Alzheimer’s disease: an integrative bioinformatics and machine learning analysis reveals glutamine metabolism-associated gene biomarkers

Abstract

Background

Alzheimer’s disease (AD), a hallmark of age-related cognitive decline, is defined by its unique neuropathology. Metabolic dysregulation, particularly involving glutamine (Gln) metabolism, has emerged as a critical but underexplored aspect of AD pathophysiology, representing a significant gap in our current understanding of the disease.

Methods

To investigate the involvement of GlnMgs in AD, we conducted a comprehensive bioinformatic analysis. We began by identifying differentially expressed GlnMgs from a curated list of 34 candidate genes. Subsequently, we employed GSEA and GSVA to assess the biological significance of these GlnMgs. Advanced techniques such as Lasso regression and SVM-RFE were utilized to identify key hub genes and evaluate the diagnostic potential of 14 central GlnMgs in AD. Additionally, we examined their correlations with clinical parameters and validated their expression across multiple independent AD cohorts (GSE5281, GSE37263, GSE106241, GSE132903, GSE63060).

Results

Our rigorous analysis identified 14 GlnMgs—GLS2, GLS, GLUD2, GLUL, GOT1, HAL, AADAT, PFAS, ASNSD1, PPAT, NIT2, ALDH5A1, ASRGL1, and ATCAY—as potential contributors to AD pathogenesis. These genes were implicated in vital biological processes, including lipid transport and the metabolism of purine-containing compounds, in response to nutrient availability. Notably, these GlnMgs demonstrated significant diagnostic potential, highlighting their utility as both diagnostic and prognostic biomarkers for AD.

Conclusions

Our study uncovers 14 GlnMgs with potential links to AD, expanding our understanding of the disease’s molecular underpinnings and offering promising avenues for biomarker development. These findings not only enhance the molecular landscape of AD but also pave the way for future diagnostic and therapeutic innovations, potentially reshaping AD diagnostics and patient care.

Introduction

The “mitochondrial cascade hypothesis” posits a compelling link between Alzheimer’s disease (AD) etiology and disruptions in the oxidative phosphorylation (OXPHOS) system, with consequences that extend beyond its primary role in cellular energy production [1]. Central to this hypothesis is the idea that impaired OXPHOS function negatively affects various biochemical pathways, notably de novo pyrimidine biosynthesis [2]. This process hinges on dihydroorotate dehydrogenase (DHODH), an enzyme located in the inner mitochondrial membrane that channels electrons to the OXPHOS electron transport chain via coenzyme Q10. A reduction in electron transfer efficiency downstream of coenzyme Q10 impairs DHODH activity, leading to slowed pyrimidine nucleotide production [2]. These nucleotides are essential for the synthesis of nucleic acids, carbohydrates, and biological membranes, with their demand particularly high during cell division, where proliferating cells rely predominantly on the de novo pathway to meet this need. In contrast, differentiated cells such as neurons, which require extensive membrane synthesis for axonal and dendritic development, rely on salvage pathways for pyrimidine production. Here, uridine-cytidine kinase 2 phosphorylates uridine and cytidine into UMP and CMP, respectively. Neurons, with their complex axonal structures, have high basal energy requirements, making them particularly vulnerable to mitochondrial dysfunction [3, 4]. The hypothesis suggests that OXPHOS dysfunction may also disrupt pyrimidine biosynthesis and membrane remodeling, providing a plausible explanation for the altered neuronal membrane composition and architecture observed in AD, particularly synaptic degeneration that precedes neuronal loss. The advent of high-throughput data analysis, driven by bioinformatics tools, has revolutionized the exploration of gene functional networks in disease models, offering novel insights into the molecular mechanisms of diseases like AD. By utilizing extensive transcriptomic data and clinical annotations from resources such as the AD Neuroimaging Initiative, researchers have gained transformative perspectives on AD pathophysiology [5,6,7,8]. However, the exploration of glutamine metabolism genes (GlnMgs) in AD remains in its early stages. This study aims to analyze AD-related datasets from the GEO database through the lens of GlnMgs, seeking to elucidate their roles in the molecular landscape of AD.

