Alzheimer’s disease (AD) is a complex multi-factorial neurodegenerative disorder with a hidden onset that affects more than 50 million people worldwide (Liu et al., 2022). AD is the leading cause of mortality in dementia and due to its non-curable reality based on current medical development status, it has brought extreme economic burden to patients’ households as well as continuous distress. The main clinical signs and symptoms of AD include subjective cognitive decline (SCD), behavioral change, and dementia (Jessen et al., 2014). However, before clinical symptoms are observed, alterations in neurons, microglia, and astroglia have already driven insidious progression of the disease (De Strooper and Karran, 2016). The preclinical phase was termed as the cellular phase of AD (Scheltens et al., 2021). However, in most cases when AD could be definitively diagnosed, the patient is already in the dementia stage. Recent studies have found that intervention and prevention before the onset of AD may delay or even prevent the occurrence and development of the disease (Feng et al., 2021; McDade et al., 2021). And, current medications and FDA-approved treatments cannot cure AD and AD-related dementia (Fang et al., 2022). Therefore, it is significant to find effective biomarkers for early diagnosis and early prevention of AD.

Researches have confirmed that the precipitating features of AD involve multiple factors, among which most studies believe that AD, as a progressive and extensive neurodegenerative disease, is characterized by extensive gliosis, and the accumulation of amyloid β (Aβ) in the form of extracellular plaques and intracellular neurofibrillary tangles, ultimately leading to neurodegeneration and dementia (Joe and Ringman, 2019). Meanwhile, existing studies have found that excitatory neurotransmitter deficiencies or dysfunctions, as well as cerebrovascular dysfunction, are also involved in the development of AD (Fang et al., 2022). And adult hippocampal neurogenesis (AHN), an important neuroplasticity process, is also involved in the development of AD. A decrease in AHN may lead to hippocampal degeneration manifested by progressive memory loss and even the development of cognitive impairment (Moreno-Jimenez et al., 2021; Kim et al., 2022). In addition to the decrease in AHN, the presence of plaques and tangles in the hippocampus is also strongly associated with cognitive decline. However, there is no consensus on the core cause of cellular dysfunction in AD. Research over the last 5 years has focused on the use of proteomics, genomics and transcriptomics to investigate the pathogenesis of key metabolic pathways and regulators of Alzheimer’s disease (Nativio et al., 2020; Horgusluoglu et al., 2022). In this study we aimed to identify novel hub genes involved in AD development in the hippocampal brain region by differential gene screening combined with WGCNA analysis, providing new targets for clinical diagnosis and treatment of AD.

AD is the most common cause of dementia, with 60–80% caused by genetic factors (Scheltens et al., 2021). Since the core mechanisms of molecular characterization and mechanistic pathways are still unknown, more precise pathogenic targets and key pathways can be obtained by using multiple integrated analytical methods to covary the histological data of AD. This may not be achievable with any single histological research method. Therefore, we aimed to jointly identify novel AD hub genes by analyzing DEGs and WGCNA for transcriptomic data information in the hippocampal region of AD.

Transcriptomic information from lesions in many brain regions should be taken into account in the screening of DEGs for AD, including: posterior cortical atrophy for visuospatial impairments (Alladi et al., 2007), frontotemporal lobe degeneration for progressive aphasia and cognitive deficits (Bergeron et al., 2018; Graff-Radford et al., 2021), frontal neuroprogenitor fiber tangles for executive deficits (Townley et al., 2020), corticobasal degeneration for motor dysfunction (Mathew et al., 2012), and hippocampus for learning and memory deficits (Mu and Gage, 2011). AD is often characterized by progressive memory loss and cognitive deterioration mostly related to hippocampal function, and the hippocampus as a specific, vulnerable brain region is often the first to be affected by AD lesions. Therefore, we chose to compare the transcriptomic information (GSE29378) of normal human hippocampal brain regions and hippocampal brain regions of AD patients to uncover AD-related hub genes.

