Modular transcriptional repertoire analyses of adults with systemic lupus erythematosus reveal distinct type I and type II interferon signatures

Abstract

Objective: The role of interferon-α (IFNα) in the pathogenesis of systemic lupus erythematosus (SLE) is strongly supported by gene expression studies. The aim of this study was to improve characterization of the blood IFN signature in adult SLE patients.

Methods: Consecutive patients were enrolled and followed up prospectively. Microarray data were generated using Illumina BeadChips. A modular transcriptional repertoire was used as a framework for the analysis.

Results: Our repertoire of 260 modules, which consisted of coclustered gene sets, included 3 IFN-annotated modules (M1.2, M3.4, and M5.12) that were strongly up-regulated in SLE patients. A modular IFN signature was observed in 54 of 62 patients (87%) or 131 of all 157 samples (83%). The IFN signature was more complex than expected, with each module displaying a distinct activation threshold (M1.2 < M3.4 < M5.12), thus providing a modular score by which to stratify SLE patients based on the presence of 0, 1, 2, or 3 active IFN modules. A similar gradient in modular IFN signature was observed within patients with clinically quiescent disease, for whom moderate/strong modular scores (2 or 3 active IFN modules) were associated with higher anti-double-stranded DNA titers and lower lymphocyte counts than those in patients with absent/mild modular scores (0 or 1 active IFN modules). Longitudinal analyses revealed both stable (M1.2) and variable (M3.4 and M5.12) components of modular IFN signature over time in single patients. Interestingly, mining of other data sets suggested that M3.4 and M5.12 could also be driven by IFNβ and IFNγ.

Conclusion: Modular repertoire analysis reveals complex IFN signatures in SLE, which are not restricted to the previous IFNα signature, but which also involve IFNβ and IFNγ.

Trial registration: ClinicalTrials.gov NCT00920114.

Copyright © 2014 by the American College of Rheumatology.

Figures

Figure 1. Modular repertoire analysis of SLE patients compared to healthy controls

A: Modular analysis at the group level. Samples from SLE patients (n=157) are compared to matched healthy controls (n=20). Each module is assigned a position on the grid. The percent difference between probes significantly upregulated and down-regulated within each module determines the color and intensity of the spot (red for upregulation, blue for down-regulation in SLE). Modules annotations are provided in the second grid. A strong upregulation in SLE was observed for 4 modules, including the 3 IFN-annotated modules M1.2 (100% probes upregulated in SLE patients), M3.4 (90% upregulated, 0% downregulated) and M5.12 (78% upregulated, 0% downregulated).B Modular IFN signature at the individual level. Each sample from SLE patients at inclusion (n=62) is compared to the average of healthy controls (n=20). The percent difference between probes up- and down-regulated (FC ≥ 2 or ≤1/2 and difference ≥ 100) determines the color and intensity of each IFN module for each sample. Modular IFN signature was present (at least one active IFN module, i.e. percent difference ≥ 20) in 54/62 SLE patients. A modular IFN score was assigned to samples (“absent”, “mild”, “moderate”, or “strong”) according to the number (0, 1, 2 or 3) of active IFN modules.

Figure 2. Repartition of SLE samples according to modular IFN score

A: The 157 SLE samples were ordered and classified according to the modular IFN score (number of active IFN modules, from 0 to 3). IFN signature was “absent” in 26 samples, “mild” in 17 samples, “moderate” in 68 samples and “strong” in 46 samples. A dynamic IFN signature, from M1.2 to M3.4 and M5.12, was observed: when only one of the three IFN modules was up-regulated, it always corresponded to M1.2. Module M3.4 appeared next, and there were no M5.12 modules up-regulated in the absence of the two others. B: The same dynamic IFN signature was observed in 2 independent cohorts of SLE patients: a cohort of 82 pediatric patients (19), with Hispanic (57%), Black (23%), White (15%) and Asian (5%) ethnicities, and a cohort of 43 adult patients (21), with Black (54%), White (44%) and Asian (2%) ethnicities. On the contrary, although a modular IFN signature was observed in active tuberculosis (19), no such gradient was observed in IFN modules.

Figure 3. Longitudinal intra-individual variation of IFN modules in SLE patients

Longitudinal analyses were obtained for 29 SLE patients with at least 3 consecutive visits. A: The level of upregulation of each IFN module at each visit is plotted on the circos figure, representing from center to periphery M1.2, M3.4, M5.12 and time elapsed since 1st visit. Spaces separate different patients (e.g., 3 visits of the same patient are framed by the dotted line). B: Coefficient of variation (mean CV ± SD), corresponding to intra-individual variability of IFN modules, indicates that while M1.2 is stable over time for a given patient (CV = 0.05 ± 0.88), M3.4 (CV = 0.39 ± 0.56) and even more M5.12 (CV = 0.91 ± 0.82) show fluctuations across time, reflecting the complexity of IFN signature. These differences of variability between modules are significant (M1.2 CV vs. M3.4 CV, p= 0.0033; M1.2 CV vs. M5.12 CV, p = <0.0001; M3.4 CV vs. M5.12 CV, p = 0.00065).

Figure 4. Effect of different types of IFN on IFN modular patterns

Modular analysis was performed on gene expression data from 2 public domain datasets (22,23). A: At the individual level, treatment with IFN-α results in the upregulation of M1.2 only, while treatment with IFN-β is associated in most patients with upregulation of both M1.2 and M3.4, as well as M5.12 in some patients. B: The responsiveness to different types of IFN of the genes from the 3 IFN modules was evaluated using the Interferome database: the log2(FC) observed in each experiment for each gene after in vitro stimulation was compared between type I and type II IFN, as well as between IFNα and β and represented on the Beeswarm plots. Transcripts belonging to M1.2 were induced significantly more by type I than type II IFN (median log2(FC) of 4.10 vs 2.60, ***p<0.0001), while transcripts belonging to M3.4 and M5.12 were similarly induced by type I and type II IFN (respectively, 2.60 vs 2.54, p=0.87 and 2.01 vs 2.04, p=0.68). In addition, M1.2 and M3.4 transcripts were induced significantly more by IFN-β than by IFN-α (5.39 versus 3.78, **p=0.0001 and 2.74 versus 2.40, *p=0.034 respectively), while transcripts belonging to M5.12 only exhibited a non significant trend (2.07 versus 1.89, p=0.051).

Figure 5. Inter-scores correlations in SLE patients

A: Modular IFN score (absent/mild/moderate/strong) is compared to 4 scores from the literature using the 157 SLE samples of the present cohort. Linear model shows an increase in the Yao score of 2.9 points per category of modular IFN score (p<0.0001), an increase in the Petri score of 1.85 points per category (p<0.0001) and an increase in the Feng score of 9.14 points per category (p<0.0001). Both IFN-α and IFN-γ Kirou’s scores were linked (chi-squared test for trend in proportions p<0.0001) with modular IFN score. B: Petri IFN score in 29 patients with ≥ 3 visits (longitudinal cohort) across time. A limited inter-individual variability of the Petri IFN score across time is observed (coefficient of variability = 0.07). This score is based on the expression of 3 IFN-inducible genes (IFI27, IFI44 and OAS3).