Glutamine (Gln), as the most copious amino acid in the bloodstream, plays an indispensable role in the metabolic exigencies of proliferating tumor cells. It fuels the tricarboxylic acid (TCA) cycle and acts as a precursor for citrate in lipid biosynthesis through the process of reductive carboxylation [9]. Its pivotal role in glutaminolysis is crucial for cellular survival, mitigating oxidative stress, and preserving mitochondrial integrity. Intriguingly, the metabolism of Gln is vital for the functionality of the M2 subset of macrophages, with its depletion leading to a transition towards a pro-inflammatory M1 phenotype. This transformation highlights the potential of Gln metabolism modulation as a means to reprogram tumor-associated macrophages from an M2 to an M1 state, thereby potentiating anti-tumor immunity [9]. Furthermore, the metabolism of Gln exerts a significant influence on the differentiation of Th1 cells and the activation of effector T cells, offering a potential avenue to remodel the immune microenvironment and enhance the efficacy of immunotherapy [10]. In the context of AD, the aberrant formation of inflammasomes, leading to an inflammatory form of cell death known as pyroptosis, emerges as a critical pathological hallmark [11]. The strategy of concurrently targeting Gln metabolism and applying immunotherapy presents a novel and promising approach to combating AD. Nonetheless, the intricacies of Gln metabolism within the purview of immune recognition, coupled with its interaction with immunotherapeutic strategies, necessitate further exploration [12]. This investigation aims to systematically delve into the GlnMgs and their synergy with immunotherapy in AD. The insights garnered from this study are anticipated to forge a groundbreaking therapeutic paradigm, emphasizing the assembly of purinosomes and the pathways of Gln metabolism. However, to fully decipher the role of Gln metabolism in modulating immunogenicity and optimizing immunotherapy in AD, a more comprehensive research endeavor is essential. This study is poised to contribute to this endeavor, potentially catalyzing significant strides in clinical outcomes and therapeutic innovations for AD.

Central to this hypothesis is the premise that impaired OXPHOS efficiency triggers cascading disruptions across various biochemical pathways, particularly affecting the de novo biosynthesis of pyrimidines [13]. This pathway is critically dependent on DHODH, an enzyme embedded in the inner mitochondrial membrane that facilitates electron transfer to the OXPHOS electron transport chain via coenzyme Q10. Reduced electron transfer downstream of coenzyme Q10 compromises DHODH activity, resulting in impaired pyrimidine nucleotide production [14]. These nucleotides are essential for synthesizing nucleic acids, carbohydrates, and biological membranes, with their demand peaking during cell proliferation, where rapidly dividing cells predominantly rely on the de novo pathway [15]. In contrast, differentiated cells, such as neurons undergoing extensive axonal and dendritic growth for membrane synthesis, primarily depend on salvage pathways, where uridine-cytidine kinase 2 converts uridine and cytidine into UMP and CMP, respectively [16]. Neurons, with their highly complex axonal structures, exhibit substantial energy demands, making them particularly vulnerable to mitochondrial dysfunction [17]. The hypothesis suggests that OXPHOS impairment not only affects pyrimidine biosynthesis but also alters membrane remodeling, offering a plausible explanation for the disrupted neuronal membrane composition and architecture observed in AD, especially synaptic loss preceding neuronal death [17]. Recent advances in high-throughput data analysis, facilitated by bioinformatics, have revolutionized the study of gene networks across various disease models, providing novel insights into the molecular mechanisms of conditions such as AD [18]. Leveraging extensive transcriptomic data and clinical annotations from initiatives like the AD Neuroimaging Initiative, these approaches have revealed transformative perspectives on AD pathophysiology. However, the investigation of GlnMgs in the context of AD remains in its infancy. This study explores AD-related datasets from the GEO database through the lens of GlnMgs, aiming to elucidate their contributions to the molecular landscape of AD (Fig. 1).

Fig. 1

Framework. To advance our understanding of AD, we conducted a comprehensive analysis using patient-derived datasets from the GEO repository. Our primary cohort included the GSE5281, GSE37263, GSE106241, and GSE132903 datasets, with the GSE63060 dataset employed for validation. By applying a rigorous matching strategy for GlnMgs, we performed differential expression analyses and constructed a prognostic risk model. This approach identified a distinct subset of GlnMgs with prognostic significance in AD, highlighting their potential as candidate biomarkers. To further explore the functional roles of these genes, we conducted an extensive array of bioinformatics analyses, including GO, KEGG, and GSEA. These analyses were supplemented with data from multiple databases, offering a multidimensional view of the implicated GlnMgs and their involvement in cellular processes, signaling pathways, and gene regulatory networks

Materials and methods

We used the approaches proposed by Zi-Xuan Wu, et al. [19].

