A systems genomics approach uncovers molecular associates of RSV severity

Abstract

Respiratory syncytial virus (RSV) infection results in millions of hospitalizations and thousands of deaths each year. Variations in the adaptive and innate immune response appear to be associated with RSV severity. To investigate the host response to RSV infection in infants, we performed a systems-level study of RSV pathophysiology, incorporating high-throughput measurements of the peripheral innate and adaptive immune systems and the airway epithelium and microbiota. We implemented a novel multi-omic data integration method based on multilayered principal component analysis, penalized regression, and feature weight back-propagation, which enabled us to identify cellular pathways associated with RSV severity. In both airway and immune cells, we found an association between RSV severity and activation of pathways controlling Th17 and acute phase response signaling, as well as inhibition of B cell receptor signaling. Dysregulation of both the humoral and mucosal response to RSV may play a critical role in determining illness severity.

Fig 1. Overview of the study design.

Measurements of the nasal epithelial cell transcriptome and nasal microbiome were generated from nasal swabs of infants. Measurements of peripheral immune cells (CD4, CD8, and CD19) were obtained from blood samples. These measurements were subsequently integrated and associated with RSV disease severity (GRSS).

Fig 2. Five potential methods of multi-omic data integration.

Methods differ in their handling of nasal microbiome data (OTU), number of levels of PCA, and the stage at which the integration occurs. Here light purple circles represent nasal gene expression profiles (NT), light blue ones represent CD4 gene expressions, and gold ones represent nasal microbiome abundance data (OTU). Dark blue ones are principal components (PCs) computed from the original features. Large white circle represents the GRSS, which is the clinical variable of interest of this study. In our assessments, the leftmost model out-performed the other models in terms of cross-validated error.

Fig 3. Integration of different data modalities improves prediction of GRSS.

The panel on the left shows the difference between the predicted and observed GRSS for a model that integrates the nasal epithelial transcriptome, peripheral blood CD4 transcriptome, and nasal microbiome, as well as for models using just one of these data types. Similarly, the panel on the right compares a model that integrates the transcriptomes of 3 immune cell types measured in peripheral blood with models using just one of these data types. In both cases, integration increases the precision of the GRSS predictions.

Fig 4. CD4 gene weights, expression, pathways, transcriptional factors and upstream regulators associated with clinical severity in integration models.

Shown are an integration model of CD4, nasal epithelial cells and microbiota (CM) and a model of lymphocytes (LM). Weights generated by integration models are shown in word-clouds. The size of word represents the absolute value of gene weight. Word-clouds of CM & LM consist of genes that have absolute weight is greater than 0.0003. Gene expression are normalized expression levels for the 454 genes selected by univariate analysis; rows represent genes and columns represent samples. Red indicates higher expression, blue indicates low/no expression, green indicates Global RSV Severity Score (GRSS), soft orange indicates mild phenotype, lime green indicates severe phenotype, purple indicates negative weight and olive indicate positive weight. Transcriptional factors associated with severity in CD4 lymphocytes were identified using a hypergeometric test. Four transcriptional factors are shown where p-values were less than 0.05. Ingenuity Pathway Analysis (IPA) was used to identify canonical pathways and upstream regulators represented by genes associated with severity in CD4 lymphocytes. Thirty pathways and upstream regulators are shown where Fisher’s exact test p-values were less than 0.05. Red and blue indicate predicted increased or decreased pathway activation (activation z-score), respectively.

Fig 5. Nasal Epithelial gene weights, expression, pathways, transcriptional factors and upstream regulators associated with clinical severity in integration models.

Shown are Integration model of CD4, nasal epithelial cells and microbiota (CM). Weights generated by integration models are shown in world-clouds. The size of word represents the absolute value of gene weight. A M1 word-cloud consists of genes with absolute weight greater than 0.0003. Gene expression are normalized expression levels for the 993 genes selected by univariate analysis; rows represent genes and columns represent samples. Red indicates higher expression, blue indicates low/no expression, green indicates Global RSV Severity Score (GRSS), soft orange indicates mild phenotype, lime green indicates severe phenotype, purple indicates negative weight and olive indicate positive weight. Transcriptional factors associated with severity in nasal epithelial cells were identified using a hypergeometric test. Four transcriptional factors are shown where p-values were less than 0.05. Ingenuity Pathway Analysis (IPA) was used to identify canonical pathways and upstream regulators represented by genes associated with severity in nasal epithelial cells. Thirty pathways and upstream regulators are shown where Fisher’s exact test p-values were less than 0.05. Red and blue indicate predicted increased or decreased pathway activation (activation z-score), respectively.

Fig 6. Common and unique pathways in lymphocytes & nasal epithelial cells of integration models.

Sankey diagram showing the common and unique pathways among lymphocytes & nasal epithelial cells of integration model of CD4, nasal epithelial cells and microbiota (M1) and model of lymphocytes (M4). Ingenuity Pathway Analysis (IPA) was used to identify canonical pathways represented by genes associated with severity in lymphocytes and nasal epithelial cells. Pathways are shown where Fisher’s exact test p-values were less than 0.05. Red and blue indicate predicted increased or decreased pathway activation (activation z-score), respectively. The width of the flow bar is proportional to absolute value of activation z-score.