Patient characteristics
In this study, 120 RA patients were enrolled (17 seronegative RA, and 103 seropositive RA, Supplementary Table 2). The proportion of the achieved remission tended to be lower in the seronegative patients compared to that in the seropositive patients (17.6% vs 28.2%, respectively) in spite of lower disease activity before treatment in the seronegative RA patients (disease activity for 28 joints, 3.79 vs 5.1, P < 0.01), indicating that patients with seronegative RA may not respond to abatacept treatment as compared to the response in patients with seropositive RA.
To investigate the biomarkers associated with the efficacy of abatacept treatment other than seropositivity, we focused on the seropositive RA patients. Based on the definition stated in methods, patients were divided into two groups consisting of 24 early RA patients and 79 established RA patients (Supplementary Table 3). The patients in the early RA group responded better to abatacept treatment than those in the established group throughout 6 months of treatment despite high disease activity at the time of abatacept treatment initiation at the baseline with CDAI at 25.1 and CDAI at 18.3 in the early RA group and the established RA group, respectively (Figure 1B). The proportion of CDAI remission at 6 months post-treatment was higher in the early RA group than the CDAI proportion in the established RA group (37.5% and 25.3%, respectively), although the difference was not significant (P = 0.30).
Predictive clinical biomarkers for evaluating response to abatacept
We compared the baseline clinical characteristics and laboratory findings between patients who achieved CDAI remission at 6 months and those who did not achieve CDAI remission, separately in the early group and in the established group (Supplementary Table 4). In the early group, patients with remission tended (P < 0.15) to show higher serum IgA levels, anti-CCP titer, and counts of white blood cells and neutrophils than patients without remission (351 mg/dL vs 289 mg/dL, P = 0.12; 304 IU/mL vs 156 IU/mL, P = 0.06; 8555/µL vs 6666/µL, P = 0.08; and 6437/µL vs 4751/µL, P = 0.10, respectively). In the established RA group, only RF titer was found to be significant (189 IU/L vs 113 IU/L, P = 0.04).
For visualization of association between those clinical biomarkers and patients, we used hierarchical clustering analysis (Figure 2A). The levels of IgA, neutrophil count, anti-CCP titers and RF titers were found to be higher in the clusters that contained more patients who achieved remission. IgA levels, anti-CCP titer and neutrophil count were significantly correlated with the change of CDAI after 6 months with treatment by abatacept in patients with early RA (R = -0.44, P = 0.03; R = -0.56, P < 0.01; R = -0.41, P = 0.04, respectively), while RF significantly correlated with those with the established RA group (R = -0.35, P < 0.01) (Figure 2B). Receiver operating curve analysis demonstrated an optimal cut-off value of baseline serum IgA level, anti-CCP titer and neutrophil count to predict CDAI remission achievement in the early group as 342 mg/dL with sensitivity of 66.7%, specificity of 86.7%, and area under the curve (AUC) of 0.659, 330 IU/mL with sensitivity of 44.4%, specificity of 86.7% and AUC of 0.741, and 6200/µL with sensitivity of 55.6%, specificity of 86.7% and AUC of 0.704, respectively (Figure 2C and 2D). RF in the established group showed a cut-off of 212 IU/mL with sensitivity of 45%, specificity of 88.1% and AUC of 0.622.
Patients in the early RA group who satisfied with each cut-off value showed better response to abatacept treatment (Figure 2E-2G). The higher the number of variables the early patients satisfied, the better response they showed to abatacept treatment (Supplementary Figure 1). Although patients in the established RA group with RF ≥ 212 IU/mL responded well to abatacept treatment shortly after initiation of the treatment compared to those without the treatment, significance was found to be diminished at 6 months post-treatment (Figure 2H).
Peripheral immune-phenotyping and cytokine analysis
Immune-phenotyping and cytokine measurement of peripheral blood was conducted prospectively in 33 patients (9 early RA and 24 established RA). The characteristics of patients who were subjected to immune-phenotyping and cytokine analysis have been summarized in Supplementary Table 5. No major difference was found between patients with immune-phenotyping and those without immune-phenotyping except for differences in the lymphocyte counts and a history of use of other biological agents. In the early RA group, the proportions of the activated Th17 cells (aTh17) and the activated Treg cells (aTreg) at baseline were found to be significantly higher in patients with remission compared to those without remission (aTh17/Th17, 2.9 % vs 1.1 %, P = 0.02; aTreg/Treg, 34.3% vs 17%, P = 0.03) (Table 1 and Figure 3A). Regarding the established RA group, no difference was found between the patients with remission and those without remission (Table 1).
To validate the importance of aTh17 and aTreg and to identify the patient subgroups in relation to these phenotypes, we screened the key features using the Cox regression model-related method [10]. Similar to the comparison between patients with remission and without remission, aTreg and aTh17 were selected as remission-related features in the early RA group, and three subgroups were identified based on the proportions of aTreg and aTh17 (group 1, n = 4; group 2, n = 2; group 3, n = 3). We observed that all, one and none of the patients in the groups 1, 3 and 2 achieved remission, respectively (Figure 3B). Although the consensus matrix appeared to be varied among the patients of group 1 compared to matrices of other groups, the Silhouette width values supported the validity of the process of making subgroups using aTreg and aTh17 as all the patients had positive silhouette width (Figure 3C and 3D). Principal component analysis using aTreg and aTh17 suggested each subgroup had different characteristics (Figure 3E). Hierarchical clustering analysis using all parameters showed heterogeneity in the cell subpopulations and cytokine profiling except for aTreg and aTh17 (Figure 3F). When focused on correlation with change in disease activity, only the proportion of aTh17 showed significant correlation with CDAI improvement (R = -0.7, P = 0.035) (Figure 3F and 3G).
