Bioinformatics and machine learning-based screening of key genes in Alzheimer's disease.

doi:10.21203/rs.3.rs-3972836/v1

Download PDF

Research Article

Bioinformatics and machine learning-based screening of key genes in Alzheimer's disease.

https://doi.org/10.21203/rs.3.rs-3972836/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Objective

To analyze differentially expressed genes (DEGs) related to mitophagy between Alzheimer's disease (AD) patients and normal controls using bioinformatics and machine learning methods and to screen for key genes to provide theoretical support for the study of AD pathogenesis and therapeutic targets.

Methods

Data microarrays related to AD were obtained from the GEO database, and DEGs were obtained by differential expression analysis of gene expression data using R. The DEGs were intersected with mitophagy-related genes to obtain the genes of interest, and then the group of genes was enriched for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) signaling pathways. A machine-learning model was constructed at the same time to screen key genes, construct risk prediction models and predict transcription factors based on key genes. In addition, consistent clustering analysis was performed on AD samples, followed by immunization and pathway analysis for each subtype.

Results

Seven key genes were finally screened from 38 mitophagy-related DEGs, and the risk prediction model constructed on the basis of these 7 genes had an AUC of 0.877. Cluster analysis classified the AD samples into two subtypes, and the immune infiltration between the two subtypes was also significantly different.

Conclusion

This study screened the key genes of AD based on bioinformatics, which provides a new perspective and potential therapeutic targets for exploring the potential mechanism of mitophagy affecting AD as well as a new idea and direction for individualized treatment of AD.

Alzheimer's disease

differential expression analysis

mitochondrial autophagy

key genes

Alzheimer's disease (AD) is a neurological disorder characterized by cerebral cortical atrophy, beta-amyloid deposition, neuronal fibrillary tangles, and the formation of senile plaques and is mainly characterized by progressive cognitive impairment with a variety of psychiatric and behavioral symptoms. As the disease progresses, patients gradually lose the ability to take care of themselves and become completely dependent on others for daily life in the later stages, which places an enormous burden on the patients themselves, their families and society. The prevalence of AD is increasing year by year. According to the survey, by 2020, the prevalence of AD patients among the population aged 60 or above in China will reach 3.94%, with a total of 9.83 million AD patients, accounting for 65.23% of all dementia patients [1, 2]. According to another survey, the annual cost of AD patients in China is $167.74 billion and is expected to reach $1.8 trillion by 2050 [3]. In recent years, although some progress has been made in the diagnosis of AD, due to the insidious onset of AD and the slow progression of the disease course, there is still a large number of missed diagnoses and misdiagnoses in the clinical setting [4]. Moreover, the pathogenesis of AD is complex, and the development of new drugs to address the causes of the disease has been slow, with a lack of effective drugs [5]. Finding simple and effective methods of prevention, diagnosis and treatment remains the focus and challenge of current AD research [6]. Currently, AD has become a challenging public health issue in the 21st century. In-depth research on the pathogenesis of AD and the discovery of new therapeutic targets may effectively alleviate the disease burden of patients, which is of great significance in promoting the development of healthy aging in the whole society.

In recent years, with the development of bioinformatics and machine learning technologies, an increasing number of researchers have utilized these methods to screen and analyze AD-related genes, proteins, metabolites, etc., to find molecular markers and potential therapeutic targets for AD. However, there are some limitations and shortcomings in the current studies on screening key AD genes by machine learning models. On the one hand, most studies have only considered a single or a few biological processes or pathways, ignoring the complex molecular networks and interactions in AD. On the other hand, most studies have focused only on the most common neuronal damage and synaptic dysfunction in AD, ignoring other cell types and mechanisms that may be related to the development of AD, such as mitophagy. Mitochondria are organelles found in most cells and are involved in biological processes such as cellular respiration and the production of reactive oxygen species. As mitochondria continue to age and become functionally impaired, cells also age or become diseased. Therefore, when mitochondria age or become functionally impaired, autophagy needs to be initiated to enable mitochondrial renewal. mitophagy, as an important pathway to maintain mitochondrial homeostasis and the stability of the intracellular environment, plays a key role in the pathogenesis of a large number of diseases [7]. Of these, the role in neurons is particularly important because neurons are very sensitive to energy metabolism and oxidative stress, and there is evidence that mitophagy is inhibited or impaired in AD, leading to mitochondrial dysfunction and increased oxidative stress, which in turn accelerates degenerative neuronal changes [8]. Another study found that there are abnormalities in mitochondrial structure and function in the brain tissue of AD patients, which are closely related to the impairment of the mitophagy process [9]. Therefore, exploring the role of mitophagy-related genes in the pathogenesis of AD and their potential mechanisms as well as screening key genes not only helps to further elucidate the pathogenesis of AD but also provides new ideas for the diagnosis and targeted therapy of AD [10].

