Background Type I IFN (IFN-I) is elevated in SLE and is thought to play a role in its pathogenesis. In this study, we utilized RNA sequencing (RNA-Seq) to evaluate whole blood gene expression profiles from a cohort of SLE patients and from age- and sex- matched African American and Hispanic controls in an effort to create more selective biomarkers associated with IFN-I activation.
Objectives We generated a transcriptional signature from SLE patient blood associated with in vitro IFN-I pathway inhibition. We then tested whether IFN-I inhibition could reverse SLE-specific gene expression signatures in an effort to predict which individuals with SLE may be most likely to respond to IFN-I inhibition treatment.
Methods Paired-end strand specific RNA-Seq was used for gene expression profiling of healthy donors and lupus patients. For each sample, almost 100 million reads were sequenced for a total of ∼50 million fragments. A Support Vector Machine (SVM), a powerful classification algorithm in machine learning, was trained on multiple datasets to classify SLE patients and identify genes responsive to IFN-I inhibition.
Results RNA-Seq analysis of 23 healthy donors (23H) versus 25 SLE patients (25L) identified ∼200 differentially expressed transcripts. Approximately 95% of these transcripts were up-regulated, and Ingenuity® Pathway Analysis (IPA) confirmed that IFN-I was the most enriched pathway in this SLE cohort. To characterize the molecular contribution of the IFN-I pathway in SLE, we treated a subgroup of SLE patient blood samples in vitro with an IFN-I inhibitor and examined the effect of treatment on the transcriptome. SVM was trained on these data and identified a gene signature in SLE donors that was normalized to a healthy donor profile after in vitro treatment. SVM identified a signature of down modulated transcripts after IFN-I inhibition. We next tested this signature in a larger cohort (23H and 25L) to verify the relevance of these genes in SLE and to examine the potential of this signature to identify lupus patients most likely to respond to IFN-I inhibition. SVM correctly classified 21 out of 23 Healthy donors and 23 out of 25 SLE patients. Of note, two misclassified Healthy donors had positive Antinuclear Antibodies titer, one of which also exhibited specific Lupus-related autoantibodies (anti-Smith, and anti-Ro). Comparison with other SLE signatures suggests that the genes identified by our approach better represent the larger cohort of SLE patients examined in our study.
Conclusions We identified ∼200 SLE-specific transcripts and showed that the IFN-I pathway was the most prevalent canonical pathway enriched in this SLE cohort. Applying a machine learning approach using an in vitro interventional dataset from this same cohort, we were able to identify SLE-specific signature that was down-modulated after treatment. This signature correctly classified ∼92% of SLE patients. Applying this approach to larger data sets may enable the development of more precise biomarkers for patient stratification and pharmacodynamic assessment for therapies targeting IFN-I.
Acknowledgements We acknowledge Tanesha Cash-Mason for her support with processing these samples
Disclosure of Interest M. Cesaroni Employee of: Janssen, J. Jordan Employee of: Janssen, J. Schreiter Employee of: Janssen, M. Chevrier Employee of: Janssen, W.-H. Shao Grant/research support from: Janssen, B. Hilliard Grant/research support from: Janssen, P. Cohen Grant/research support from: Janssen, R. Caricchio Grant/research support from: Janssen, J. Benson Employee of: Janssen