. We employed an online tissue-specific expression analysis (TSEA) tool (http://doughertytools.wustl.edu/TSEAtool. html/) for examining this hypothesis to ascertain in which specific tissues MIND-related genes are overrepresented. Overall, the probability of specificity index [probability of specificity index (pSI) = 0.05, 0.01, 0.001, and 0.0001, permutation-rectified] served to determine how likely it is that a gene was exhibited particularly in a certain tissue in relation to all others. We undertook enrichment analysis, involving Gene Ontology (GO) analysis, using Metascape software [34] towards further knowledge of the functional features of these essential PLS +/- genes. In addition, we employed DVAID to recognize Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, which encompassed the majority of recognized metabolic pathways and regulatory pathways with the ability to discern higher level functions of the genome [35]. For gene enrichment analysis significance was calculated via fisher’s exact test and rectified by FDR method, p < 0.05.

Protein–protein interaction analysis

PPI analysis was performed using STRING (version 12.0, http://stringdb.org) to create a PPI network of genes linked to alterations in MIND based on a high-confidence interaction score of 0.7. The level of top 5 hub genes expressed brain-wide was as hub genes. Later, the brain got split into 116 subregions using automated anatomical labeling atlases, and the average gene expression level was calculated for each subregion. The gene expression levels of the top ten ranked subregions are displayed and correlation analysis was used to reveal the association between brain MlND changes and hub gene expression. Using the brain atlas constructed by Schaefer et al., we correlated the whole-brain gene expression profiles of t values with those of overlapping genes. A permutation test with 10,000 iterations was performed, and statistical significance was set at p < 0.05.

Neurotransmitter receptors and transporters analysis

We evaluated the association between NMDA and neurotransmitter system MlND in all brain areas. For that purpose, we have acquired data from PET receptor images of brain volumes of more than 1,200 healthy subjects to create a comprehensive map covering 19 diverse neurotransmitter receptors and transporters in nine distinct neurotransmitter systems (https:github. com/netneurolab/hansen receptors/tree/main/data/ PET_nifti_images). Drawing on prior research [36], this study analyzed three neurotransmitters related to NMDA, consisting of glutamate [metabotropic glutamate receptor 5 (mGluR5), NMDAR] and Gamma-Aminobutyric Acid (GABAa). Distinct neurotransmitter receptors and transporters belonging to the given category are equally processed to yield the respective neurotransmitter system atlas. Each of the three neurotransmitter system atlases was allocated to 500 areas with the mean values of all voxels inside the corresponding region considering as neurotransmitter values. Following Z-standardization of neurotransmitter grams for 500 brain districts, we employed a multiple linear regression model to investigate the effect of PLSl scores and neurotransmitter systems on NMDA in MIND. Explained MlND t-statistic map variance for each independent variable is tested by comparison with 10,000 spatially continuous random models. Materiality indicators could be applied to handle linear regression issues with multicollinearity [37]. The model was described below:

Oυ=β0 + β1 × PLS1 + β2 ×NMDA + β3 ×GABAa + β4 × mGluR5 + ε

Among them, Ov (overlap value) is 500 regions of interest (ROIs) values from the MIND in anti-NMDAR encephalitis. Having identified the percentage of variance explained by these polling covariates, we used bootstrapping to evaluate the comparative contribution to the variance explained by each neurotransmitter system. Multiple comparisons were adjusted using FDR correction, with corrected p-values less than 0.05.

Assigning NMDA-related genes to cell types.

Adhering to the methodology of previous studies [38], we classified cell types into seven classic categories: inhibitory neurons, microglia, astrocytes, oligodendrocyte precursors, endothelial cells, oligodendrocytes and excitatory. We overlaid the gene collections for each cell type with the PLS1-ranked gene list to allocate NMDA related genes gained through PLS analysis to cell types. The p value of the amounts of superimposed genes per cell type was determined by permutation testing [39], and corrected by FDR with p < 0.05. We additionally computed the mean expression levels of gene sets for each cell type contained within the 250 regions of the AHBA partition. We conducted enrichment analysis to discover enriched pathways among genes participating in each cell type. All pathways were screened with a criticality threshold of 5% and corrected using FDR.

Null model

We employed a null model spin testing technology to solve the confounding influence caused by spatial autocorrelation [40]. Randomly rotating spherical projections of spatial maps may yield a group of null spearman correlation coefficients while maintaining spatial connections. The null distribution in this study was modeled by conducting 10,000 spin-exchange tests over cortical areas. Subsequently, the pspin value was counted as the percentage of null correlations over real ones.

Reproducibility verification

Aiming to ascertain the soundness and reliability concerning our results, we have evaluated stability of the case-control t-maps in MIND with the effect of TIV. The data were further validated by comparing the correlation between the two groups of t-values in case-control t-maps in MIND with removal versus retention of TIV.

Demographic characteristics

This study employed a combination of various cortical features and transcriptional data to establish a link between gene expression and altered MIND in anti NMDAR encephalitis patients (Figure 1). The demographic characteristics, including sex, age, education level, and TIV, exhibited no statistically significant differences between anti-NMDAR encephalitis patients and HC (Table 1).