The aim of this study was to screen key genes related to mitophagy in AD and to explore their roles and mechanisms in AD by combining bioinformatics methods and machine learning approaches. The study first performed gene differential expression analysis using publicly available datasets, then analyzed the significance of the differentially expressed genes by machine learning, and finally screened seven genes related to mitophagy function: FXR1, TUBB4B, MAP4K4, IDH3G, TUBB, RTN3, and ATP5F1B, which were expressed in the AD group and the normal control group with significant differences, which may be involved in the developmental process of AD and provide potential targets for the diagnosis and treatment of AD. In addition, in this study, AD samples were divided into two subtypes by clustering analysis, and then immunoassay and GSEA analyses were performed, which showed that the immune cell infiltration of the two subtypes was also significantly different. These different analyses provided important clues for further research on the functional status of different subtypes of cells and their regulatory mechanisms. Finally, the study constructed a risk prediction model based on seven key genes with a model AUC value of 0.877. The technology roadmap is shown in Fig. 1.

2.1 Data collection and preprocessing

The GEO database (https:// www.ncbi.nlm.nih.gov/geo/) is a public repository created and maintained by the National Center for Biotechnology Information (NCBI) in 2020 for storing and sharing gene expression data and is the most comprehensive public gene expression database available today [11]. The expression matrix and platform information of AD-related datasets GSE5281 [12], GSE9770 [13] and GSE28146 [14] were obtained from the GEO database using the "GEOquery" package in RStudio 4.3.1 (where all subsequent analyses were performed), and the expression data were extracted for normalization and normalization correction. Then, ID transformed using the "hgu133plus2.db" package was used to convert the IDs of the genes, and finally, the three sets of data were merged to remove the batch effect.

GeneCards (https://www.genecards.org/) is an online database that integrates human genetic information from multiple authoritative data sources. From this database, we searched for “mitophagy” and screened and downloaded genes with scores > 1.

2.2 Acquisition of DEGs and intersecting genes

Gene expression values were statistically analyzed using the "limma" package of RStudio to screen out differentially expressed genes. P. Values < 0.05 and |log2FC| > 1 were used as the inclusion criteria, where log2FC > 1 was upregulated DEGs and log2FC < -1 was downregulated DEGs. DEGs were intersected with mitophagy-related genes with a score > 1 to obtain intersecting genes of interest. Volcano plots and heatmaps of DEGs were plotted, and Venn diagrams and heatmaps of intersecting genes were plotted.

2.3 GO and KEGG enrichment analysis

The DEGs were analyzed for GO and KEGG pathway enrichment using the "clusterProfiler" package of R. A P value < 0.05 was considered significant, and the top 15 significance levels were plotted in bubble and bar graphs. The intersecting genes were then analyzed for GO enrichment with a P value < 0.05 indicating significance, and a chord diagram was drawn for showcase.

2.4 Screening for key genes

To screen key genes, importance scoring of intersecting genes was performed using Lasso regression, XGBoost, Boruta, and RF. Based on Lasso's feature selection and regression modeling, the intersecting gene expression data and classification labels were processed and modeled, and the results associated with the model were saved and visualized. The XGBoost model was constructed based on the intersecting genes, the maximum depth was set to 6, the learning rate was 0.5, 25 rounds of training were performed, and the importance score was calculated. Visualize the feature importance scores using the functions in the "xgboost" package. The Boruta algorithm was used to select features for intersecting genes, set the maximum number of iteration rounds to 500, and eliminate genes that were irrelevant or uncertain to the classification results to reduce the feature dimensions and improve the model's interpretive and predictive performance. A random forest model was created, the number of decision trees was set to 1000, the importance score was calculated, the 15 most important genes were selected, and the training error plot and feature importance score plot of the random forest model were drawn. Finally, the four results were intersected to filter out the key genes and plot the Venn diagram.

2.5 Consensus clustering

Consensus clustering analysis was conducted on 143 AD samples using 'ConsensusClusterPlus.' Gene expression data were clustered to generate a consensus clustering plot and the corresponding cluster results. To observe the similarity and distinctiveness of subtypes among the samples, principal component analysis (PCA) was applied to the clustered gene expression data, reducing high-dimensional data to a two-dimensional plane for visualization, with coloring based on the clustering results. The optimal number of clusters was determined by calculating the fuzzy clustering coefficient. Box plots of key gene expression data were created using the 'ggplot2' and 'ggpubr' libraries. The box plots were colored based on the clustering results, and statistical significance labels were added to display differences between groups.

