(692e) Transcription-Based Functional Modeling of the Host Response to Influenza Virus Infection: Defining a Healthy Immune Response

Jason E Shoemaker, Satoshi Fukuyama, Saori Sakabe, Hiroaki Kitano, and Yoshihiro Kawaoka

                Recent evidence suggests that an overly ambitious immune response may be the primary determinant of pathology during an influenza virus infection[1], yet the specific host functions involved, the timing and the magnitude of these host response functions invoked during a healthy immune response remains poorly defined. Thus, we collected microarray samples from the lungs of mice infected with a seasonal infection (KUKT4), the new Pandemic Swine flu (SOIV) or a mouse adapted Avian influenza (VN1203) with each strain representing an increasingly pathogenic infection. The microarray data was clustered using the Weighted Gene Coregulation Network algorithm (WGCNA). Each resulting cluster was analyzed to determine, first, if the cluster was highly enriched in a specific biological function, and, secondly, which transcription factors primarily regulate the genes within the cluster. This resulted in 14 clusters with clear functional roles and 19 transcription factors, several known to be canonical to influenza RNA sensing and immune activation, which can regulate said clusters. Finally, a linear model was constructed to relate transcription factor activity to the functional outcome of an influenza virus infection, and the model has been validated for its ability to predict functional behavior for additional influenza strains. 

Yearly, seasonal influenza results in approximately 30,000 deaths and $87 billion dollars in associated economic losses in the United States [2]. There is a strong concern in the influenza community that a highly pathogenic Avian influenza, which has a 66% mortality rate [3], may become human to human transmissible and result in a pandemic comparable to the 1918 Spanish influenza during which over 50 million people lost their lives [4]. To best prepare for future pandemics, strain independent therapies must be designed to manage the infection by targeting the host machinery usurped by the virus. Previous studies have shown that selective targeting of interferon stimulated signaling pathways can effectively limit inflammation and mortality during highly pathogenic infections [1]. Thus, we first sought to identify which host functions are differentially regulated during infection and which of those functions are strain dependent.

Clustering determined which genes had highly similar regulation, and, via the “eigengene” for each cluster, also allowed for rapid visualization of strain dependent transcriptional behavior. The WGCNA package [5] was applied to microarray data collected at 14 time points from KUTK4, SOIV, VN1203 or mock infected mice. Given the large numbers of comparisons (42 conditions for ANOVA analysis when comparing against mock data), standard approaches to analyze differential expression failed to minimize the data to an interpretable size. Clustering, on the other hand, resulted in 30 statistically unique clusters, 14 of which were highly enriched in a variety of biological processes. Several of the biological functions which emerged have a clear relationship with the immune response (Toll-like Receptor signaling, inflammation, B cell/T cell activation), while others may represent host defense mechanisms whose activation was strain dependent (suppression of mRNA processing). Since the genes within each cluster are highly correlated, the first eigenvector, i.e. the eigengene, of the transcriptional response provides rapid means of visualizing gene behavior within the cluster. In all, clustering allowed us to define the timing of activation, the duration and the apparent magnitude of 14 biological functions.

Having summarized the microarray data to 14 biological functions, we sought to identify which transcription factors may be regulating each cluster. The Genbank association numbers were loaded into the GATHER tool which combines promoter sequence analysis and gene set enrichment to identify the top transcription factors. The most enriched transcription factor promoter sequences belong to interferon response factor 7 (IRF7), MAZ (a well known inflammation-related protein), NFκB and several other immune signaling proteins.  

Taking the highly enriched transcription factors from the GATHER analysis, we fitted a linear model to the transcriptional data so that functional regulation during an infection could be predicted by the behavior of the 19 top transcription factors. All model construction and fitting was performed in MatLab in which the eigengene for each cluster was used as the model output while the transcription factors’ transcriptional profile acted as the inputs. Initially, model parameterization was performed by fitting the responses to the moderately pathogenic infection (SOIV) and tested for its ability to reproduce functionality in the low and high pathogenic infections (KUTK4 and VN1203, respectively). In terms of the timing and magnitude of eigengene for each biological function, the model predictions were very accurate. We then improved predictive capability by parameterizing the model for data from all three infections, so that behaviors representing a low, medium and highly pathogenic infections would be included. The new model was validated against data from mice infected with different high-pathogenic Avian strain, VD5. The model well predicted gene/functional behavior and identifies several relationships between immune cell signatures which differs greatly depending on the degree of pathology for a given infection.

In conclusion, a mathematical model was constructed to predict from transcription factor data the genetic behavior at the functional level of lung tissue infected with influenza virus.  The cluster results were surprisingly robust to the parameters applied, and, using the cluster eigengenes, allowed for simple temporal description of the functional behaviors differentially regulated during influenza infection. The model itself offers multiple opportunities to explore the host response to determine which biological functions may be suitable targets for therapeutic intervention. Furthermore, temporal aspects of immune activation can be explored to determine when particular immune events occur, and if delays or advances of particular biological functions may be associated with enhanced pathology.

1.            Peiris, J.S., et al., Innate immune responses to influenza A H5N1: friend or foe? Trends Immunol, 2009. 30(12): p. 574-84.

2.            Molinari, N.A., et al., The annual impact of seasonal influenza in the US: measuring disease burden and costs. Vaccine, 2007. 25(27): p. 5086-96.

3.            Kobasa, D., et al., Aberrant innate immune response in lethal infection of macaques with the 1918 influenza virus. Nature, 2007. 445(7125): p. 319-23.

4.            Tan, S.L., et al., Systems biology and the host response to viral infection. Nat Biotechnol, 2007. 25(12): p. 1383-9.

5.            Langfelder, P. and S. Horvath, WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics, 2008. 9: p. 559.