Figure 1: Process flow of this research.
(a) Construction of the MIND network. A 500 × 500 matrix of MIND was created using multiple macrostructural features (cortical thickness, average curvature, surface area, gray matter volume and gyral depth). Subsequently, the regional MIND values were constructed through mean computation over 500 cortical regions, without any thresholding. (b) (b) Transcriptome analysis. Then, PLS regression was employed to recognize the link between imaging and transcriptome. The correlation between MIND alterations and whole-brain gene expression among patients with anti-NMDAR encephalitis was assessed via PLS regression, neurotransmitter, and PLS2 weighted gene functional enrichment analysis. Abbreviation: NMDAR: N-methyl-D-aspartate receptor, HC: health control, GM: gray matter volume, SA: surface area, SC: sulcal depth, MC: mean curvature, CT: cortical thickness, PLS1: first partial least squares, ROIs: regions of interest, GO: Gene Ontology, KEGG: Kyoto Encyclopedia of Genes and Genomes

Table 1: Demographic characteristics and clinical features of the healthy control and anti-NMDAR encephalitis patients groups.

Abbreviations: anti-NMDAR: anti-N-methyl-D-aspartate receptor, TIV: total intracranial volume, MMSE: Mini-Mental State Examination, HAMA: Hamilton Anxiety Scale, HAMD: Hamilton Depression Scale

MIND-related changes in anti-NMDAR encephalitis

The global MIND value with computing mean of all areas revealed a remarkable discrepancies (t = 2.652, p = 0.010) between anti-NMDAR encephalitis patients and HC (Figure 2a). Patients with anti-NMDAR encephalitis presented MIND similar to that of healthy controls (HC), yet marked extremes were identified in the temporal-parietal gyrus and sensorimotor network (Figure 2b). Furthermore, anti-NMDAR encephalitis patients showed a notable growth regional MIND distribution in the left paracentral cortex, left insula, left cingulate cortex, right parahippocampal cortex and middle frontal cortex (Figure 2c; Supplementary Table 1). The t-values for anti-NMDAR encephalitis patients and case-control subjects demonstrated a positive spatial correlation (r(500) = 0.346, pspin = 0.004) (Figure 2d), suggesting that areas with a larger number of connections tend to show more pathological variations. Furthermore, we contrasted the unusual MIND patterns among patients with anti NMDAR encephalitis and healthy controls (HC) within the functional Yeo 7 network (Figure 2e). Across the Yeo 7 network, individuals with anti-NMDAR encephalitis showed dramatically higher MIND values in limbic (t = -2.95, pFDR = 0.025), somatomotor (t = -2.45, pFDR = 0.025),dorsal attention (t = -2.19, pFDR = 0.032), default mode networks (t = -2.43, pFDR = 0.025), visual (t = -2.43, pFDR = 0.025), ventral attention (t = -2.71, pFDR = 0.025) and frontaloparietal control network (t = -2.21, pFDR = 0.032).

Figure 2: Regional changes of anti-NMDAR encephalitis in MIND.
(a) Frequency distributions of regional MIND between anti-NMDAR encephalitis patients and HC, with age, sex, and TIV as covariates. (b) Mean MIND distributions of 500 brain regions in anti-NMDAR encephalitis patients and HC. The regional distributions of MIND in anti-NMDAR encephalitis patients showed patterns analogous to those in HC, with pronounced extremes in the temporoparietal gyrus and insula. (c) Case control comparison of regional MIND between anti-NMDAR encephalitis patients and HC. (d) Scatter plot of mean control MIND values and case-control t-values. positive spatial correlation existed between case-control t-values and mean control MIND values. (e) MIND discrepancy between patients with anti-NMDAR encephalitis and HC within the functional Yeo 7 network. The MIND has grown considerably among anti NMDAR encephalitis patients in the limbic, somatomotor, visual, dorsal attention, frontoparietal, default mode networks and ventral attention. Abbreviations: MIND: morphometric inverse divergence, NMDA: N-methyl-D-aspartate

Diagnosis and symptom prediction

Four metrics were utilized for performance evaluation: AUC, specificity, sensitivity and accuracy (ACC). SVM (AUC = 0.786, ACC = 0.786, sensitivity = 0.714 specificity = 0.854) and XG Boost (AUC = 0.797, ACC = 0.812, sensitivity = 0.838, specificity = 0.757) exhibited good classification performance (Figure 3a and Supplementary Table 2). Based on the SVM model weights, we assessed the contribution of features from the seven canonical brain networks (Figure 3b). Among them, the default mode network (DMN) exhibited the highest predictive contribution for distinguishing anti-NMDAR encephalitis (Figure 3c).

Figure 3: Classification performance and network contribution analysis based on MIND features.
(a) ROC curves of the SVM and XGBoost models, illustrating their performance in distinguishing patients with anti-NMDAR encephalitis from healthy controls. The AUC values and their 95% confidence intervals are marked in the figure. (b) Brain surface maps showing the regional contributions to classification based on SVM model weights. (c) Radar plot displaying feature weight scores and the contribution distribution across Yeo’s seven functional networks, with DMN showing the highest contribution. Abbreviation: SVM: support vector machines, AUC: Area Under the Curve, XG Boost: extreme gradient boosting, SVR: support vector regression, SMN: sensorimotor network, ROC: receiver operating characteristic curve, DAN: dorsal attention network, DMN: default mode network, VAN: ventral attention network, LN: limbic network, VIS: visual cortex, FPN: frontoparietal network

Patterns of transcription relevant to brain region variations

PLS1 and PLS2 validly interpreted 12% and 10% of the macrostructural differences in patients with anti NMDAR encephalitis, respectively, markedly surpassing the random expectation value (pspin < 0.001). Weighted gene expression profiles showed genes in the cingulate gyrus and temporal cortex significantly (Figure 4a). In addition, PLS1 scores and case-control t-value plots in MIND exhibited remarkable positive spatial correlation, demonstrating that PLS1-weighted genes were co-overexpressed in areas with reduced MIND in patients with anti-NMDAR encephalitis (Pearson’s r = 0.378, pspin =0.007) (Figure 4b). We successfully recognized PLS1 weighted genes by univariate single-sample Z-test and further grouped them into two gene lists, incorporating 846 PLS1+ genes (Z > 3.40, pFDR < 0.005) and 1229 PLS- genes (Z < -3.40, pFDR < 0.005) (Supplementary Table 2). Based on the ranking of gene weights, we selected the top 10 genes that were significantly associated with the regional changes in MIND (p < 0.05 after FDR correction), including 5 genes with negative correlations (i.e., JADE1, KLF9, MAP3K13, FES, NAGPA) and 5 genes with positive correlations (i.e. ANP32E, PPILA, GPD2, NTSR2, CEP70) (Figure 4c).

