Introduction
Alzheimer's disease (AD) is a progressive neurodegenerative disease and the main cause of dementia, accounting for 60–80% of dementia patients. Dementia refers to the decline of many brain functions, including memory, reasoning and language. The progression of AD can last 15–25 years. AD is mainly characterized by memory loss and cognitive impairment, and the patient's ability to independently carry out daily activities is reduced. The biggest risk factors for AD are advanced age (over 65 years) and carrying at the apolipoprotein E ε4 (APOE ε4) allele [
1,
2]. New biomarkers including PET scans and plasma measurements of amyloid β (Aβ) and phosphorylated tau (p-tau) hold great help for AD diagnosis. In addition to the biochemical amyloid and tau pathology that are core features of AD, microglia responses, the vascular system, blood–brain barrier, the peripheral immune system, glymphatic and other clearance systems, and potentially the gastrointestinal microbiome influence the clinical progression of the disease [
2]. Current treatments for AD include cognitive improvement therapy, treatment of neuropsychiatric symptoms, and disease modification therapy. But some drugs are still being studied and are not very effective. Therefore, new progress in the diagnosis and treatment of AD remains an urgent task.
It is widely believed that neuroinflammation in AD is mediated by microglia and astrocytes [
3]. Substantial evidence has emerged for the involvement of innate and adaptive immune responses in the development or progression of AD [
4,
5]. More recently, it has been found that increased T cell infiltration promotes crosstalk between T cells and microglia, leading to further acceleration of neuroinflammation [
6,
7]. Similarly, peripheral B lymphocytes can enter the central nervous system of AD patients, break through the blood–brain barrier, and promote the activation of immune response by interacting with the stationed brain cells [
8]. Phenotypic changes of circulating neutrophils at different stages of AD affect systemic chronic inflammation and the rate of cognitive decline [
9]. Natural killer cells, peripheral dendritic cells, and mast cells were all associated with an increased risk of AD injury [
10,
11]. These findings highlight the critical role of immune cells in AD.
Mitochondria are indispensable organelles to maintain cell energy metabolism and signal organelles to maintain cell biological functions. Compounds that reduce ROS levels, regulate mitochondrial metabolism, enhance mitochondrial biogenesis may be potential methods to delay aging and treat neurodegenerative diseases by restoring mitochondrial homeostasis [
12,
13]. Auwerx et al. have shown that amyloid-β proteotoxic could be reduced by increasing mitochondrial proteostasis [
13]. In addition, Fang et al. found that impairment of mitophagy induces cognitive deficits by affecting Aβ and Tau accumulation through increased oxidative damage and mitochondrial energy defects [
14]. Meanwhile, mitochondrial miRNAs have also been shown to affect ATP production, oxidative stress, mitochondrial dynamics, and thus regulate AD progression [
15]. It must be mentioned that mitochondria dysfunction plays an important role in exacerbating inflammatory responses in multiple ways [
16]. Dysfunctional mitochondria release mitochondrial components [including mitochondrial DNA (mtDNA)] through a variety of mechanism that induce inflammatory responses through pattern recognition receptors (PRRs). For example, impaired mitophagy contributes to inflammation, and likewise, an increase in mitochondrial ROS and mtDNA stimulates the cyclic GMP-AMP synthase (cGAS)—stimulator of interferon genes (STING) pathway to increase interferon signaling [
17,
18]. In ageing microglia, the reduced glucose flux and mitochondrial respiration lead to maladaptive proinflammatory responses [
19]. Drugs to prevent AD by improving mitochondrial function has been widely studied, such as mitochondrial fission protein inhibitors, drugs that promote fusion, as well as antioxidants including MitoQ, vitamin E and curcumin [
20,
21]. The mitochondrial gatekeeper protein, VDAC1, has been shown to be a promising drug candidate for AD [
22]. However, the crosstalk between mitochondrial genes and the immune microenvironment has been little studied in AD.
Based on the important role of mitochondria in the occurrence and development of AD, we will further understand the role of mitochondria-related genes in regulating mitochondrial function and immune progression, so as to provide certain reference value for the pathogenesis and diagnosis of AD. We explored the potential molecular mechanism by searching GEO database, and analyzed the role of mitochondria-related genes in AD development and their relationship with immune infiltration, which contributed to a better understanding of the immune metabolism in the development of AD.
