Gene Expression has Distinct Associations with Brain Structure and Function in Major Depressive Disorder

Abstract Major depressive disorder (MDD) is associated with structural and functional brain abnormalities. MDD as well as brain anatomy and function are influenced by genetic factors, but the role of gene expression remains unclear. Here, this work investigates how cortical gene expression contributes to structural and functional brain abnormalities in MDD. This work compares the gray matter volume and resting‐state functional measures in a Chinese sample of 848 MDD patients and 749 healthy controls, and these case‐control differences are then associated with cortical variation of gene expression. While whole gene expression is positively associated with structural abnormalities, it is negatively associated with functional abnormalities. This work observes the relationships of expression levels with brain abnormalities for individual genes, and found that transcriptional correlates of brain structure and function show opposite relations with gene dysregulation in postmortem cortical tissue from MDD patients. This work further identifies genes that are positively or negatively related to structural abnormalities as well as functional abnormalities. The MDD‐related genes are enriched for brain tissue, cortical cells, and biological pathways. These findings suggest that distinct genetic mechanisms underlie structural and functional brain abnormalities in MDD, and highlight the importance of cortical gene expression for the development of cortical abnormalities.


Replication dataset (UK Biobank)
We used UK Biobank data with brain scans as replication dataset. [1] All participants provided written informed consent and the project was granted ethical approval (reference: 11/NW/0382) by the National Health Service North West Centre for Research Ethics Committee. As shown in Figure S19, a total of 802 individuals were identified as depressed individuals using the following criteria: (1) Participants reported the presence of one of the core symptoms of depression as frequency of depressed mood (Data Field: 2050) or unenthusiasm / disinterest (Data Field: 2060) in the last 2 weeks prior to brain scanning that has been present "more than half the days" or "nearly every day", and (2) Data for these participants were available for the following variables: T1 and restingstate functional MRI data as well as covariates including fluid intelligence score (Data Field: 20016), educational attainment (Data Field: 6138), sex (Data Field: 31), and age when attended assessment centre (Data Field: 21003).
It should be noted that the two items about depression mentioned above were the only two symptoms of depression available in the UK Biobank that reflect the mood state during the MRI assessment, [2] whereas a total of five out of nine symptoms are required to get a full diagnosis of major depressive disorder (MDD). The controls had MRI data and covariates available, but reported no presence ("not at all") of negative mood symptom or a loss of disinterest in the last 2 weeks prior to brain scanning. Moreover, the controls did not have a history of MDD, and never diagnosed to have any mental health problems. Of this control group we randomly selected 802 individuals matching age, sex, intelligence, and educational attainment with those of the depressed individuals (Table S6).

Imaging procedures in UK Biobank
As replication dataset, we used T1-weighted MRI and rsfMRI data from UK Biobank. The brain imaging scanners used were standard Siemens Skyra 3T running VD13A SP4 with a 32-channel head coil. T1 scanning lasted about 5 minutes with the following parameters: repetition time = 2000 ms; echo time = 2.1 ms; flip angle = 8°; matrix size = 256 × 256 mm; voxel size = 1 × 1 × 1 mm; number of slices = 208. The acquisition parameters for the rsfMRI data were TR = 735 ms, TE = 39 ms, flip angle = 52°, matrix size =88 × 88 mm, voxel size = 2.4 × 2.4 × 2.4 mm, number of slices = 64, and volumes = 490. A series of preprocessing procedures were applied for T1 and rsfMRI data (http://biobank.ctsu.ox.ac.uk/crystal/crystal/docs/brain_mri.pdf). UK Biobank provided cortical surface area (CSA) and cortical thickness (CT) of 66 regions based on the Desikan-Killiany (DK) atlas. [3] For rsfMRI data, we used the BRANT toolbox [4] to estimate the amplitude of low-frequency fluctuation (ALFF) and regional homogeneity (ReHo) to detect the regional intensity of spontaneous fluctuations in the BOLD signal, and extracted region values based on the DK atlas. High-pass temporal filtering has been applied to the rsfMRI data in UK Biobank which can cause calculation bias for fractional ALFF (fALFF), so we did not include fALFF in statistical analysis.

Statistical analysis in UK Biobank
We first used two-sample t tests to compare the neuroimaging measures for cortical thickness (CT), ALFF, and ReHo based on the DK atlas between depressed individuals and controls. The t-values were then converted to Cohen's d effect sizes for ease of interpretation. The covariates, including age, sex, education, intelligence, and head motion, were regressed out of the brain measures. Principle component analysis (PCA) was used to extract the first components of effect sizes for ALFF and ReHo (funcPC1). Moreover, AHBA transcriptome gene expression data were mapped to the DK atlas, and the first component pf gene expression matrix (genePC1) was also extracted. To identify the global transcriptome-neuroimaging relationships, we related the genePC1 to funcPC1 and CT differences, respectively.

Statistical analysis in sex differences
We split the whole REST-meta-MDD dataset into the male group and female group. The male group consisted of 314 MDD patients (age range: 18 -65 years old) and 327 healthy controls (age range: 18 -64 years old). The female group was comprised of 534 MDD patients (age range: 18 -65 years old) and 437 healthy controls (age range: 18 -63 years old). The covariates, including age, education, and head motion, were regressed out of the brain measures. We then used two-sample t test to examine the structural and functional case-control differences in the male and female group, respectively, and associated case-control differences with genePC1.

