Methods

Subjects and data collection

Ninety-four patients satisfying four or more of the revised 1997 American College of Rheumatology (ACR) classification criteria for SLE12 were recruited from the University of Toronto Lupus Clinic. Demographic, clinical and treatment information for the patients and controls is shown in table 1.

Blood and clinical data were obtained, enabling calculation of disease activity using the SLE disease activity index (SLEDAI-2K).13 SLEDAI-2K scores were calculated prospectively and independently from expression assays. The mean (SD) SLEDAI-2K was 7.7 (6.6) (range 0−30, median 6). Clinically relevant high disease activity was defined as SLEDAI-2K ⩾7, given that >60% of physicians would alter treatment with scores of this magnitude.14 Persistent significant activity was defined as two or more consecutive assessments over a minimum of 3 months with a SLEDAI-2K ⩾7. The modified SLEDAI-2K was calculated by subtracting the contribution of hypocomplementaemia and anti-dsDNA positivity from the clinical score. The SLEDAI-2K descriptors were used to define active renal disease. Items necessary to calculate the Systemic Lupus International Collaborative Clinics (SLICC)/ACR index, time from last episode of sustained disease activity, and ACR criteria were obtained from the prospectively acquired database. Control blood samples (n = 11) were obtained from healthy donors with no family history of SLE.

RNA isolation and real-time PCR

Total RNA was isolated from blood archived in PAXgene tubes using the PAXgene Blood RNA Kit (Qiagen, Basel, Switzerland) with the following modifications to improve RNA yield and quality: addition of RNAse inhibitor, an off-column DNAse I digestion and final ethanol precipitation. A first-strand complementary DNA was produced using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, California, USA). Quantitative real-time PCR amplification was performed using a custom TaqMan Low Density Array (Applied Biosystems) with genes printed in duplicate. Normalisation and quantification of the PCR signals was performed by comparing the cycle threshold value of the gene of interest with GAPDH. Identification numbers for the five genes tested were Hs00356631_g1 (IFIT1), Hs00182073_m1 (MX1), Hs00192713_m1 (ISG15), Hs00158942_m1 (LY6E) and Hs00242943_m1 (OAS1).

Serological testing

Anti-dsDNA was measured by FARR assay and anti-Ro, -La, -Sm, and -RNP antibodies were measured by ELISA in the hospital laboratory.

Statistical analysis

The Mann-Whitney non-parametric test was used for comparisons between various groups. For analysis of correlations between variables, a linear regression was performed using Pearson’s correlation coefficient analysis; p values of <0.05 were considered statistically significant. Two multivariate linear regression models were built. The first model used the global interferon score as outcome variable and clinical parameters identified in preliminary analyses as predictor variables. The second model was developed to test whether the global IFN score added independent information to traditional biomarkers (C3, anti-dsDNA antibodies) in predicting clinical disease activity as measured by the modified SLEDAI-2K. All values used were obtained from the initial clinical encounter.

Results

Increased IFN-induced gene expression in patients with SLE correlates with disease activity at a single point in time

Quantitative expression of five IFN-responsive genes (LY6E, OAS1, IFIT1, ISG15 and MX1), reported to be induced in SLE,10,11 was determined for 94 patients with SLE and 11 controls. The expression of the five genes was significantly higher in SLE than in controls (fig 1A). Since expression of the five genes strongly correlated with each other by linear regression analysis, their expression levels were summed to generate a global IFN score (IFN5). This composite score was markedly higher in patients with SLE than in controls. To determine whether expression of IFN-induced genes was associated with disease activity, patients were segregated into a low/moderate (SLEDAI-2K <7) or a high (SLEDAI-2K ⩾7) disease activity group. Analysis revealed that LY6E, OAS1 and IFIT1 expression and the global composite score were significantly increased in patients with high disease activity (fig 1B). To further explore the association with disease activity, linear regression analysis was performed with the expression of LY6E (r = 0.244, p = 0.0176), OAS1 (r = 0.261, p = 0.0111), IFIT1 (r = 0.288, p = 0.0049) and the composite score (r = 0.253, p = 0.0318, fig 1C) positively correlated with the SLEDAI-2K.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Expression level of five interferon (IFN)-inducible genes (IFN-induced protein with tetratricopeptide repeats 1 (IFIT1), myxovirus resistance 1 (MX1), IFN-α-inducible protein (clone IFI-15K) (ISG15), lymphocyte antigen 6 complex, locus E (LY6E) and 2′-5′-oligoadenylate synthetase 1, 40/46 kd (OAS1)) in patients with systemic lupus erythematosus (SLE) and healthy controls. The relative quantity for each gene was calculated as described in the Methods section with normalisation of individual genes to GAPDH, a housekeeping gene. (A) Scatter plots showing the relative quantity (RQ) for each IFN-inducible gene. The global IFN score (RQ IFN5) was calculated by summing the individual RQ scores for each gene. (B) Patients were segregated into low and high disease activity groups based on the SLEDAI-2K, with a score ⩾7 defined as high disease activity. The RQ for each individual gene in the two patient populations is shown. Each symbol represents the determination from a single individual, with the mean value for each group indicated by a horizontal line. For (A) and (B), significance levels were determined by the Mann-Whitney non-parametric test. Statistically significant p values (<0.05) are shown. (C) Linear regression analysis of the global IFN score versus the SLEDAI-2K score.

