Improving the analysis of surveillance antigenic data to support seasonal influenza vaccine composition recommendations

Seasonal influenza is estimated to cause 3-5 million infections and 300,000-650,000 deaths worldwide per year. The main countermeasure is vaccination. As the virus evolves to evade the immune system, ongoing surveillance is required to determine if circulating viruses differ substantially from the vaccine. The World Health Organization (WHO) convenes a committee to issue recommendations on which viruses to include in the influenza vaccines used internationally. One of the key inputs to this decision is the antigenic phenotype, measured by antibody titres using haemagglutination inhibition and virus neutralisation assays with ferret and human sera. Analyses of these data using antigenic cartography to produce a quantitative visualisation of antigenic diversity is a core component of the WHO process. Since 2016, I have prepared and presented the reports of these antigenic cartography analyses at the WHO vaccine composition committee which meets twice a year, and the two teleconferences in the two months prior to each meeting. Compared with typical research data, antigenic data from global surveillance characterises a large number of viruses, using fewer distinct antisera. Hence there are unique challenges in reliable and accurate determination of the antigenic diversity in these data. The aim of my thesis is to understand the cause of these difficulties, to devise methods address them, and consequently to improve the resolution and accuracy of the antigenic data generated for the WHO influenza vaccine strain selection process by: * developing methods to assess the reliability of the data * identifying, characterising and addressing instability in antigenic maps—the phenomemon where the addition of a small amount of data can cause a substantial change in the map configuration. * using existing human serology datasets, to validate the backboost of prior immunity when using antigenically advanced vaccines To this end, by splitting the antigenic data sets into smaller groups, I have identified four different types of antigenic map issues in these data. I developed a series of tests and applied these to the antigenic data generated by the WHO surveillance system. Three of the types of issues were ameliorated by decreasing the column bases (normalisation factors used when converting titres to antigenic distances) and using different optimisation methods. The cause of the final issue was excluded data, data which had been excluded because the standard deviation of repeated measurements was above the default threshold. By examining previous algorithms, analysing antigenic data, and performing simulations, I showed that this threshold for excluding data should be increased. I further developed an alternative method for assessing repeated titres measurements using statistical process control rules that identified anomalous data with greater accuracy. To address the first three issues, I investigated alternative methods of merging and optimising antigenic maps, some of which showed improvements when tested using a large antigenic map typical of large-scale surveillance data. Finally, I changed the focus to pre-vaccination and post-vaccination human serology. I identified the factors that affect the backboost, an increase in titres to older viruses when the exposure is more antigenically advanced. I investigated an example where, in some interpretations, a backboost was not observed because the post-vaccination to the old and new antigen were similar. I demonstrated that a backboost, in terms of equivalence of post-vaccination titre or fold change between the old and new antigens, did occur - an important finding as relying on the backboost has become a key element of vaccine strain selection. These results form a solid foundation for substantially improving the analysis of antigenic data from the WHO surveillance system: easier identification of subtle antigenic changes, and the generation of antigenic maps spanning longer time periods - useful not only for vaccine strain selection but for the fundamental understanding of antigenic evolution.

PHE PhD fellowship MRC CRTF