Raw data

The mRNA expression data for analysis were sourced from publicly available GEO datasets, including GSE5281, GSE37263, GSE106241, GSE132903, and GSE63060. These datasets, derived from different platforms (GPL570, GPL5175, GPL6947, GPL24170, and GPL10558), were designated for various roles in our study. GSE5281, GSE37263, GSE106241, and GSE132903 served as both training and test groups, while GSE63060 functioned as an independent test group. The search strategy utilized a MeSH term for Alzheimer’s disease, specifying mRNA data from Homo sapiens. The MSigDB comprised 105 GlnMgs (Table. S1).

Analysis of DEGs

To ensure accurate mRNA data, Perl scripting facilitated the matching and sorting of transcription data. Subsequently, a standardized approach was applied to the GSE5281, GSE37263, GSE106241, and GSE132903 datasets. Differential expression analysis employed criteria of FDR<0.05 and|log2FC|≥1 for GlnMgs, referred to as DEGs. Pearson’s correlation coefficient, implemented through the corrplot package, was employed to identify statistically significant and highly associated genes within modules.

GO and KEGG analysis

To elucidate the biological functions and pathways linked to differentially expressed genes (DEGs), we conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses. These analyses provided a comprehensive framework for understanding the molecular mechanisms underlying the observed gene expression changes. Leveraging the power of R programming, we systematically investigated the impact of dysregulated GlnMgs on key Biological Processes (BP), Molecular Functions (MF), and Cellular Components (CC), offering valuable insights into the broader biological landscape. The integration of GO and KEGG analyses allowed for a deeper characterization of the functional roles of GlnMgs, revealing significant disruptions in cellular pathways and biological activities. This approach enabled the identification of critical molecular functions and cellular components that may contribute to disease etiology or progression.

Model construction and immune cell infiltration

For model construction, the glmnet package was employed for Lasso regression analysis along with cross-validation. Additionally, the support vector machine recursive feature elimination algorithm (SVM-RFE) was utilized to build a machine learning model using the e1071 package. Cross-validation was used to assess the model’s error and accuracy. Furthermore, Lasso and SVM algorithms were used to construct a model to rank the significance of feature genes. Immune cell composition was analyzed using the CIBERSORT method [20].

SVM-RFE represents a sequential backward selection algorithm predicated on the maximum margin principle inherent to SVMs. In its inaugural iteration, the algorithm trains an optimized SVM model utilizing the full feature set available in the given dataset. Subsequent to this, each feature is scored, and these scores are ordered in a descending manner. The feature associated with the lowest score is then identified and eliminated from the dataset. This iterative process is perpetuated until a solitary feature remains, culminating in a refined feature subset optimized for model performance. By operationalizing this approach, SVM-RFE effectively navigates the high-dimensional feature space, mitigating the risk of overfitting while honing in on the most salient features. Such a procedure serves as a robust strategy for feature selection, particularly in complex datasets where discerning the most informative features is non-trivial. The methodology thus facilitates enhanced generalizability and predictive accuracy in machine learning models, making it particularly applicable to bioinformatics, finance, and other domains requiring intricate feature selection mechanisms.

Analysis of GSEA and GSVA

To identify relevant functions and pathway alterations across multiple samples, we employed GSEA [21] and GSVA [22]. These methods allowed for a nuanced and detailed exploration of dynamic biological processes and pathway fluctuations among distinct risk groups, offering a robust framework for characterizing the underlying molecular mechanisms. The resulting scores and visualizations provided deep insights into the temporal and spatial variations in biological activity, further elucidating the functional and structural heterogeneity present within the studied cohorts. This integrative approach facilitated a refined understanding of how GlnMgs contribute to disease pathophysiology, uncovering key pathway deviations and potential therapeutic targets. These computational tools allowed us to assess the dynamic activities and pathway changes within different risk subcategories by analyzing associated scores and visualizations. Furthermore, we utilized R to investigate the impact of differentially expressed PMGs on BP, MF, and CC, and pathways.