2.6 Immune analysis

Gene expression data from 143 AD samples were scored for infiltration of immune cell types using the ssGSEA method, and a heatmap of immune score correlations was plotted using the "corrplot". Boxplots of the differences in the scores of each immune cell between clusters of different samples were also plotted, and significance tests were performed to compare the infiltration of each immune cell between different clusters. The gene set of the IL-6/IL-7 pathway was extracted from the MSigDB database, and the gene expression data of 143 AD samples were scored for the activity of the IL-6/IL-7 pathway using ssGSEA and integrated with the sample clustering information. A violin plot was drawn to compare the differences in the pathway scores among different clusters.

2.7 GSEA

Based on the results of sample clustering and typing, differential expression analysis was performed using the "limma" package to find genes with significant differences in expression between subtypes. Based on the list of differential genes, enrichment analysis of KEGG pathways and GO functions was performed using the GSEA algorithm. The key pathways with differences between subtypes were highlighted by visualizing the enrichment curves.

2.8 Constructing risk prediction models

Based on the key genes, a logistic regression model was fitted using the lrm() function to predict AD risk, and the predictive performance, calibration performance, and utility of the model at different prediction thresholds were evaluated by plotting the nomogram, ROC curve, and calibration curve.

2.9 Evaluation of key genes

The ROC curve was plotted based on the expression data of the key genes using the roc() function, and the AUC was calculated to assess the effectiveness of key genes as classifiers for predicting AD, while violin plots were drawn to compare the differences in the distribution of expression of the key genes between the control and AD groups.

2.10 Prediction of transcription factors

Using the NetworkAnalyst (https://www.networkanalyst.ca/) online database, key genes were uploaded, the JASPAR database of transcription factors was selected for analysis, and finally, the data were imported into Cytoscape software to draw a network map.

3.1 Acquisition of DEGs and intersecting genes

The three gene chips contained a total of 225 samples, including 82 in the normal control group and 143 in the AD group. These samples were combined after removing the batch effect (Fig. 2A, B) and analyzed and screened to obtain 254 DEGs (Log2FC > 1, P.value < 0.05), which contained 99 upregulated genes and 155 downregulated genes (Fig. 2C), and the expression of these genes was significantly different between the two groups (Fig. 2D). Intersecting these 254 DEGs with mitochondrial autophagy-related genes with score > 1 yielded 38 intersecting genes of interest (Fig. 3A), and the expression of the intersecting genes also differed significantly between the two groups (Fig. 3B).

3.2 GO and KEGG enrichment analysis

The bar graph of GO enrichment analysis of the DEGs (Fig. 2E) showed that the DEGs were highly enriched in functions such as vesicular transport, membrane and proton pump/ATPase. The KEGG pathway enrichment analysis bubble plot (Fig. 2F) showed that the most significant enrichment was found for pathways related to neurodegenerative diseases, such as Alzheimer's disease, synaptic transmission, Parkinson's disease, and Huntington's disease, which indicated that the DEGs were highly correlated with these diseases. GO analysis of the intersecting genes (Fig. 3C) showed that the intersecting genes were enriched in the mitochondrial respiratory chain complex and the inner mitochondrial membrane, suggesting that the intersecting genes are related to mitochondrial function and cellular metabolism.

3.3 Screening for key genes

After scoring the importance of 38 genes using each of the four methods, LASSO regression analysis screened out 13 important genes (Fig. 4A, B), the top 15 highest-scoring genes were selected as important genes in the random forest model (Fig. 4C, D), Boruta screened out 31 important genes (Fig. 4E, F), and XGBoost screened 37 important genes (Fig. 4G), intersected the results of the four methods (Fig. 4H), and took the overlapping seven genes as key genes: FXR1, TUBB4B, MAP4K4, IDH3G, TUBB, RTN3, and ATP5F1B.

3.4 Consensus clustering

The consensus cluster analysis results (Fig. 5A-E) showed that the consensus clustering matrix was most clearly differentiated and had the best cluster stability when k = 2 (Fig. 5F). Principal component analysis scatter plots (Fig. 5G) showed significant differences between the 2 subtypes. Therefore, the 143 AD samples were categorized in this study into two subtypes, C1 (n = 72) and C2 (n = 71). The expression of seven key genes was also significantly different between the two subtypes (Fig. 5H).

3.5 Immune analysis

The correlation analysis heatmap of different immune cell types in the AD group (Fig. 6A) showed that there were complex patterns of correlation between immune cells in AD patients; for example, natural killer cells were negatively correlated with effector-memory CD4 T cells but positively correlated with effector-memory CD8 T cells, and these patterns may be related to their coordinated behaviors and functions in the pathogenesis of AD. There were also significant differences in some immune cell infiltrations between the two subtypes (Fig. 6B), such as effector memory CD4 T cells and natural killer T cells. The PID_IL6_7_PATHWAY gene set also differed significantly between Cluster 1 and Cluster 2 (Fig. 6C), with Cluster 1 having an overall higher level of IL-6 pathway gene expression than Cluster 2.