Materials and methods
Data acquisition
The series matrix files and platform’s annotation files were obtained from the NCBI GEO public database. GSE122063 contained RNA expression data generated by the GPL16699 platform, including 56 AD samples and 44 non-demented (ND) controls from humans. The data were background corrected, normalized between arrays, and log2 transformed [
23]. Samples from three datasets GSE132903, GSE33000 and GSE44770 were selected to validate the hub genes. GSE132903 is annotated by GPL10558 and includes 97 AD samples and 98 ND samples from humans. GSE44770 is annotated by GPL4372, comprising 129 AD samples and 101 ND samples from humans. GSE33000 is annotated by GPL4372 and comprises 310 AD samples and 157 ND samples.
Differential expression genes (DEGs) analysis
Differential expression genes of expression series matrix was identified using the “limma” R package in R software [
24]. And principal component analysis (PCA) was obtained with R package “factoextra” [
25]. Genes with adj.p.val (false discovery rate (FDR)-adjusted) < 0.05 and |log2 (Fold-change)|> 0.58 (Fold change > 1.5) [
26,
27] were identified as DEGs. The “pheatmap” [
28] and “ggplot2” [
29] packages were used to visualize the DEGs results, creating heat maps and volcano plots.
Functional enrichment analysis of gene
Gene Set Enrichment Analysis (GSEA) [
30] was performed using R package “clusterProfiler” [
31] and “GSEABase” [
32]. The Molecular Signatures Database (MSigDB) “c2.all.v2023.1.Hs.entrez” and “c5.all.v2023.1.Hs.entrez” as the reference gene set. The results were visualized with R package “enrichplot” [
33].
There are 1136 mitochondria-related genes in the MitoCarta3.0 database [
34]. MitoDEGs were obtained by crossing the above DEGs with mitochondria‑related genes, which were visualized via Venn diagram [
35]. Then, the top 50 MitoDEGs with the most significant differences were visualized using R packets “pheatmap” [
28].
Identification of hub MitoDEGs
Based on the MitoDEGs obtained from the above analysis, the “glmnet” package [
36] was used to identify hub MitoDEGs by performing the Least Absolute shrinkage and selection operator (LASSO) logistic regression. The specific parameters: family = “binomial”, nfolds = 10, and tenfold cross-validation was used to adjust the optimal value of the parameter λ. The minimum lambda was defined as the optimal value. In this way, more accurate prediction models can be obtained by this method.
Support vector machine (SVM) is a powerful binary classifier which establishes a classification hyperplane as a decision surface. SVM-RFE (recursive feature elimination) was used to optimize the prediction model by reducing the eigenvectors generated by SVM, using the “1071” R package [
37] with the specific parameters: halfve. above = 20 and k = 10. And tenfold cross-validation was used to make the algorithm more accurate. The feature MitoDEGs are obtained by overlapping the results of the LASSO and SVM-RFE.
Analysis of protein–protein interactions
PPI networks were obtained with the STRING database and visualized using Cytoscape 3.9.1 [
38]. The plug-ins CytoHubba and MCODE provided by Cytoscape were used to obtain key MitoDEGs.
Random forests (RF) to screen hub genes
Random forest is an algorithm of recursive partition based on the construction of binary tree. We screened the hub genes with the R package “randomForest” [
39] with the following parameters: ntree = 500, mtry = 3, importance = T, and the Gini index was used as an important measure. Random forest algorithm was used to sort the DEGs according to the decrease in Gini index, and the top six genes with significant values greater than 3 were selected for downstream analysis.
Immune infiltration analysis
To better identify the characteristics of immune cells in the brain tissue of normal population and AD patients, we conducted a single-sample gene-set enrichment analysis (ssGSEA) algorithm to estimate the differential composition of infiltration abundance of 28 immune cell types between the two groups [
40] based on gene expression profiles from microarray. The correlation between the expression of hub MitoDEGs and the distribution of immune cells was revealed by spearman correlation analysis.
