Objective: To determine whether peripheral blood gene expression of patients with systemic lupus erythaematosus (SLE) correlates with disease activity measured using the SLE Disease Activity Index 2000 (SLEDAI-2K).
Methods: RNA was isolated from peripheral blood of 269 patients with SLE and profiled on a custom microarray. Hierarchical clustering and a heat map were used to categorise samples into major clusters based on gene expression pattern. Correlates, including demographic and disease-related characteristics such as SLEDAI-2K score, of the major sample clusters were compared using multivariate regression models.
Results: A set of 31 interferon (IFN)-regulated genes were seen to be driving the separations of samples into two clusters, one characterised by a relatively high IFN-regulated gene signature (n = 150) and the other by a relatively low IFN-regulated gene signature (n = 119). Disease activity measured using the SLEDAI-2K was significantly correlated with the high IFN gene signature. In multivariate regression analysis the immunological component of the SLEDAI-2K was a significant correlate of the high IFN gene signature as was presence of antibodies to U1RNP. There were no discernable correlates of the 156 non-IFN regulated genes profiled on the custom array.
Conclusion: Peripheral blood gene expression profiling (GEP) in SLE allows patients to be categorised into two groups based on a high or low IFN gene signature. Disease activity measured using the SLEDAI-2K is correlated with the high IFN gene signature, indicating that GEP may be a useful biomarker of disease activity in SLE.
Statistics from Altmetric.com
Systemic lupus erythaematosus (SLE) is a heterogeneous multi-system autoimmune disease characterised by intermittent episodes of increased disease activity that require treatment with immunosuppressive agents.1 The current approach to assessment of disease activity in SLE incorporates measurement of multiple clinical and laboratory variables.2 3 Anti-dsDNA antibody level measured using radioimmunoassay is arguably the best laboratory marker of disease activity currently available. However, anti-dsDNA antibodies are present in only 70% of patients at diagnosis and up to 25% of patients have a clinically and serologically discordant course.4–8 Therefore there is a need for more sensitive biomarkers of SLE disease activity.
Several studies using oligonucleotide microarrays have shown differences in white blood cell gene expression profiles of patients with SLE compared with healthy controls as well as patients with other autoimmune diseases such as multiple sclerosis and rheumatoid arthritis.9–14 Many of the genes that are upregulated in patients with SLE are involved in type I interferon-mediated immune responses and comprise the interferon (IFN) gene signature.15–17 While gene expression profiling (GEP) has provided insight into the genetic mechanisms of SLE pathogenesis and possible therapeutic targets, there is also emerging evidence that peripheral blood gene expression patterns enable differentiation of patients with SLE with active disease from those with inactive disease.10 11 18–20 Baechler et al identified 161 genes that were differentially expressed in peripheral blood mononuclear cells (PBMCs) of 48 patients with SLE compared to 42 healthy controls.11 Many of these genes had known or suspected roles in the immune response. Importantly, the IFN regulated gene cluster was preferentially expressed in patients with SLE, and the INF gene score correlated with the number of SLE disease criteria and with the presence of renal or CNS disease. Similarly, Kirou et al compared the expression of 3 interferon-inducible genes in PBMCs from 77 patients with SLE, 22 disease controls and 28 healthy donors using RT-PCR. They found that activation of the IFNα pathway was associated with increased disease activity including renal disease.20 Although many of the findings have been consistent across studies, to date sample sizes have been relatively small and there have been differences among studies in the correlation between gene expression pattern and specific clinical features or autoantibody profiles.
In this study of a large sample size of 269 patients with SLE, our objectives were to determine whether peripheral blood gene expression profiled using microarray technology correlates with disease activity measured using the SLE Disease Activity Index 2000 (SLEDAI-2K), and to determine the demographic and disease correlates of major gene expression clusters using multivariate regression models.3
Patients and collection of clinical data
We recruited 269 consecutive patients attending the University of Toronto Lupus Clinic for follow-up over a 6-month period. All participants fulfilled four or more of the 1971 or 1982 American College of Rheumatology (ACR) classification criteria for SLE, or had three criteria and a typical lesion of SLE on renal or skin biopsy.21 22 The study was approved by the research ethics board of the University Health Network and written informed consent was obtained from all participants. At recruitment patients underwent a detailed clinical evaluation where all clinical data required for determination of SLEDAI-2K score were obtained as well as data on medications and doses.