3.6 GSEA

The results of GSEA on the samples of each subtype showed that KEGG_AUTOIMMUNE_THYROID_DISEASE, KEGG_HEMATOPOIETIC_CELL_LINEAGE and KEGG_PRIMARY_IMMUNODEFICIENCY were significantly enriched in Cluster 2 (Fig. 6D), suggesting a more pronounced immune activation profile of Cluster 2. In contrast, GO_MITOCHONDRIAL_GENE_EXPRESSION, GO_MITOCHONDRIAL_TRANSLATION, and GO_MITOCHONDRIAL_TRANSLATIONAL_TERMINATION were significantly enriched in Cluster 1 (Fig. 6E), suggesting a higher relevance of Cluster 1 to mitochondrial oxidative phosphorylation and translation.

3.7 Constructing risk prediction models

An AD risk prediction model was established based on the key genes FXR1, TUBB4B, MAP4K4, IDH3G, TUBB, RTN3, and ATP5F1B (Fig. 7A), which had an area under the receiver operating characteristic curve (AUC) of 0.877 (Fig. 7B) and a high degree of overlap of the calibration curve with the ideal curve (Fig. 7C), suggesting that the calibrated nomogram better predicts AD risk. The decision curve analysis (Fig. 7D) showed that when the prediction threshold probability was set at 0.2–0.8, more net benefit could be obtained by using this AD nomogram model for prediction compared with no prediction or using a fixed threshold. This means that for AD patients whose risk of morbidity is intermediate, applying this model for disease prediction can improve the accuracy of judgment and help in clinical decision-making.

3.8 Evaluation of key genes

The comparative plots of the expression distribution of key genes (Fig. 8A-G) showed that the expression of the key genes FXR1, TUBB4B, MAP4K4, IDH3G, TUBB, RTN3, and ATP5F1B differed significantly between the AD group and the normal control group, with upregulation of MAP4K4 and FXR1 expression and downregulation of TUBB4B, IDH3G, TUBB, RTN3, and ATP5F1B expression. The ROC (Fig. 8H) showed that the AUCs of the key genes were as follows: ATP5F1B: 0.854, FXR1: 0.756, TUBB4B: 0.808, MAP4K4: 0.788, IDH3G: 0.820, TUBB: 0.801, and RTN3: 0.821. These results suggested that these seven key genes might have important diagnostic value. The AUC of the nomogram based on the 7 key genes = 0.877 (Fig. 7B), which is larger than the AUC of the individual key genes, suggests that when these key genes are considered together, the diagnostic value is greater and is more favorable for the prediction of clinical AD.

3.9 Prediction of transcription factors

Based on the key genes, this study identified a series of potential transcription factors totaling 36 (Fig. 8I), including NR3C1, TP53, and NFKB1. These transcription factors may regulate each other with key genes and participate in critical biological processes. The prediction provides a promising candidate set of transcription factors but awaits further experimental validation.

AD was discovered in 1906, and for a hundred years, the question of how to treat the disease has plagued the scientific community. Although decades of research have identified features of the disease, such as the presence of amyloid plaques between neurons and the buildup of another toxic protein, tau protein, inside neurons, there are still a great deal of questions about what causes the onset of the disease and how best to treat it. The study of AD genes is important for understanding their pathogenesis and developing treatments. For example, by analyzing the epigenomic and gene expression changes that occur in AD, one research team identified cellular pathways that could be targets for new drugs [15–18]. Another study revealed new genetic loci for AD [19]. The discovery of these key genes and loci offers the possibility of a deeper understanding of AD pathogenesis and the development of new therapeutic strategies.