Logistic regression model
In order to establish a diagnostic model for AD classification, logistic regression algorithm was used, and the variance inflation factor (VIF) values for multicollinearity check were all less than 5 and shown in Additional file
1: Table S1. The area under the ROC curve (AUC) was used to evaluate the accuracy of five hub MitoDEGs and logistic regression model. Therefore, we calculated the AUC values of the five hub MitoDEGs to evaluate the accuracy of the diagnostic model using the R package “ROCR” [
41]. A nomogram was established to predict the risk of AD based on the feature genes using the “rms” R package [
42]. The prediction efficacy of the nomogram was estimated using calibration curves.
Validation of a diagnostic model
The AUC values of five hub MitoDEGs as diagnostic models were calculated on dataset GSE132903, GSE44770 and GSE33000, respectively, which verified the effectiveness of the diagnostic model.
Animals
Male WT and APP/PS1 mutant mice aged 10 months old [
43] were obtained from the Chinese Academy of Military Sciences (Beijing, China) and had free access to food and water with a comfortable environment. All protocols were approved by the Animal Care and Use Committee of Tianjin Medical University and were performed according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals.
PC12 neuron culture and handing
PC12 neuron cells were obtained from infrastructure cell line resources in China and cultured in high glucose Dulbecco’s modified Eagle’s medium with 5% fetal bovine serum, and appropriate amounts of penicillin and streptomycin at 37 ℃ and 5% CO
2. We induced PC12 cells with Aβ1-42 for 12 h, and the final concentration of Aβ1-42 was 7 μM [
44].
Immunofluorescence staining of Aβ1-42 in PC12 cells
PC12 cells were exposed to HiLyte Fluor™488-labeled Aβ1-42 [
45] (Table
1) for 12 h. Subsequently, the cells were rinsed twice with PBS and fixed with a 4% paraformaldehyde solution (PFA) for 20 min. Following three washes with PBS, the cells were stained with DAPI (Abcam, UK). Fluorescence images were acquired using a fluorescence microscope (CarlZeiss, Oberkochen, Germany).
The plasmid transfection
Optic atrophy 1 (OPA1) plasmids and the control empty plasmid vector (GenePharma, Shanghai, China) were transfected into cells using jetPRIME® (Polyplus-transfection S.A, Illkirch, France) according to the manufacturer’s instructions. 2.0 µg DNA plasmid was mixed with 200 µL jetPRIME® buffer, and 4 µL jetPRIME® reagent was added to the above solution, vortex for 1 s, spin down briefly, and incubated for 15 min at room temperature. Then, the transfection mix was added to the 6-well plates with the serum-containing medium and incubated in an incubator for 24-48 h for further experiments.
RNA extraction and quantitative real-time polymerase chain reaction (qRT-PCR)
Total RNA was extractazed from cultured PC12 cells or the cortex in AD mice using the TransZol Up Plus RNA Kit (TransGEN, Beijing, China) following the manufacturer’s instructions. RNA concentration and quality was measured by Nanodrop Spectrophotometer (Thermo Scientific, Waltham, MA, USA).
Reverse transcription and RT-PCR (mRNA) were performed with corresponding primers (Table
2) using the TransScript
® One-Step gDNA Removal and cDNA Synthesis SuperMix (AT311, TransGEN, Beijing, China) and PerfectStart
® Green qPCR SuperMix (AQ601, TransGEN, Beijing, China), respectively. The relative value of mRNA transcription was calculated using the 2
−∆∆CT formula, and U6/GAPDH was used as the internal control for normalization.
Detection of mitochondrial reactive oxygen species (mtROS)
To label the mtROS, 1 ml working solution (5 µM) of MitoSOXTM Red (M36008, Invitrogen, Carlsbad, CA, USA) was added into PC12 cells cultured in a 6-well plate and incubated for 10 min at 37 °C in the dark. The cells were washed three times with PBS. To stain the nuclei, a 1 × working solution of Hoechst 33342 (C1027, Beyotime, China) was prepared by mixing 10 μl of the stain with 1 ml of DMEM medium. The working solution was added to the PC12 cells and the cells were incubated in the incubator at 37 °C for 10 min. Then, the cells were washed three times with warm PBS. Images were photographed using a a fluorescence microscope (CarlZeiss, Oberkochen, Germany).
