Airway gene-expression classifiers for respiratory syncytial virus (RSV) disease severity in infants

Abstract

Background: A substantial number of infants infected with RSV develop severe symptoms requiring hospitalization. We currently lack accurate biomarkers that are associated with severe illness.

Method: We defined airway gene expression profiles based on RNA sequencing from nasal brush samples from 106 full-tem previously healthy RSV infected subjects during acute infection (day 1-10 of illness) and convalescence stage (day 28 of illness). All subjects were assigned a clinical illness severity score (GRSS). Using AIC-based model selection, we built a sparse linear correlate of GRSS based on 41 genes (NGSS1). We also built an alternate model based upon 13 genes associated with severe infection acutely but displaying stable expression over time (NGSS2).

Results: NGSS1 is strongly correlated with the disease severity, demonstrating a naïve correlation (ρ) of ρ = 0.935 and cross-validated correlation of 0.813. As a binary classifier (mild versus severe), NGSS1 correctly classifies disease severity in 89.6% of the subjects following cross-validation. NGSS2 has slightly less, but comparable, accuracy with a cross-validated correlation of 0.741 and classification accuracy of 84.0%.

Conclusion: Airway gene expression patterns, obtained following a minimally-invasive procedure, have potential utility for development of clinically useful biomarkers that correlate with disease severity in primary RSV infection.

Keywords: Classification; Gene expression; RNA-seq; Respiratory severity score; Respiratory syncytial virus.

Fig. 1

Correlating NGSS1 (severity score predicted by Model 2) with GRSS. Left: naïve Pearson correlation between GRSS and NGSS1 is $\rho =0.935$. Right: cross-validated Pearson correlation between GRSS and NGSS is $\rho =0.813$. Solid dots are subjects with severe symptoms (defined by GRSS > 3.5) and empty dots are those with mild symptoms (GRSS ≤ 3.5)

Fig. 2

Paired comparisons between visit 1 and visit 2 using NGSS1 (panel a) and NGSS2 (panel b). A total of n = 54 subjects with samples in both visits were used. Solid dots represent severe subjects and empty dots represent mild subjects. The solid line represents the mean trend of severe subjects and the broken line represents the mean trend for mild subjects. a At visit 1, there was a significant difference in mean NGSS1 between the severe (n = 29) and mild (n = 25) groups (6.22 vs. 1.96, p < 0.001). Mean NGSS1 of the mild group was virtually unchanged between two visits (1.96 vs. 2.31, p = 0.45). In comparison, mean NGSS1 of the severe group declined significantly at visit 2 (6.22 vs. 2.82, p < .001). b In contrast to NGSS1, the differences in NGSS2 was virtually unchanged between the two visits, due to the fact that NGSS2 were built with stable genes

Fig. 3

a Diagram indicating the stable genes for the mild (GRSS ≤ 3.5) and severe (GRSS > 3.5) groups and the 689 intersecting stable genes common to both groups. b Correlating NGSS2 (severity score predicted by Model 4) with GRSS. Naïve Pearson correlation between GRSS and NGSS2 is $\rho =0.800$. Right: cross-validated Pearson correlation between GRSS and NGSS is $\rho =0.741$. Circles are subjects with correct cross-validated classification based on NGSS2; solid triangles are misclassified subjects