In this study, bioinformatics and machine learning methods were combined to jointly analyze three gene microarrays from the GEO database, from which differences in gene expression between AD patients and the normal control group were demonstrated, and 254 DEGs were selected. To understand the functions and pathways of these DEGs, GO analysis and KEGG enrichment analysis were performed, and the results of GO analysis showed that DEGs were highly enriched in functions and structures such as vesicular transport, vesicular membranes, and proton pumps/ATPases, which implied that these genes play important roles in membrane transport and energy metabolism processes in neuronal cells. This may be closely related to the pathophysiological process of AD, which is usually accompanied by abnormal metabolism of proteins and disturbances in vesicular transport within neurons. This finding is consistent with previous findings, such as those discussing the relationship between amyloid and defects in synaptic vesicle dynamics and neurotransmitter release, and proposing the treatment of Alzheimer's disease through interventions targeting synaptic vesicle dynamics [20]. KEGG enrichment analysis, on the other hand, showed the most significant enrichment of pathways related to neurodegenerative diseases, such as Alzheimer's disease, synaptic transmission, Parkinson's disease, and Huntington's disease, suggesting that AD may share certain pathological mechanisms with other neurological disorders and that these DEGs may play a key role in these shared mechanisms. GO analysis of intersecting genes (mitophagy-related DEGs) showed significant enrichment of intersecting genes in mitochondrial respiratory chain complexes and endosomes, suggesting that these genes may be involved in the regulation of mitochondrial function. Mitochondria are important organelles for cellular energy production, and their dysfunction leads to neuronal damage [21]. It has been reported in the relevant literature that mitochondrial dysfunction may be an important link in the early stages of Alzheimer's disease and may be involved in the development of AD through the generation of oxidative stress [22]. Another study indicated that mitochondrial autophagy maintains cellular homeostasis by degrading damaged mitochondria, which plays a key role in maintaining normal mitochondrial structure and function, cell survival, and neurological development and is closely related to the development of AD [23, 24]. These results indicate the important role of mitochondrial functional impairment and autophagy in the pathogenesis of AD, and the present study searched for DEGs associated with mitophagy in AD, which is of great significance for the development of diagnostic markers and drugs.

In addition, this study revealed the two main subtypes of AD samples in this study by consistency cluster analysis and analyzed the variability of the different sets of pathway genes among the subtypes. Among them, the IL-6 pathway gene expression levels were overall higher in Cluster 1 than in Cluster 2. This result may reflect a more severe inflammatory pathologic process in patients with Cluster 1, or it may be associated with a specific subtype or clinical presentation. It has been suggested that IL-6 controls the development of neurological disorders by binding to receptors that activate transcriptional activators in signaling, which can be used as a basis for the diagnosis of AD, as well as for the screening of patients at high risk for AD and the early detection of patients with mild and severe AD [25]. Similarly, the immune cell infiltration, immune activation, and mitochondrial oxidative phosphorylation of the two subtypes were significantly different, and these differential analyses provide directions for further research on the functional status of the different subtypes of cells and their regulatory mechanisms, which are conducive to the research and implementation of personalized therapy.

Machine learning was then used to screen seven key genes from 38 DEGs associated with mitophagy: FXR1, TUBB4B, TUBB, MAP4K4, IDH3G, RTN3, and ATP5F1B. Among them, FXR1 is a member of the Fragile X gene family, which, together with FXR2 and FMRP, forms the Fragile X gene family that is highly expressed in neurons and associated with schizophrenia, bipolar disorder, and emotion regulation. It has been shown that members of the fragile X gene family are able to regulate synaptic structure and function [26]. Another study indicated that adhesion molecules of synaptic cells can regulate Aβ production through interactions with key enzymes that promote Aβ formation, and Aβ-dependent synaptic adhesion affects synaptic function and integrity, suggesting that altered adhesion molecules at synapses play a key role in the disruption of the AD neuronal network [27]. Therefore, there may also be an association between FXR1 and AD, especially in terms of synaptic function and integrity, and FXR1 may be involved in regulating synaptic function in neuronal networks, in which synaptic adhesion molecules may play a key role. Both TUBB4B and TUBB are microtubule-encoding genes. Microtubules are cytoskeletal polymers of ⍺/β-tubulin heterodimers that are essential for a wide range of cellular processes. Pathogenic variants in microtubule-encoding genes (e.g., TUBB4B, which encodes a β-4B microtubule protein isoform) are responsible for multiple brain malformations [28]. A study [29] showed that microtubule proteins isolated from the brains of AD patients could polymerize in vitro to form structurally abnormal microtubules, suggesting that disturbances in the self-regulation of the microtubule system in human brain cells may lead to structural defects in microtubules and may result in the destabilization of the functioning of the preferential metabolic system within the cell, which contributes to the onset of AD. Tau is also a microtubule protein, and aberrant aggregation of tau proteins into neuroprogenitor fiber tangles is thought to be an important pathogenetic mechanism in AD [30]. Moreover, a study [31] reported direct evidence that overexpression of tau proteins leads to axonal transport disorders in neuronal cells and suggested that this is associated with disorders of microtubule proteins. TUBB4B and TUBB, as genes coding for microtubule proteins, are likely to be involved in the development of AD as well. The MAP4K4 gene encodes mitogen-activated protein kinase kinase kinase kinase kinase 4 protein (MAP4K4 protein). Endothelial MAP4K4 enhances endothelial permeability, thereby promoting inflammation and atherosclerosis progression [32]. In addition, recombinant MAP4K4 protein aggravates neurological damage, brain edema, and blood‒brain barrier damage [33]. Although there is no direct evidence for a clear relationship between MAP4K4 and AD, inflammation plays an important role in the pathogenesis and progression of AD, and MAP4K4, a gene encoding an inflammation-associated protein kinase, may be related to the inflammatory process in AD to some extent.