Detection of mitochondrial membrane potential
Mitochondrial membrane potential (MMP, ∆Ψm) was detected using a JC-1 assay kit (C2003S, Beyotime, China). After the treatment, cultured PC12 cells were washed with PBS. 5 µl JC-1 reagent was mixed with 1 ml JC-1 staining buffer and added to PC12 cells cultured in 6-well plates containing 1 ml medium. The cells were then incubated at 37 ℃ for 20 min. PC12 cells were washed with JC-1 buffer for two times, and fresh medium was added for detection. Images were captured using a fluorescence microscope (CarlZeiss, Oberkochen, Germany) and analyzed using the ImageJ software. When the mt∆Ψ is high in cells, JC-1 aggregates in the mitochondria with a red fluorescence. In cells with low mt∆Ψ, JC-1 is unable to accumulate in mitochondria and remains in the cytoplasm as a monomer, which shows green fluorescence. The value of mt∆Ψ are represented by the ratio of red/green fluorescence intensity.
Immunofluorescence (IF)
The mice were killed by transcardial perfusion of cold phosphate-buffered saline (PBS) and 4% PFA. Subsequently, the brain tissue was removed completely and fixed with 4% PFA overnight. After gradient dehydration with sucrose, the brain was embedded within the optimal cutting temperature (Sakura, Torrance, CA, USA). The brain samples were cut into slices of appropriate thickness using a − 20 °C frozen slicer for IF staining.
The brain tissue sections were placed at room temperature for 15 min from the – 20 ℃ refrigerator, washed with PBS for 3 times, then treated with 0.3%Triton for 30 min at room temperature, and then incubated with 3% BSA for 60 min. They were then incubated with primary antibodies (Table
1) overnight at 4 ℃. On the second day, the sections were incubated with secondary antibody for one hour after washing with PBS, and then DAPI (Abcam, UK) was used to stain the nucleus. Images were captured using a fluorescence microscope (CarlZeiss, Oberkochen, Germany) and analyzed using the ImageJ software.
Immunoblotting for protein evaluation
Western blotting for OPA1, caspase3, β-actin was performed, as a previous description [
46,
47] (Table
1). The band gray values were measured with ImageJ (National Institutes of health, Bethesda, MD, USA).
Statistical analysis
R software (version 4.2.2) was used for all bioinformatics statistical analysis and visualization. Values in experiment were expressed as mean ± standard deviation (SD). The t-test was selected to measure the data of the two groups, and one-way ANOVA followed by LSD post hoc test or Tukey’s post hoc test was used for comparisons of multiple groups. Shapiro–Wilk test was used to check the normality of the data. All statistical analyses of experiments were performed by SPSS 26 or GraphPad Prism 9. All experiments were independently repeated at least three times. Significance was defined as P < 0.05.
Discussion
There are two forms of Alzheimer's disease: early-onset (familial) and late-onset (sporadic). Familial AD typically occurs between the ages of 30 and 50, and accounts for 1–2% of all AD cases. Familial AD is caused by mutations in the Aβ precursor protein, presenilin-2and presenilin-1 genes, resulting in overproduction of Aβ plaques [
51,
52]. Late-onset AD involves a variety of factors including lifestyle, traumatic brain injury, obesity, hypertension, diabetes, depression, and epigenetic factors [
53]. Similarly, APOE4 genotype and age are important risk factors for AD [
1,
54]. It has been pointed out that the presence of the toxic proteins is thought to activate microglia to remove excess proteins and dead cells [
55]. When microglia cannot maintain their balance, chronic inflammation occurs in the brain. Therefore, inflammation is largely a biomarker of AD.
Given the close relationship between the immune microenvironment status and the occurrence and progression of AD [
5,
10], as well as the alterations of mitochondrial metabolism and mitoophagy processes in immune cells [
17,
18,
22,
56], it is crucial to identify potential targets of mitochondrial genes to guide the treatment of AD. In our study, we used bioinformatics methods to comprehensively investigate the DEGs of AD and ND from the GEO database and found that the immune response and brain aging were enriched in AD patients. Moreover, neuron to neuron synapse and the transmission across chemical synapses were inhibited in AD patients. Further analysis revealed that IL-1β, IL-18 inflammatory pathways, oxidative damage were activated, and mitochondrial OXPHOS and fatty acid oxidation were decreased in AD samples.