Figure 4: Transcriptional signatures associated with case-control regional changes in MIND.
(a) Case - control t - map of the regional MIND gradient scores in the left hemisphere. Left: The coincident distribution between case control t-map of regional changes in MIND and weighted gene expression-map of PLS1 scores in left hemisphere. Right: The distribution of the weighted gene expression map showed a pattern of gradual increase around the cingulate gyrus and temporal cortex. (b) The scatter plot displayed a striking positive spatial correlation between PLS1 scores in MIND and case-control t-values in patients with anti-NMDAR encephalitis (Spearman’s r = 0.378, pspin =0.007). (c) A total of 846 PLS1+ genes and 1229 PLS1- genes were identified by ranked Z scores, and only genes with top5 expression were listed. Abbreviation: MIND: morphometric inverse divergence, PLS: partial least squares

Genes related to MIND concrete expression in brain

The pSI values (0.05, 0.01, 0.001, and 0.0001) were used to identify the likelihood that the specific expression of a gene in a particular tissue versus all other tissues involved in the analysis. Genes involved in MIND showed marked enrichment in brain tissue, even under the strictest conditions (pSI = 0.0001) (Figure 5).

Figure 5: Analysis of the specific expression of MIND-related genes in different tissues.
Integrated tissue-specific expression analysis showed that genes related to MIND were specifically expressed in the brain rather than in other tissues (pSI = 0.05, 0.01, 0.001, and 0.0001). The dotted circles indicate the −log10 (p) values, which reflect the statistical importance of specific gene expression in various organizations. Abbreviation: MIND: morphometric inverse divergence

PLS weighted gene functional annotation associated with brain alterations

To explore the molecular mechanisms underlying the changes of MIND in anti-NMDAR encephalitis patients compared with healthy controls, we conducted a spatial Pearson correlation analysis on the statistical differences between MIND and the AHBA profiles to identify relevant genes. Enrichment analysis of the PLS1+ genes revealed that KEGG pathways were engaged in metabolism (fatty acid degradation, histidine metabolism) and synapse-related biological processes (GABAergic synapses, axon guidance, tight junctions) (Figure 6a,b). Enrichment analysis of the PLS1- genes demonstrated that KEGG pathways were implicated in signaling pathway (Adrenergic hormone signaling pathway, Calcium signaling pathway, MAPK signaling pathway) and synapse-related biological processes (dopaminergic synapses) (Figure 6c,d). PPI network analysis recognized hub genes that might perform a crucial role in regulating anti-NMDAR encephalitis (Figure 6e).

Figure 6: Functional enrichment and PPI analysis of PLS1 weighted genes related to MIND alterations among anti-NMDAR encephalitis patients. (a) Results of KEGG pathway enrichment analysis with MIND and PLS1+ related genes. (b) Metascape enrichment network representing GO enrichment term network colored by cluster identity, demonstrating the intra - and inter - cluster resemblance of enriched terminations weighted with MIND and PLS1+. Nodes possessing the same cluster identification are typically adjacent to each other. Every term consists of a circular node, whose size is directly proportionate to the amount of input genes contained in that term, while its color denotes its clustering identity. (c) Results of KEGG pathway enrichment analysis with MIND and PLS1- related genes. (d) Metascape enrichment network representing GO enrichment term weighted with MIND and PLS1-. (e) PPI network. Left: constructed for MIND and PLS1+ weighted genes using the STRING database. Right: constructed for MIND and PLS1- weighted genes. Top 10 most highly expressed genes degree values calculated by the maximum clique centrality algorithm in the network were identified as hub genes and visualized using Cytoscape software. Abbreviation: KEGG: Kyoto encyclopedia of genomes

Whole-brain expression levels of the Hub genes

The whole-brain expression profiles of the top 5 hub genes identified as related to MIND with PLS1+ and PLS1-, with high expression levels in frontal, occipital, insula, inferior temporal gyrus supplementary motor area (Figure 7a). Among all the hub genes, HSP90AA1 and HISR1H4A identified by MIND were the most highly expressed, and we further analyzed their distribution in the top ten expressed brain regions (Figure 7b). The HSP90AA1 gene was highly expressed in the insula and precentral gyrus,and its expression level was positively correlated with MIND changes (r = 0.170, pperm < 0.0001). The HISR1H4A gene was highly expressed in the inferior temporal gyrus, middle occipital gyrus, middle frontal gyrus, and its expression level was negatively correlated with MIND changes (r = -0.173, pperm < 0.0001) (Figure 7c).

Figure 7: Expression levels of whole-brain overlapping hub genes.
(a) Top 5 hub genes in whole-brain expression levels as related to MIND with PLS1+ and PLS1-. (b) Displays the whole-brain gene expression of the overlapping genes HSP90AA1 and HISR1H4A and the top 10 brain areas with the highest expression levels. (c) A remarkable spatial association was found between the gene expression of HSP90AA1 and HISR1H4A and the changes in MIND (pperm < 0.0001).