Exploration of drug-gene interactions

As the field of bioinformatics progresses, the identification of potential biomarkers has become increasingly important for the development of biological models and effective diagnostic strategies in various diseases. However, it is crucial to understand how to effectively translate these biomarkers into clinical applications. Therefore, accurate prediction of drug responses based on informative markers is paramount for future prevention and treatment strategies in AD. Validated biomarkers serve as crucial reference points for targeted therapies. In this study, we utilized the DGIdb database (https://dgidb.genome.wustl.edu/) to predict drug interactions with the identified hub genes, enabling us to explore potential therapeutic interventions for AD.

Construction of mRNA-miRNA-lncRNA network

Non-coding RNA transcripts, including microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), play pivotal roles in genetic regulation. MiRNAs can modulate gene expression by either enhancing mRNA degradation or inhibiting translation. On the other hand, lncRNAs are non-coding RNA molecules typically composed of approximately 200 nucleotides. They regulate various physiological and biochemical cellular processes by mediating chromosomal changes, transcriptional activation, and interference. Recent studies have highlighted extensive crosstalk between miRNAs and lncRNAs, involving competition for binding between miRNAs, lncRNAs, and other regulatory targets. Notably, certain competitive endogenous RNAs (ceRNAs) have been identified, where an lncRNA functions by sequestering miRNAs. Thus, in this study, we aim to investigate whether specific miRNAs and lncRNAs exhibit shared regulatory mechanisms and participate in developmental processes relevant to AD. To obtain information regarding target genes for the identified common miRNAs and lncRNAs, we utilized empirically validated databases, including miRTarbase and PrognoScan, which provide comprehensive miRNA-lncRNA-target relationships. By intersecting the target genes of the common mRNA-miRNAs-lncRNA and the shared genes identified in AD, we constructed a regulatory network. The network was visualized using Cytoscape software, allowing for a comprehensive understanding of the intricate interactions between mRNA, miRNA, and lncRNA genes involved in AD.

Mendelian randomization analysis

To ensure the independence of exposure and outcome variables in our genome-wide association study (GWAS) summary data, we engaged in an association analysis via the TwoSampleMR package in R. Designating GLUL-related expression as the exposure and ovarian function diminution as the outcome, we aimed to explore potential causal relationships. The analysis entailed: (1) Instrumental Variables (IVs) Configuration: DDAH2 and GOT1-related expressions were screened with a P-value threshold of < 5 × 10^-8 to identify strongly associated exposures. (2) Independence Configuration: Linkage disequilibrium (LD) between SNPs was calculated using the PLINK clustering method, excluding SNPs with LD coefficient r^2 > 0.001 and within 10,000 kb to ensure SNP independence and reduce pleiotropic biases. (3) Statistical Strength Configuration: The robustness of instrumental variables was assessed using the F-statistic (F = β^2/SE^2), with variables having F < 10 deemed inadequate to mitigate confounding effects.

Leveraging GWAS data, SNPs associated with the instrumental variables were identified, and through the “harmonise_data” function within TwoSampleMR, we aligned allelic directions of exposure and outcome, excluding incompatible SNPs. The inverse variance-weighted (IVW) method served as the cornerstone for causal inference, employing the variance of instrumental variables as weights to determine causal dynamics, thereby advancing our understanding of the genetic architecture underlying disease states.

Results

DEG identification and principal component analysis

Among the 34 GlnMgs investigated, several GlnMgs exhibited significant differences. Notably, certain genes demonstrated clustering tendencies within the treatment group, while others clustered within the control group. Specifically, the treatment group included genes such as GLS2, GFPT1, ALDH5A1, GLS, GOT1, etc. On the other hand, the control group comprised SLC7A11, GLUL, HAL, ARG2, OAT, ATCAY, GAD1, SLC25A12, etc. (Fig. 2a). To assess the correlation among these GlnMgs, a correlation matrix was constructed and visualized (Fig. 2b) (Table S2).

Fig. 2

Principal component analysis. a Analysis of differential expression. b Analysis of correlation

Enrichment analysis of GlnMgs

GO enrichment analysis identified 368 core targets, encompassing BP, MF, and CC. The MF category primarily involved acyltransferase activity (GO:0016746), lyase activity (GO:0016829), carboxylic acid binding (GO:0031406). Within the CC category, notable enrichments included transferase complexes involved in the mitochondrial inner membrane (GO:0005743), presynapse (GO:0098793), mitochondrial matrix (GO:0005759). In terms of BP, significant enrichments were observed in lipid transport (GO:0006869), purine-containing compound metabolic process (GO:0072521), response to nutrient levels (GO:0031667). Additionally, KEGG enrichment analysis revealed that the overexpressed genes were predominantly associated with Arginine and proline metabolism (hsa00330), Glutathione metabolism (hsa00480), Type I diabetes mellitus (hsa04940) (Fig. 3 and Table S3a-b).