In the GSE29378 dataset, we compared AD and non-AD expression data with the help of differential genes to find AD-related DGEs (p < 0.05, |log2FC| ≥ 1). In the exploration of DEGs associated with previous diseases, the screening criteria for DEGs in brain tumors such as gliomas were mostly set at p < 0.05, |log2FC| ≥ 1.0 (Huang et al., 2020), and the screening criteria for neurological autoimmune diseases such as multiple sclerosis were also mostly set at this criterion (Li et al., 2021). However, after downloading and standardizing the data, it was difficult to obtain a sufficient number of DEGs associated with AD even at p < 0.05. When screening for differential genes, we initially chose to use p < 0.05, |log ₂ FC| ≥ 1, but only got 8 differential genes. This suggests that we may have missed some important genes, and at the same time we cannot guarantee the novelty of those genes found. This is likely because mRNAs in the hippocampus are not as up-regulated or down-regulated in AD as in tumors and autoimmune diseases, or AD is a chronic process with complex gene involvement and ack of large gene expression in quantity. It is difficult for the genes to be involved in the same time to directly regulate the AD process and lack specificity. The screening criteria for the AD dataset have not yet been determined. A and B use |log2FC| ≥ 0.5, p < 0.05 as the criteria for screening DEGs (Wang et al., 2021; Zhang et al., 2021). In order to ensure that a sufficient number of DEGs were obtained, we also chose this criterion for screening DEGs. However, this kind of screening will inevitably miss some important genes affecting the occurrence and development of AD. Biological pathways involved in altered synaptic vesicle cycling, GABAergic synapses, and glutamatergic synapses are associated with altered traits between groups of AD and non-AD samples in GSE29378. Previous studies have shown that the function of neuronal axonal transport and synaptic vesicle release in neurons determines neuronal plasticity (Lazarevic et al., 2017; Palikaras and Tavernarakis, 2020). In contrast, the regulation of Aβ through CDK5, calmodulin phosphatase signaling, and increased synaptic vesicle cycling in the AD state disrupts synaptic function and homeostasis, ultimately leading to cognitive decline and neurodegeneration (Wu et al., 2022). At the same time, APP, the precursor protein of Aβ, interacts with the neuron-specific potassium chloride (K + -Cl-) cotransporter KCC2/SLC12A5 and regulates GABAergic synaptic levels and activity affecting synaptic vesicle release (Tang, 2019). Compensatory enhancement of excitatory signaling or remodeling in early AD correlates with an early and pronounced loss of glutamatergic synapses in its progression (Mitew et al., 2013). This suggests that the differences we defined between the AD and normal groups are indeed associated with AD, and after screening for DEGs, these DEGs were mainly enriched in the pathways of calcium-dependent phospholipid binding, glutamatergic synapse, and cellular response to calcium ion in line with the GESA analysis, which also suggests that these DEGs are indeed associated with AD.

However, a single approach still struggles to explain the importance of these 57 DEGs for AD, and we used the WGCNA analysis to avoid using p-values and logFC alone as the sole measure of hub genes. WGCNA is not only a systems biology approach for describing patterns of genetic correlation among microarray samples, using module signature genes to summarize these sample clusters, but also a signature gene network approach to correlate modules with each other and with external sample traits, as well as for calculating module membership metrics (Langfelder and Horvath, 2008; Zhao et al., 2010). Thus, this methodology is more of a holistic methodology that measures the co-expressed gene modules among samples and explores the linkages between these modules and sample traits. The performance of WGCNA in hub discovery for AD and other neurological disorders was well supported by the literature (Mukherjee, 2021; Lin et al., 2022). However, we must also realize that there are limitations in WGCNA analysis: 1. The samples in the selected dataset should be larger than 15 in order to focus on the formation of effective modules. 2. WGCNA mostly uses default parameters to control the construction of the network and the extraction of the different modules, and does not preset the differentially expressed genes in advance. After our WGCNA analysis of the AD dataset, we focused on a few specific modules: the dark green module and the bright yellow module, which are positively correlated with AD status, the gray 60 module and the steel blue module, which are positively correlated with CA1, as well as the navy blue, steel blue, and gray 60 modules, which are negatively correlated with the CA3 brain region. However, we considered the hub genes should be more used for AD diagnosis and treatment, the relationship between the module genes and hippocampal subdivisions and their role on AD will be further elucidated in subsequent studies. One of the dark green modules we extracted 124 genes, these are mainly related to the response to unfolded proteins. The genes in bright yellow module, on the other hand, were mostly involved in intercellular adhesion, organ development, and other energies. This showed the consistency of processing with the development of AD occurrence. Finally, we obtained a total of 52 genes by intersecting these module genes with the DEGs of AD, and further analysis showed that these genes were associated with protein folding and intercellular adhesion, corroborating the previous analysis. Finally, we obtained 10 hub genes among them: HSPA1B, HSPB1, HSPB8, HSPA1A, DNAJB1, ANXA2, ANXA1, SOX9, YAP1 and AHNAK.