Blood samples and RNA isolation
Whole peripheral blood (10 ml) was collected from each patient using the PAXGene blood RNA collection system (Qiagen, Valencia, California, USA) and stored at −70°C. RNA was purified in batches using the PAXGene blood RNA kit or the PAXGene 96 blood RNA kit (Qiagen). To prepare for microarray hybridisation, the extracted RNA was converted to cDNA, in vitro transcribed to cRNA in the presence of amino-allyl UTP and labelled with Cy5 fluorochrome using dye-coupling (Amino Allyl MessageAmp Kit, Ambion, Austin, Texas, USA).
Additional serum samples were simultaneously obtained from patients and tested for blood cell counts, C-reactive protein (CRP), creatinine, complement 3 (C3) and 4 (C4), autoantibodies including anti-dsDNA, anti-Ro, anti-La, anti-Sm, anti-U1RNP as well as anti-phospholipid antibodies. Anti-dsDNA antibodies were measured using radioimmunoassay. Other antibodies were measured using commercial ELISA kits. CRP was measured using an immunoturbidimetric assay with high sensitivity in low range. Urinalysis was also performed. SLEDAI-2K scores were determined for each patient using relevant clinical and laboratory data.3 The Systemic Lupus International Collaborative Clinics (SLICC)/ACR Damage Index (SDI) was also recorded.23
The genes represented on the custom array were selected from an initial pilot study comparing white blood cell pan-genome expression profiles generated on Affymetrix U133 GeneChips from 72 SLE, 26 rheumatoid arthritis, 27 Wegener granulomatosis and 28 healthy samples retrieved from the Gene Logic database (GeneLogic, Gaithersburg, Maryland, USA). Genes with 1.8-fold or more increased expression in patients with SLE compared with comparison groups were included in the focus array as well as genes related to disease activity and genes previously shown to be important in SLE susceptibility and pathogenesis10 24 Overall, the focus array was comprised of 423 oligonucleotide probes (mean length of 42 nucleotides), representing 329 unique genes, and 20 control probes. Quantitative RT-PCR was used to validate results generated on the custom microarray and to confirm results obtained from Affymetrix microarrays. Gene expression profiles generated on the custom microarray were shown to be reproducible, with an inter-assay coefficient of variation of 15%.25
Data processing, quality filtering and normalisation
One microgram of Cy5-labelled cRNA target was hybridised to the microarray (42°C for 18–20 h) and the resultant hybridised array imaged using a Tecan LS200 scanner (Tecan, San Jose, California, USA). All assays were performed in duplicate. The fluorescent signal was extracted for each probe–target hybrid using Imagene V.5.6 imaging software (BioDiscovery, El Segundo, California, USA). Quality filtering was performed such that probe spots that had a background corrected signal intensity of less than one standard deviation of the local background signal intensity were considered uninformative. All spots were normalised to the median signal intensity of a set of pooled probes representing the housekeeping genes β-actin and GAPDH. Gene probes were selected for subsequent analysis if at least 20% of the samples had probe spots that passed quality filtering. This resulted in 337 of the 423 probes being selected for subsequent data analysis, corresponding to 256 unique genes.
Unsupervised hierarchical clustering was used to identify gene expression patterns that identified subsets of patients with SLE. Spearman correlation with average linkage was used to perform hierarchical clustering on the samples and probes, and a heat map was generated to visualise the relationship between the samples using Cluster 3.0 and Java Treeview 126.96.36.199
A smaller set of 37 highly correlated gene probes (Spearman rho >0.75) representing 31 unique genes were selected to refine the clustering of the samples following the same methodology (table 1). The results from this cluster were used to categorise a sample as either having a low or high level of expression of IFN-regulated genes. This binomial classification was then used for comparison of patient demographics and clinical characteristics.