Fig. 3

GO and KEGG analyses were conducted for GlnMgs. a: The GO circle illustrates the scatter map of logFC (logarithm of fold change) values of the selected genes. b: The KEGG barplot and bubble chart illustrate the scatter map of logFC values for the indicated genes

Model construction

To establish a gene signature, we employed LASSO and Cox regression analyses, determining the optimal value (Fig. 4a-b). Subsequently, SVM-RFE was utilized to construct a machine learning model, validating its accuracy and reliability. The model exhibited an accuracy of 0.842 and an error of 0.158 (Fig. 4c-d). A total of 20 GlnMgs were identified through the intersection of LASSO and SVM (Fig. 4e). Comparative analysis of the model with these 14 hub genes revealed consistently high accuracy: GLS2 (AUC=0.783), GLS (AUC=0.657), GLUD2 (AUC=0.644), GLUL (AUC=0.644), GOT1 (AUC=0.720), HAL (AUC=0.613), AADAT (AUC=0.557), PFAS (AUC=0.673), ASNSD1 (AUC=0.687), PPAT (AUC=0.658), NIT2 (AUC=0.641), ALDH5A1 (AUC=0.679), ASRGL1 (AUC=0.680), ATCAY (AUC=0.574) (Fig. 4f). Moreover, the area under the curve (AUC) was 0.918 (95% CI 0.889−0.942) in GSE58331, demonstrating the superior accuracy and stability of the prediction model (Fig. 4g) (Table S4).

In addressing concerns regarding the AUC value, it is imperative to highlight the observed AUC of 0.842, as evidenced in Fig. 4. This metric unequivocally attests to the high accuracy of our predictive model. While individual genetic variability may contribute to fluctuations in AUC values, it is crucial to emphasize that the aggregated AUC values for the evaluated genes consistently approximate or exceed the notable threshold of 0.7. This synthesis of results substantially bolsters the validity, precision, and resilience of our predictive framework, affirming its potential utility in clinical and research settings (Table 1).

Table 1 The characteristics of model

Fig. 4

Development of the GlnMgs signature. a: LASSO regression for the 14 AD-related genes. b: Parameter selection through cross-validation in LASSO regression. c-d: Accuracy and error of the model. e: Venn diagram displaying the intersected genes. f: AUC values for the 14 hub genes. g: AUC of the training group

GSEA of analysis

Based on literature evaluation and the sensitivity of hub genes in the model, we identified DDAH2 and GOT1 as the most relevant genes associated with AD. In GO analysis, DDAH2 was primarily associated with biological processes such as CC glutamatergic synapse, CC postsynapse, CC presynapse. On the other hand, GOT1 was predominantly involved in MF cyclosporin a binding, MF histone binding, MF olfactory receptor activity (Fig. 5a). Regarding KEGG analysis, DDAH2 was primarily associated with glycine serine and threonine metabolism, hematopoietic cell lineage, leukocyte transendothelial migration. Meanwhile, GOT1 was mainly involved in olfactory transduction, progesterone mediated oocyte maturation, t cell receptor signaling pathway (Fig. 5b) (Table S5).

Fig. 5

GSEA of analysis in DDAH2 and GOT1. a: GO. b: KEGG

Immune cells

The immune milieu plays a crucial role in the initiation and progression of AD. To investigate the immune cell components in adipose tissue, we employed CIBERSORT. The results were visually represented using barplots and corplots, highlighting the expression patterns of immune cells. Notably, B cells naive demonstrated high expression levels in the AD group. While, Dendritic cells activated, NK cells activated, Macrophages M0, T cells follicular helper, B cells memory, T cells CD8demonstrated high expression levels in the control group (Fig. 6a). Furthermore, we performed correlation analysis between these genes and immune cells to gain insights into their expression patterns (Fig. 6b).