10 AD-related hub genes were eventually identified, and these were supported in the literatures that some of them are involved in the development of AD. Five of these 10 hub genes (HSPA1B, HSPA1A, HSPB1, HSPB8, DNAJB1) are from the heat shock protein family. Heat shock proteins (HSPs) are molecular chaperones which can be categorized into nine subfamilies (HSP10 (HSPE), HSP20 (HSPB), HSP40 (DNAJA, DNAJB, DNAJC), HSP60 (HSPD), HSP70 (HSPA), HSP90 (HSPC), and the large HSPs) with forms and functions capable of being expressed and induced by stress to promote proper folding of newly synthesized polypeptides, regulate assembly and disassembly of multiprotein complexes, and regulate intracellular protein trafficking and transmembrane transfer (Tower, 2009; Hu et al., 2022). It was found that the A2 allele of HSPA1B was able to quantitatively affect the mental state of patients and attenuate the anti-cellular oxidative stress effect of HSPA1B, increasing the incidence of AD (Clarimón et al., 2016; Ramos et al., 2018). HSPA1A is an isoform of HSP70. HSPA1A was found to be highly expressed in cerebrospinal fluid extracellular vesicles of AD patients directly involved in AD progression. HSPA1A expression was down-regulated in the prefrontal cortex of patients with advanced AD, and these may be directly related to its inhibition of the aggregation of tau protein isoforms (Voss et al., 2012; Kitzlerová et al., 2018; Muraoka et al., 2020). Moreover, it has also been shown that HSPA1A is also a gene that mediates the interaction between AD and depression (Mathys et al., 2019). HSPB8 also has an important effect on the development of AD. In cell lines expressing amyloid precursor protein (AβPP), HSPB1 expression alters the expression and processing of AβPP and directly reduces the amount of Aβ42 released by the cell line protecting the cells from the potential toxic effects of Aβ (Conway et al., 2014). At the same time HSPB1 can segregate toxic Aβ oligomers and convert them into large non-toxic aggregates to eliminate the toxicity of Aβ oligomers to cells, potentially reducing plaque deposition in AD patients (Ojha et al., 2023). Inhibition of the lncRNA SNHG14/UPF1 axis promotes HSPB8 expression, which inhibits apoptosis in AD neuronal cells (Tan et al., 2023). Meanwhile, HSPB8 can inhibit the production of D-Aβ1-40 and the formation of β-folds, and completely inhibit D-Aβ1-40-mediated cerebrovascular cell death (Wilhelmus et al., 2006). The DNAJB (Hsp40) family functions in protein folding or defolding, membrane trafficking, synaptic regulation, and mitochondrial function affect not only dopaminergic neurotransmission, but also Parkinson’s-associated neuropathological changes (Hasegawa et al., 2018). At the same time this chaperone protein interacts with β-proteins to promote intracellular aggregation of β-peptides and facilitates their translocation to mitochondria to exert toxic effects on AD patients (Ring et al., 2022). Annexin A1 (ANXA1) is a glucocorticoid anti-inflammatory mediator in the peripheral system that efficiently and selectively removes apoptotic neuron-like cells (McArthur et al., 2010). In early AD, ANXA1 is increased in the brain, where it is able to reduce Aβ levels by increasing the enzymatic degradation of neprilysin in N2a cells and to stimulate phagocytosis of Aβ by microglia to reduce inflammatory mediators produced as a result of Aβ stimulation (McArthur et al., 2010). ANXA1 levels were reduced in the peripheral plasma of patients with behavioral variants of AD and showed a correlation between ANXA1 and the production and abatement of its peripheral inflammatory mediators (Fraga et al., 2019). We found that ANXA2 did not exhibit a direct association with AD, but it has been shown that ANXA2 develops calcium-regulated membrane-cytoskeletal junctions that exhibit interactions with tau proteins in the context of Ca²⁺ segregation or elimination of differential capture of tau proteins by knockdown of ANXA2 (Gauthier-Kemper et al., 2011). SRY-box transcription factor 9 (SOX9) is a positive regulator of astrocyte formation, and activation of miR-22-3p levels in the hippocampus of mice improves their cognitive performance through SOX9-mediated activation of the NF-κB signaling pathway (Xia et al., 2022). Yes-associated protein 1 (YAP1), a transcriptional regulator that promotes tissue growth and regeneration, is also a potential regulator of AD (Xu et al., 2017). YAP was down-regulated and inactivated in hippocampal astrocytes of AD model mice in a hippocampal pathway-dependent manner, whereas activation of the YAP-CDK6 pathway improved cognitive function in both AD model mice and senescent mice (Ries et al., 2016). In the study of cognitive improvement by dexmedetomidine (Dex) in AD patients, the miR-129/YAP1/JAG1 axis may be a potential mechanism by which Dex protects against cognitive impairment in AD patients (Sun et al., 2020). However, we found that only 9 of these 10 hub genes were found to be directly associated with AD, among which the role of HNAK in AD has not yet been elucidated. Meanwhile, existing studies on the mechanisms of these molecules for the development of AD are still small and vary in depth, and these molecules still have great potential for research.

