Stress-induced altered expression of hippocampal nuclear and mitochondrial encoded genes in rats and cross-species genetic associations reveal molecular links to depression

Abstract

Background

Mitochondria play a pivotal role in energy production, and their dysfunction not only hampers cells’ ability to meet energy requirements but also contributes to the impairment of neural plasticity, a critical feature of depressive disorders. In this study, mitochondrial cross-omics analysis was carried out in the hippocampus of restraint rats to understand the role of mitochondria in depression pathophysiology.

Methods

The expression profiles of hippocampal mitochondrial and nuclear-encoded genes in mitochondrial fractions from restraint and handled control rats were obtained using high-throughput RNA sequencing. Weighted gene co-expression network analysis (WGCNA) was used to identify the gene co-expression and pathways associated with the restraint phenotype. Mutual Information Network algorithm tools Arance, CLR, and MRNET were additionally used to screen the functional modules and hub genes and their similarity with the WGCNA-based network analysis. Finally, cross-species homology followed by gene association analysis was conducted to obtain SNPs and haplotypes related to depression phenotype.

Results

A significant proportion of mitochondrial and nuclear-encoded genes showed differential regulation in the hippocampus of restraint rats. WGCNA and Mutual Information Network analysis yielded distinct functional modules significantly related to restraint phenotype. Further network analysis revealed distinct co-expression patterns associated with differentially expressed genes associated with these modules. Cross-species analysis showed 39 significantly associated SNPs with the depression phenotype, where the most significant SNP, rs10899570, was located within the TENM4 gene. Further, rs1573529 and rs10899570 were distributed into the linkage disequilibrium block where SNPs were highly correlated. Subsequent haplotype analysis showed that rs1573529 and rs10899570 were significantly associated with depressive behavior.

Conclusions

The study demonstrates a significant impact of restraint stress on mitochondrial functions and genetic association, suggesting their critical role in depression pathophysiology.

Significance Statement.

Major depressive disorder (MDD) is a global health issue. Mitochondria play a pivotal role in energy production, and their dysfunction not only hampers the brain’s ability to meet energy requirements but also contributes to the impairment of neural plasticity, a critical feature of MDD. In this study, we used restraint stress as an animal model of depression-like behavior using Sprague–Dawley rats and examined the molecular characteristics of nuclear and mitochondrial-encoded genes in the hippocampus. Additionally, we explored the evidence of functional genes in humans through cross-species research methods. Our data show that a significant proportion of mitochondrial and nuclear-encoded genes are differentially regulated and co-expressed in restrained rats. Cross-species analysis revealed several significantly associated SNPs; among them, rs1573529 and rs10899570 were significantly associated with depressive behavior. Altogether, our study demonstrates a significant impact of restraint stress on mitochondrial functions and genetic association, suggesting their critical role in depression.

INTRODUCTION

Depression is a common neuropsychiatric disorder, affecting up to 5% of the population globally.¹ Depressed individuals have a higher incidence of other physical illnesses (ie, comorbid cardiovascular disease, stroke, etc.), decreased social functioning, and a high mortality rate.² The complexity of this disorder is further exacerbated by the fact that it often co-occurs with other psychiatric disorders,³ which indicates a more severe form of depression, longer recovery time, higher risk of relapse, higher disability, and more suicide attempts.⁴ It is generally accepted that both environmental and genetic factors influence the development of depression.^5,6 Additionally, disruption of neural plasticity caused by stress and other negative stimuli plays a significant role in the onset and development of depression.^7–9 However, despite extensive research, a fundamental understanding of the specific biological changes that cause depressive symptoms remains lacking.

The brain, being highly reliant on energy, contains a high density of mitochondria.^10,11 Mitochondria are the only organelles outside the nucleus that possess their own genome: the circular, double-stranded mitochondrial DNA (mtDNA), which is inherited maternally.^12,13 Evidence from autopsy,^14–18 imaging,¹⁹ genetic,^14,20 and cellular²¹ studies has emerged to link mitochondrial dysfunction to mood disorders. Mitochondrial mutations have been observed in individuals diagnosed with depression,^22,23 and mitochondrial dysfunction and depression co-morbidity exist.²⁴ Several recent studies have implicated multiple specific mitochondrial genes in depression. Using a systematic evaluation of mitochondrial PCR array analysis, our group identified 16 genes that were differentially expressed in the dorsolateral prefrontal cortex of postmortem brains from patients with depression versus controls.²⁵ The genes identified are known to regulate oxidative stress and neuronal ATP levels, suggesting that mitochondrial genes are altered in the brains of major depressive disorder (MDD) patients. Similarly, the mitochondrial ATP6V1B2 gene has been implicated in MDD, possibly through its effects on neurotransmission and receptor-mediated endocytosis.²⁶ Less direct evidence comes from the observation that mice with mutations in the POLG gene (encoding a subunit of mitochondrial DNA polymerase) exhibit depressive-like symptoms,²⁷ and polymorphisms in genes encoding mitochondrial enzymes, such as MTHFD1L, are associated with negative rumination, a precursor to depression.²⁸ This polymorphism is also associated with high levels of homocysteine, which is associated with hippocampal volume and depression.²⁹