Combining genetic and neurotransmitter contributions in a multivariate model

To investigate the mechanisms underlying the observed psychological variations, we built a multiple linear regression model with the following predictor variables: PLS1, NMDA, GABA, and mGluR5 receptor maps. The model accounted for 11.2% of the variation in MlND t-statistic maps (R2 = 0.112, F = 16.6, p < 0.001) (Figure 8a).All individual elements All individual factors remarkablely predicted the variance in MlND shifts, with PLS scores comprising the largest proportion of the variance (PLS weight = 0.948, p < 0.001 corrected by FDR, Figure 8c).

Figure 8: Relationship between different neurotransmitter systems and MIND in patients with anti-NMDAR encephalitis.
(a) Multivariable linear regression model was applied to explore the relationship between various neurotransmitter systems and MIND among anti-NMDAR encephalitis patients. (b) Fitting plot of MIND variations and scatter graph of predicted versus observed values. (c) The relative contribution of each individual predictor variable to the explained variance was computed using the partial correlation coefficient with respect to the alteration in MIND. Asterisks indicate that the p values remained significant after FDR correction. Abbreviation: PLS: partial least squares, MIND: morphometric inverse divergence, NMDA: N-methyl-D-aspartate, GABA: glutamate γ-aminobutyric acid, GluR: glutamate receptor 

Transcriptional characteristics of classic cell types

Considering the cellular diversity in the brain, we analytically recognized particular cell types abundant in MIND mutations and visualized the spread in gene expression among various cell types (Figure 9a). Many of the genes enriched for PLS were significantly involved in astrocytes (n = 154, pFDR < 0.001), excitatory neurons (n = 237, pFDR < 0.001), and inhibitory neurons (n = 158, pFDR < 0.001) (Figure 9b). Cell type-specific gene enrichment analysis revealed that MIND alterations in patients with anti-NMDAR encephalitis were prominently enriched in inflammatory biological processes associated with excitatory neurons and neuronal cells. MIND alterations characterized in neuronal cells became enriched in GO terms, comprising “synaptic signaling”, “inflammatory response” and “negative regulation of immune system process” (Figure 9c).

Figure 9: Specific expression of MIND-related gene alterations in cell types.
a Gene expression profiles of overlapping genes between the PLS1 gene list and cell type-specific genes in each cell type region. b The amounts of overlapping genes in each cell type (excitatory neurons: number = 237, adjusted pperm < 0.001; inhibitory neurons number = 158, adjusted pperm < 0.001; astrocytes: number = 154, adjusted pperm < 0.001; oligodendrocytes: number = 61, adjusted pperm  = 1.000; microglia: number = 77, adjusted pperm  = 1.000; OPCs: number = 22, adjusted pperm = 0.405; endothelial: number = 84, adjusted pperm  = 1.000). All permutated pperm values were corrected using FDR and identified based upon one-sided tests. An asterisk indicates a p-value that remains less than 0.05 after FDR correction. c Gene ontology term enrichment analysis of genes related to MIND undergoing variations in diverse cell types. d Chord diagram showing the enrichment and overlap of NMDA-related genes among seven canonical brain cell types: microglia, endothelial cells, OPCs, oligodendrocytes, astrocytes, excitatory neurons, and inhibitory neurons. Purple ribbons represent the degree of gene set overlap between cell types, with thicker connections indicating higher overlap. The analysis was based on aggregated single-cell transcriptomic data from five postmortem human cortical studies. Gene sets were derived via PLS-based ranking, and significance was determined using permutation testing and FDR correction (p < 0.05). Abbreviation: OPCs: oligodendrocyte precursor cells

Following the assignment of PLS derived NMDA-related genes to the seven canonical brain cell types, we observed significant enrichment overlap among several cell types. Notably, excitatory and inhibitory neurons exhibited substantial gene set overlap, suggesting potential cooperative roles in the pathophysiology of anti-NMDAR encephalitis. Additional overlaps were observed between oligodendrocytes and their precursors.Enrichment analysis revealed cell-type-specific pathways that may underlie NMDA receptor-mediated neuroimmune responses.

Reproducible analysis of brain function and transcriptional pattern changes associated with anti NMDAR encephalitis

To validate the robustness of our results, we launched an investigation to determine whether case-control differences in major MIND were affected by TIV. Notably, the differences between the case and control groups in MIND (with TIV as a covariate) showed a strong association with the ones unadjusted for TIV (r = 0.955, pspin <0.001) (Figure 10).

Figure 10: Validation of TIV effect on case-control MIND discrepancies.
(a) Differences in case-control studies of regional MIND following controlled TIV. (b) Regional case-control MIND variance without treatment for TIV. (c) Spearman’s correlation for 500 t-statistic regional values for case-control differences with and without controlling for TIV (r = 0.955, pspin <0.001). Abbreviation: TIV: total intracranial volume

This study aims to comprehensively investigate abnormal MIND alterations in patients with anti-NMDAR encephalitis and systematically integrate transcriptomics,neurotransmitter system, and cell type-specific analyses to unravel the underlying molecular mechanisms and pathophysiological associations of the disease. Thirty seven patients with anti-NMDAR encephalitis and 30 healthy controls matched for age, sex, and educational level were enrolled. Multidimensional analyses revealed that patients exhibited significant global and regional MIND abnormalities. Using PLS regression analysis, we identified genes significantly associated with MIND changes, which are involved in multiple key biological processes and signaling pathways. Additionally, we found that neurotransmitter systems play a crucial role in brain functional abnormalities in patients with anti-NMDAR encephalitis: the spatial distribution of the norepinephrine and GABAergic systems recapitulates NMDAR-related MIND patterns, providing novel insights for unifying heterogeneous neuroimaging studies. PPI network analysis recognized hub genes that might perform a central role in disease mechanisms. These findings offer new perspectives on the complex relationships between macroscale morphometric structural abnormalities and microscale transcriptional patterns during the onset and progression of anti-NMDAR encephalitis.