Fig. 6

Expression of immune cells. a Expression profiles of immune cells in different clusters. b Correlation between GlnMgs and immune cells

GSVA of analysis

In GO analysis, DDAH2 exhibited significant involvement in biological processes such as CC borc complex, BP regulation of presynaptic membrane potential, BP vestibulocochlear nerve formation, MF phosphatidylserine flippase activity, BP regulation of maintenance of sister chromatid cohesion, BP negative regulation of interleukin 6 mediated signaling pathway. On the other hand, GOT1 was primarily associated with biological processes such as CC wash complex, BP interleukin 21 production, BP cardiac muscle myoblast proliferation, CC borc complex, BP regulation of maintenance of sister chromatid cohesion, MF phosphatidylserine flippase activity (Fig. 7a). Regarding KEGG analysis, DDAH2 was primarily associated with valine leucine and isoleucine biosynthesis, o glycan biosynthesis, glycosylphosphatidylinositol gpi anchor biosynthesis, type ii diabetes mellitus, riboflavin metabolism, taurine and hypotaurine metabolism. GOT1, on the other hand, was mainly involved in bladder cancer, vegf signaling pathway, olfactory transduction, linoleic acid metabolism, glycerophospholipid metabolism, mtor signaling pathway (Fig. 7b).

Fig. 7

GSVA of analysis in DDAH2 and GOT1. a: GO. b: KEGG

Drug-gene interactions

Based on the analysis, the 14 hub genes were found to predict the response to 79 drugs. Some of these drugs include carglumic acid, methionine, riluzole, methadone, telaglenastat, purpurogallin, cefquinome, glutamine, etc. (Table S6). Furthermore, the interactions between drugs and genes were visualized using Cytoscape 3.7.1, providing a comprehensive overview (Fig. 8).

Fig. 8

Visualization of drug-gene interactions. Note Up-regulated genes are represented by red circles, down-regulated genes by green hexagons, and associated drugs by blue squares

Identification of common RNAs and construction of miRNAs-LncRNAs shared genes network

Through an extensive search across three databases, a total of 610 miRNAs and 535 lncRNAs associated with AD were identified (Table S7a-b). To establish a comprehensive network, the intersection of these miRNAs, lncRNAs, and shared genes (identified through Lasso regression and SVM-RFE) was taken into account. Ultimately, the network consisted of miRNAs, lncRNAs, and several common genes, including 20 hub genes (SLC7A11, ALDH5A1, GLUL, GOT1, CPS1, PPAT, etc.) (Fig. 9).

Fig. 9

Network of miRNAs-LncRNAs shared genes. Note mRNA nodes are represented by red circles, miRNA nodes by blue quadrangles, and lncRNA nodes by green triangles

Validation of hub genes

To enhance the confidence and predictive accuracy of the hub genes, GSE63060 dataset was utilized for validation (Fig. 10). This observation supports the reliability of our model and the validation model.

Fig. 10

Fourteen hub genes were validated

Model verification

The Boxplots depicted the residual expression patterns of these genes in AD (Fig. 11a). There are some differences in the proportions of the four different modes (Fig. 11b). The GlnMgs’ diagnostic capacity in distinguishing AD from control samples revealed a satisfactory diagnostic value, with an AUC of RF: 0.904; SVM: 0.907; XGB: 0.905; GLM: 0.892 (Fig. 11c). An AUC of 1.000 (95% CI 1.000-1.000) in GSE63060 (Fig. 11d).

Fig. 11

Model verification. a Residual expression patterns. b-c Model expression patterns d AUC of model. e AUC of test group

Mendelian randomization analysis

In our exploration of the intrinsic connection between GLUL and OS forest plots were meticulously employed to visually articulate the associations. The SNP rs10752870, rs149007767, rs115388715 conspicuously positioned itself to the right of the confidence interval, indicating a positive association. Conversely, rs56330463 was observed to the left, reinforcing the credibility of our findings (Fig. 12), suggesting a similar trend of association with AD. Further dissecting the heterogeneity inherent in our analysis, the funnel plot tailored to AD revealed a deviation from the expected symmetrical distribution, albeit maintaining a general symmetry. This nuanced observation was further scrutinized through sensitivity analysis, employing a “leave-one-out” approach. Remarkably, the omission of any individual SNP from the analysis had a negligible effect on the results of the Inverse Variance Weighted (IVW) analysis, indicating that the remaining SNPs consistently mirrored the outcomes of the aggregate dataset. Substantiating the validity of our findings, the MR-Egger regression analysis was invoked, providing a solid foundation that bolsters both the robustness and authenticity of our results and the methodologies applied. This Mendelian randomization analysis unequivocally confirms the intimate association of GlnMgs with AD.