The IDH3G protein, encoded by the IDH3G gene and working in conjunction with IDH3A and IDH3B, forms the IDH3 complex (isocitrate dehydrogenase). Located within the mitochondria, this complex catalyzes the conversion of isocitrate to α-ketoglutarate, playing a vital role in intracellular energy metabolism and the regulation of oxidative stress. It serves as a core component of the tricarboxylic acid cycle and is closely associated with cellular metabolism and energy homeostasis [34, 35]. Meanwhile, disruptions in neuronal energy metabolism play a pivotal role in the onset and progression of AD [36]. Mitochondrial energy metabolism dysfunction is one of the main pathological features of AD, and mitochondrial dysfunction and energy metabolism abnormalities in the brains of AD patients precede Aβ deposition and Tau protein phosphorylation [37]. Abnormal mitochondria can be found in the brain tissue of several AD model animals and AD patients [38], and early on, they can exhibit dysfunction, such as ultrastructural changes, abnormal mitochondrial enzyme activity and oxidative stress [39], which ultimately leads to reduced ATP synthesis, free radical production and oxidative damage, leading to neuronal dysfunction. Due to the high energy demands of the nervous system and the important role of mitochondria in neuronal homeostasis, ATP synthase deficiency may lead to neurological impairment. Although ATP5F1B is known to be a core gene encoding the v β subunit of the ATP synthase complex, the exact mechanism by which these variants are pathogenic remains incompletely understood [40]. RTN3 belongs to the RTN family, is localized in the endoplasmic reticulum and is widely expressed in various tissues of the body, with the highest expression in brain tissue [41]. The functions of most of the family members are still unknown. Some scholars speculate that they are involved in some important functions related to the endoplasmic reticulum, such as the formation of vesicles, the packaging of secretory products, and the regulation of intracellular calcium levels. They also found that this gene is widely expressed in the nonglial cells of the central nervous system of the mouse, especially in the hippocampus, cerebral cortex, hypothalamus, and some neuronal nuclei [42]. Another study showed that RTN3 was sufficient to negatively regulate BACE1 activity by blocking the binding of BACE1 to amyloid precursor proteins [43], whereas BACE1-deficient mice were able to survive healthily and completely abrogate Aβ production and that increasing the expression of RNT3 markedly reduced amyloidogenic peptides, whereas inhibition of RTN3 increased amyloidogenic peptide production [44]. Although the mechanism of action of RTN3 in AD is still not well understood, these results suggest that RTN3 plays an important role in neurodegenerative diseases.

Finally, based on these 7 key genes, a nomogram model for AD risk prediction was developed. In the dataset of this study, the AUC of the seven pivotal genes was greater than 0.75, which has good diagnostic value. The AUC of the nomogram model even reached 0.877, which has good disease prediction accuracy. In summary, FXR1, TUBB4B, MAP4K4, IDH3G, TUBB, RTN3, ATP5F1B and the nomogram model of these 7 key genes have great potential in the diagnosis and treatment of AD.

In this study, a combination of bioinformatics and machine learning was used to screen key genes, and some preliminary valuable results were obtained. However, the study also has some shortcomings. First, although the three gene chips were combined in this study, the sample size was still small, with 143 AD samples and 82 control samples, and enlarging the sample size might enhance the reliability of the results. Second, due to the limitation of experimental conditions, the results obtained in this paper could not be further verified, and we expect other researchers to further analyze and verify these key genes and the transcription factors associated with them.

The study combined bioinformatics approaches with machine learning methods to study the differentially expressed genes associated with mitochondrial autophagy in AD and screened seven key genes from them, which provides new perspectives and potential therapeutic targets for exploring the potential mechanisms by which mitochondrial autophagy affects AD. Cluster analysis and difference analysis between subtypes were also performed on AD samples to provide new ideas for studying individualized AD treatment.

Statements & Declarations

Funding

This work was supported by Key Research Center for Humanities and Social Sciences in Hubei Province(Hubei University of Medicine)(No. 2021YB005); the Start-up Foundation of Hubei University of Medicine (No. 2019QDJRW02); Cooperative Education Program of the Ministry of Education(No.202101142004); Medical education Research Project of Chinese Medical Association(No.2020B-N02353); Teaching and research project of Hubei University of Medicine(No.YHJ2020012;)

Competing Interests

The authors have no relevant financial or non-financial interests to disclose.