Peripheral immune disorders and peripheral-central immune crosstalk have been investigated for their important roles in the pathogenesis and progression of AD [
8]. Several researchers have demonstrated cytotoxic effects of CD8
+ efferent memory CD45RA
+ (T
EMRA) cells, Th1 and Th17 in CD4 T cells in AD pathology, which is consistent with our study [
57,
58]. Natural killer (NK) T cells have the ability to rapidly induce cell apoptosis through cytotoxic granules and release of inflammatory factors such as TNF-α, INF-γ. When stimulated, NK cells can activate other immune cells, triggering an immune cascade reaction. Furthermore, alterations in NK cell subsets are closely associated with the progression of Alzheimer's disease (AD) [
11,
59]. Similarly, inhibition of neutrophils also reduces systemic chronic inflammation and cognitive decline at various stages of AD progression [
9,
60].
In this study, we found that compared with the ND group, the AD group had significant changes in the proportion of immune cell infiltration, including macrophage, regulatory T cell, activated CD8 T cell, memory B cell, activated dendritic cell, activated CD4 T cell, natural killer T cell, type 17 T helper cell, Neutrophil, MDSC. CD4 T cells play an important role in neurodegenerative diseases, such as Parkinson's disease (PD) [
61], AD [
62], stroke [
63], and Lewy body dementia [
64]. CD4 T cells mainly influence the function of mature microglia to neuronal synapses [
65]. Recently, researchers have found that there are many more T cells, especially cytotoxic T cells, in the mice of tauopathy than in the mice with amyloid deposition or control mice. T cells are most abundant in areas with the most severe tau pathology and the highest concentration of microglia. Activated microglia release molecular compounds that activate and draw T cells from the blood into the brain, and T cells release compounds that push the microglia toward a more pro-inflammatory phenotype. Together, these two types of immune cells create an inflammatory environment that is primed for neuronal damage [
66]. In AD, Aβ deposition and p-tau accumulation induce parenchymal innate immune activation, affecting the integrity of the blood–brain barrier (BBB), CSF/ISF flow and lymphatic drainage, further leading to the antigen-presenting microglia, expansion of IFN-reactive, the increase of inflammatory cytokines and antigen accumulation, as well as parenchymal T-cell infiltration, T cell receptor (TCR) clonal expansion in the brain parenchyma and border areas [
4]. In the innate immune response, TLRs recognize molecular patterns associated with microbial pathogens and damage, promoting NF-kB signaling and the activation of inflammation [
4,
67]. The increase in TLRs signaling pathway in AD samples was also found in our study.
Subsequently, we discovered the strongly relationship between hub mitoDEGs and immune cells, as well as immune microenvironment and AD pathology, and finally identified five hub MitoDEGs (BDH1, TRAP1, OPA1, DLD, SPG7) through in-depth analysis of three machine learning algorithms, which was of great significance for the search for pathological biomarkers and promising therapeutic intervention targets for AD. Although there have been some studies on the immune microenvironment related to AD [
10], and the gene sets (CXCR4, PPP3R1, HSP90AB1, CXCL10, and S100A12) that are responsible for immune filtration have been indicated. Inflammatory gene markers activated in immune cells could induce harmful neuroinflammatory programs and contributed to neurodegenerative environments [
68]. However, we investigated mitochondrial genes associated with the immune microenvironment. These hub MitoDEGs (OPA1) regulated mitochondrial ROS, MMP, and neuronal apoptosis. In the same way, epigenetic silencing may also promote neuron survival by eliminating the potential mediator of neuron death [
69]. In short, under different physiological and pathological conditions, unique patterns of altered gene expression may reduce or promote disease progression by affecting immune cells, epigenetic inheritance, or simultaneously activating multiple synergistic regulatory mechanisms.