Interestingly, stress, one of the causal factors in depression, has been shown to be associated with alterations in mitochondrial functions.^30,31 Chronic stress can cause mitochondrial fragmentation, mtDNA damage, oxidative stress, and cell death.^32,33 Recently, we showed that early-life stress can alter mtDNA copy number along with telomere shortening, which can be associated with MDD and suicidality.³⁴ Animal studies also suggest that increased glucocorticoid levels can reduce mitochondrial membrane potential, respiratory chain enzymatic activity, and increased production of reactive oxygen species.^35,36 Research on mice indicates that chronic unpredictable stressors can trigger depression-like behaviors linked to reduced mitochondrial respiration rates.³⁷ As the mitochondrial genome contains a restricted number of genes (13 proteins, 22 tRNAs, two rRNAs), it cannot support its normal functions alone. Thus, mitochondria import the majority of proteins that are encoded by the nuclear genome, setting up a stage for mitonuclear cross-talk,^38,39 which is modulated under stressful conditions.⁴⁰ In the present study, we used restraint stress as an animal model of depression-like behavior and examined the molecular characteristics of nuclear and mitochondrial-encoded mRNA omics in the hippocampus, a brain region highly relevant to neural plasticity and whose functions are altered in MDD patients.^7,41–43 Additionally, we explored the evidence of functional genes in transcriptomics and genomics through cross-species research methods. Restraint is a well-established source of stress that leads to depressive behaviors in rats. These rats exhibit hippocampal atrophy,^44–47 increased corticosteroid levels, heightened aggressive behaviors,⁴⁸ and depressed behavior. Also, rats exposed to restraint stress show genomic changes that are associated with depression. This model enables research into a primary, progressive form of depression without confounds from resilience or other factors. We have used this stress model in our previous study.⁴⁹ We used the entire hippocampus (CA1-3 and dentate gyrus) for the study, as interconnected subfields work together to support complex processes such as learning and memory. Studying the entire hippocampus can provide a more comprehensive understanding of its overall role in these functions, including how its subfields interact.

MATERIALS AND METHODS

Animals

Adult male Sprague–Dawley rats (250–300 g body weight) were obtained from Envigo (Indianapolis, USA) and housed in similar cages (2 rats/cage) within the same room under standard laboratory conditions (temperature 21 ± 1°C, humidity 5 ± 5%). Animals were given free access to food and water and adapted to the laboratory environment for one week prior to the experiment. Rats were randomly assigned to the handled control group and a restraint stress group. Restraint stress was given to rats during the light cycle (08:00–12:00). All the experiments were carried out according to the National Institutes of Health guide for the care and use of Laboratory animals and were approved by the Animal Care Committee of the University of Alabama at Birmingham.

Restraint Stress

Rats were given restraint stress, as detailed earlier.⁴⁹ Briefly, rats were placed individually in clear acrylic tubes (21.59 cm long, 6.35 cm internal diameter, air vents in the cap and along the tube) with the tail extending from the rear of the tube. The cap was placed inward enough to prevent the rat from moving forward or backward inside the tube. Rats were restrained for 2 hours each day for 14 consecutive days. Control rats were handled daily but not restrained. All the studies were done in seven handled controls and seven restraint rats. Twenty-four hours after the final restraint session, rats were decapitated and their brains removed. Rat brains were dissected and flash-frozen with liquid nitrogen and stored at -80°C. A serial dissection using a cutting block was performed following the reported protocol,⁵⁰ and the hippocampus was dissected out and stored at -80°C until analysis. All the experiments were done in a blinded fashion, and the rats were randomly assigned to two groups. In this study, we used the entire hippocampus, comprising CA1-3 and the dentate gyrus.

Corticosterone Assay

Trunk blood from the rat was collected on ice during the light phase (12:00 p.m. to 4:00 p.m.) when corticosterone levels are naturally low and more stable. This timing helps us clearly detect any increases caused by chronic stress, without the influence of normal daily hormone fluctuations. Blood samples were centrifuged to obtain serum, which was then stored at -80°C until use. The serum corticosterone levels were measured in platelet-free plasma using an enzyme-linked immunosorbent assay (Enzo Life Sciences, USA).

Mitochondria Isolation

The frozen hippocampi were used to isolate the mitochondria with a mitochondria purification protocol. Briefly, ~40 mg tissue was homogenized with a pestle in 500 μL of chilled NKM buffer (NKM buffer composed of 1 mL Tris–HCl pH 7.4, 0.13 M NaCl, 5 mM KCl, and 7.5 mM MgCl₂) and centrifuged at ~400 g for 10 minutes. The supernatant was decanted, and the pellet was washed with 1 mL of NKM buffer, followed by centrifugation at 400 g for 10 minutes. The pellet was resuspended in 600 μL of homogenization buffer (10 mM Tris HCl pH 6.7, 10 mM KCl, 0.15 mM MgCl₂, 1 mM PMSF, 1 mM DTT) and incubated on ice for 10 minutes. 600 μL of 1.5 M sucrose solution was added to each tube and mixed gently. The samples were centrifuged at 1200 g for 10 minutes, after which the supernatant was transferred to another tube, discarding the pellet. The supernatant was further centrifuged at 7000 g for 10 minutes. The pellet (mitochondria) was suspended in 300 μL of mitochondrial suspension buffer (10 mM Tris–HCl pH 6.7, 0.15 mM MgCl₂, 0.25 M sucrose, and 1 mM DTT). The supernatant was discarded, and 1 mL of TRIzol was added to isolate RNA, as described earlier.⁵¹

RNA Extraction

Total RNA was extracted using TRIzol reagent. The RNA quality was assessed using Nanodrop (260/280 nm; cutoff ≥1.8), and its integrity was evaluated using agarose gel electrophoresis. The RNA Integrity Number for all samples was maintained within an acceptable range of >7.

RNA Sequencing

RNA sequencing libraries were prepared using the SEQuoia RNA-Seq Library Prep Kit, following the manufacturer’s instructions. The library was initially fragmented, and an end repair enzyme was employed for 3’ end repair. Following the polyA tailing step, the kit utilizes a proprietary SEQzyme, which simplifies the library preparation process by merging cDNA synthesis and adapter ligation into a single step. After cDNA purification and cleanup using magnetic beads and SPRIselect reagent, the PCR library amplification was performed. The post-amplification cleanup also employed SPRIselect reagent. Following library preparation, ribosomal RNA was depleted using the SEQuoia RiboDepletion Kit. The quality of the prepared library was assessed using Agilent 2100 Bioanalyzer and quantified by qPCR absolute quantification method. Finally, single-end library sequencing was performed on the Illumina HiSeq 4000 according to the manufacturer’s protocol.

Identification of Significantly Differentially Expressed Genes

The SEQuoia Complete Kit adapters feature a single cytosine nucleotide base located immediately upstream of the insert, along with a poly(A) sequence positioned directly after the insert. To trim the cytosine base, the poly(A) tail, and any reads shorter than 15 bases, we employed Cutadapt⁵² with the following parameters: -u 1, -a A{10}, and -m 15. For the alignment of reads against the rat reference genome (GCF_036323735.1), we utilized the STAR aligner.⁵³ The read counts were generated through the rsem-calculate-expression function from the RSEM package.⁵⁴ Furthermore, the sequencing reads that did not align with the nuclear genome were mapped to the mitochondrial reference genome of Rattus norvegicus (NC_001665.2) using the STAR aligner, and transcript counts were subsequently calculated with rsem-calculate-expression.