Author Contributions

MT.H. and J.B. wrote the main manuscript text and XY.L. prepared figures 1-4, JX.L. Prepared figures 5-8. All authors reviewed the manuscript.

Ethics approval

The study used data from publicly available databases

Consent to participate

not applicable

Consent to publish

not applicable

Data Availability Statement

The data used in this study are from the GEO public database. PMID: 21756998. PMID: 17077275. PMID: 18332434. PMID: 29937276. PMID: 18270320.

Jia L, Du Y, Chu L (2020) Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: a cross-sectional study. J Lancet Public health 5(12):e661–e67
Jia L, Quan M, Fu Y (2020) Dementia in China: epidemiology, clinical management, and research advances. J Lancet Neurol 19(1):81–92
Jia JP, Wei CB, Chen SQ (2018) The cost of Alzheimer's disease in China and re-estimation of costs worldwide. J Alzheimer's Dementia: J Alzheimer's Association 14(4):483–491
Jack CR, Bennett DA, Blennow K (2018) NIA-AA Research Framework: Toward a biological definition of Alzheimer's disease. J Alzheimer's Dementia: J Alzheimer's Association 14(4):535–562
Jia JP, Wang SH (2023) Advances in the pathogenesis and treatment of Alzheimer's disease. J J Stroke Neurol Disease 40(05):387–390
Makin S (2018) The amyloid hypothesis on trial. J Nat 559(7715):S4–S7
Ding WX, Yin XM (2012) Mitophagy: mechanisms, pathophysiological roles, and analysis. J Biol Chem 393(7):547–564
Li Y, Xia X, Wang Y (2022) Mitochondrial dysfunction in microglia: a novel perspective for pathogenesis of Alzheimer’s disease. J J Neuroinflammation 19(1):248–248
Kerr JS, Adriaanse BA, Greig NH (2017) Mitophagy and Alzheimer’s Disease: Cellular and Molecular Mechanisms. J Trends Neurosciences 40(3):151–166
Fang FE, Yu JH, Konstantinos P (2019) Mitophagy inhibits amyloid-β and tau pathology and reverses cognitive deficits in models of Alzheimer's disease. J Nat Neurosci 22(3):401–412
Barrett T, Troup DB, Wilhite SE (2009) NCBI GEO: archive for high-throughput functional genomic data. J Nucleic Acids Res 37(Database):D885–D890
Liang WS, Dunckley T, Beach TG (2007) Gene expression profiles in anatomically and functionally distinct regions of the normal aged human brain. J Physiological Genomics 28(3):311–322
Readhead B, Haure-Mirande J, Funk CC (2018) Multiscale Analysis of Independent Alzheimer’s Cohorts Finds Disruption of Molecular, Genetic, and Clinical Networks by Human Herpesvirus. J Neuron 99(1):64–82
Blalock EM, Buechel HM (2011) Microarray analyses of laser-captured hippocampus reveal distinct gray and white matter signatures associated with incipient Alzheimer's disease. J J Chem Neuroanat 42(2):118–126
Mathys H, Boix PZ (2023) Single-cell atlas reveals correlates of high cognitive function, dementia, and resilience to Alzheimer’s disease pathology. J Cell 186(20):4365–4385
Xiong X, James BT, Boix CA (2023) Epigenomic dissection of Alzheimer’s disease pinpoints causal variants and reveals epigenome erosion. J Cell 186(20):4422–4437
Sun N, Victor MB, Park YP (2023) Human microglial state dynamics in Alzheimer’s disease progression. J Cell 186(20):4386–4403
Vishnu D, Hansruedi ABC (2023) Neuronal DNA double-strand breaks lead to genome structural variations and 3D genome disruption in neurodegeneration. J Cell 186(20):4404–4421
Wightman DP, Jansen IE, Savage JE (2021) A genome-wide association study with 1,126,563 individuals identifies new risk loci for Alzheimer’s disease. J Nat Genet 53(9):1276–1282
Marsh J, Alifragis P (2018) Synaptic dysfunction in Alzheimer's disease: the effects of amyloid beta on synaptic vesicle dynamics as a novel target for therapeutic intervention. J Neural regeneration Res 13(4):616–623
Reddy PH (2007) Mitochondrial dysfunction in aging and Alzheimer's disease: strategies to protect neurons. J Antioxid redox Signal 9(10):1647–1658
Devi L, Ohno M (2012) Mitochondrial dysfunction and accumulation of the β-secretase-cleaved C-terminal fragment of APP in Alzheimer's disease transgenic mice. J Neurobiol Disease 45(1):417–424