One of the important pathogeneses of AD is the disorder of mitochondrial metabolism, which can control synaptic transmission and information exchange. Mitochondrial metabolism has an enormous impact on the fate and function of immune cells. Microglial lipid metabolism is involved in microglial activation, phenotypic transitions and functions, such as phagocytosis, inflammatory signaling, and migration [
70]. In addition, the interaction of Aβ and P-tau with dynamin-related protein 1 (Drp1) is the key factors in mitochondrial fragmentation, damage of mitochondria and synapsis, eventually possibly resulting in neuronal damage and cognitive deficit [
71,
72]. OPA1 is an important intima fusion protein regulated by two membrane proteases, OMA1 and YME1L1. Studies have pointed out that inhibition of OPA1 enhances the progression of neurodegenerative diseases, and OPA-1 regulates apoptosis resistance of OXPHOS IL-17-producing CD4 T cells by regulating mitochondrial fusion and limiting mitophagy [
73]. In conclusion, the OPA1 may part acts on immune cells to maintain mitochondrial homeostasis in patients with AD. Overexpression of OPA1 is associated with adipocyte browning, which is beneficial to improve glucose tolerance and insulin sensitivity [
48]. However, in the AD model, OPA1 may not affect cognitive function by altering tau's phosphorylation [
74]. Our study demonstrated that up-regulation of OPA1 restored mitochondrial membrane potential and reduced neuronal apoptosis, which was consistent with previous studies [
50], suggesting that OPA1 may alter AD disease progression by affecting neuronal apoptosis rather than tau's phosphorylation.β-hydroxybutyrate dehydrogenase (BDH1), as a main rate-limiting enzyme of ketone metabolism, controls the transformation between acetoacetic acid (AcAc) and β-hydroxybutyrate (BHB). BHB acts as an alternative carbon source for maintaining the redox balance, including the production of amino acids, OXPHOS, and glutathione. The ability of BHB to enhance CD4 + T cell metabolism and promote T cell responses depends on BDH1 [
75]. In addition, it was found that BDH1 is significantly down-regulated in glioblastoma [
76] and has an important role in metabolic regulation in the liver [
77]. Tumor necrosis factor receptor-associated protein 1 (TRAP1), a member of the chaperone family of heat shock protein 90 (HSP90), is predominantly present in mitochondria. TRAP1 is a regulator factor of oxidative stress-induced cell death, redox homeostasis and unfolded protein response [
78]. Giffard et al. demonstrated that TRAP1 overexpression reduced ROS production, maintained mitochondrial membrane potential and increased preservation of ATP levels in oxygen-glucose deprived neurons and astrocytes [
79,
80]. Moreover, TRAP1 may be a causative gene in PD, which needs to be further confirmed [
81,
82]. Therefore, it is very meaningful to find this target in AD and explore it in depth. Spastic paraplegia type 7 (SPG7) mutations are a common cause of hereditary spastic paraplegia, resulting in mitochondrial dysfunction, including decreased mitochondrial membrane potential, reduced OXPHOS, decreased ATP, and increased mitochondrial stress [
83]. The lack of paraplegin, a protein encoded by SPG7, impairs the opening of mitochondrial permeability transition pore (mPTP) by increasing the expression and activity of sirtuin3, thereby increasing the concentration of Ca
2+ and reactive oxygen species in the matrix and destroying mitochondrial homeostasis, leading to the disorders of synaptic transmission disorders [
84]. In the present study, we found a downward trend in SPG7, but there was no statistical difference, which may be caused by different species of samples or the heterogeneity of samples. Dihydrolipoamide dehydrogenase (DLD), as a cuproptosis-related gene [
85], is associated with steatosis and plays an important role in nonalcoholic fatty liver disease [
86]. DLD may serve as a therapeutic target for energy metabolism in AD [
87]. Together, the research of these genes reinforces the importance of mitochondria in AD. The results of this study may contribute to a better understanding of whether the interaction between mitochondrial gene signatures and immune cells influences AD progression.
This study is the first to identify the relationship between mitochondrial related genes and immune microenvironment in AD through bioinformatics analysis. BDH1, TRAP1, OPA1, DLD and SPG7 have been tested well as potential molecular targets for the diagnosis and prediction of AD risk. The expression levels of the five hub MitoDEGs have also been verified in cell and animal experiments. Of course, our research also has some limitations: 1. The brain region of the AD model we selected for analysis is the frontal and temporal cortex, which has not been verified in multiple other brain regions; 2. The specific roles of mitochondria-related genes in immune cells and their involvement in AD require further high-quality animal experimental verification. The possible signaling pathways affected by these hub genes and their functions in immune cells are our next research directions.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.