MIND Abnormalities and Brain Network Dysfunction

MIND is a metric for quantifying similarity via divergences between multivariate distributions with multiple degrees of freedom, and has demonstrated efficacy in detecting individual variations in the human connectome attributed to developmental changes and genetic diversity [7]. In our study, anti-NMDAR encephalitis patients exhibited increased global morphometric similarity, with region-specific significant elevations in MIND particularly in the left paracentral cortex (involved in somatosensory and motor regulation), left cingulate cortex (responsible for cognitive and emotional modulation) [41], right parahippocampal cortex (engaged in memory processing), and middle frontal gyrus (associated with higher cognitive functions) [42]. These regional abnormalities are highly consistent with the clinical manifestations commonly observed in anti-NMDAR encephalitis, such as cognitive impairment, emotional disturbances, and memory deficits. Critically, MIND alterations were significantly correlated with lower MMSE scores, directly linking these structural abnormalities to cognitive impairment. The cingulate cortex, particularly its anterior subdivision, serves as a hub for attention allocation, conflict monitoring, and cognitive control. Disruption of its structural integrity, as reflected by elevated MIND values, may impair the ability to sustain attention and suppress task-irrelevant information, contributing to deficits in executive function commonly reported in patients. Similarly, the parahippocampal cortex is critically involved in memory encoding and retrieval, and its structural abnormality provides a plausible neural basis for the episodic memory deficits frequently observed in anti-NMDAR encephalitis. The middle frontal gyrus, a key node of the dorsolateral prefrontal cortex, supports working memory and cognitive flexibility; its involvement further underscores the multidimensional nature of cognitive dysfunction in this disease, encompassing not only memory but also executive and attentional domains. Further analyses revealed that the abnormal MIND patterns in patients broadly involve the Yeo 7 functional networks (including visual, somatomotor, and default mode networks, among others) [28], with significantly elevated MIND values across all networks. This finding suggests that anti-NMDAR encephalitis is not confined to damage within a single functional system but involves coordinated dysfunction across multiple brain networks. Abnormalities in DMN MIND may be associated with impairments in self-cognition and memory function in patients [43], whereas alterations in the somatomotor network may be linked to motor symptoms. Such widespread network abnormalities may arise from compensatory remodeling of whole-brain networks triggered by impaired synaptic transmission following antibody-mediated blockade of NMDARs. In the present study, the diagnostic value of MIND features was validated using support vector machine (SVM) and XGBoost models, which exhibited favorable classification performance. Further analyses indicated that the DMN made the greatest predictive contribution to the models, consistent with its central role in higher cognitive functions, suggesting that it may serve as the most sensitive neuroimaging biomarker for cognitive impairment in anti-NMDAR encephalitis. Additionally, our reproducibility analyses demonstrated that case-control difference t-values were unaffected by total intracranial volume (TIV), thereby confirming the robustness of our results.

Transcriptomic Associations and Molecular Mechanisms

Through partial least squares (PLS) regression analysis, this study identified gene expression patterns significantly associated with regional MIND changes. Among these, PLS1 accounted for 12% of MIND variance, with its weighted genes significantly enriched in brain regions exhibiting MIND abnormalities, such as the cingulate gyrus and temporal cortex. PLS1+ genes (Z > 3.40) are primarily involved in metabolic pathways including fatty acid degradation and histidine metabolism, learning and memory-related processes, as well as synapse-related processes such as GABAergic synapses and axon guidance. In contrast, PLS1- genes (Z < -3.40) are enriched in signaling pathways such as the MAPK signaling pathway and calcium signaling pathway, along with dopaminergic synaptic functions. These findings reveal that the molecular mechanisms of anti-NMDAR encephalitis involve not only immune-mediated inflammatory responses but also the synergistic effects of metabolic disorders and synaptic dysfunction. Notably, PLS1 weighted genes exhibit highly specific expression in brain tissues—they remain significantly enriched even under the strict threshold of pSI = 0.0001—confirming their close association with brain function. PPI network analysis recognized core hub genes HSP90AA1 and HIST1H4A, which are highly expressed in key brain regions such as the frontal lobe, occipital lobe, and insula. Specifically, the hub gene HSP90AA1 is highly expressed in the insula and precentral gyrus, and positively correlated with MIND changes. As a heat shock protein, it may participate in neuroinflammatory responses by regulating protein folding [44], potentially perturbing somatosensory and motor-related circuits to exacerbate sensory abnormalities and motor dysfunction. HIST1H4A is highly expressed in the middle frontal gyrus and middle occipital gyrus, and negatively correlated with MIND changes; it may regulate neuronal function by influencing chromatin structure, and its downregulation may exacerbate central inflammation as well as cognitive and visual impairments through epigenetic imbalance [45]. The identification of these hub genes provides critical clues for understanding the molecular targets of the disease.