Fig. 12

Mendelian randomization analysis. a Correlation rectangle plot. b Heatmap. c The expression patterns of Correlation Coefficient. d SNP effect on Insomnia

Discussions

AD, a neurodegenerative disorder, predominantly affects individuals over the age of 65, marking its onset with memory impairments that evolve into irreversible cognitive decline and comprehensive functional incapacities [3]. This progression culminates in a profound degradation of life quality for patients [23]. The demographic shift towards an older population is amplifying the incidence of AD, thereby exerting considerable economic strain on both families and societal structures [24]. Within this context, the modulation of gene expression emerges as a pivotal element. Gln, a non-essential amino acid abundant in the bloodstream, is instrumental in the biosynthetic pathways of proliferative cells. It provides nitrogen for the synthesis of nucleotides and plays a vital role in the production of proteins and glutathione [25]. Its significance is further underscored in the metabolic processes of cancer cells, which often rely on Gln not only as a pivotal energy source but also as an essential molecule for maintaining cellular function [26, 27]. The enzyme glutaminase, which catalyzes the conversion of Gln to glutamate, serves as a key mediator in cellular metabolism [28]. Emerging research, including seminal works by Dai et al. and Liu et al., has begun to elucidate the role of GlnMgs within the context of diseases such as hepatocellular carcinoma and lung adenocarcinoma, highlighting a critical area of metabolic dependency exploited by various cancer types [29]. However, the implications of GlnMgs in the realm of non-malignant diseases have yet to be fully delineated. As our comprehension of tumorigenesis expands, the scientific inquiry is progressively venturing into the study of non-cancerous conditions [30]. The exploration of Gln metabolism in the progression of AD presents a novel frontier, promising to unlock significant insights into its pathogenic mechanisms. Such an endeavor holds the potential to not only deepen our understanding of AD’s molecular underpinnings but also to pave the way for innovative therapeutic interventions targeting these newly identified pathways [31]. This shift in focus towards the metabolic aspects of AD underscores a broader paradigm shift in disease research, moving beyond oncology to address complex neurodegenerative disorders through the lens of metabolic dysfunction.

In the realm of AD research, our comprehensive analysis identified 34 DEGs intricately linked to GlnMgs. Through an integrative approach combining Lasso regression and SVM-RFE, we have pinpointed specific GlnMgs that play critical roles in AD pathogenesis. Further refinement through crossover analysis revealed 14 key hub GlnMgs, including GLS2, GLS, GLUD2, GLUL, GOT1, HAL, AADAT, PFAS, ASNSD1, PPAT, NIT2, ALDH5A1, ASRGL1, and ATCAY. The diagnostic relevance of these hub genes was validated against external datasets, highlighting their potential roles in AD progression. Despite these advances, the complex interactions between these genes and specific transcription factors within the GlnMgs network demand further exploration. A detailed literature review highlighted the critical roles of DDAH2 and GOT1 in AD pathology. Investigations into their biological functions revealed involvement in lipid transport, purine metabolism, and nutrient response, shedding light on how GlnMgs influence AD pathophysiology through immune-related pathways. These findings not only deepen our molecular understanding of AD but also open new avenues for therapeutic innovation targeting these molecular mechanisms. This study expands the traditional boundaries of AD molecular research, offering a refined perspective on the disease and paving the way for the development of novel treatment strategies.

In the exploration of the intricate pathophysiology of AD, contemporary research has increasingly illuminated the critical link between gene expression regulation and the onset and progression of this disorder. Notably, the genes DDAH2 and GOT1 have been identified as playing pivotal roles in the development of AD. DDAH2, a crucial enzyme involved in the metabolism of the endogenous inhibitor asymmetric dimethylarginine (ADMA), plays a significant role in vascular health. Elevated levels of ADMA, a known endothelial function inhibitor, are associated with the pathogenesis of various cardiovascular diseases [32]. Recent studies have discovered a downregulation of DDAH2 expression in the brains of AD patients, leading to increased levels of ADMA [33]. This elevation could exacerbate cerebral vascular endothelial dysfunction, contributing to insufficient cerebral blood perfusion and oxygen supply, thereby advancing the pathological process of AD [34]. Furthermore, DDAH2’s regulation of nitric oxide synthase (NOS) activity and, consequently, nitric oxide (NO) production may engage in the neuroinflammatory and neuroprotective mechanisms within AD [35]. GOT1 is involved in key pathways of amino acid and energy metabolism, particularly bridging glycolysis and the tricarboxylic acid (TCA) cycle in mammals [36]. Changes in the expression of GOT1 in the brains of individuals with AD suggest an impact on the energy metabolism processes within the disease’s pathology [37]. Anomalies in GOT1 could lead to the accumulation or deficiency of metabolic products, affecting cellular energy production and utilization, thus impairing neuronal cell survival and function and exacerbating cognitive decline in AD [38]. By integrating metabolomics, proteomics, and genomics data, researchers are unveiling the mechanisms by which DDAH2 and GOT1 influence AD development. This involves their roles in modulating metabolic pathways and signaling cascades. These insights not only enhance our understanding of the complex pathophysiology of AD but also provide a scientific basis for developing therapeutic strategies targeting these novel markers. Future treatments based on DDAH2 and GOT1 may offer new hope for individuals with AD, especially in terms of early diagnosis and intervention.