The differential analysis was performed on nuclear and mitochondrial genes using the read counts obtained from control and restraint samples through the DESeq2 package.⁵⁵ Significant dysregulated genes were identified with a P < .05 and a fold change threshold of 2 for nuclear genes and 1.2 for mitochondrial genes. Significantly altered nuclear-encoded genes in restrained rats were correlated with corticosterone levels using Spearman correlation.

Weighted Gene Co-Expression Network Analysis

To identify the gene co-expression patterns (modules), hub genes, and pathways associated with the restraint phenotype, we performed weighted gene co-expression network analysis (WGCNA) as described earlier⁵⁶ with some modifications. To reduce the correlation differences caused by the size of the module gene, we introduced the balance cluster function so that each module had a similar size. The specific steps are as follows: we extracted the differentially expressed gene (DEG) expression values as the expression analysis matrix and performed WGCNA. The function pickSoftThreshold was used to determine the appropriate number as the soft threshold, which can enhance the correlation of genes. Then, after calculating the adjacency, the adjacency matrix was converted into a topological overlap matrix using the function TOMsimilarity. Cluster analysis was performed based on the hclust function and cluster_balance, which helped to classify a group of genes with a high topological overlap index into one module and marked with one color name. Considering the number of DEGs and the default value, we set the module number to 15. Subsequently, the Spearman correlation coefficient in the association analysis between each module and phenotypic traits was used to analyze the correlation between the module and the phenotypic characteristics, and the key modules with the most significant correlation coefficient and the smallest P-value were selected for further study. Finally, the chooseTopHubInEachModule function was used to identify the hub genes in the association module, and the clusterProfile package was used to enrich the genes in the module into the signaling pathway.

Gene Ontology Analysis

Human-specific orgDb package org.Hs.eg.db was used, which is most suitable for converting gene IDs or obtaining GO information for the current genome construction. This package contains all the information needed for GO analysis, that is, gene list, gene list of interest, gene score (if any), and the GO ontology part (GO graph). Each gene symbol maps to a named vector containing the corresponding entrez gene identifier. The name of the vector corresponds to the gene symbol. GO contains three orthogonal ontologies, namely molecular function, biological process (BP), and cellular component.⁵⁷ The biological functions of the significantly up-and down-regulated genes were analyzed based on topGO.⁵⁸

Other Network Analysis

In this study, in addition to the WGCNA described previously, we also constructed a mutual information (MI)–based network using the MINET (Mutual Information Network) software package.⁵⁹ MINET is a widely used R package that enables the inference of gene regulatory networks from high-dimensional omics datasets, such as transcriptomic data. A key advantage of MINET is its ability to capture both linear and non-linear dependencies between genes or molecular entities, making it particularly suitable for reconstructing real biochemical regulatory networks. By leveraging MI-based inference methods, MINET provides valuable insights into the complex and often nonlinear interactions that underlie biological systems. For this analysis, we focused on three well-established algorithms available within the MINET framework: ARACNE (Algorithm for the Reconstruction of Accurate Cellular Networks),⁶⁰ CLR (Context Likelihood of Relatedness),⁶¹ and MRNET (Minimum Redundancy Network).⁶² (1) ARACNE reconstructs gene regulatory networks using an information-theoretic approach that evaluates all possible three-node (triplet) interactions within the dataset. By applying the Data Processing Inequality (DPI), ARACNE eliminates indirect associations, retaining primarily direct regulatory relationships. This property makes ARACNE particularly effective in reducing the number of false-positive edges in dense networks.⁶⁰ (2) CLR extends the traditional MI framework by incorporating adaptive background correction. Specifically, CLR evaluates the MI value of each gene pair relative to the empirical distribution of MI values for each gene, enhancing the specificity of inferred interactions. This approach adjusts for variations in the connectivity degree across different nodes, making it especially useful for detecting context-dependent gene–gene relationships.⁶¹ (3) MRNET employs a feature selection strategy known as Maximum Relevance Minimum Redundancy. This method prioritizes gene pairs that exhibit high MI while simultaneously minimizing redundancy among selected variables. The MRNET algorithm thus aims to balance the selection of informative interactions and reduce the inclusion of spurious or redundant links in the reconstructed network.⁶² All three algorithms utilize non-parametric estimators to calculate MI from the data. The MINET package offers multiple estimation techniques, such as empirical estimators and kernel density estimators, allowing users to tailor the inference process to the characteristics of their specific dataset.⁵⁹ By integrating these complementary algorithms, we aimed to comprehensively capture both direct and indirect, as well as linear and non-linear, regulatory relationships within our target biological system. The combination of WGCNA and MINET-based network construction provided a multi-layered and high-resolution understanding of gene regulatory interactions, enhancing the robustness and interpretability of our network analysis.

Cross-Species Genetic Association Analysis of Differentially Expressed Genes

Based on the screening of differentially expressed genes, we used the Rgd analysis software⁶³ to conduct cross-species homology analysis between rats and humans. We obtained the human homology genes of the differentially expressed genes related to restraint stress. Then we conducted gene association analysis based on the human homology genes. In the genomic study, we focused on mitochondria SNP, top DEGs, and SNP sites within the hub gene. A total of 894 SNP sites in 24 genes (including the top 10 significant genes and 15 hub genes in each module) were retrieved. We obtained the genotype data of dbGAP and retrieved the genotypes of 891 SNPs with the help of plink⁶⁴ and bcftools.⁶⁵ At the same time, association analysis and haplotype association analysis were carried out in 1882 MDD samples and 1891 control samples. The analysis method used the SHEsis software analysis platform.⁶⁶ to screen out SNP sites and haplotypes related to the phenotype, providing further genomic evidence for the screened differentially expressed genes.

