Background Systemic Lupus Erythematosus (SLE) is a chronic inflammatory autoimmune disease characterized by production of pathogenic autoantibodies. These autoantibodies form immune complexes that can deposit in tissues causing organ damage. The clinical heterogeneity of SLE, and its manifestations that overlap other diseases, pose a diagnostic challenge (1–3). Early identification of disease and disease activity is key to curtailing permanent organ damage (3–5). More accurate diagnosic tools than those in current clinical use would have a significant impact on patient care.
HealthTell has developed a high density peptide array platform to diagnose a disease based on differential Ab binding called ImmunoSignatures (IS). The IS technology uses arrays of hundreds of thousands of unique peptides designed to survey an individual's Ab repertoire from a drop of blood, plasma, or serum (6). Differential binding profiles of the Ab on the array between patients with the disease of interest and control groups are selected as input to develop classification algorithms. The technology holds promise for detecting the presence of any disease that generates a specific B-cell response. Furthermore, for autoimmune diseases the technology has the potential to provide accurate differential diagnosis as well as detection of disease activity and progression.
Objectives To establish the feasibility of the IS technology in identifying subjects with SLE.
Methods Two well-annotated cohorts of serum samples were used representing i) 45 SLE cases and ii) 45 age, gender and race-matched normal donors. All SLE patients met ACR criteria, and as their primary clinical manifestation had joint (53%), renal (38%), skin (2%), or inactive disease (7%). Serum samples were diluted and applied on to the arrays. Ab-peptide binding was detected with fluorescently-labeled goat anti-human IgG. Peptide binding signal was quantified and features with significant differences in intensity between cases and controls were identified by Bonferroni adjusted t-test and log odds ratios. Support vector machine classification algorithms were trained using the most distinguishing peptides. Classifier model performance was evaluated by applying 100 iterations of a four-fold cross-validation routine that included feature selection, model training, and model testing.
Results The cross-validated performance metrics of the 500-peptide IS classifier included an area under the Receiver Operator Curve of 0.99 (95% confidence interval (CI) 0.98–0.99) and an overall accuracy of 93% (95% CI 91%>96%) at equal sensitivity and specificity.
Conclusions ImmunoSignature identification of SLE is feasible and highly accurate based on the preliminary cross-validated performance. These results need verification in larger cohorts and validation in blinded studies. Future studies will include control samples relevant to SLE differential diagnosis such as inflammatory and non-inflammatory rheumatic diseases, as well as active and inactive SLE.
Lateef A et. al. Arthritis Res Ther. 2012;14.
Bertsias GK et.al. Nat Rev Rheumatol 2013;9:687–94.
Tunnicliffe D et. al. Arthritis Care Res (Hoboken). 2015 Oct;67(10):1440–52.
Ceccarelli F et.al. Autoimmun Rev. 2015 Jul;14(7):601–8.
Sprangers B et.al. Nat Rev Nephrol. 2012 Dec;8(12):709–17.
Legutki JB et.al. Nat Commun. 2014 Sep 3;5:4785.
Disclosure of Interest B. Nayak Employee of: None, C. Putterman Consultant for: None, R. Gerwien Employee of: None, K. Sykes Employee of: None, T. Tarasow Employee of: None