Numerous studies have demonstrated that the infiltration of peripheral immune cells into the brain parenchyma post-injury, along with dysregulation of the brain’s immune environment, is closely associated with the progression of AD [39]. Given the inherent challenges in collecting intracerebral data, machine learning approaches are increasingly essential to elucidate the complex relationships between peripheral and central immune responses and their contributions to AD pathology [40]. Recently, neuroinflammation has been recognized as a critical factor in therapeutic failures [41]. Following injury, neuroinflammation is marked by a dysregulated balance in the production and release of pro- and anti-inflammatory cytokines from both central and peripheral sources. Microglial activation is a key indicator of neuroinflammation. Under physiological conditions, activated microglia release neuroprotective and anti-inflammatory factors [42]. However, in pathological states, such as chronic stress or infection, excessive microglial activation leads to heightened production of inflammatory mediators, culminating in neuronal injury and loss [43]. Addressing neuroinflammation and restoring neurotrophic and neurotransmitter functions require the identification of novel diagnostic biomarkers that capture the intricate crosstalk between the nervous and immune systems, a critical focus for both basic and clinical AD research. Building on our previous studies, we further investigated the expression patterns of immune-related genes, particularly GlnMgs, within the immunological microenvironment. In the AD group, naive B cells showed notably high expression levels, while activated dendritic cells, activated NK cells, M0 macrophages, follicular helper T cells, memory B cells, and CD8+ T cells were elevated in the control group. The increased expression of naive B cells, CD8+ T cells, and neutrophils in AD suggests a potential role for these immune populations in the disease’s pathogenesis.

The nexus between metabolic processes and AD represents a frontier that, despite its critical importance, has been only marginally explored within the contemporary research landscape [44,45,46]. A growing body of literature, bolstered by bioinformatics analyses, is beginning to unravel the intricate connections between metabolic dysregulation and AD. Zhang et al. highlighted five key biomarkers, with ZNF384 emerging as a potential therapeutic target for both psoriasis and AD due to its dual involvement in inflammatory responses and metabolic pathways. Simultaneously, Gu et al. developed a predictive model centered on iron metabolism dysregulation, identifying 520 genes closely linked to iron homeostasis. Despite these advancements, the role of GlnMgs in AD remains underexplored. Our study addresses this gap by investigating the influence of cellular metabolism on AD, specifically through an expanded analysis of GlnMgs derived from the GEO database, thus offering a more comprehensive perspective. While we contribute to the growing discourse on metabolic regulation in AD, we recognize the preliminary nature of these findings, reflecting an incomplete understanding of AD’s underlying biological complexity. This underscores the necessity for further empirical validation through in vivo and in vitro studies, which hold the potential to significantly deepen our understanding. The interplay between prognostic genes and GlnMgs presents a compelling area for future research, offering promising therapeutic implications. This study not only advances the current understanding of AD’s metabolic landscape but also advocates for a broader interdisciplinary approach in addressing this multifaceted neurodegenerative disorder.

Conclusions

AD manifests through intricate interactions among various targets, pathways, and regulatory mechanisms, constituting a complex landscape in its development and progression. These regulatory processes exhibit bidirectional and synergistic characteristics. Pivotal among them are GlnMgs, comprising GLS2, GLS, GLUD2, GLUL, GOT1, HAL, AADAT, PFAS, ASNSD1, PPAT, NIT2, ALDH5A1, ASRGL1, ATCAY.