Longitudinal study of the association between changes in the IFN score and disease activity

Given the association between disease activity and the global IFN score, a longitudinal study was performed to determine whether changes in disease status are reflected in the IFN5 score. Twenty-seven patients had at least one subsequent assessment with follow-up encounters at least 3 months and no more than 12 months from initial enrolment. The changes in SLEDAI-2K and IFN5 indices were calculated by subtracting the initial score from that obtained at the following encounter. Thirteen patients had a marked change in their |SLEDAI-2K| of ⩾6 (relapse/remission), with this fluctuation considered clinically meaningful since it would prompt most clinicians to alter treatment.15 Fourteen patients remained clinically stable with a change in the |SLEDAI-2K| of ⩽2. Of the patients with a marked clinical change, the majority (n = 10) had a minimal fluctuation (⩽1SD) in their global IFN score (fig 2A). This immutability of the global IFN score was independent of clinical disease exacerbation or remission. The change in the IFN score (ΔIFN) had no predictive utility since no clear correlation was noted between the ΔIFN and the magnitude of the change in the clinical index (ΔSLEDAI-2K) by linear regression analysis (fig 3A). Of the three patients with a major change in their global IFN scores, one had an increase in their IFN signature despite marked clinical improvement, while the remaining two had changes in their IFN5 scores that paralleled those in their clinical status.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Longitudinal changes in the global interferon (IFN) score compared with changes in clinical activity. All patients were assessed on a yearly basis with a minority of individuals, enrolled in an intensively studied cohort, assessed every 3 months. Patients enrolled in this latter cohort were selected on the basis of limited disease duration and significant disease activity at the time of recruitment. This strategy was designed to monitor patients likely to have significant changes in disease activity within the next 5 years. The global IFN score was calculated for 27 patients with more than one clinical assessment. Changes in the systemic lupus erythematosus disease activity index (ΔSLEDAI-2K) and IFN5 (ΔIFN5) were calculated by subtracting the score at the initial visit from that obtained at follow-up. For (A) and (B) the directionality of the bar indicates the net change in the global IFN signature. The change in the SLEDAI-2K for each individual patient is represented above the corresponding bar with a gain (+) or loss (−) indicated. (A) Change in global IFN score for 13 patients with marked changes in their clinical index (Δ|SLEDAI-2K| ⩾6). (B) Change in global IFN score for 14 patients with minimal change in their clinical activity (Δ|SLEDAI-2K|⩽2). (C) Longitudinal tracking of the IFN5 signature and SLEDAI-2K for three patients. Lower panels show the IFN5 for the initial and two subsequent assessments; the upper panel shows the corresponding SLEDAI-2K. RQ, relative quantity.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Lack of correlation between changes in the global interferon (IFN) score and changes in (A) the clinical disease index (SLEDAI-2K), (B) complement (C3) and (C) anti-dsDNA antibodies by linear regression analysis. The global IFN score was calculated for 27 patients with more than one clinical assessment. The change in the interferon score (ΔIFN5) was calculated by subtracting the initial score from the score obtained at the subsequent visit. ΔSLEDAI-2K, ΔC3 and ΔdsDNA antibodies were calculated for all clinical assessments with the Δ variable defined as the value at the initial visit subtracted from the value obtained at the subsequent encounter.

For the majority of patients with stable disease (n = 12) the global IFN score did not fluctuate between clinical assessments. Examination of the two exceptions revealed that one patient was active and had a marked decrease in the IFN5 score without a commensurate change in the clinical index (SLEDAI-2K = 20 to 18), while the other patient was clinically inactive throughout the study period (SLEDAI-2K = 0 to 2) despite a substantial change in the IFN5 score. These findings highlight the lack of association between the IFN score and rapid variations in clinical disease.