Correlation Analysis of Nuclear and Mitochondrial Genes

The correlation of upregulated and downregulated nuclear genes with mitochondrial gene expression was assessed using the rcorr function from the Hmisc R package. Only those correlations demonstrating statistical significance with P < .05 were retained for further analysis.

RESULTS

Corticosterone Levels

The corticosterone level was measured in the serum of restraint and handled control rats. The corticosterone level (pg/mL) was significantly higher in the restraint group compared to the control group (restraint group: 12502.78 ± 2810.41; control group: 5168.04 ± 1074 [data mean ± SEM], P = .019).

Differential Regulation of Nuclear-Encoded Genes in the Mitochondrial Fraction of Hippocampus

The comparison analysis between restraint and control groups identified a total of 3405 transcripts with a P-value < .05, irrespective of the fold change (FC) (Figure 1A; Table S1). Based on FC as cutoff (FC=/>2 for upregulation and FC=/<0.5 for downregulation) and significance of P < .05, the expression of 165 genes was up-regulated (Table S2), and 52 genes was down-regulated (Table S3). Heatmap of the top 50 significantly up and downregulated genes in each sample are shown in Figure 1B and C, respectively. Table 1 lists the gene information and chromosomal locations of the top 10 significant DEGs, significantly upregulated and significantly downregulated genes. Among these genes, we found multiple stress-related hotspot genes, such as NEGR1 (P = 6.81E-11), SCL7A8 (P = 4.61E-09), and GABRA3 (P = 1.16E-08). Out of the 165 upregulated nuclear protein-coding genes, 35 demonstrated a positive correlation with blood corticosterone levels (R² ≥ 0.4, P < .05). Conversely, among the 51 downregulated genes, 22 exhibited a negative correlation with corticosterone levels (R² ≤ -0.4, P < .05, Table S4).

Figure 1.

Differentially expressed nuclear DNA-encoded gene identification and function enrichment analysis. (A) Volcano plot showing the relationship between log₂ fold change (x-axis) and -log₁₀ (P-value) (y-axis), visually displaying DEGs in the hippocampus of restraint rats (n = 7) compared to handled control rats (n = 7). Each dot represents a gene. The vertical dashed line marks the log₂ fold change cutoff, while the horizontal dashed line indicates the significance threshold. Genes with significant differential expression are marked. (B) Heatmap showing top 30 differentially expressed genes based on P-values. Rows correspond to individual genes, and columns represent samples. Gene expression levels are color-coded, with red representing upregulated genes, blue representing downregulated genes, and intermediate colors representing moderate expression levels. (C) Heatmap showing top 30 significantly up-regulated genes; (D) Heatmap showing top 30 significantly down-regulated genes. (E, F) Figures showing functional enrichment analysis of differentially expressed genes based on gene ontology terms and/or KEGG pathways, ranked by importance. The x-axis represents a functional category or pathway, and the y-axis represents -log10 (P-value) or enrichment score. (E) The gene enrichment analysis for overall significant DEGs. (F) The gene enrichment analysis for significantly up-regulated genes. (G) The gene enrichment analysis for significantly down-regulated genes.

Table 1.

Top 10 significant nuclear and mitochondrial encoded differentially expressed genes in the prefrontal cortex of restraint rats.

Group	Gene	Description	Chromosome location	FC	P-value
Nuclear-Encoded Genes
Differentially Expressed Genes (based on P-values)
	Atp2b4	ATPase plasma membrane Ca2+ transporting 4	chr13:47708157-47 807 389	1.33051834	8.03E-14
	Negr1	neuronal growth regulator 1	chr2:248283164-249 015 538	0.61832348	6.81E-11
	Pcdh11x	protocadherin 11 X-linked	chrX:90279191-90 974 671	1.11826241	1.79E-10
	Hnrnpa2b1	heterogeneous nuclear ribonucleoprotein A2/B1	chr4:81867354-81 875 886	−0.274888	2.48E-09
	Slc7a8	solute carrier family 7 member 8	chr15:32153016-32 212 715	0.43375156	4.61E-09
	Gabra3	gamma-aminobutyric acid type A receptor subunit alpha 3	chrX:155301979-155 543 870	1.03325871	1.16E-08
	Brpf1	bromodomain and PHD finger containing 1	chr4:148011977-148 028 431	−0.3619391	1.78E-08
	Sstr1	somatostatin receptor 1	chr6:81567237-81 571 759	1.13109166	4.05E-08
	Timp2	TIMP metallopeptidase inhibitor 2	chr10:104041604-104 089 214	0.70626209	4.58E-08
	Tenm4	teneurin transmembrane protein 4	chr1:157699611-160 674 549	0.69625366	4.74E-08
Upregulated (based on fold-change)
	LOC134483215	uncharacterized LOC134483215	chr19:19774430-19 816 888	3.80902195	.01653588
	Shisal3	shisa like 3	chr1:143705991-143 707 951	3.71481528	.01105096
	Ly6k	lymphocyte antigen 6 family member K	chr7:108572732-108 575 450	3.59828667	.03852228
	Gcnt3	glucosaminyl (N-acetyl) transferase 3 mucin type	chr8:79570081-79 577 389	3.58471664	.00487004
	LOC134483248	disks large homolog 5-like	chr15:22398079-22 401 551	3.55439648	.02720698
	Ly6l	lymphocyte antigen 6 family member L	chr7:109101360-109 103 440	3.21308546	.00674559
	Cckar	cholecystokinin A receptor	chr14:61505270-61 513 618	3.19019579	.00432864
	LOC134482878	vomeronasal type-2 receptor 116-like	chr11:72442310-72 459 556	3.18959227	.02634328
	Nox3	NADPH oxidase 3	chr1:46643277-46 708 109	3.15414368	.00161669
	Gcm1	glial cells missing transcription factor 1	chr8:87805993-87 819 229	3.09088964	.03806875
Downregulated (based on fold-change)
	Tcf21	transcription factor 21	chr1:24520499-24 523 358	−4.1224769	.00745091
	Bhmt2	betaine-homocysteine S-methyltransferase 2	chr2:26630502-26 647 450	−3.7276914	.004757
	LOC120098765	histocompatibility antigen 60b-like	chr1:3249867-3 271 579	−3.604159	.03436852
	Cel	carboxyl ester lipase	chr3:32281518-32 289 019	−3.3471134	.00825236
	Crnn	cornulin	chr2:181427384-181 432 142	−3.2798607	.00750832
	Cnmd	chondromodulin	chr15:61450617-61 475 382	−2.9411404	.01685308
	Nphs2	NPHS2 stomatin family member podocin	chr13:70993622-71 011 585	−2.8997289	.01936452
	Cdh26	cadherin 26	chr3:185898185-185 950 892	−2.8013347	.04086282
	LOC120101052	glyceraldehyde-3-phosphate dehydrogenase-like	chr15:210258693-210 259 085	−2.568329	.03575145
	Tmprss2	transmembrane serine protease 2	chr21:50403707-50 443 224	−2.493707	.0124881
Differentially Expressed Mitochondrial-Encoded Genes
	mt-Rnr2	16S ribosomal RNA	NC_001665.2:1094-2664	0.528030	.000827671
	mt-Rnr1	12S ribosomal RNA	NC_001665.2:68-1025	0.476760	.014465135
	COX2	cytochrome c oxidase subunit II	NC_001665.2:7006-7689	0.49768	.046695506
	Trnq	tRNA-Gln	NC_001665.2:3761-3831	−0.497354	.012409109
	Trnl1	tRNA-Leu	NC_001665.2:2665-2739	−0.347355	.042188036

