A novel gene signature for forecasting time to next relapse in multiple sclerosis using peripheral blood mononuclear cells

Abstract

Aim: The purpose of this research study was to develop and validate a gene signature based on peripheral blood mononuclear cells (PBMCs) for predicting the time to the next relapse in multiple sclerosis (MS).

Methods: The GSE15245 dataset (N = 94) was divided into a training set (N = 65) and a testing set (N = 29). First, the training set was analyzed using weighted gene co-expression network analysis (WGCNA) to identify key modules that were highly correlated with the timing of the next acute relapse. Subsequently, the hub genes within these key modules were subjected to univariate Cox regression analysis, and genes related to the recurrence time of MS were identified. The least absolute shrinkage and selection operator (LASSO) Cox regression was used to refine the extraction further. Then, the gene signatures were constructed using multivariate Cox regression. The efficacy of the model that was based on the training set database was evaluated using receiver operating characteristic (ROC) curves and validated using an independent testing set. Additionally, gene signatures were also validated for differential expression using an external independent dataset, GSE21942 (N = 29), along with experimental verification.

Result: Two key modules were identified with WGCNA. Univariate Cox regression analysis yielded 30 genes related to the relapse time of MS from these two modules, and then LASSO regression analysis further refined the selection to four genes, namely, BLK, P2RX5, GP1BA, and PF4. These four genes were used within the training dataset to build a Cox regression model, and this showed high prediction performance in the training as well as the testing datasets. Both external dataset analysis and experimental validation corroborated the differential expression of BLK and P2RX5 in patients with MS.

Conclusion: BLK, P2RX5, GP1BA, and PF4 emerge as potential predictors of future disease activity in individuals with MS.

Keywords: Cox regression; Gene signature; Multiple sclerosis; Peripheral blood mononuclear cells; Relapse prediction.

Fig. 1

Flow chart of the study

Fig. 2

Process of module identification through WGCNA. (A) Comparison of scale-free fit indexes for soft-threshold powers. (B) Soft-threshold connection. (C) Module cluster dendrogram and assignment. Symbolically, each color denotes one module, with the gray module being one among the 15 identified. (D) For each row, a clinical trait is represented by a column. The module-to-clinical-trait correlation coefficient is stored in each cell. Positive correlations are shown in red, while negative correlations are represented in blue

Fig. 3

Analysis of functional enrichment of genes in key modules. (A-B) GO and KEGG are used to analyze the genes contained within the green-yellow module. (C-D) GO and KEGG are also used to explore the functional annotations and pathway associations of genes contained in the yellow module

Fig. 4

The process of identifying genes that are most significantly connected with a patient’s prognosis. (A) Utilizing univariate Cox regression, hazard ratio forest plots were generated to reveal the prognostic correlations of all genes based on their respective P values. (B) LASSO Cox model fitting procedures. Each gene is represented by a curve, depicting the trend of the coefficient as it changes by a given amount. (C) In the training dataset, the regularization parameter affects the partial probability bias. The dashed line on the left indicates the smallest possible mistake, while the dashed line on the right indicates the largest value that can occur within one standard deviation. As seen in the graph, each condition had a distinct set number of genes (non-zero coefficients) at the time of shrinkage and selection. Notably, when opting for the lowest error, four genes with non-zero coefficients (BLK, P2RX5, GP1BA, and PF4) were identified

Fig. 5

A metric that indicates how accurately the risk model predicts outcomes in both hypothetical and actual situations. (A) High-risk and low-risk survival curves for the training set are displayed. (B) High-risk and low-risk survival curves for the testing set are displayed. (C) ROC analysis estimates the 1-year, 2-year, and 3-year recurrence-free survival rates based on the training set risk score. (D) ROC analysis predicts 1-year, 2-year, and 3-year recurrence-free survival rates based on the testing set’s risk score

Fig. 6

Differential expression analysis of the four genes in prognostic models of healthy individuals and patients with MS. (A, B, C, D) are external data set validations of BLK, P2RX5, GP1BA, and PF4, respectively, while (E, F, G, H) are RT-PCR experimental validations of BLK, P2RX5, GP1BA, and PF4, respectively