Three patients had two encounters following their initial assessment. In two of these patients, normalisation of their SLEDAI-2K did not impact on the global IFN5 score, which remained static over the study period (fig 2C). The third patient had an increase in SLEDAI-2K from an initial score of 0, through a score of 2 (6 months), culminating in a score of 12 (9 months). The global IFN score did not mirror this clinical exacerbation with the 6-month assessment showing a marked increase in the IFN signature, although the clinical change was minimal. Although this transient increase in the IFN5 score may be interpreted as heralding the clinical exacerbation, this conclusion is not supported by the decrease at the time of overt disease.

Association between clinical and serological variables and the IFN5 score

Analysis of the global IFN score stratified for the presence or absence of serological variables revealed a statistically significant association between an increased IFN signature and the presence of anti-RNP, anti-Sm or any anti-RNA binding protein (RBP) antibody (table 2).

The presence of anti-dsDNA antibodies was also associated with an increased IFN5 score, with a positive correlation between the global IFN score and the level of anti-dsDNA antibodies demonstrated by linear regression analysis (r = 0.27, p = 0.01).

A decreased C3 was significantly associated with an increased IFN5 score, although no correlation between C3 levels and the global IFN score was noted by linear regression analysis (r = −0.16, p = 0.12). Since decreased C3 and anti-dsDNA positivity are associated with an increased global IFN score, we examined whether longitudinal fluctuations in the global IFN score correlated with changes in the level of these two serological variables. Fluctuations in neither variable correlated with changes in the IFN5 score (fig 3B, C).

A multivariate linear regression analysis was performed to explore the association between variables significantly associated with the IFN5 score (eg, SLEDAI-2K, C3, anti-dsDNA, anti-RNP or anti-Sm positivity, use of immunosuppressives). As the levels of C3 and anti-dsDNA were highly correlated (r = −0.59, p<0.001), they were not used simultaneously. The only variable found to be associated with the IFN5 score was the use of immunosuppressive agents. Patients on immunosuppressives had, on average, 22 more units in their IFN5 score with a wide 95% confidence interval (2 to 43 units).

A modified SLEDAI-2K score was calculated to remove the impact of anti-dsDNA positivity and decreased C3 on the correlation between the global IFN score and the clinical index. Linear regression analysis showed a positive correlation between the modified clinical index and the global IFN score (r = 0.23, p = 0.026). Multivariate linear regression analysis to determine whether the global IFN score independently contributes to the modified SLEDAI-2K after accounting for the C3 and anti-dsDNA levels was performed. This model revealed that C3 levels were significantly associated with the modified SLEDAI-2K (p<0.001), with similar findings noted when anti-dsDNA replaced C3. The IFN5 score was not independently associated with the modified SLEDAI-2K when either C3 or dsDNA were included in the model.

Analysis of patients with severe active renal disease (⩾3 SLEDAI-2K renal descriptors) showed a statistically significant association with the IFN5 score (table 2), and linear regression analysis showed a positive correlation between the number of renal SLEDAI-2K criteria and the IFN signature (r = 0.253, p = 0.014). Although a positive association between increased expression of LY6E and proteinuria has been reported,11 we found no association between proteinuria and LY6E expression or the IFN5 score. To further explore this issue, 29 patients with a 24 h urinary protein determination performed within 2 weeks of the clinical assessment were identified. Linear regression analysis again showed no association between this quantitative urinary protein measurement and LY6E or global IFN expression.

Since fluctuations in disease activity were not reflected by alterations in the IFN5 score, we questioned whether the increased IFN5 score associated with renal disease persists in patients achieving clinical remission. From our initial cohort, 34 patients had low/moderate disease activity with no recent (<2 years) history of persistent disease activity. Comparison of patients with (n = 19) and without (n = 15) a history of nephritis did not show a statistically significant difference (fig 4A). This finding suggests that the IFN5 score in patients with nephritis declines over time to the levels noted in patients without a history of nephritis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Correlation between the global interferon (IFN) score and a history of nephritis or timing of the last disease flare. (A) Scatter plot comparing the global IFN signature in patients with and without a history of nephritis. Thirty-four patients with no recent history of disease flare (<2 years) were stratified on the basis of a prior positive (LN) or negative (No LN) history of nephritis. (B) Fifty-three patients were stratified on the basis of their global IFN signature. The highest quartile (HQt, n = 13) and lowest quartile (LQt, n = 13) were selected and plotted against the time from their most recent flare. The mean value for each group is shown by a horizontal line. Significance levels were determined by Mann-Whitney non-parametric testing. Statistically significant p values (<0.05) are shown.