We performed functional annotation analysis on the total DEGs (3405 genes), up-regulated genes (165 genes), and down-regulated genes (52 genes) and found that the total DEGs mainly mediated different protein binding related pathways (Figure 1E and Table S5), up-regulated genes primarily mediated monoatomic related activity (Figure 1F and Table S6), and down-regulated genes mainly mediated neurotrophin-related biological functions (Figure 1G and Table S7).

Weighted Gene Co-Expression Network Analysis

With the help of balance_clustering, we conducted WGCNA on 3405 DEGs and obtained 15 functional modules, of which three were significantly related to the restraint stress group (Figure 2A). These modules were blue (P = .01, r = -0.69) (Table S8), magenta (P = .001, r = 0.81) (Table S9), and turquoise (P = -.03, r = -0.63) (Table S10). We performed functional annotation analysis on the above modules and found that the blue module mediates the biological functions of ribonucleoproteins (Table S11), the magenta module mediates the biological functions of the synapse and synaptic-related pathways (Table S12), and the turquoise module is related to synaptic signaling and monoatomic ion related pathways (Table S13). At the same time, with the help of network analysis, we identified Tennm4 gene in the turquoise module (Figure 2B), Yju2 gene in the magenta module (Figure 2D), and Il1rapl2 gene (Figure 2D) in the blue module as hub genes.

Figure 2.

Module-trait association analysis in WGCNA. (A) The figures illustrate the correlation between gene co-expression modules and phenotypic traits using WGCNA. Each row represents a different gene module, while each column corresponds to a specific trait. The color-coded heat map indicates the strength and direction of the correlation. The numbers in each cell indicate the correlation coefficient, with the corresponding P-values ​​in parentheses. Three modules, including magenta, blue, and turquoise, showing significant association with the phenotype. (B–D) These figures depict network diagrams of genes, where nodes represent individual genes and edges represent interactions between them. The red dots represent the hub gene. (B) Turquoise module; (C) blue module; (D) Magenta module.

Construction of Mutual Information Network and Expression Analysis

In this study, in addition to the WGCNA mentioned above, we also constructed a MI network with the help of the MINET software package. This network analysis is suitable for real biochemical regulatory networks and has attracted widespread attention. As stated in the method, we used the ARACNE, CLR, and MINET algorithms to construct related networks. The ARACNE algorithm focuses on all node triples, the CLR algorithm focuses on the MI matrix, and the MINET algorithm relies on estimator parameters to infer the MI between all variables. Subsequently, the correlation analysis of different modules with restraint stress groups was carried out as shown in Figure 3A for CLR, Figure 3B for ARACNE, and Figure 3C for MINET. Overall, the co-expression network based on the CLR algorithm had more significant correlation with restraint stress. We also analyzed three modules of CLR, including pink (Table S14), tan (Table S15), and yellow module (Table S16) with higher significance coefficients. From the above, the three key network hub genes obtained from each module were Epb4122 (Figure 3D), Herc2 (Figure 3E), and calm3 genes (Figure 3F), respectively. At the same time, we conducted a Venn diagram analysis on the positive and negative modules obtained from the network analysis of the CLR algorithm and the WGCNA network analysis. We found that 7% of the positive modules were overlapping (Figure 3G), and 12% of the negative modules were overlapping (Figure 3H), indicating the two kinds of algorithms have different biological meanings.

Figure 3.

Construction of MI network and expression analysis. Each row represents a different gene module, while each column corresponds to a specific trait. The color-coded heat map indicates the strength and direction of the correlation. (A) CLR algorithm illustrating the correlation between gene co-expression modules and phenotypic traits. (B) ARACNE algorithm illustrating the correlation between gene co-expression modules and phenotypic traits. (C) MINET algorithm illustrating the correlation between gene co-expression modules and phenotypic traits. (D–F) Network diagrams of genes derived from modules in the CRL algorithm, where nodes represent individual genes and edges represent interactions between them. The hub genes are highlighted in the network. (D) Tan module, (E) Pink module, (F) Yellow module. (G, H) Venn diagram analysis on the negative modules obtained from the network analysis of the CLR algorithm and the WGCNA network analysis. Overlapping circles show how the groups are similar or different.

