Unveiling Key Genetic Drivers of Brain Network Disruptions in Major Depressive Disorder
Poster No:
553
Submission Type:
Abstract Submission
Authors:
Yuan Liu1, Bo Yan2, Meijuan Li1, Bin Zhang1, Ying Gao1, Jie Li1
Institutions:
1Mental Health Center of Tianjin Medical University, Tianjin, China, 2Tianjin Medical University General Hospital, Tianjin, China
First Author:
Co-Author(s):
Introduction:
Major depressive disorder (MDD) is a prevalent psychiatric condition, contributing significantly to global disease burden. Neuroimaging studies have shown that MDD is linked to disruptions in the brain's functional topology, including changes in global and nodal network metrics. However, the molecular mechanisms underlying these abnormalities remain unclear.
The Allen Human Brain Atlas (AHBA) provides a valuable transcriptional map for investigating gene expression patterns related to brain network disruptions in MDD. While genes have been implicated in MDD-related network dysfunction, specific causal genes are still unidentified.To address this, we employ imaging transcriptomics, bioinformatics, and machine learning to identify key genes involved in MDD-associated brain network abnormalities.
The Allen Human Brain Atlas (AHBA) provides a valuable transcriptional map for investigating gene expression patterns related to brain network disruptions in MDD. While genes have been implicated in MDD-related network dysfunction, specific causal genes are still unidentified.To address this, we employ imaging transcriptomics, bioinformatics, and machine learning to identify key genes involved in MDD-associated brain network abnormalities.
·Figure1. Overview of the Study Design
Methods:
2.1 Dataset and Preprocessing
We analyzed brain imaging data from the REST-meta-MDD consortium, including 544 MDD patients (HAMD ≥ 14) and 569 age- and sex-matched HCs. Ethical approval and informed consent were obtained. Imaging preprocessing followed standard protocols for motion correction, normalization, and bandpass filtering.
2.2 Network Topology and Gene Expression
The brain was parcellated into 90 regions using the AAL atlas, and functional connectivity was computed based on BOLD signal correlations. Network metrics (degree centrality, nodal efficiency) were analyzed using the GRETNA toolbox. Gene expression data from the AHBA were mapped to brain regions and analyzed for topological changes using partial least squares (PLS) regression. Differentially expressed genes (DEGs) were identified from GSE98793 using the "limma" R package (p < 0.05, |log2 FC| > 0.2).
2.3 Machine Learning for Key Gene Identification
Key genes were selected by intersecting DEGs with genes identified in PLS regression. Seven machine learning models (LASSO, SVM-RFE, RF, GLM, GBM, XGB, KNN) were used for feature selection, and a logistic regression model was developed for diagnostic prediction. Model performance was evaluated using AUC, with validation from the GSE52790 dataset.
2.4 Validation
We validated the robustness of network topology changes and DEGs by analyzing different MDD subgroups (first-episode, severe MDD) and confirming consistency with primary results using left hemisphere data.
We analyzed brain imaging data from the REST-meta-MDD consortium, including 544 MDD patients (HAMD ≥ 14) and 569 age- and sex-matched HCs. Ethical approval and informed consent were obtained. Imaging preprocessing followed standard protocols for motion correction, normalization, and bandpass filtering.
2.2 Network Topology and Gene Expression
The brain was parcellated into 90 regions using the AAL atlas, and functional connectivity was computed based on BOLD signal correlations. Network metrics (degree centrality, nodal efficiency) were analyzed using the GRETNA toolbox. Gene expression data from the AHBA were mapped to brain regions and analyzed for topological changes using partial least squares (PLS) regression. Differentially expressed genes (DEGs) were identified from GSE98793 using the "limma" R package (p < 0.05, |log2 FC| > 0.2).
2.3 Machine Learning for Key Gene Identification
Key genes were selected by intersecting DEGs with genes identified in PLS regression. Seven machine learning models (LASSO, SVM-RFE, RF, GLM, GBM, XGB, KNN) were used for feature selection, and a logistic regression model was developed for diagnostic prediction. Model performance was evaluated using AUC, with validation from the GSE52790 dataset.
2.4 Validation
We validated the robustness of network topology changes and DEGs by analyzing different MDD subgroups (first-episode, severe MDD) and confirming consistency with primary results using left hemisphere data.
Results:
3.1 Network Topological Alterations in MDD
MDD patients exhibited lower nodal efficiency in the left superior parietal gyrus and higher efficiency in the bilateral thalamus, right caudate nucleus, and middle frontal gyrus. Degree centrality was increased in the right thalamus, and local efficiency was decreased in MDD patients compared to controls.
3.2 Gene-Related Network Alterations
PLS1 explained 30.63% of the variance in network changes (p = 3 × 10⁻³), with significant correlations to degree centrality and nodal efficiency. Of 15,633 genes, 3,247 were significant, with 882 showing positive and 2,365 negative weights.
3.3 DEGs and Key Genes
215 DEGs were identified from the GSE98793 dataset (p < 0.05). Among them, 24 key genes were linked to network changes, enriched in pathways like Rap1 signaling. FKBP5, PTX3, and APCDD1 were identified as key genes. A logistic regression model with these genes achieved an AUC of 0.65, and a nomogram showed an AUC of 0.728.
3.4 Validation of Network and Gene Findings
Network alterations were consistent across first-episode and severe MDD patients. Using left-hemisphere AHBA data, 17 key genes were identified, with 82.35% overlapping with the main results, confirming their robustness.
MDD patients exhibited lower nodal efficiency in the left superior parietal gyrus and higher efficiency in the bilateral thalamus, right caudate nucleus, and middle frontal gyrus. Degree centrality was increased in the right thalamus, and local efficiency was decreased in MDD patients compared to controls.
3.2 Gene-Related Network Alterations
PLS1 explained 30.63% of the variance in network changes (p = 3 × 10⁻³), with significant correlations to degree centrality and nodal efficiency. Of 15,633 genes, 3,247 were significant, with 882 showing positive and 2,365 negative weights.
3.3 DEGs and Key Genes
215 DEGs were identified from the GSE98793 dataset (p < 0.05). Among them, 24 key genes were linked to network changes, enriched in pathways like Rap1 signaling. FKBP5, PTX3, and APCDD1 were identified as key genes. A logistic regression model with these genes achieved an AUC of 0.65, and a nomogram showed an AUC of 0.728.
3.4 Validation of Network and Gene Findings
Network alterations were consistent across first-episode and severe MDD patients. Using left-hemisphere AHBA data, 17 key genes were identified, with 82.35% overlapping with the main results, confirming their robustness.
Conclusions:
Our study highlights significant brain network alterations in MDD and identifies 24 genes associated with these changes. Notably, FKBP5, PTX3, and APCDD1 emerge as key diagnostic markers, offering potential for more precise, personalized approaches to diagnosing and treating MDD.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Genetics:
Transcriptomics
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 2
Neuroinformatics and Data Sharing:
Databasing and Data Sharing
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Affective Disorders
FUNCTIONAL MRI
Machine Learning
Psychiatric Disorders
1|2Indicates the priority used for review