Correlates, including demographic and disease-related characteristics of the major gene clusters were compared using univariate and multivariate statistical models. Categorical data were analysed using the two-tailed Fisher exact test and continuous variables were compared using the Mann–Whitney U test; multivariate analysis was performed using logistic regression (SAS JMP 6.0, Cary, North Carolina, USA).
Among the 269 study participants, 241 (89.6%) were female and 28 (10.4%) male. There were 194 (72.1%) Caucasian, 32 (11.9%) African–American, 29 (10.8%) Asian and 14 (5.2%) “other” ethnicities. Mean (SD) age and disease duration were 44.5 (13.7) and 16.1 (9.8) years respectively. Mean SLEDAI-2K score was 4.4 (4.20). Mean SLICC/ACR damage index (SDI) was 1.7 (1.94). Mean prednisone dose at the time of study was 10.3 (7.6) mg daily. In all, 178 (66.2%) patients were on antimalarials and 117 (43.5%) were on other immunomodulatory agents including methotrexate (n = 31), azathioprine (n = 64) and mycophenolate mofetil (n = 25). Two patients were taking ciclosporine and one patient was receiving monthly intravenous cyclophosphamide at the time of the study. Six patients were taking more than one immunomodulatory agent.
Hierarchical clustering resolved two major sample clusters; SLE samples characterised by a relatively high type 1 IFN-regulated gene signature (n = 150) and SLE samples characterised by a relatively low type 1 IFN-regulated gene signature (n = 119) (fig 1A). However, there were no major sample clusters based on the expression of 156 non-IFN-regulated genes. Further refinement of clustering using a set of 37 gene probes representing 31 unique genes was performed and confirmed the presence of two distinct clusters (fig 1B).
Univariate comparison of clinical and demographic variables in low and high IFN-regulated gene signature groups is summarised in table 2. Patients in the high IFN-regulated gene signature group were generally younger (40.7 (12.8) vs 49.3 (13.3) years, p<0.001), with shorter disease duration (14.6 (9.2) vs 18.0 (10.3) years, p = 0.006) than patients with low IFN-regulated gene signature. Patients with a high IFN-regulated gene signature had significantly greater mean disease activity (SLEDAI-2K score 5.11 (4.15) vs 3.47 (4.10), p<0.001) than patients with a low IFN gene signature. Although SLEDAI-2K scores of 5 or less were equally associated with high or low IFN gene signature, SLEDAI-2K scores of 6 or more (reflecting moderate to severe disease activity) were mostly associated with the high IFN gene signature (fig 2). A significantly greater proportion of patients in the high IFN gene signature group had increased anti-dsDNA antibodies (50.7% vs 23.5%, p<0.001) or low complement (56.0% vs 28.6%, p<0.001). Overall 105 (70%) patients in the high IFN gene signature group and 51 (42.9%) patients in the low IFN gene signature group had immunological manifestations of SLEDAI-2K (increased anti-dsDNA and/or low complement). Anti-U1RNP antibodies were more common in patients with high IFN gene signature (57.3% vs 42.0%, p = 0.01) but there was no difference overall in the proportion of patients positive for one or more antibodies to RNA binding proteins, including U1RNP, Ro, La and Sm.
Musculoskeletal manifestations defined as arthritis or myositis at the time of study were more common in patients with high IFN gene signature (15.3% vs 8.4%) as was skin (32.0% vs 21.9%) and renal (18.7% vs 14.3%) involvement at the time of study. However, these differences did not reach statistical significance. Although a larger proportion of patients in the high IFN gene signature group were taking prednisone at the time of the study (62.4% vs 37.6%, p = 0.01), there was no significant difference in mean daily dose of prednisone or percentage of patients taking antimalarial or immunosuppressive therapy between the low and high IFN gene signature groups.