SNP Variant Analysis of Key Genes

First, we conducted a homology analysis of human and rat genes. We analyzed the homology of 3405 DEGs from the restraint group and found that 3173 DEGs were homologous with human genes (Table S17). We then conducted an association analysis of gene sites on 854 SNPs (Table S18) in the top 20 genes and four hub genes (Table S19) and obtained a total of 39 significantly associated SNP sites (Table S20). We also found that rs10899570 (P = .000959, OR = 1.259), and rs1941379 (P = .001363, OR = 1.262) are located within the TENM4 gene and had the highest significance. The SNP analysis results are detailed in Table 2, and those significant SNPs were mainly located within the genomic region of three genes (TENM4, KSR2, and NEGR1). We then conducted haplotype association analysis of the TENM4 gene because it shows multiple significant SNPs. We found that ss69321287 (rs1573529) and ss69321288 (rs10899570) within TENM4 gene were in linkage disequilibrium (LD) and were distributed into one LD block (Figure 4A, B). This suggests that they are inherited as a unit with little recombination (r = 0.99. We then conducted haplotype association analysis and found that the C-C haplotype within TENM4 gene had a significant difference between depression and control subjects (P = .0053) (Table S21). From the results of single SNP association analysis and haplotype analysis of the TENM4 gene, we believe that the TENM4 gene may be a potential target gene related to depression.

Table 2.

Top significant SNPs associated with restraint stress key genes.

Genes	CHR	SNP	BP	A1	F_A	F_U	A2	CHISQ	P-value	OR
TENM4	11	ss69321288	78 168 966	C	0.1478	0.1211	T	10.9	.000959	1.259
TENM4	11	ss69321298	78 185 438	G	0.134	0.1092	A	10.25	.001363	1.262
NEGR1	1	ss68770086	71 909 719	A	0.06528	0.08421	G	9.194	.002428	0.7595
TENM4	11	ss69321522	78 730 508	G	0.1236	0.102	C	8.314	.003934	1.242
TENM4	11	ss69321510	78 703 267	A	0.3883	0.3561	T	7.909	.004918	1.148
TENM4	11	ss69321380	78 449 282	T	0.4977	0.4655	C	7.383	.006586	1.138
TENM4	11	ss69321426	78 535 086	T	0.1035	0.08508	C	6.927	.008492	1.242
KSR2	12	ss69122366	1.17E+08	T	0.117	0.1375	C	6.718	.009547	0.831
TENM4	11	ss69321393	78 467 501	G	0.3541	0.3251	T	6.658	.009871	1.138
TENM4	11	ss69321394	78 471 750	T	0.4235	0.3942	C	6.256	.01238	1.129
TENM4	11	ss69321285	78 157 859	C	0.1287	0.1095	G	6.246	.01245	1.201
TENM4	11	ss69321514	78 707 601	C	0.4576	0.4282	A	6.161	.01306	1.127
TENM4	11	ss69321509	78 700 069	G	0.1751	0.1533	A	6.001	.0143	1.172
NEGR1	1	ss68770093	72 004 812	G	0.1448	0.1657	A	5.892	.01521	0.8524
TENM4	11	ss69321436	78 548 333	T	0.1129	0.09594	C	5.496	.01906	1.2
KSR2	12	ss69122365	1.17E+08	G	0.2762	0.3011	A	5.345	.02078	0.8857
TENM4	11	ss69321367	78 407 628	T	0.3995	0.4263	C	5.048	.02466	0.895
TENM4	11	ss69321370	78 419 732	C	0.4546	0.4811	T	5.004	.02529	0.8991
TENM4	11	ss69321259	78 061 680	T	0.252	0.2746	C	4.672	.03065	0.89

Figure 4.

Linkage disequilibrium (LD) plots of selected single nucleotide polymorphisms (SNPs) of the Tennm4 gene. (A) LD heatmap displaying the pairwise LD values (D’) SNPs: rs10899567, rs1573529, rs10899570, and rs17309544. The intensity of color represents the degree of LD, with darker shade indicating stronger LD. (B) This LD heatmap displays the pairwise LD values (r²) between four SNPs.

Differential Regulation of Mitochondria-Encoded Genes

We also mapped the sequence reads of the mitochondrial genome. As shown in the volcano plot, a significant dysregulation of mitochondrial genes was observed in the hippocampus of restraint rats (Figure 5A). Specifically, the expression of two tRNA family genes, Trnl1 and Trnq, was downregulated. Furthermore, the COX2 gene, which encodes cytochrome c oxidase subunit II, along with two mitochondrial ribosomal RNAs, Rnr1 and Rnr2, exhibited increased expression levels (Figure 5B). Although not statistically significant, other mitochondrial genes, such as COX3, COX1, ND5, ATP6, ND1, ND2, ND6, ND4L, and ND4, demonstrated a trend towards upregulation, with fold changes ranging from 1.3 to 1.4 (P-value <.08) (Table S22).

Figure 5.

The differential analysis of the mitochondrial DNA-encoded genes. (A) Volcano plot illustrating differential gene expression. The x-axis represents the log2 fold change (Log2FC) of gene expression between experimental conditions, while the y-axis depicts the negative log10-transformed P-values (-log10(P-value). Vertical dashed lines indicate the Log2FC thresholds. (B) Heatmap displaying normalized gene expression across samples (E1-E6: Control, H1-H6: Restraint stress). The color gradient indicates relative expression levels. The hierarchical clustering of genes based on expression profiles. (C) The distribution of pairwise Pearson correlation coefficients for all mitochondria-encoded genes and the upregulated nuclear-encoded genes. (D) Boxplot depicting the distribution of pairwise Pearson correlation coefficients for all mitochondria-encoded genes and the downregulated nuclear-encoded genes.

Correlation Analysis of Nuclear and Mitochondrial Genes

Among the 165 upregulated nuclear-encoded genes identified within the mitochondrial fraction, 115 exhibited a significant positive correlation with all 37 mitochondria-encoded genes. However, only five nuclear-encoded genes correlated with mitochondrial protein-coding genes (Figure 5C, Table S23). The highest number of correlating genes was identified mostly for the tRNA gene family. The upregulated mitochondrial genes COX2, mt-Rnr1, and mt-Rnr2 showed a positive correlation with 4, 9, and six nuclear genes, respectively. Specifically, COX2 correlated with C1qtnf7, Helt, LOC102556287, and Calhm5, and these nuclear genes also correlated with other mitochondrial protein-coding genes. The mt-Rnr1 gene showed correlations with Gimd1, Calhm5, Cckar, LOC120095953, Myo3a, Napsa, Gabrq, Calca, and Aanat whereas mt-Rnr2 was associated with Calhm5, Cckar, Reeld1, C1qtnf7, LOC102556287, and Gimd1. Furthermore, the nuclear gene Reeld1 strongly correlated with the mitochondrial protein-coding genes COX3, ND5, and ND6.

