Predictors of response of rituximab in rheumatoid arthritis by weighted gene co-expression network analysis.

Abstract

The purpose of this study was to identify a biomarker that can predict the efficacy of rituximab (RTX) in the treatment of rheumatoid arthritis (RA) patients. Utilized weighted gene co-expression network analysis (WGCNA) and LASSO regression analysis of whole blood transcriptome data (GSE15316 and GSE37107) related to RTX treatment for RA from the GEO database, the critical modules, and key genes related to the efficacy of RTX treatment for RA were found. The biological functions were further explored through enrichment analysis. The area under the ROC curve (AUC) was validated using the GSE54629 dataset. WGCNA screened 71 genes for a dark turquoise module that were correlated with the efficacy of RTX treatment for RA (r = 0.42, P < 0.05). Through the calculation of gene significance (GS) and module membership (MM), 12 important genes were identified; in addition, 21 important genes were screened by the LASSO regression model; two key genes were obtained from the intersection between the important genes. Then, BANK1 (AUC = 0.704, P < 0.05) was identified as a potential biomarker to predict the efficacy of RTX treatment for RA by ROC curve evaluation of the treatment and validation groups. BANK1 gene expression was significantly decreased after RTX treatment, and a statistically significant difference was found (log FC = - 2.08, P < 0.05). Immune cell infiltration analysis revealed that the infiltration of CD4 + T cell memory subset was increased in the group with high BANK1 expression, and a statistically significant difference was found (P < 0.05). BANK1 can be used as a potential biomarker to predict the response of RTX treatment in RA patients. Key Points • Identifying the hub genes BANK1 as a potential biomarker to predict the response of RTX treatment in RA patients and confirming it in validation data. • Using the WGCNA approach and LASSO analyses to identify the BANK1 in a data set consisting of two GEO data merged and assessing the correlations between BANK1 and immune infiltration by CIBERSORT algorithm.