Development of systems biology-oriented biomarkers by permuted stepwise regression for the monitoring of seasonal allergic rhinitis treatment effects

Background: The immune system, a complex set of integrated responses, often cannot be explained,predicted, or monitored by examining its separate components as biomarkers. Combining different components may therefore be a suitable approach to develop relevant biomarkers reflecting immune system functioning in an appropriate way. Methods: Here we compute and test pattern variables that should reflect immune system functioning on the systems level. Computation was based on a dataset (from a randomized controlled trial comparing two routes of administration) of allergen-specifically induced expression levels of cytokines (IL-1β, IL-5, IL-10, IL-12, IL-13, IL-17, IFN-γ and TNF-α) and symptom severity scores from 22 seasonal allergic rhinitis (SAR) patients measured before and after sixweeks of treatment with medicinal products containing Citrus and Cydonia. By means of stepwise regression analyses we explored and tested pattern variables of the immunological data using permuted stepwise regression (PStR) to distinguish optimally between (immunological) baseline and post-baseline data for the whole treatment group (22 patients) and the two separate treatment groups (11 patients in each group). The validity of the stepwise selection method for the computed pattern variables was tested by means of random permutation tests and evaluated with the cross-validated correct rate of classification (CV correct). Results: For the total group a pattern variable was computed with three variables: IL-10 (day 7), TNF-α (day 1) and IL-10 (day 1) (CV correct: 0.91; pb0.001; R2=0.66), demonstrating a small improvement from the model with IL-10 (day 7) only (CV correct: 0.84; pb0.001; R2=0.47). For the subcutaneous injection group a pattern variable was computed with four variables: IL-10 (day 7), IL-10 (day 1), IL-17 (day 7) and IFN-γ (day 7) (CV correct: 0.90; pb0.01; R2=0.78), demonstrating a very small improvement from the model with IL-10 (day 7) only (CV correct: 0.86; pb0.01; R2=0.58). For the nasal spray group a pattern variable was computed with three variables: IL-10 (day 7), TNF-α (day 1) and IL-10 (day 1) (CV correct: 0.95; pb0.01; R2=0.79), demonstrating a moderate improvement from the model with IL-10 (day 7) only (CV correct: 0.79; pb0.05; R2=0.37). Conclusion/discussion: In this study three robust systems biology-oriented biomarkers for the monitoring of SAR were computed that demonstrated small to moderate improvement compared to monitoring of a single cytokine (IL-10 (day 7)) (CV correct improvement: 0.07 (total group), 0.04 (subcutaneous injection group), 0.16 (nasal spray group)). Further computation and biomarker validation with larger datasets, including data from healthy persons and SAR patients, are indicated.