Statistical analysis in age stratification
We split the whole REST-meta-MDD dataset into an older group and younger group respective to the age of 32 years old. The younger group was comprised of 428 MDD patients and 361 healthy controls. The older group consisted of 420 MDD patients and 433 healthy controls ( Figure S10). The covariates, including age, sex, education, and head motion, were regressed out of the brain measures. We then used two-sample t test to examine the structural and functional brain case-control differences in the older and younger group, respectively, and associated case-control differences with genePC1

Bin-based correlation analysis
As shown in Figure S20, we conducted the bin-based correlation analysis to link transcriptional correlates of brain abnormalities to differential gene expression (DGE) values. For all AHBA genes, we obtained their relationships of interregional expression levels with brain abnormalities (Figure S20A). According to the study by Gandal et al., we also obtained the differential gene expression (DGE) values for psychiatric disorders. We overlapped the AHBA genes and DGE genes ( Figure S20B). The overlapping genes (N) were then sorted by their correlations between gene expression and brain abnormalities. These ordered genes were then clustered into 100 bins (bin size n = N/100). In detail, the top n genes were included as the first bin, the second n genes were grouped as the second bin, and so on. For all n genes in a bin, we calculated average DGE values, as well as average correlations between gene expression and abnormalities, and associated them using Spearman's correlation analysis ( Figure S20C).

The robustness of GMV case-control differences
We obtained the CT differences between MDD patients and controls across cortical regions based on meta-analysis in ENIGMA dataset [5] and associated them with our identified GMV case-control differences. A significant positive correlation was observed between structural brain abnormalities from two studies ( Figure S21). The consistency demonstrated that GMV case-control differences were robust and represented the distribution of structural brain abnormalities, although no significant differences were observed.

The effects of GMV on case-control functional brain differences
While testing the case-control functional brain differences, the covariates, including age, sex, education, and head motion were regressed out of functional brain measures. However, changes in GMV may artificially inflate or mask changes in functional measures in the same brain region. To test the effects of GMV, we corrected the functional brain measures for total GMV (TGMV). Compared to original case-control differences without correcting for TGMV, the effect sizes for TGMV-corrected functional brain measures hardly changed ( Figure   S22A-C). In addition, for functional brain measures of each region, we further corrected them for their regional GMV, and still found small changes on effect sizes (Figure S22D-F). These results indicate that small changes in functional measures caused by GMV changes do not influence the case-control brain differences. Figure S1. Case-control functional brain differences based on the Brainnetome (BN) atlas. (A) Differences (Cohen's d) for amplitude of low-frequency fluctuation (ALFF), and statistically significant regions; (B) Differences (Cohen's d) for fractional ALFF (fALFF), and statistically significant regions; (C) Differences (Cohen's d) for regional homogeneity (ReHo) and statistically significant regions; (D) the correlations between effect sizes for functional brain measures.             differences. X axis shows the chromosomes, and Y axis shows the -log10(p) value, which indicates the significance of the association of the gene expression with cortical differences. The horizontal dotted line indicates the significance threshold after correcting for multiple testing. Figure S15. The correlations of the transcriptional correlates of brain abnormities with differential gene expression (DGE) values for major depressive disorder (MDD) when all genes are individually included. The funcPC1 is the first principal component of effect sizes for functional brain measures, and GMV indicates the gray matter volume. Figure S16. Enrichment analysis for 500 up-and down-regulated genes for psychiatric disorders. pFDR is the adjusted p value after FDR multiple testing correction. FuncPC1 is the first principal component of effect sizes for functional brain measures, and GMV indicates gray matter volume. Genes positively and negatively related to funcPC1 and GMV differences are defined as funcPC1+, funcPC1-genes, GMV+, and GMV-genes. Major depressive disorder, MDD; autism spectrum disorder, ASD; bipolar disorder, BP; schizophrenia, SCZ. (F) 800. pFDR is the adjusted p value after FDR multiple testing correction. FuncPC1 is the first principal component of effect sizes for functional brain measures, and GMV indicates the gray matter volume. Genes positively and negatively related to funcPC1 and GMV differences are defined as funcPC1+, funcPC1-genes, GMV+, and GMV-genes. Major depressive disorder (MDD), autism spectrum disorder (ASD), bipolar disorder (BP), and schizophrenia (SCZ). Figure S18. Sample selection of REST-meta-MDD dataset copied from the study by Yan et al.. [6] (The open access article has been distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).)   Figure S21. The relationship of structural brain abnormalities between REST-meta-MDD and ENIGMA dataset. Figure S22. The effects of gray matter volume (GMV) on case-control functional brain differences. (A) Amplitude of low-frequency fluctuation (ALFF) corrected for total GMV (TGMV); (B) Fractional ALFF (fALFF) corrected for TGMV; (C) Regional homogeneity (ReHo) corrected for TGMV; (D) ALFF corrected for TGMV and regional GMV; (E) fALFF corrected for TGMV and regional GMV; (F) ReHo corrected for TGMV and regional GMV.