Neurotransmitter Imbalance as a Driver of Cognitive Impairment

As a key receptor in the glutamatergic system, NMDAR is closely associated with neural plasticity and plays a central role in synaptogenesis, synaptic maturation, long term plasticity, neuronal network activity, and cognitive functions, serving as a critical molecular basis for mediating learning and memory [46]. Autoantibodies in patients can directly inhibit NMDAR function; such inhibition reduces Ca²? influx, impairing synaptic plasticity as well as learning and memory [47], This NMDAR hypofunction is hypothesized to be a primary driver of cognitive impairment in anti-NMDAR encephalitis. Abnormal interactions between mGluR5 and NMDAR may disrupt the excitation-inhibition balance, triggering epilepsy and neurodegenerative changes. Dysregulation of the norepinephrinergic system (NAT) exacerbates cognitive and emotional impairments [48,49]. As the primary inhibitory neurotransmitter in the central nervous system, abnormal changes in GABAergic inhibitory function disrupt the balance of neuronal excitability, potentially inducing excessive neuronal excitation, which in turn leads to symptoms such as seizures, attentional and executive dysfunction and psychobehavioral abnormalities [50]. A multiple regression model validated the associations of PLS1 pathways, as well as NMDA, GABA, and mGluR5 receptors, with MIND changes, suggesting that neurotransmitter network imbalance is a core link, and that the PLS1 pathway may dominate their coordinated regulation.

Cell Type Specificity and Pathological Basis

The type and distribution of cellular abnormalities are critical in the pathological progression of autoimmune encephalitis. Dysfunction of specific cell types may represent the core biological mechanism linking abnormal gene expression to changes in brain structure and function. In this study, PLS-associated genes were found to be significantly enriched in astrocytes, excitatory neurons, and inhibitory neurons, suggesting that these cell types may be key effector cells in the pathological changes of anti NMDAR encephalitis. Cell type-specific genes are enriched in processes such as “synaptic signaling” and “inflammatory response” [51], indicating that MIND alterations involve both abnormal neural conduction and cell-autonomous inflammation. Astrocyte-mediated neuroinflammation can impair synaptic function and contribute to cognitive deficits through release of pro-inflammatory cytokines that negatively impact neuronal plasticity. Moreover, the imbalance between excitatory and inhibitory neurons may directly affect synaptic transmission efficiency, leading to both cognitive dysfunction and abnormal structural similarity of MIND in brain regions [7]. Such functional disturbances at the cellular level may constitute the microstructural basis for macroscale MIND changes and their association with cognitive impairment. Notably, the overlap of NMDA-related genes between excitatory and inhibitory neurons suggests cooperative dysregulation of these cell types in driving cognitive deficits, possibly through disrupted excitation-inhibition balance within local circuits. Taken together, this study reveals multidimensional associations between MIND abnormalities and gene regulation, neurotransmitter imbalance, as well as cell type dysfunction in anti-NMDAR encephalitis, providing a theoretical basis for therapeutic strategies targeting metabolism, synaptic function, and inflammation.

A few limitations of the present study should be acknowledged. First, although the novel MIND method was used to explore the potential association between brain structural changes and the molecular mechanisms of anti-NMDAR encephalitis, the patient sample size was still relatively small. Consequently, the features of macro structural MIND alterations and specific transcription patterns in anti-NMDAR encephalitis patients necessitate repeated estimation and validation through independent consortium studies. Secondly, while the MIND metric captures structural similarity, it is insufficient to directly delineate dynamic cellular-level changes, thus warranting further interrogation via techniques such as single-cell sequencing. Finally, transcriptomic data were derived from 6 post-mortem brain specimens in the AHBA database, which are restricted to the left hemisphere. This may introduce inter-individual variability and hemispheric bias, precluding a comprehensive reflection of gene expression profiles in in vivo brain tissues of patients.

In summary, through multidimensional integrative analyses, the present study demonstrates that patients with cognitive impairment of anti-NMDAR encephalitis exhibit significant MIND abnormalities, and these changes are closely associated with specific gene expression, neurotransmitter system dysfunction, and cell-type functional dysregulation. At the molecular level, PLS regression identified gene expression patterns that link MIND alterations to synaptic plasticity, learning/memory pathways, and excitation-inhibition balance. Hub genes HSP90AA1 and HIST1H4A, along with neurotransmitter systems (glutamate, GABA) and specific cell types (astrocytes, excitatory/inhibitory neurons), were implicated in the structural abnormalities underlying cognitive deficits. These findings support a mechanistic model in which NMDAR hypofunction and excitation inhibition imbalance drive both network disruption and cognitive impairment. The results not only offer novel inspirations into the pathological mechanisms of the disease but also lay a foundation for the development of imaging- and molecular-based diagnostic biomarkers as well as targeted therapeutic strategies.

We sincerely appreciate the cooperation of all participants in our study, the University of Science and Technology of China and the Anhui Provincial Mental Health Center for providing support.

Funding

This study was financially grant from the National Natural Science Foundation of China (No. U24A20702), the Anhui Provincial Institute of Translational Medicine Research Foundation (No. 2021zhyx-B10) and the Anhui Province Clinical Medical Research Transformation Special Project (Grant Nos.202204295107020028 and 202204295107020006).

Data and Code Availability

The Allen Brain Atlas (human.brain-map.org/static /download) contains human gene expression data in favor of the results in this study. The brain maps were displayed via SurfIce (v1.0.20190902, https://www.nitrc. org/projects/surfice/). Computational code for the spatial alignment test could be obtainable at the github: https:// github.com/frantisekvasa/rotate_parcellation.The code for PLS profiling is readily obtainable at the github: https:// github.com/SarahMorgan/Morphometric_Similarity_SZ. Other data will be made available on request.

Consent to Participate declaration

Each human participant consent to participate in this study.

