Introduction
Rheumatoid arthritis (RA) is a systemic autoimmune disease characterized by chronic joint inflammation which may lead to cartilage and bone destruction. It is a heterogeneous disease, as reflected by differences in severity, pathogenesis and treatment outcome. From diagnosis onwards, RA patients often receive immunosuppressive treatment with non-biologic disease-modifying anti-rheumatic drugs (DMARDs) and/or glucocorticoids (GCs). When patients no longer benefit from the non-biologic therapy, they usually start on treatment with biologics, such as TNFα-blockers and B-cell depletion therapy using rituximab (RTX) [
1]. Approximately 30% to 50% of patients do not achieve a favorable response to biologics. To increase treatment efficacy and to develop personalized treatment, predictors of therapy response are needed.
Independent studies have shown that activation of the type I interferon (IFN) system is associated with the clinical outcome of RTX therapy [
2,
3]. This so-called ‘IFN signature’ represents a response program consisting of genes that are activated by type I IFNs and is present in approximately 50% of RA patients [
4]. Induction of type I IFN response genes (IRG) is triggered via activation of the JAK-STAT signaling pathway, more specifically via JAK1, TYK2, STAT1 and STAT2, followed by recruitment of IRF9 and formation of the ISGF3 transcription factor complex [
5]. It was shown that patients with a good response to RTX have low IRG expression prior to the start of treatment, whereas non-responders display relatively high IRG expression. Potential clinical utility of IRG expression reflected as an IFN-score to predict the clinical outcome of RTX treatment was demonstrated by an area under the receiver operating characteristics (ROC) curve of 87% [
3]. Hence, knowledge of IRG expression in a RA patient before the start of RTX treatment is of crucial importance to predict the success of the clinical outcome.
It has been reported that GCs can interfere with the type I IFN system by modulation of IFN induction as well as downstream IFN signaling [
6,
7]. GCs were initially prescribed to RA patients in high doses (≥10 mg/day) to suppress flares of inflammation, but nowadays long-term treatment with low-dose GCs is commonly used as well [
8]. Since use and dose of GCs are highly variable among patients prior to the start of treatment with RTX [
2,
3], we aimed to determine what the effect of GC use is on IRG expression in relation to the clinical response to RTX.
Discussion
Previous studies have shown that the IFN-score has clinical relevance by predicting the outcome of RTX therapy; a high IFN-score reflecting increased IRG expression at baseline is associated with a poor clinical response to RTX [
2,
3]. In the present study, we demonstrated that the average IFN-score was consistently lower in prednisone-using patients compared to patients not using prednisone. As a consequence, RTX response prediction based on the IFN-score was considerably improved when stratifying patients based on prednisone use. ROC analysis of the PREDN
− group based on an IFN-score of three IRGs yielded an almost perfect AUC of 0.975, compared to 0.848 and 0.797 in all patients or prednisone users alone, respectively. This means that a test based on the three-IRG IFN-score would correctly classify 98% of two PREDN
− patients of two randomly drawn pairs, which is considered ‘excellent.’ Based on these data, non-response to RTX could be predicted with a specificity of 100% and a sensitivity of 88% in PREDN
− patients.
At the moment, the IFN-score-based RTX prediction model seems to be the most discriminative test for RTX response prediction and has already demonstrated clinical utility [
3,
11]. The current data show that the model could be further optimized via stratification for prednisone use. GC therapy has proven to be a vital part in the management of RA [
1] and is often prescribed as bridging therapy in between biologics to prevent or suppress inflammatory flares. The observation that the RTX response prediction reached optimal predictive value in patients without current prednisone treatment suggests that the prednisone-related suppression of the IFN-score obscures the ‘genuine’ intrinsic IRG expression, leading to the lower accuracy of prediction. Since elimination of prednisone use in RA patients would be practically intolerable, implementation of prednisone use and/or dose into the IFN-score-based prediction model might be an approach to optimize prediction of treatment outcome. Eventually, it might be considered to taper or temporarily stop prednisone treatment to measure the ‘genuine’ intrinsic IRG expression to predict the response to RTX treatment, if the clinical condition of the patient allows that. The current data provide the first indication of the effect of medication history on response prediction. The study consisted of cross-sectional data and results need to be confirmed in a larger prospective cohort of RA patients who are sampled before and after GC treatment. Moreover, besides medication history, the influence of cumulative dosing and term of prednisone treatment should also be analyzed.
Our findings on the
in vivo suppressive effect of prednisone on IRG expression in RA corroborates results from mechanistic studies that reported an effect of GCs on the type I IFN system. In systemic lupus erythematosus (SLE), methylprednisolone injection coincided with a decrease in plasmacytoid dendritic cells (pDCs), which are considered to be the main producers of IFNα in SLE [
12,
13]. In RA, evidence is available for a role of both IFNα and IFNβ [
14,
15], indicating a broader cellular origin for these IFNs, making it unlikely that the prednisone-related IRG suppression in RA is caused solely by a decrease in pDCs.
