(658f) Lung Immunodynamics during pH1N1 Influenza Virus Infection

Lung Immunodynamics during pH1N1 Influenza
Virus Infection

Emily E. Ackerman¹,
Ericka Mochan^1,2, Jason E. Shoemaker^1,3,4

¹Dept
of Chemical & Petroleum Engineering, University of Pittsburgh, ²Dept
of Mathematics, Carlow University,³Dept of Computation and Systems Biology,⁴McGowan
Institute

Deadly influenza infections are often associated with
distinct immune system dynamics. Distinct dynamics of key immune cells and
inflammation-associated signaling pathways were observed in mouse and macaque
studies of infection with a reconstructed strain of the 1918 pandemic influenza
virus, with human-isolated H5N1 virus and with the H1N1 virus responsible for
the 2009 pandemic (pH1N1). Moreover, studies in mice depleted of select immune
cell groups have demonstrated improved survival to select infections. Together,
all of these observations suggest that immunomodulatory treatments may help
protect high risk groups during pandemics. But it is also clear that complete
ablation of the immune response leads to uncontrolled virus replication and
severe outcomes. Treatments must be optimized to promote healthy immune
response profiles which minimize dangerous secondary processes, e.g.
inflammation.

Our group has developed a method to maximally exploit
available mouse host response data to develop an ensemble of models that characterize
immune response dynamics in animals infected with pH1N1 virus. Previous models
of influenza induced host responses focus primarily on lung cytokines and
immune cell counts, but do not use gene expression data to inform model
development. Here, we used an iterative, coexpression
analysis approach to infer the log fold changes of key effector signaling
molecules (e.g. IFNα, IFNβ, IRF3). The coexpression analysis allows us to dynamically characterize
a larger number of molecules and improve time-course resolution for other model
components, including changes in immune cell counts. After data normalization,
we performed a series of parameterization training exercises to address several
questions on influenza virus induced immune responses.

We used ensembles of parametrized models to determine if the
kinetics of the immune response are conserved in mild and severe infections. We
collected lung gene expression, cytokine and cell count data from animals
infected with a mild, seasonal H1N1 virus or the more severe pH1N1 virus. We
start with an initial ordinary differential equation model of the immune system
that describes the chemical responses of epithelial cells, B cells, T cells,
macrophages and several key cytokines (14 states in all). Using a Markov Chain
Monte Carlo (MCMC) parameterization approach, we trained the model to the data
derived from H1N1 infected animals. MCMC produces ensembles of parameterized
models all of which may suitably characterize the data. We selected the top
1000 models which had the best fit and the least correlated parameters. We then
challenged each model to correctly predict the immunodynamics observed in
pH1N1-infected animals. To do so, we performed another MCMC parameterization
but only allowed parameters associated with virus replication to vary. None of
the models were capable of predicting the immunodynamics of pH1N1-infected
animals. This suggests that either the kinetics driving immunodynamics are
unique to each infection or that MCMC has simply not found the optimal solution.
We then performed two more computational experiments: we applied the MCMC
formulation to the pH1N1 data and then to the combined H1N1/pH1N1 data. After
10⁸ parameter iterations, solutions were found which fit the pH1N1 data
but the combined data never satisfactorily converged (based on the normalized
sum squared errors and visual confirmation). With the caveat that the results
are dependent on the model architecture, we concluded that pH1N1 and H1N1
induce unique host response kinetics. A direct comparison of the
parameterizations between H1N1- and pH1N1-infected animals suggest that pH1N1
induces an unexpectedly strong interferon response. Further system
architectures will be considered in the future and the modeling will be
expanding to include data from more severe infections (e.g. H5N1 and H7N9
viruses).