Identification and immunological characterization of cuproptosis-related molecular clusters in Alzheimer's disease

Abstract

Introduction: Alzheimer's disease is the most common dementia with clinical and pathological heterogeneity. Cuproptosis is a recently reported form of cell death, which appears to result in the progression of various diseases. Therefore, our study aimed to explore cuproptosis-related molecular clusters in Alzheimer's disease and construct a prediction model.

Methods: Based on the GSE33000 dataset, we analyzed the expression profiles of cuproptosis regulators and immune characteristics in Alzheimer's disease. Using 310 Alzheimer's disease samples, we explored the molecular clusters based on cuproptosis-related genes, along with the related immune cell infiltration. Cluster-specific differentially expressed genes were identified using the WGCNA algorithm. Subsequently, the optimal machine model was chosen by comparing the performance of the random forest model, support vector machine model, generalized linear model, and eXtreme Gradient Boosting. Nomogram, calibration curve, decision curve analysis, and three external datasets were applied for validating the predictive efficiency.

Results: The dysregulated cuproptosis-related genes and activated immune responses were determined between Alzheimer's disease and non-Alzheimer's disease controls. Two cuproptosis-related molecular clusters were defined in Alzheimer's disease. Analysis of immune infiltration suggested the significant heterogeneity of immunity between distinct clusters. Cluster2 was characterized by elevated immune scores and relatively higher levels of immune infiltration. Functional analysis showed that cluster-specific differentially expressed genes in Cluster2 were closely related to various immune responses. The Random forest machine model presented the best discriminative performance with relatively lower residual and root mean square error, and a higher area under the curve (AUC = 0.9829). A final 5-gene-based random forest model was constructed, exhibiting satisfactory performance in two external validation datasets (AUC = 0.8529 and 0.8333). The nomogram, calibration curve, and decision curve analysis also demonstrated the accuracy to predict Alzheimer's disease subtypes. Further analysis revealed that these five model-related genes were significantly associated with the Aβ-42 levels and β-secretase activity.

Conclusion: Our study systematically illustrated the complicated relationship between cuproptosis and Alzheimer's disease, and developed a promising prediction model to evaluate the risk of cuproptosis subtypes and the pathological outcome of Alzheimer's disease patients.

Keywords: Alzheimer's disease; cuproptosis; immune infiltration; machine learning; molecular clusters; prediction model.

Figure 1

The study flow chart.

Figure 2

Identification of dysregulated CRGs in AD. (A) The expression patterns of 13 CRGs were presented in the heatmap. (B) Boxplots showed the expression of 13 CRGs between AD and non-AD controls. ****p < 0.0001, ns, no significance. (C) The location of 13 CRGs on chromosomes. (D) Correlation analysis of 12 differentially expressed CRGs. Blue and Red colors represent positive and negative correlations, respectively. The correlation coefficients were marked with the area of the pie chart. (E) Gene relationship network diagram of 12 differentially expressed CRGs. (F) The relative abundances of 22 infiltrated immune cells between AD and non-AD controls. (G) Boxplots showed the differences in immune infiltrating between AD and non-AD controls. *p < 0.05, ***p < 0.001, ****p < 0.0001, ns, no significance. (H) correlation analysis between 12 differentially expressed CRGs and infiltrated immune cells.

Figure 3

Identification of cuproptosis-related molecular clusters in AD. (A) Consensus clustering matrix when k = 2. (B–E) Representative cumulative distribution function (CDF) curves (B), CDF delta area curves (C), the score of consensus clustering (D), and heatmap of non-negative matrix (E). (F) t-SNE visualizes the distribution of two subtypes.

Figure 4

Identification of molecular and immune characteristics between the two cuproptosis clusters. (A) Clinical features and expression patterns of 13 CRGs between two cuproptosis clusters were presented in the heatmap. (B) Boxplots showed the expression of 13 CRGs between two cuproptosis clusters. ***p < 0.001, ****p < 0.0001, ns, no significance. (C) the relative abundances of 22 infiltrated immune cells between two cuproptosis clusters. (D) Boxplots showed the differences in immune infiltrating between two cuproptosis clusters. *p < 0.05, **p < 0.01 ***p < 0.001, ****p < 0.0001, ns, no significance. (E) Boxplots showed the estimated immune score between the two cuproptosis subtypes.

Figure 5

Co-expression network of differentially expressed genes in AD. (A) The selection of soft threshold power. (B) Cluster tree dendrogram of co-expression modules. Different colors represent distinct co-expression modules. (C) Representative of clustering of module eigengenes. (D) Representative heatmap of the correlations among 10 modules. (E) Correlation analysis between module eigengenes and clinical status. Each row represents a module; each column represents a clinical status. (F) Scatter plot between module membership in turquoise module and the gene significance for AD.

Figure 6

Co-expression network of differentially expressed genes between the two cuproptosis clusters. (A) The selection of soft threshold power. (B) Cluster tree dendrogram of co-expression modules. Different colors represent distinct co-expression modules. (C) Representative of clustering of module eigengenes. (D) Representative heatmap of the correlations among 11 modules. (E) Correlation analysis between module eigengenes and clinical status. Each row represents a module; each column represents a clinical status. (F) Scatter plot between module membership in turquoise module and the gene significance for Cluster1.

Figure 7

Identification of cluster-specific DEGs and biological characteristics between two cuproptosis clusters. (A) The intersections between module-related genes of cuproptosis clusters and module-related genes in the GSE33000 dataset. (B) Differences in hallmark pathway activities between Cluster1 and Cluster2 samples ranked by t-value of GSVA method. (C) Differences in biological functions between Cluster1 and Cluster2 samples ranked by t-value of GSVA method.

Figure 8

Construction and evaluation of RF, SVM, GLM, and XGB machine models. (A) Cumulative residual distribution of each machine learning model. (B) Boxplots showed the residuals of each machine learning model. Red dot represented the root mean square of residuals (RMSE). (C) The important features in RF, SVM, GLM, and XGB machine models. (D) ROC analysis of four machine learning models based on 5-fold cross-validation in the testing cohort.

Figure 9

Validation of the 5-gene-based RF model. (A) Construction of a nomogram for predicting the risk of AD clusters based on the 5-gene-based RF model. (B,C) Construction of calibration curve (B) and DCA (C) for assessing the predictive efficiency of the nomogram model. (D,E) ROC analysis of the 5-gene-based RF model based on 5-fold cross-validation in GSE5281 (D) and GSE122063 (E) datasets.

Figure 10

Validation of correlation analysis based on GSE106241 dataset. (A-E) Correlation between MYT1L (A), PDE4D (B), SNAP91 (C), NPTN (D), KCNC2 (E), and Aβ-42 levels. (F–J) Correlation between MYT1L (F), PDE4D (G), SNAP91 (H), NPTN (I), KCNC2 (J), and β-secretase activity.