Who gets infected and why: Confronting models with data to determine drivers of pathogen susceptibility at the individual and population-level
Host susceptibility is a foundational concept in infectious disease dynamics. Susceptible individuals are the fuel that allows outbreaks to grow and spread. Once an epidemic takes hold, depletion of susceptible hosts (through new infections) eventually drives the effective reproduction number (Reff) below 1, causing the outbreak to stutter and fade. Eventually, the demographic buildup of new susceptible hosts (via new births) creates conditions hospitable to a new epidemic cycle. Ultimately, the fraction of the population susceptible to a given pathogen, and heterogeneity in individual susceptibility by age, by birth year or by physiological status, determine whether a pathogen can spread and persist in a given host population.
Despite its crucial importance, understanding how host susceptibility is distributed across populations is a perennial challenge. Many pathogens of humans and animals have complex strain structure, with partial cross-protection acting among a variety of serotypes. Immunity to other pathogens may wane over time, or may reduce disease severity without entirely preventing infection. For the myriad pathogens with these characteristics, host susceptibility can be difficult to model and difficult to measure, even when serological data on antibody titers is available.
Individual susceptibility is an emergent property of within-host interactions between pathogens and immune effectors. The specific immune interactions that determine susceptibility are often pathogen-specific, and difficult to observe. However, individual-level data on infection outcomes, or population-level epidemiological data are abundant. Statistical analysis of these existing data can help identify host-level factors that govern individual susceptibility. In turn these insights can be used to improve our understanding of how susceptibility is distributed across the population, and predictions of epidemic spread. These inferences can also provide clues to the underlying molecular drivers of host immunity against specific pathogens.
In chapter 1, I compile publicly available data on two avian influenza viruses, H5N1 and H7N9, which have each spilled over to cause hundreds of human cases. I perform likelihood-based model comparison on these data to show that individuals gain exceptionally strong, lifelong protection against avian influenza subtypes in the same phylogenetic group as the first influenza virus encountered in childhood. These results show that susceptibility varies systematically with birth year, and challenge the long-standing assumption that antigenically novel, zoonotic or pandemic influenza viruses escape pre-existing immunity when they spill over to cause cases in humans. These results can help explain why certain birth years have been spared during past influenza pandemics, and may help improve birth year-specific forecasts of future pandemic risk. Further, these results suggest an antigenic basis for naturally-occurring, broadly protective influenza immunity.
In chapter 2, I analyze a large epidemiological surveillance data set to ask whether the same patterns of broadly protective childhood immune imprinting shape birth year-specific risk from the seasonal influenza viruses that cause wintertime epidemics in humans. Model selection shows that seasonal influenza risk from subtypes H1N1 and H3N2 is indeed tied to birth year, and shaped by childhood immune imprinting. However, unlike for avian influenza, immune cross-protection acts more narrowly. Individuals only gain imprinting protection against seasonal influenza viruses of the very same antigenic subtype as the first virus encountered in childhood.
Together, results from chapters 1 and 2 provide a partial proof of concept for development of universal influenza vaccines. Chapter 1 illustrates that the sort of broadly protective immune responses that universal vaccines would aim to elicit can already act naturally in human populations, and in certain epidemic contexts, already seem to shape population susceptibility. But chapter 2 highlights the difficulty of deploying these broadly protective immune responses against familiar, high-burden seasonal strains. Taken alongside recent immunological evidence, these results suggest that the breadth of immune cross-protection against influenza viruses is not fixed, but instead is an emergent property of within-host competition between B cell (antibody-producing) clones. On exposure to a familiar, seasonal influenza virus, narrowly-protective B cell clones competitively exclude broadly protective clones, and the antibody response provides only narrow immune cross-protection, against a single viral subtype. But on exposure to a novel, avian influenza virus, the host may only recognize conserved viral epitopes, and more broadly protective B cell clones are transiently released from competition.
In chapter 3, I shift my focus from childhood imprinting history to explore another dimension of host susceptibility, the role of physical immune barriers in infection resistance. I develop a mechanistic dose-response model to identify factors that limit the spillover of an environmentally abundant bacterial pathogen, Leptospira interrogans. Hosts living in contaminated environments may be exposed to low doses of Leptospira on a daily basis, yet not all become infected. Using data from animal challenge experiments, we show that broken skin is most likely necessary for low-dose environmental exposures to cause infection.
Together, these studies illustrate that heterogeneity in host susceptibility can be linked to measurable, underlying drivers. Demographic factors like year of birth, and immune history predictably modulate susceptibility to specific influenza virus subtypes. Physiological factors, like the presence of wounds and abrasions, predictably modulate susceptibility to environmentally persistent bacterial pathogens like Leptospira. By developing models based on biological principles and then confronting those models with data, we can identify specific risk factors that govern individual susceptibility against specific pathogens. Scaling these insights up to the population level can improve our ability to estimate key epidemiological parameters and can help guide the distribution of limited treatment or prevention resources during outbreaks.