In the downregulated gene set, 16 nuclear genes exhibited negative correlations with 30 mitochondrial genes, comprising both protein-coding and non-coding genes (Figure 5D). The nuclear genes heterogeneous nuclear ribonucleoprotein C-like 1 (Hnrnpcl1) and LOC134483549 demonstrated negative correlations with the mitochondrial protein-coding genes COX2 and ND3. Furthermore, Hnrnpcl1 was found to be negatively correlated with additional coding genes, including ATP6, ATP8, COX3, CYTB, ND1, ND2, ND4, ND4L, and ND5. The non-coding genes, encompassing various tRNA genes and mt-Rnr1, exhibited negative correlations with several nuclear genes, including Asprv1, Il11, Hdc, Plaat5, LOC134484183, Slc25a6, Sytl3, LOC134481801, Dazl, Depdc1b, Cnmd, Kcp, Ndufb4l4, and Rpl19l (Table S23).

DISCUSSION

Mitochondria are the primary energy source for neurons and are essential for various processes vital to brain function.^67,68 Multiple lines of evidence suggest that mitochondrial functions are altered in depressed patients.^24,69,70 Using mitochondrial PCR arrays, we earlier screened differentially expressed genes in the dlPFC of depressed subjects. We identified 16 mitochondrial genes, and several mitochondria-associated signaling pathways were significantly altered in the MDD group compared to nonpsychiatric controls. In the present study, we employed restraint stress as a depression-like behavioral model and examined the expression of mitochondrial genes using RNA sequencing from RNAs isolated from hippocampal mitochondria. We observed differential regulation of a significant number of genes in restrained rats. Our functional analysis revealed that these differentially expressed genes primarily influenced biological pathways highly relevant to depression. Additionally, SNP association and haplotype analysis identified TENM4 as a potential gene associated with depression.

Mitochondrial proteins are key components of mitochondrial functions.⁷¹ About 1500 proteins are present in rat mitochondria,⁷² acting as components of the electron transport chain and metabolic pathways. To date, 1158 mitochondrial proteins have been identified in humans⁷³ and almost all are derived from nuclear sources.^74–76 Only 13 proteins are encoded from the mitochondrial genome. Initially, we focused on nuclear-encoded genes derived from mitochondria and their functional annotations to understand their contribution to stress-related disorders. For transcriptional data analysis, bioinformatics approaches to transcriptomic datasets have generally focused on identifying genome-wide significantly differentially expressed genes. In recent years, other tools have been developed to provide more information about transcriptional regulation and organization in tissues. A popular approach is gene co-expression network analysis. The biological principle of this analysis is that genes cannot function in isolation and may require co-regulated transcriptional control. A standard tool for this type of analysis is WGCNA,⁷⁷ which has earlier been used to study co-expression patterns in normal brain tissue⁷⁸ and psychiatric disorders such as depression⁷⁹ and PTSD.⁸⁰ We have earlier used WGCNA analysis to examine the comorbidity between depression and bipolar disorder.⁸¹ Additionally, MI networks have been successfully applied to transcriptional network inference,^59,82 which generally relies on estimating MI between all pairs of variables. Mutual information networks focus not only on gene transcript data but also on biochemical regulatory networks, that is, gene-to-gene networks. Deep learning techniques have also been applied to transcriptome screening in identifying gene–gene relationships.⁸³ In the present study, we designed a novel process to encode gene expression data of single cell types and then performed deep neural network analysis. This framework allowed us to detect various related biological phenomena, including transcription factor target prediction and identification of disease-related causal and co-expressed genes. As shown in the results, the use of these tools allowed our mitochondrial-derived RNA sequencing data to identify several new target genes and signaling pathways based on nuclear-encoded genes, particularly those related to synaptic signaling and monoatomic ions affected by restraint stress. Moreover, we identified different co-expression modules that may serve as the core networks associated with stress-related disorders such as depression. For example, two key hub genes in the magenta module and the blue module were Yju2 and IL1RAPL2, respectively. Yju2 gene is involved in pre-mRNA splicing and is crucial in synaptic plasticity and neuronal functions,^84,85 key aspects of depression and stress-related diseases. On the other hand, IL1RAPL2 (Interleukin-1 receptor accessory protein-like 2) is a member of the interleukin-1 receptor family, which plays a critical role in inflammatory signaling and neurodevelopment.⁸⁶ While IL1RAPL2 has been primarily associated with synapse formation and neurodevelopmental disorders,^87,88 its potential involvement in stress-related psychiatric disorders like depression has never been explored. Our construction of a MI network and expression analysis (MINet) yielded three different key genes: Epb41l2, Herc2, and Calm3. Epb41l2 plays a role in neuronal cytoskeleton organization and synaptic stability.⁸⁹ It is abnormally expressed in the PFC and hippocampus of individuals with depression. Herc2 is an E3 ubiquitin ligase that regulates protein degradation, synaptic plasticity, and neural development. In depression studies, Herc2 has been implicated in the oxidative stress pathway, which in turn contributes to depressive-like behavior.⁹⁰ Calm3 encodes calmodulin, a calcium-binding protein that regulates multiple neurotransmitter systems, including glutamate, dopamine, and serotonin. A study on chronic corticosterone-induced depression in rodents found that Calm3 expression in the hippocampus was downregulated. Calcium-dependent signaling pathways play a crucial role in brain-derived neurotrophic factor regulation. Interestingly, SSRI and ketamine treatments reversed the changes in Calm3 expression, suggesting its involvement in synaptic remodeling.^91,92 As mentioned earlier, while MINet is well-suited for modeling gene regulatory networks and capturing nonlinear dependencies, WGCNA is more effective for large datasets and provides insights into functional gene clusters and biological pathways. As noted in the results section, the two network analyses exhibited minimal overlap (7%–12%), indicating that they captured distinct co-expression gene network patterns (with distinct hub genes), which could be valuable in psychiatric genomics.

