Abstract:

Influenza viruses cause regular epidemics and occasional pandemics that have substantial economic and public health costs. The World Health Organization's global influenza surveillance program predicts the identities of future epidemic and pandemic influenza viruses in order to enable the timely development of effective influenza vaccines. While this surveillance program has been fairly successful, it has failed on many occasions to predict epidemic influenza viruses, resulting in increased influenza-associated morbidity and mortality, especially among the elderly. Failure to predict a pandemic influenza virus could have much more dire consequences. There is, therefore, a great need for improvements to influenza virus surveillance. To this end, I developed as part of this thesis new quantitative approaches to influenza virus surveillance.

The success of influenza virus surveillance depends crucially on the accurate interpretation of (antigenic) data collected for this purpose. Yet, not much is currently known about the fundamental nature of those data. In Chapter 2, I present a biophysically grounded mathematical equation that sheds light on how antigenic data are influenced by key experimental variables, such as the affinity of antibody for virus and the avidity of virus for red blood cells. I use the equation to predict, among other things, that the most commonly used method of interpreting antigenic data could yield inaccurate predictions about antigenic differences between influenza viruses, while an alternative method could be more accurate. I present empirical support for this prediction in Chapter 3. I also show in Chapter 3 that while antigenic data are predictive of the effectiveness of influenza vaccines in healthy adults, influenza virus sequence data have poor predictive accuracy with respect to vaccine effectiveness. Therefore, such sequence data should be used with caution when selecting influenza vaccine viruses.

In Chapter 4, I introduce mathematical and statistical approaches to (i) computing orthogonal superpositions of the variation found in antigenic data, in order to enable the filteration of antigenically irrelevant information from these data, (ii) quantifying antigenic differences between influenza viruses, and (iii) constructing low-dimensional, probabilistic embeddings of antigenic differences, in order to aid the understanding of influenza virus' antigenic structure and evolution. I also introduce a new estimate of the statistical significance of antigenic differences that can be used as an additional criterion for selecting influenza vaccine viruses. In Chapter 5, I use genetic, antigenic, and structural data on influenza viruses to propose and empirically test a new mechanism by which influenza viruses may escape from vaccine-induced antibodies. The results provide insights that could guide the future development of more effective vaccines against influenza viruses and other pathogens. In addition to influenza viruses, the quantitative approaches described here could be applicable to the surveillance of a variety of other pathogens.