Since GCs are able to interfere with the IRF3 and IRF9 pathways, thereby affecting IFNα/β induction and/or downstream IFN receptor (IFNAR) signaling, this could lead to suppression of both type I IFN production as well as downstream IRG induction. Such suppression is caused by the interaction of GRIP1/NCOA2 –a cofactor of GC signaling– with IRF3 and IRF9, and subsequent interference between GR signaling and TLR signaling and IFNAR signaling, respectively [
6,
7]. Additionally, it was demonstrated that GCs are able to induce expression of SOCS1 [
16], a well-known inhibitor of JAK-STAT signaling, including type I IFN signaling [
17]. Because both TLR and JAK-STAT signaling are implicated in the regulation of type I IFN activity in RA [
18,
19], this may be an additional mechanism of the observed prednisone-related IRG suppression. However, our study was not aimed at unravelling the mechanisms of GC-mediated type I IFN suppression, which is the objective of future studies.
Our observations raise questions regarding the relation between high baseline IRG expression and a poor response to RTX. It is yet unclear whether high IRG expression is (in)directly causative for RTX non-response or whether it is a related epiphenomenon. In the case of a causative relation between high baseline IRG expression and RTX non-response, it would be expected that prednisone use, as a suppressor of IRG expression, would lead to more responders. This was not observed in our cohort, as reflected by the described absence of a direct relation between prednisone use and the clinical response to RTX [
3]. Moreover, we did not observe any bias in clinical parameters between the subgroups of prednisone use and RTX response. Since the numbers of patients per subgroup are rather small, this could be due to a lack of power. However, our data indicate that the difference in prediction accuracy between PREDN
− and PREDN
+ patients is selectively due to prednisone-related IRG suppression in RTX non-responders, resulting in false-positive good responders in the PREDN
+ group, whereas responders are almost perfectly distinguishable from non-responders in the PREDN
− group. Altogether, these observations indicate that IFN
high patients using prednisone might appear as IFN
low patients due to the prednisone-related IRG suppression, but still turn out to be non-responders to RTX. This would in turn imply that the relation between high IRG expression and RTX non-responders is not a directly causative one.
Besides the association between baseline IRG expression and RTX response, there are indications of pharmacodynamic differences during RTX therapy as well. Vosslamber
et al. provided evidence that RTX responders, that is, patients with low baseline IRG expression, exhibited IRG upregulation after three months of therapy, whereas RTX non-responders did not [
20], suggesting that type I IFN dynamics is related to the clinical outcome of RTX treatment. It was hypothesized that high IRG expression before RTX treatment could reflect an over-stimulated type I IFN system, incapable of further inducing the IRG expression that would be essential to reach a favorable response to RTX. With regard to prednisone interference, one could speculate that this process of pathway saturation, possibly caused by extensive negative feedback or shortage of signaling proteins, could be synergistically enhanced by prednisone. This would then result in the absence of IRG induction during RTX therapy, despite relatively low IRG expression at baseline. Interestingly, the majority of patients in the study of Vosslamber
et al. was using prednisone (82%), and patients were allowed to continue using it during RTX therapy [
20]. Moreover, the observed pharmacological induction of IRG expression during RTX therapy was described to be irrespective of clinical parameters, such as prednisone use [
20], suggesting it was persistent despite GC interference. This could in turn imply that the IRG expression as induced during RTX treatment occurs via a different mechanism than the IRG expression at baseline, which has appeared sensitive to prednisone interference.
Our data might also be useful for other treatment regimens, as the relation between the IFN system and treatment response does not seem to be restricted to RTX. For example, for anti-TNF therapy with infliximab, the dynamics of IRG expression appeared to be related to the clinical response, as non-responders showed an IRG upregulation during treatment whereas good responders did not [
21,
22]. Furthermore, a genome-wide expression study revealed that high IRG expression before the start of treatment with tocilizumab, an IL-6R blocker, is associated with a favorable response [
23]. The response patterns observed for these biologics are not in line with that for RTX. Although the clinical relevance for these results needs to be validated in independent studies, it may indicate that the status of the IFN system might have different clinical consequences in RA depending on the specific biologic that is used, that is, the immune pathway that is modulated. Our findings on
in vivo interference of the IFN-system by prednisone may be equally relevant for the other biologic therapies and indications that are characterized by differential IFN activity. In these cases separate analysis of PREDN
− and PREDN
+ patients could provide supportive value for these claims.
Competing interests
CLV, SV and TDJ are inventors on a patent application wherein the use of the information on the interference of GCs to modulate the IFN system to improve outcome predictions on the use of biologics, such as rituximab in chronic inflammatory and other conditions, is claimed. MB, GW, MTN, CJL, GJ and AEV declare they have no competing interests.
Authors’ contributions
All authors were involved in drafting the article or revising it critically for important intellectual content. CLV had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: SV, TDJ and CLV. Acquisition of patient material and of data: GW, MTN, AEV, SV, MB. Analysis and interpretation of data: TDJ, SV, CJL, GJ, MTN, AEV and CLV. All authors have read and approved the final manuscript.