Approval Committee or the Internal Review Board in the Ethics Approval declaration

Ethics Committee of Anhui Medical University

Ethics Number

PJ2023-11-25

1. Dalmau J, Gleichman AJ, Hughes EG, Rossi JE, Peng X, Lai M, et al. Anti-NMDA-receptor encephalitis: case series and analysis of the effects of antibodies. Lancet Neurol. 2008; 7: 1091-1098.
2. Titulaer MJ, McCracken L, Gabilondo I, Armangué T, Glaser C, Iizuka T, et al. Treatment and prognostic factors for long-term outcome in patients with anti-NMDA receptor encephalitis: an observational cohort study. Lancet Neurol. 2013; 12: 157-165.
3. Long Q, Lv Z, Zhao J, Shi K, Li C, Fan B, et al. Cerebral gray matter volume changes in patients with anti-N-methyl-D-aspartate receptor encephalitis: A voxel-based morphometry study. Front Neurol. 2022; 13: 892242.
4. Peer M, Prüss H, Ben-Dayan I, Paul F, Arzy S, Finke C. Functional connectivity of large-scale brain networks in patients with anti-NMDA receptor encephalitis: an observational study. Lancet Psychiatry. 2017; 4: 768-774.
5. Cai L, Liang Y, Huang H, Zhou X, Zheng J. Cerebral functional activity and connectivity changes in anti-N-methyl-D-aspartate receptor encephalitis: A resting-state fMRI study. Neuroimage Clin. 2020; 25: 102189.
6. Bazinet V, Hansen JY, Misic B. Towards a biologically annotated brain connectome. Nat Rev Neurosci. 2023; 24: 747-760.
7. Sebenius I, Seidlitz J, Warrier V, Bethlehem RAI, Alexander-Bloch A, Mallard TT, et al. Robust estimation of cortical similarity networks from brain MRI. Nat Neurosci. 2023; 26: 1461-1471.
8. Wang GZ, Belgard TG, Mao D, Chen L, Berto S, Preuss TM, Lu H, Geschwind DH, Konopka G. Correspondence between Resting-State Activity and Brain Gene Expression. Neuron. 2015; 88: 659-666.
9. Zhu D, Yuan T, Gao J, Xu Q, Xue K, Zhu W, et al. Correlation between cortical gene expression and resting-state functional network centrality in healthy young adults. Hum Brain Mapp. 2021; 42: 2236-2249.
10. Anderson KM, Collins MA, Kong R, Fang K, Li J, He T, Chekroud AM, et al. Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder. Proc Natl Acad Sci U S A. 2020; 117: 25138-25149.
11. Yang J, Chen K, Zhang J, Ma Y, Chen M, Shao H, et al. Molecular mechanisms underlying human spatial cognitive ability revealed with neurotransmitter and transcriptomic mapping. Cereb Cortex. 2023; 33: 11320-11328.
12. Yao G, Luo J, Zou T, Li J, Hu S, Yang L, et al. Transcriptional patterns of the cortical Morphometric Inverse Divergence in first-episode, treatment-naïve early-onset schizophrenia. Neuroimage. 2024; 285:120493.
13. Dale RC, Church AJ, Surtees RA, Lees AJ, Adcock JE, Harding B, et al. Encephalitis lethargica syndrome: 20 new cases and evidence of basal ganglia autoimmunity. Brain. 2004; 127: 21-33.
14. Dalmau J, Lancaster E, Martinez-Hernandez E, Rosenfeld MR, Balice-Gordon R. Clinical experience and laboratory investigations in patients with anti-NMDAR encephalitis. Lancet Neurol. 2011; 10: 63-74.
15. McCracken L, Zhang J, Greene M, Crivaro A, Gonzalez J, Kamoun M, et al. Improving the antibody-based evaluation of autoimmune encephalitis. Neurol Neuroimmunol Neuroinflamm. 2017; 4: e404.
16. Endres D, Perlov E, Stich O, Rauer S, Maier S, Waldkircher Z, et al. Hypoglutamatergic state is associated with reduced cerebral glucose metabolism in anti-NMDA receptor encephalitis: a case report. BMC Psychiatry. 2015; 15: 186.
17. Kulkarni J, Thomas N, Hudaib AR, Gavrilidis E, Grigg J, Tan R, et al. Effect of the Glutamate NMDA Receptor Antagonist Memantine as Adjunctive Treatment in Borderline Personality Disorder: An Exploratory, Randomised, Double-Blind, Placebo-Controlled Trial. CNS Drugs. 2018; 32: 179-187.
18. Perry A, Hughes LE, Adams N, Naessens M, Murley AG, Rouse MA, et al. The neurophysiological effect of NMDA-R antagonism of frontotemporal lobar degeneration is conditional on individual GABA concentration. Transl Psychiatry. 2022; 12: 348.
19. Hansen JY, Shafiei G, Markello RD, Smart K, Cox SML, Nørgaard M, et al. Mapping neurotransmitter systems to the structural and functional organization of the human neocortex. Nat Neurosci. 2022; 25: 1569-1581.
20. Jeong H, Mason SP, Barabási AL, Oltvai ZN. Lethality and centrality in protein networks. Nature. 2001; 411: 41-42.
21. Yu X, Chen K, Ma Y, Bai T, Zhu S, Cai D, et al. Molecular basis underlying changes of brain entropy and functional connectivity in major depressive disorders after electroconvulsive therapy. CNS Neurosci Ther. 2024; 30: e14690.
22. HAMILTON M. The assessment of anxiety states by rating. Br J Med Psychol. 1959; 32: 50-55.
23. HAMILTON M. A rating scale for depression. J Neurol Neurosurg Psychiatry. 1960; 23: 56-62.
24. Wang T, Xiao S, Li X, Wang H, Liu Y, Su N, et al. Reliability and validity of the Chinese version of the neuropsychiatric inventory in mainland China. Int J Geriatr Psychiatry. 2012; 27: 539-544.
25. Saczynski JS, Inouye SK, Guess J, Jones RN, Fong TG, Nemeth E, et al. The Montreal Cognitive Assessment: Creating a Crosswalk with the Mini-Mental State Examination. J Am Geriatr Soc. 2015; 63: 2370-2374.
26. Fischl B. FreeSurfer. Neuroimage. 2012; 62: 774-781.
27. Hedges EP, Dimitrov M, Zahid U, Brito Vega B, Si S, Dickson H, et al. Reliability of structural MRI measurements: The effects of scan session, head tilt, inter-scan interval, acquisition sequence, FreeSurfer version and processing stream. Neuroimage. 2022; 246: 118751.
28. Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol. 2011; 106: 1125-1165.
29. Hawrylycz MJ, Lein ES, Guillozet-Bongaarts AL, Shen EH, Ng L, Miller JA, et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature. 2012; 489: 391-399.
30. Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Elife. 2021; 10: e72129.
31. Abdi H, Williams LJ. Partial least squares methods: partial least squares correlation and partial least square regression. Methods Mol Biol. 2013; 930: 549-579.
32. Li J, Seidlitz J, Suckling J, Fan F, Ji GJ, Meng Y, et al. Cortical structural differences in major depressive disorder correlate with cell type-specific transcriptional signatures. Nat Commun. 2021; 12: 1647.
33. Xue K, Guo L, Zhu W, Liang S, Xu Q, Ma L, et al. Transcriptional signatures of the cortical morphometric similarity network gradient in first-episode, treatment-naive major depressive disorder. Neuropsychopharmacology. 2023; 48: 518-528.
34. Zhou Y, Zhou B, Pache L, Chang M, Khodabakhshi AH, Tanaseichuk O, et al. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat Commun. 2019; 10: 1523.
35. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 2000; 28: 27-30.
36. Alexopoulos H, Dalakas MC. The immunobiology of autoimmune encephalitides. J Autoimmun. 2019; 104: 102339.
37. Li J, Wang D, Xia J, Zhang C, Meng Y, Xu S, et al. Divergent suicidal symptomatic activations converge on somato-cognitive action network in depression. Mol Psychiatry. 2024; 29: 1980-1989.
38. Seidlitz J, Nadig A, Liu S, Bethlehem RAI, Vértes PE, Morgan SE, et al. Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders. Nat Commun. 2020; 11: 3358.
39. Romero-Garcia R, Seidlitz J, Whitaker KJ, Morgan SE, Fonagy P, Dolan 