Table 3 is a summary of the multivariate logistic regression analysis comparing the low and high IFN-regulated gene signature groups. In this model, SLEDAI-2K, age and disease duration were entered as continuous variables while sex, race, corticosteroids, antimalarials and immunosuppressives were entered as categorical variables. SLEDAI-2K score, age and prednisone use were found to be significant correlates of high IFN-regulated gene signature. For every one-point increase in SLEDAI-2K score, the odds of having the high IFN-regulated gene signature was seen to increase by 7.3%. For every 1 year increase in age the odds of having the high IFN-regulated gene signature was seen to decrease by 4.7%. Treatment with prednisone was associated with 40.3% increase in odds of having the high IFN gene signature.
Multivariate regression analysis was repeated to include individual components of SLEDAI-2K score as categorical variables, CRP level and autoantibody profile, as well as other variables previously entered, with the exception of the total SLEDAI-2K score. In this model the presence of the immunological component of SLEDAI-2K was seen to increase the odds of having the high IFN gene signature by 60%, while the presence of anti-U1RNP antibodies increased the odds of having the high IFN gene signature by 60.4% (table 4). In this model age retained its significance.
We sought to examine the association between the gene expression profile of patients with SLE, determined using microarray technology, and disease activity measured using the SLEDAI-2K. We demonstrated that peripheral blood gene expression profiling in SLE allows patients to be categorised into groups based on high or low interferon-regulated gene expression signature. The custom microarray used in this study included IFN-regulated and non-IFN-regulated genes, which we had previously found to be differentially expressed in patients with SLE compared with healthy and disease controls.24 However, there were no discernable sample clusters based on the expression of non-IFN-regulated genes in this study. Furthermore these non-IFN-regulated genes did not represent any major functional gene families. Although all 100 IFN-regulated genes in our custom array were seen to be coordinately expressed, in refined hierarchical clustering we used 31 of these genes to define major sample clusters. As the genes profiled on the custom array were selected from an initial study comparing pan-genome expression profiles of patients with SLE with healthy as well as RA and Wegener granulomatosis disease controls, some inflammatory response genes that were upregulated in patients with SLE and disease controls may have been excluded. However, care was taken to include genes previously shown to be important in SLE pathogenesis and disease activity.
In contrast to studies previously published where either white blood cell (WBC) or PBMC gene expression profiles of patients with SLE were compared with controls, in this study we performed microarray GEP on whole peripheral blood. Intercurrent viral infections as well as certain vaccinations and medications have been shown to induce the expression of IFN-regulated genes.27–29 The effect of these potential confounders needs further evaluation in a longitudinal study.
Since the study participants were consecutive patients seen in the lupus outpatient clinic, overall mean disease activity was relatively low, with mean SLEDAI-2K score of 4.4 (4.2). In our study, the difference in mean SLEDAI-2K score between patients with high and low IFN regulated gene signature was 1.64. Although seemingly small, this difference must not be trivialised in view of the fact that mean disease activity among all participants was relatively low. Our participants’ SLEDAI-2K scores ranged from 0 to 21, with a median score of 4. Although SLEDAI-2K scores of 5 or less were equally associated with high or low IFN gene signature, SLEDAI-2K scores of 6 or more, reflecting moderate to severe disease activity were mostly associated with the high IFN gene signature.
With the exception of the immunological component of SLEDAI-2K, there was no significant association between the organ-specific components of SLEDAI-2K and the high IFN gene signature. This finding contrasts with that of other studies that have shown an association between presence of SLE renal disease and the IFN gene signature.11 20 One possible explanation for the lack of such association in our study is that although the clinical features exhibited by the patients in this study were representative of patients with SLE at large, there were relatively few patients with each of the individual organ manifestations of disease activity. Alternatively, the high IFN gene signature may be reflective of disease activity overall rather than specific for any individual organ manifestation.
