The interferon type I signature towards prediction of non-response to rituximab in rheumatoid arthritis patients

Abstract

Introduction: B cell depletion therapy is efficacious in rheumatoid arthritis (RA) patients failing on tumor necrosis factor (TNF) blocking agents. However, approximately 40% to 50% of rituximab (RTX) treated RA patients have a poor response. We investigated whether baseline gene expression levels can discriminate between clinical non-responders and responders to RTX.

Methods: In 14 consecutive RA patients starting on RTX (test cohort), gene expression profiling on whole peripheral blood RNA was performed by Illumina® HumanHT beadchip microarrays. Supervised cluster analysis was used to identify genes expressed differentially at baseline between responders and non-responders based on both a difference in 28 joints disease activity score (ΔDAS28 < 1.2) and European League against Rheumatism (EULAR) response criteria after six months RTX. Genes of interest were measured by quantitative real-time PCR and tested for their predictive value using receiver operating characteristics (ROC) curves in an independent validation cohort (n = 26).

Results: Genome-wide microarray analysis revealed a marked variation in the peripheral blood cells between RA patients before the start of RTX treatment. Here, we demonstrated that only a cluster consisting of interferon (IFN) type I network genes, represented by a set of IFN type I response genes (IRGs), that is, LY6E, HERC5, IFI44L, ISG15, MxA, MxB, EPSTI1 and RSAD2, was associated with ΔDAS28 and EULAR response outcome (P = 0.0074 and P = 0.0599, respectively). Based on the eight IRGs an IFN-score was calculated that reached an area under the curve (AUC) of 0.82 to separate non-responders from responders in an independent validation cohort of 26 patients using Receiver Operator Characteristics (ROC) curves analysis according to ΔDAS28 < 1.2 criteria. Advanced classifier analysis yielded a three IRG-set that reached an AUC of 87%. Comparable findings applied to EULAR non-response criteria.

Conclusions: This study demonstrates clinical utility for the use of baseline IRG expression levels as a predictive biomarker for non-response to RTX in RA.

Figure 1

Cluster diagrams of genes that were differentially expressed between 14 patients with RA before the start of rituximab treatment in relation to the ΔDAS28 response outcome at six months. A. Supervised (one-way) hierarchical cluster analysis of baseline gene expression levels of a set of 124 genes that differed at least two-fold in at least three patients. Genes (rows) that are increased relative to the mean are indicated in red, decreased in green and genes that show no difference are indicated in black. Patients were stratified based on changes in ΔDAS28 at six months after the start of treatment in responders (indicated by the orange bar) and non-responders (indicated by light blue bar) ('change in ΔDAS28' from low (left) to high (right)). The supervised analysis revealed five gene clusters of which one, consisting of IRGs, was associated with clinical outcome. B. An expanded view of the subcluster of eight IRGs that is associated with clinical responder status. C. Cluster of eight IRGs associated with ΔDAS28 clinical responder status. D. Cluster of 8 IRGs associated with EULAR clinical responder status. DAS28, 28 joints disease activity score; EULAR, European League against Rheumatism; IRG, interferon response genes.

Figure 2

Differential expression of type I interferon (IFN)-response gene activity in relation to clinical responder status to RTX. The expression levels of eight IRGs (LY6E, HERC5, IFI44L, ISG15, MxA, MxB, EPSTI1 and RSAD2) were determined by quantitative (q)PCR analysis in peripheral blood cells of patients with RA before RTX treatment. Average expression in Log2 of the eight IRGs was calculated and used as IFN-score. Data are shown as box plots; each box shows the 25^th to 75^th percentiles. Student t test analysis revealed a significantly higher IFN-score in non-responders compared to responders based on ΔDAS28 (P = 0.0074) (A), and EULAR good/intermediate responders (1, 2) versus non-responders (0) (P = 0.0599) (B). PCR, polymerase chain reaction.

Figure 3

Receiver operating characteristics (ROC) curves for the IRGs as predictor for nonresponse upon RTX treatment in the validation cohort (n = 26). A. AUC (0.82) for the eight IRG set based on ΔDAS28 response criteria, B. AUC (0.87) for the three IRG set based on ΔDAS28 response criteria, C. AUC (0.78) for the eight IRG set based on EULAR response criteria (responders and intermediate responders vs. non-responders) and D. AUC (0.83) for the three IRG set based on EULAR response criteria (responders and intermediate responders versus non-responders). On the y-axis sensitivity and on the x-axis 1-specificity is indicated. AUC, area under the curve; DAS28, 28 joints disease activity score; EULAR, European League against Rheumatism; IRGs, interferon response genes; RTX, rituximab.