Expression profile meta-analysis
To systematically analyze the expression profile of brain tissues from patients with MDD, we integrated the expression profile datasets from anterior cingulate cortex and prefrontal cortex respectively to increase confidence in the biological reality. The integrated datasets of ACC contained 152 samples, including 76 disease samples and 76 healthy controls. While integrated datasets of PFC included 126 samples, 63 of them were disease samples and 63 healthy controls. The “limma” R package was applied for the identification of differentially expressed genes (DEGs). As a result, 586 DEGs were obtained in ACC (Fig 1a) and 616 DEGs in PFC (Fig 1b). Comparative analysis between the two brain tissues showed that the gene expression patterns in these two tissues were highly consistent, with most of the DEGs down-regulated in both tissues (Fig 1c). Up to 50 % DEGs were dysregulated in both of the two tissues. These results indicated that during depression development, there is a certain relevance in the abnormal alterations between the two brain tissues.
Co-expression network construction
The dependencies between genes and their relationships to other co-expressed genes reflected closely associations of these genes in biology (33). It is speculated that the significantly co-expressed genes may have similar biological functions and associate with similar or same biological pathways. We utilized Pearson correlation coefficient to identify the co-expression relationships of genes from the two brain tissues in both normal and disease conditions. Co-expression networks were constructed based on these relationships by cytoscape software (Fig 2).
The network topological properties were analyzed (Table 1). As compared to normal condition, the number of nodes had no significant change in the co-expression network under disease condition, but gene connections obviously loss in the network, which may result in the increase of unconnected nodes in the disease network (Table 1). The network clustering coefficient, density and centralization were significantly reduced, suggesting that the centrality and robustness of the disease network were significantly decreased as compared to the normal network. We statistically calculated the significance of the gene connections in both normal and disease conditions using Wilcox test. In PFC, Wilcox test P=0.04, and in ACC, P=1.227e-09. The results revealed that gain or loss of gene connections showed significant statistical significance in both tissues under disease condition, and more significant in ACC.
Comparative analysis on difference of networks
Through the network topology analysis, we found that as compared to normal control, in the brain tissues of patients with depression, the reduced connectivity and centrality of the disease co-expression network and the loss of gene connections were important features along with the disease development. To further compare the correlation of the two tissue lesions and the disease, we performed comparative analysis on the co-expression networks under disease condition. The nodes with gain of connections (Fig 3a) and loss of connections (Fig 3b) were compared in the two disease networks. We found that the number of nodes with gain of connections were nearly balanced with that of nodes with loss of connections in PFC (Fig 3c). Thus, in the disease condition, stability of PFC network is better than that of ACC network, and the PFC network showed better resistance to the disease signal. Our data also showed that the number of nodes with gain of connections in PFC network was higher than that in ACC network with ratio of 1.72:1, while the numbers of nodes with loss of connections tended to be similar in the two network with ratio of 1.02:1. Some genes had gain or loss of connections in both ACC and PFC cortex, which may be bridge genes connecting the two tissues by regulating similar biological functions in both parts.
To get an overall view of the gene expression pattern of the two tissues in disease condition, we statistically calculated the probability density distribution of the co-expression networks in diseased tissues compared to that in normal tissues (Fig 3d and 3e). The results showed that in ACC of depression, variance of density distribution was markedly increased (Fig 3d), suggesting that the volatility of the network was increased; while in PFC of depression, the connectivity density distribution of network tended to normal, there were no obvious deviation on mean and variance (Fig 3e). This observation indicated that in disease states, the PFC network state tended to be normal, whereas the ACC network showed drastic fluctuations, which was in line with the previous results of Wilcox test. These results revealed that the lesions of brain tissues in patients with depression were not synchronized and alterations of biological functions were not consistent either, ACC showed a greater degree of abnormal as compared to PFC suggesting a higher correlation with disease progression. Therefore, ACC is more likely to exist important therapeutic targets.
Functional pathway analysis
To gain a knowledge about the DEGs influenced biological effects, functional enrichment analysis is conducted in up- and downregulated DEGs, respectively. We found that DEGs in ACC mainly associated with circulatory system related pathways (Fig 4a), while DEGs in PFC were enriched in metabolic system related pathways (Fig 4b). The gene count and pathway P values were compared respectively (Fig 4c). The results showed that along with the increase of gene count, the P value of pathway was elevated. Thus, the pathways enriched by more DEGs were more likely to associate with disease. Moreover, the genes participated in more signaling pathways were more likely to play important roles during disease progression. We further performed cross talk analysis on these DEGs enriched pathways, to identify the cross talk genes that regulate multiple signaling pathways.
Cross talk analysis
A regulatory network was constructed using signaling pathways and genes enriched in the pathways. Two types of nodes were contained, signaling pathways and genes; three types of connections were included, signaling pathway-gene (belong), gene-gene (co-express) and gene-gene (interact). The relationship of “belong” was extracted from KEGG pathway database; the “co-express” relationship was from co-expression network; the “interact” relationship was extracted from HPRD database (http://www.hprd.org/). As a result, a pathway-gene complex network was established including 219 relationship pairs, 16 signaling pathways and 70 genes (Fig 5a).
The network topological properties were analyzed and the degree distribution was obtained. The signaling pathways and genes with highest degrees ranking as top 10 were extracted (Table 2). These pathways were associated with more DEGs suggesting a significant correlation between the abnormal of these functions with the occurrence of depression. While, DEGs with high degrees interacted or co-expressed with each other participated in multiple signaling pathways, they may act as cross talk genes dysregulated in disease progression, and may also be potential novel therapeutic targets.
We further statistically analyzed the significance of the top10 genes through group comparison between ACC and PFC in both disease and normal conditions (Table 3). We found 9 genes significantly differentially expressed in PFC of patients with depression, while 6 genes significantly differentially expressed in ACC of patients with depression, excluding ITGA3, MAPK11, PAK6 and DUSP8 (Table 3). The results exhibited that significant difference exists between ACC and PFC during development of depression, reflecting the specificity of alteration in each tissue, but the sharing 6 DEGs in both tissues revealed the consistency of changes in both tissues during disease progression. Receiver operating characteristic (ROC) values of genes were calculated in the two tissues. All the ROC values of genes were higher than random state (0.5) (Fig 5b and Table 3). To examine whether these top10 genes are capable to be potential novel therapeutic targets, we queried these top10 genes in Drugbank database. As shown in Table 4, five genes were known targets that targeted by several drugs. Literature mining results showed that except Blinatumomab and KC706, other drugs have been reported to be related with brain tissue injury and cerebral nervous system diseases. Moreover, MAPK11 and PAK6 only differentially expressed in PFC, while CACNA1A, PTDSS2 and CD19 differentially expressed in both ACC and PFC. Therefore, these three genes may correlate with depression progression or other cerebral nerve system related diseases, are more likely to become the new drug targets.