We performed the same analysis with the longitudinal data from patients at pre-treatment stage to 6 months post-treatment stage in the early RA group. The total CD4 T cells and naïve CD4 T cells (NCD4) populations significantly increased within 6 months in patients with remission compared to those without remission (5% vs -6%, P = 0.02; 14.7% vs -10.9%, P = 0.02, respectively), while the population of effector CD4 T cells (ECD4) and activated Th2 cells (aTh2) significantly decreased in the patients with remission compared to those without remission (-29.2% vs 11.7%, P = 0.01; -59.1% vs -5.9%, P = 0.03, respectively) (Table 2).
During the process of feature selection based on the Cox regression model, the changes in the total CD4 T cells, NCD4, ECD4, aTreg, aTh2 and CD86+ B cells were identified as remission-related features. The patients were divided into three subgroups, and the patients in group 1 achieved remission earlier than those in group 2 (Figure 3H). Notably, the data from all three patients in this group 1 overlapped with the pre-treatment data of patients in group 1. The consensus matrix, the Silhouette width and the principal component analysis demonstrated that the three subgroups were clearly distinguishable (Figure 3I, 3J and 3K). The heat-maps show that the total CD4 T cells, in particular NCD4, increased after treatment in group 1, and the ECD4 T cells decreased after treatment in groups 1 and 2 (Figure 3L). Regarding the correlation with disease activity, the changes in the total CD4 cells, NCD4 and NCD8 T cells, CD86+ B cells, ECD4, and aTh2 correlated with CDAI improvement (R = -0.88, P < 0.01; R = -0.87, P < 0.01; R = -0.7, P = 0.04; R = -0.72, P = 0.04; R = 0.87, P < 0.01; R = 0.72, P = 0.03, respectively) (Figure 3L). However, comparison of these parameters among patients with remission and those without remission throughout the 6 months duration after abatacept treatment initiation, levels of NCD8 and CD86+ B cells fluctuated and there was no difference at 6 months post-treatment (Figure 3M and Table 2). In accordance with the cell-surface markers considered for this study, the increase in the levels of NCD4 may reflect a decrease in the levels of ECD4.
Although the patients in the established RA group were divided into six subgroups using the pre-treatment data set and four groups using the longitudinal data set, both the Cox regression models were found to be insignificant (Supplementary Figure 2A and 2F), and the patient subgroups using selected features were not observed to make clusters in the principal component analysis (Supplementary Figure 2D and 2I). Only lower IFN-γ levels before treatment was associated with better CDAI improvement (Figure 4A and Supplementary Figure 2E).
In summary, the comparison between remission group and the non-remission group showed that aTh17 levels and aTreg levels before treatment and decrease in the levels of ECD4 and aTh2 may be used as biomarkers for evaluating the response to abatacept treatment in the early RA group. Using the Cox regression model-related method, we internally validated their efficacy to distinguish between the feature of remission and non-remission, and we found subgroups of patients based on remission-related features. In the established RA group, only IFN-γ before treatment was associated with response to abatacept treatment.
Association among key features in the early RA group
We investigated association between clinical biomarkers (Figure 2), cell subpopulations, and cytokines by visually analyzing the similarity between these features using multi-dimensional scaling plots in the early RA group (Figure 3N). Using pre-treatment data, anti-CCP titer was found to be in the positive MDS1 axis and in the negative MDS2 axis along with cytokines and functional T cell subpopulations including aTh17 and aTreg, while IgA levels and neutrophil counts were in a different cluster. MDS plot by longitudinal datasets also demonstrated that IgA levels were separated from remission-related key cell subpopulations, suggesting that IgA levels may be an independent biomarker for peripheral cell subpopulations and for cytokines. Interestingly, neutrophil count was close to the change in the levels of NCD4 and on the opposite side of the levels of ECD4, indicating that the change in the NCD4/ECD4 ratio was associated with the neutrophil count.
Correlation between IFN-γ and activated CD56bright NK cells
We had previously reported the association between inadequate response to abatacept treatment with the population of NK cells and specific-CD56 signature genes [17]. As NK cells, particularly the CD56bright subset, are potent IFN-γ producers, we hypothesized that the concentration of IFN-γ in the established RA group may be associated with the NK cell population. To confirm this, we additionally analyzed the NK cell subpopulations in the patients with established RA (Figure 4). Our analysis found that the IFN-γ levels significantly correlated with the NKG2D+ CD56bright NK cells, which is the activated CD56bright NK cell subpopulation (R = 0.43, P = 0.04) (Figure 4B and 4C). In addition, decrease in the NKG2D+ CD56bright NK cell population after abatacept treatment was found to be significantly larger in patients with remission compared to those without remission, while no difference was seen in any other NK cell subpopulations (-10.1% vs -2%, P = 0.03) (Figure 4D).