Our study of mitochondria-encoded genes in the hippocampus of restraint stress rats also uncovered insights into how mitochondria adapt to stressful situations. Notably, we observed that the COX2, mt-Rnr1, and mt-Rnr2 genes were upregulated, contrasting with the downregulation of the Trnl1 and Trnq genes. This indicates a nuanced regulatory landscape governing mitochondrial function and cellular resilience. COX2, or cytochrome c oxidase subunit II, plays a crucial role in the mitochondrial electron transport chain, facilitating the production of ATP through aerobic metabolism. Significant increases in COX2 expression were noted after exposure to restraint stress, which suggests an adaptive response aimed at enhancing mitochondrial energy production under stress conditions.⁹³ This finding is supported by previous studies indicating that COX2 is upregulated in response to various stressors, reflecting its role in mitochondrial homeostasis and bioenergetics during adverse conditions.⁹⁴ It is equally noteworthy that mt-Rnr1 and mt-Rnr2 exhibited upregulation, as these genes play a critical role in mitochondrial protein synthesis by encoding mitochondrial ribosomal RNAs. This elevation reinforces the hypothesis that mitochondria are primed to enhance protein production in response to stress, thereby augmenting the translational machinery essential for sustaining mitochondrial function.^95,96 Enhanced expression of these ribosomal components possibly reflects a necessity to restore or maintain mitochondrial homeostasis and support energy production pathways. As mentioned earlier, we observed downregulation of Trnl1 and Trnq. Trnl1 encodes tRNA for leucine, whereas Trnq encodes tRNA for glutamine. Their marked reduction suggests a potential resource-conservation strategy that limits the availability of tRNA for protein synthesis. This downregulation could impair mitochondrial translation efficiency, indicating an operational limit under stress that may compromise mitochondrial functions.

The significant correlations between the dysregulated nuclear-encoded and mitochondrial-encoded genes emphasize the interconnectedness of these two genetic systems in maintaining cellular homeostasis.⁹⁷ In restraint rats, the positive correlation of upregulated mitochondrial genes such as COX2, mt-Rnr1, and mt-Rnr2 with nuclear genes suggests that these nuclear genes may play critical roles in modulating mitochondrial functions. For instance, COX2, a vital component of the electron transport chain, ensures efficient ATP production, while its correlation with nuclear genes like C1qtnf7 and Helt suggests a supportive role in mitochondrial respiration and energy availability. The relevance of C1qtnf7 in this context is further supported by its role in maintaining immune homeostasis, which is crucial during stressful conditions.⁹⁸ The significant correlation observed among the upregulated mitochondrial genes and Calhm5 serves to modulate neuronal excitability through calcium ion channels, thereby establishing a connection between calcium signaling and mitochondrial bioenergetics.⁹⁹ In contrast, the findings from the downregulated geneset highlighted potential vulnerabilities in mitochondrial function during restraint stress. Notably, the Hnrnpcl1 gene exhibits significant negative correlations with several mitochondrial coding genes. Although its role in stress responses is somewhat limited, we can draw valuable insights from the broader family of heterogeneous nuclear ribonucleoproteins, which are known to play a crucial role in various cellular stress mechanisms, including the formation of stress granules.^100–102

In this study, we also employed a cross-species analysis method that uniquely examined the contribution of specific genes and transcriptional programs from human cadaveric brain tissue, thereby reproducing the transcriptional features associated with MDD more accurately.^79,103 We were successfully able to map the differentially expressed genes of rats to the human genome through the similarity analysis of genes between species and identified relevant risk genes that could relate to MDD. Based on this dataset, the LD analysis showed that rs1573529 and rs10899570 within the TENM4 gene were distributed into the haplotype block, a genomic region where SNPs are highly correlated. Also, the haplotypic frequency of rs1573529 and rs10899570 showed a global significance of 0.0053 between the depressed subjects and healthy controls, suggesting that TENM4 is associated with depression. TENM4 is a gene that encodes a transmembrane protein and is expressed predominantly in neurons. It is involved in neural development and establishing appropriate connections within the nervous system by regulating axon guidance and central myelination.^104,105 Recently, Ament et al.¹⁰⁶ sequenced the genomes of 200 individuals from 41 families with bipolar disorder and found that rare variants in neuronal excitability genes affected the risk of bipolar disorder, and TENM4 is a significant gene for bipolar disorder. Another study identified possible pathogenic mutations in the TENM4 gene through target sequencing of TENM4 in 68 SCZ families and demonstrated that aberrant expression of Ten-m leads to lower learning ability, sleep reduction, and increased aggressiveness in animal models.¹⁰⁷ Our findings, based on candidate gene, haplotype analysis, and transcriptomic data analysis, indicate that TENM4 may be a common gene across various psychiatric disorders associated with stress. More studies are needed to confirm this finding.

Overall, we identified significant abnormalities in the transcription of both nuclear and mitochondria-encoded genes within the hippocampal mitochondria of restraint rats. This transcriptional dysregulation may lead to changes in various biological pathways, which could be relevant to stress-related disorders like depression. Furthermore, our cross-specific genetic association study indicates that TENM4 might play a crucial role in the pathophysiology of depression. This study was conducted in male rats. In the future, it will be interesting to see if a similar phenomenon occurs in female rats or if the observed changes are sex-specific.

Author Contributions

Ellie Hulwi (Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Writing—original draft [equal]), Qingzhong Wang (Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Writing—original draft [equal]), Aleena Francis (Data curation [equal], Formal analysis [equal], Investigation [equal], Writing—original draft [equal]), Anuj K Verma (Data curation [equal], Investigation [equal], Methodology [equal], Validation [lead], Writing—original draft [equal]), Yogesh Dwivedi (Conceptualization [lead], Funding acquisition [lead], Investigation [equal], Project administration [equal], Supervision [lead], Writing—original draft [equal], Writing—review & editing [lead]).

Funding

This work was supported by funding from the National Institute of Mental Health (R01MH130539, R01MH124248, R01MH118884, R01MH128994, and R56MH138596) to Dr. Dwivedi.

Conflicts of Interest

The authors report no financial relationships with commercial interests.

Data Availability

All data generated or analyzed during this study are included in this article [and its supplementary information files].