In the past literature, there is the identification of AD-related hub genes by using only the modular genes in DEGs or WGCNA to identify AD hub genes, which is still slightly insufficient in terms of in data selection and processing (Wu et al., 2021; Liang et al., 2022). In comparison, the strengths in our study: 1. The combination of two common methods can minimize the selection bias on hub genes. 2. Step-by-step analysis and validation, and try to be biologically functional and clinically phenotypically oriented in screening. 3. Supplementary dataset as a validation cohort for secondary validation of clinical value. However, this study still has some limitations that are difficult to avoid. Firstly, due to the difficulty in obtaining tissue samples, the active state of the tissue cells cannot be guaranteed in subsequent studies. Secondly, the present study was limited to the hippocampal region only, whereas the pathogenesis of AD is related to multiple brain regions, and a single role of the hippocampus cannot explain the multifactorial problem of AD. Meanwhile, since our criteria for screening hub genes were decided based on the existing analysis, there may still be some neglected important factors involved in the development of AD. In our subsequent studies we will combine different brain regions for further and more favorable analysis and validation. By targeting some new key targets and pathogenic pathway pathways (vascular accidents and specific functional synapses), effective treatment of AD may be realized (Shekhar et al., 2021). Specifically, the follow-up studies, as shown in Figure 8, consisted of combining a multi-omics approach to probe AD-related hub genes within multiple brain regions based on the present study. Analyze and validate the roles of these hub genes in the dominant pathways and synapses. Based on this, experimental models were established to investigate the molecular mechanisms of AD regulation by these molecules. We will also explore the role of these molecules in predicting the course of the disease in the clinic and develop possible target drugs.

In conclusion, we used a combination of differential gene screening methods and WGCNA sorting of modular genes to jointly identify hub genes that mediate AD onset and development in hippocampal brain regions. Ten hub genes, HSPA1B, HSPB1, HSPA1A, DNAJB1, HSPB8, ANXA2, ANXA1, SOX9, YAP1, and AHNAK, were identified by extracting genes intersecting the two modular genes in the DEGs and WGCNA. an external dataset validated the diagnostic significance of these molecules for AD. This study identifies new AD-related genes in the hippocampus and provides new potential therapeutic biomarkers, and molecular pathways.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary material.

YC: Conceptualization, Data curation, Methodology, Software, Writing – original draft, Writing – review & editing. ZL: Investigation, Methodology, Writing – review & editing, Writing – original draft. XG: Formal analysis, Project administration, Writing – review & editing. HL: Data curation, Validation, Writing – review & editing. ZG: Funding acquisition, Project administration, Resources, Writing – review & editing.