    RJ, et al. Schizotypy-Related Magnetization of Cortex in Healthy Adolescence Is Colocated With Expression of Schizophrenia-Related Genes. Biol Psychiatry. 2020; 88: 248-259.

40. Váša F, Seidlitz J, Romero-Garcia R, Whitaker KJ, Rosenthal G, Vértes PE, et al. Adolescent Tuning of Association Cortex in Human Structural Brain Networks. Cereb Cortex. 2018; 28: 281-294.
41. Aziz Q, Thompson DG, Ng VW, Hamdy S, Sarkar S, Brammer MJ, et al. Cortical processing of human somatic and visceral sensation. J Neurosci. 2000; 20: 2657-2663.
42. McNab F, Klingberg T. Prefrontal cortex and basal ganglia control access to working memory. Nat Neurosci. 2008; 11: 103-107.
43. Menon V. 20 years of the default mode network: A review and synthesis. Neuron. 2023; 111: 2469-2487.
44. Astillero-Lopez V, Villar-Conde S, Gonzalez-Rodriguez M, Flores-Cuadrado A, Ubeda-Banon I, Saiz-Sanchez D, et al. Proteomic analysis identifies HSP90AA1, PTK2B, and ANXA2 in the human entorhinal cortex in Alzheimer’s disease: Potential role in synaptic homeostasis and Aβ pathology through microglial and astroglial cells. Brain Pathol. 2024; 34: e13235.
45. Tremblay MW, Green MV, Goldstein BM, Aldridge AI, Rosenfeld JA, Streff H, et al. Mutations of the histone linker H1-4 in 

    neurodevelopmental disorders and functional characterization of neurons expressing C-terminus frameshift mutant H1.4. Hum Mol Genet. 2022; 31: 1430-1442.

46. Dupuis JP, Nicole O, Groc L. NMDA receptor functions in health and disease: Old actor, new dimensions. Neuron. 2023; 111: 2312-2328.
47. Zhang M, Feng J, Xie C, Song N, Jin C, Wang J, Zhao Q, et al. Assembly and architecture of endogenous NMDA receptors in adult cerebral cortex and hippocampus. Cell. 2025; 188: 1198-1207.e13.
48. Rafe MR, Saha P, Bello ST. Targeting NMDA receptors with an antagonist is a promising therapeutic strategy for treating neurological disorders. Behav Brain Res. 2024; 472: 115173.
49. Tao C, Guang-WZ, Wen-Jian S, Junxiang JH, Li IZ, Huizhong WT, et al. Excitation-inhibition imbalance in medial preoptic area circuits underlies chronic stress-induced depression-like states. Nat Commun. 2024; 15: 8575.
50. McArdle CJ, Arnone AA, Heaney CF, Raab-Graham KF. A paradoxical switch: the implications of excitatory GABAergic signaling in neurological disorders. Front Psychiatry. 2024; 14: 1296527.
51. Zhao Y, Huang Y, Cao Y, Yang J. Astrocyte-Mediated Neuroinflammation in Neurological Conditions. Biomolecules. 2024; 14: 1204.