Kerr JS, Adriaanse BA, Greig NH (2017) Mitophagy and Alzheimer’s disease: cellular and molecular mechanisms. J Trends neurosciences 40(3):151–166
Pickles S, Vigié P, Youle RJ (2018) Mitophagy and quality control mechanisms in mitochondrial maintenance. J Curr Biology 28(4):R170–R185
Shi ZK, Zhang ZH Advances in the regulation of IL-6 expression in neurological disorders. J Occup health 32(14), 2004–2008 (2016)
Edbauer D, Neilson JR, Foster KA (2010) Regulation of synaptic structure and function by FMRP-associated microRNAs miR-125b and miR-132. J Neuron 65(3):373–384
Leshchyns KI, Sytnyk V (2016) : Synaptic cell adhesion molecules in Alzheimer’s disease. J Neural Plast. 1–9
McFadden JR, Tolete CDP, Huang Y (2023) Clinical, genetic, and structural characterization of a novel TUBB4B tubulinopathy. J Mol Genet Metabolism Rep 36:100990–100990
Shevtsov PN, Shevtsova EF, Burbaeva GS (2006) Disturbed assembly of human cerebral microtubules in Alzheimer’s disease. J Bull Experimental Biology Med 141(2):265–268
Iqbal K, Liu F, Gong C (2010) Tau in Alzheimer disease and related tauopathies. J Curr Alzheimer Res 7(8):656–664
Mandelkow E (2003) Clogging of axons by tau, inhibition of axonal traffic and starvation of synapses. J Neurobiol Aging 24(8):1079–1085
Pannekoek W, Linnemann JR, Brouwer PM (2013) : Rap1 and Rap2 antagonistically control endothelial barrier resistance. J PloS one 8(2), e57903
Zou Z, Dong Y, Liu D (2021) MAP4K4 induces early blood-brain barrier damage in a murine subarachnoid hemorrhage model. J Neural Regeneration Res 16(2):325
Jiang J, Huang D, Jiang Y (2021) Lactate Modulates Cellular Metabolism Through Histone Lactylation-Mediated Gene Expression in Non-Small Cell Lung Cancer. J Front Oncol 11:647559–647559
Krell D, Assoku M, Galloway M (2011) : Screen for IDH1, IDH2, IDH3, D2HGDH and L2HGDH mutations in glioblastoma. J PloS one 6(5), e19868
Demetrius LA, Magistretti PJ, Pellerin L (2015) Alzheimer's disease: the amyloid hypothesis and the Inverse Warburg effect. J Front Physiol 5:522
Mosconi L (2005) Brain glucose metabolism in the early and specific diagnosis of Alzheimer’s disease: FDG-PET studies in MCI and AD. J Eur J nuclear Med Mol imaging 32:486–510
Yu H, Lin X, Wang D (2018) Mitochondrial molecular abnormalities revealed by proteomic analysis of hippocampal organelles of mice triple transgenic for Alzheimer disease. J Front Mol Neurosci 11:74
An CQ, Yang YH, Wei Z (2021) Impact of mitochondrial structural and functional alterations on the development of Alzheimer's disease. J Life Sci 33(02):154–165
Nasca A, Mencacci NE, Invernizzi F (2023) Variants inATP5F1B are associated with dominantly inherited dystonia. J Brain 146(7):2730–2738
Oertle T, Schwab ME (2003) Nogo and its paRTNers. J Trends Cell Biology 13(4):187–194
Liu J, Huang XW, Peng XZ (2003) : Distribution pattern of mouse RTN3 gene mRNA and protein in the mouse central nervous system. J KeXueTongBao (17), 1851–1856
He W, Lu Y, Qahwash I (2004) Reticulon family members modulate BACE1 activity and amyloid-β peptide generation. J Nat Med 10(9):959–965
Chen R (2012) Reticulon 3 negatively regulates autophagy to prevent clearance of cytoplasmic prion aggregates. D. Wuhan University

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Bioinformatics and machine learning-based screening of key genes in Alzheimer's disease.

Status:

Version 1

Abstract

Objective

Methods

Results

Conclusion

Figures

1 Introduction

2 Methods

2.1 Data collection and preprocessing

2.2 Acquisition of DEGs and intersecting genes

2.3 GO and KEGG enrichment analysis

2.4 Screening for key genes

2.5 Consensus clustering

2.6 Immune analysis

2.7 GSEA

2.8 Constructing risk prediction models

2.9 Evaluation of key genes

2.10 Prediction of transcription factors

3 Results

3.1 Acquisition of DEGs and intersecting genes

3.2 GO and KEGG enrichment analysis

3.3 Screening for key genes

3.4 Consensus clustering

3.5 Immune analysis

3.6 GSEA

3.7 Constructing risk prediction models

3.8 Evaluation of key genes

3.9 Prediction of transcription factors

4 Discussion

5 Conclusion

Declarations

References

Additional Declarations

Status:

Version 1