In multivariate analysis, the immunological component of SLEDAI-2K was a significant but not the sole correlate of high IFN-regulated gene expression, with 45 (30%) patients in the high IFN gene signature group having no evidence of raised anti-dsDNA antibody or low complement. Of these 18 (12%) were clinically active serologically quiescent (CASQ), defined as presence of one or more non-immunological components of SLEDAI-2K along with normal serology at the time of the study and in the two prior visits within the previous 9 months. This finding suggests that microarray GEP may have a role in assessment of disease activity in patients with SLE that display clinical–serological discordance. Overall 12% of patients with SLE have a clinically active, serologically quiescent course, while an additional 12% are serologically active in times of clinical quiescence. Furthermore, laboratory markers of SLE disease activity such as anti-dsDNA antibodies, C3 and C4 level often do not display a close temporal relationship with clinically detectable change in disease activity. In general, assessment of disease activity is more difficult in serologically discordant patients as reliance is placed solely on clinical manifestations, which may at times be non-specific and difficult to interpret. Additional information regarding disease activity obtained from GEP could enable more accurate assessment of disease activity in patients with SLE and result in more appropriate and timely change to treatment.5–8
High dose corticosteroids have previously been shown to suppress the expression of IFN-regulated genes.30 Overall our patients were on a relatively low mean dose of prednisone (10.3 (7.6) mg daily) at the time of the study, and there was no statistically significant difference in mean prednisone dose between patients in the high and low IFN gene signature groups. Despite these observations, in multivariate analysis, prednisone use was a significant correlate of the high IFN gene signature. However, in the second of the multivariate analyses the contribution of prednisone to the model was overshadowed by the immunological component of SLEDAI-2K and anti-U1RNP antibodies, with prednisone losing its significance. Overall, the likely explanation for these findings is the close association between prednisone use and presence of active disease.
The association between antibodies to U1RNP and the high IFN gene signature is intriguing. To date there have been no studies linking anti-U1RNP antibodies to disease activity or severity in SLE. Kirou et al found an association between anti-RNA binding protein (RBP) antibodies and high IFNα gene score.20 They speculated that the subgroup of patients with SLE with anti-RBP antibodies may be a distinct subset of SLE with a common underlying pathogenetic mechanism that involves autoreactivity to RNA-associated proteins and activation of the type I IFN pathway. However, in our multivariate analyses, in contrast to the study by Kirou et al, we found that in addition to anti-U1RNP antibodies, anti-dsDNA antibodies were also significantly associated with high IFN gene signature, suggesting that the IFN gene signature may reflect a subgroup with greater propensity to autoantibody production in general.
In this study younger age was correlated with the high IFN gene signature. There are several possible explanations for the seemingly protective effect of older age. In general, newly diagnosed younger patients tend to have greater disease activity than older patients with established disease.31 32 However, in our analyses the effect of age was seen independently of disease duration. Alternatively, it is possible that the IFN gene signature may be a better reflection of disease activity in younger patients with SLE. When the analysis was repeated using Euclidean distance as the similarity metric for clustering, subsequent logistic regression analysis produced similar results (data not shown).
We defined each sample as belonging to either high or low IFN regulated gene signature groups based on gene expression levels relative to other samples, rather than a predefined minimum fold increase in expression of individual genes or gene groups. Future studies will need to examine the role of microarray GEP in predicting lupus flares and focus on deriving an overall gene expression score that may serve as a potential biomarker for disease activity.
Overall, the demonstration of association between peripheral blood gene expression and lupus disease activity in this large cross-sectional study makes a compelling case for further investigating the role of GEP in predicting disease flare in a longitudinal study. This study also adds to the existing literature by delineating important clinical and serological associations of the IFN gene signature and suggesting a role for GEP in monitoring disease activity in patients who have negative or discordant serology.
We thank the patients at the University of Toronto Lupus Clinic and study coordinators Anne Mackinnon and Joanna Kokalovski. We also thank Richard Abram, Dan Wilson and Karen Rose for performing the sample preparation and microarray experiments and Dominique Ibanez for help with statistical analysis.
Funding: This study was supported by Xceed Molecular. MN is supported by the Geoff Carr Lupus Fellowship.
Competing interests: None declared.
Ethics approval: The study was approved by the research ethics board of the University of Toronto Health Network and written informed consent was obtained from all participants.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.