Abstract
Influenza A viruses evolve rapidly to escape host immunity, such that individuals may be infected multiple times with the same subtype. The form and duration of protective immunity after each influenza infection are poorly understood. Here, we quantify the dynamics of protective immunity against influenza A virus infections by fitting individual-level mechanistic models to longitudinal serology from children and adults in a household cohort study. We find that most protection in children is explained by antibody titers measured by the hemagglutination inhibition (HI) assay. In contrast, in adults, HI antibody titers explain a smaller fraction of protection. Protection against circulating strains wanes to approximately 50% of peak levels 2-4 years after infection in both age groups, and wanes faster against influenza A(H3N2) than A(H1N1)pdm09. Our results suggest that the focus of influenza immune responses changes over time from the highly mutable HA head to other epitopes. This work underscores the need for longitudinal data on multiple components of the immune response to better understand differences in susceptibility within populations.
Like many antigenically variable pathogens, influenza viruses continuously evolve to escape host immunity. As a consequence, they cause frequent epidemics and infect people repeatedly during their lives. The details of these processes—which are vital to influenza epidemiology, evolution, and the design of effective vaccines—have nonetheless remained surprisingly difficult to pin down despite nearly 70 years of study.
A major challenge is uncertainty about the nature of acquired immunity. Antibodies are the primary means of protection against influenza and impose strong selection on its surface proteins [1, 2]. Antibody responses to influenza are highly cross-reactive, in that antibodies induced by infection or vaccination with one strain often protect against infections with related strains [3, 4]. The duration and specificity of protection have been difficult to estimate, partly because the relationship between antibody titer and protection appears complex, and also because longitudinal observations of antibody titers and infections are rare. The most common measure of anti-influenza antibody is provided by the hemagglutination inhibition (HI) assay, and HI antibody titers are an established correlate of protection [5]. The HI titer corresponding to 50% protection against infection, commonly cited as 40 [6, 7], may vary by influenza A subtype and host age [8, 9], although measurement error, long intervals between titer measurements, and small titer changes after infection complicate inferences. Recent models have made progress by incorporating measurement error [10, 11], representing infections as latent states [10, 12, 13], and using titers to historic strains to measure the intervals between infections [10], attack rates [11, 12], and the breadth of the response over time [10, 13]. But the relatively short periods of observation in these studies have made it difficult to estimate some basic quantities in the response to infection, namely, how long protection lasts, and whether antibody titers adequately reflect the strength of immunity against infection in individuals over time.
Longitudinal cohorts provide an opportunity for nearly direct observations of the dynamics of infection and protection, and mechanistic models allow hypotheses about these dynamics to be tested. We fit stochastic mechanistic models to influenza antibody titers collected over five years from a large household cohort study including children and adults. These models account for pre-existing immunity, variation in the response to infection, and the possibility that the HI titer is not a good correlate of protection after infection. Their flexibility allows many previous assumptions to be relaxed. For both influenza A subtypes, we estimated the duration of within-subtype and cross-subtype protection, the relationship between HI titer and protection, and the effect of childhood influenza exposures on infection risk later in life. The dynamics inferred from these individual-level models are remarkably consistent with the dynamics of the larger population, and they also support immunological theory of how the antibody response to influenza changes with age.