Given our inability to detect this fluctuation, we hypothesised that changes in the IFN5 score are not coincident with rapid changes in disease activity but rather occur over longer periods of time. We postulated that higher IFN5 scores would be associated with a history of recent sustained high disease activity when compared with lower scores. To examine this question, 53 patients with low/moderate disease activity for whom sufficient retrospective clinical data were available were stratified into quartiles based on their IFN5 score. Patients in the highest quartile had a significantly shorter mean (SD) time from the last episode of sustained high disease activity than patients in the lowest quartile (1.58 (1.52) years vs 6.69 (8.34) years; fig 4B). Linear regression analysis also showed a negative correlation between the IFN5 score and the time from last period of sustained disease activity (r = −0.32, p = 0.021). Forty-eight patients with low/moderate disease activity at the initial assessment were segregated into thirds (n = 16) on the basis of IFN5 score and followed prospectively (low IFN: mean 21.9 months (range 17–27); high IFN: mean 19.0 months (range 16–29)). A high IFN score did not predict an imminent flare, with only a single episode of disease exacerbation occurring within 6 months of assessment. However, patients with raised scores were more likely to flare (high IFN, 5 flares vs low IFN, 0 flares; p = 0.043, Fisher exact test) when tracked over longer periods of time. These observations suggest that raised IFN scores identify patients at risk of disease exacerbations but do not fluctuate in time with disease activity.

Discussion

In agreement with previous findings, the peripheral blood expression level of five IFN-responsive genes was found to be preferentially expressed in patients with SLE compared with controls.10,11 The association between an increased global IFN score and disease activity, renal involvement, decreased C3 and the presence of anti-dsDNA or anti-RBP antibodies was also confirmed. Thus, overexpression of IFN-induced genes is a dominant and reproducible feature of SLE.

Although the IFN5 score correlates with disease activity, the candidacy of this expression signature as a biomarker remains in question. In multivariate analysis, the IFN signature did not perform as well as traditional biomarkers in predicting clinical disease activity as defined by the modified SLEDAI-2K. No correlation was noted between changes in the IFN5 score and changes in accepted biomarkers of disease activity (C3, anti-dsDNA antibodies).2 Furthermore, although an increased IFN signature correlated with disease activity at a single point in time, longitudinal analysis revealed that it does not fluctuate with disease activity, remaining static despite marked clinical changes. Given the limited number of patients assayed longitudinally, the potential for a type II error exists. However, the observation that traditional biomarkers fluctuated concordantly with disease activity in our study, whereas the global IFN score did not, suggests that the IFN5 score does not outperform current biomarkers. Taken together, these observations suggest that the IFN5 score is unsuitable as a biomarker to predict imminent disease flares.

Although the IFN5 score was not synchronous with acute changes in disease activity, our results suggest that the IFN signature can fluctuate over longer periods of time and that an increased IFN5 score predisposes to disease exacerbations. This conclusion is supported by the following observations. There was a negative correlation between time from last episode of sustained disease activity and the IFN signature, suggesting that normalisation of the IFN5 score occurs over a timeline that extends beyond the presence of symptoms. Furthermore, patients with a higher IFN5 score were more likely to have a disease exacerbation remotely during the subsequent follow-up period. Finally, although active renal disease is associated with significantly increased IFN5 scores, no difference in the IFN5 score was noted between quiescent patients with and without a history of nephritis, arguing that the IFN5 score in patients with nephritis must eventually decline to the levels noted for patients without nephritis. Taken together, these observations suggest that an increased IFN5 score is linked to disease flares and gradually normalises over time.

Models of the pathogenesis of SLE propose that engagement of Toll-like receptors by snRNPs stimulates plasmacytoid dendritic cells (PDCs), increasing production of type I IFN.16,17 Thus, long-lived anti-RBP antibodies would promote activation of PDCs and production of IFNα.18,19 In keeping with this model, plasma from patients with SLE has been shown to induce expression of IFN-regulated genes, and the level of expression correlates with the titre of anti-RBP antibodies.20 These observations are compatible with the slowly fluctuating nature of the IFN signature in which persistence of antibodies, beyond the disease flare, would support IFN production. An association between anti-Sm antibodies and lupus nephritis has been noted.21 We and others have reported a correlation between increased IFN scores and lupus nephritis.11 These data suggest that anti-RNA antibodies promote IFN production, creating a biochemical environment that supports renal disease. This contention is supported by the observation that, in patients with a high IFN signature, there is a dysregulation of IFN-mediated chemokines that are associated with lupus nephritis.22

Increased IFN levels appear to support disease flares, although fluctuations in the IFN signature do not mirror rapid clinical changes. Perhaps a robust IFN signature supports a pro-flare environment but is insufficient to trigger acute disease exacerbation. It is likely that additional factors precipitate disease flares in the context of IFN-mediated biochemical changes. This conclusion is in keeping with the discordance between the beneficial impact of treatment on symptoms and its lack of effect on the IFN5 score, and suggests that current immune modulators probably impact on factors that promote disease activity without modulating the IFN signature.