Models of infectious disease dynamics typically rely on estimates of when individuals were infected. Yet it is notoriously difficult to accurately pinpoint individual time of infection (TOI), especially for wildlife when sampling is sparse and uneven. One promising approach has been to use models of antibody response: if the concentration of antibodies against a pathogen changes predictably after infection, the concentration in a single sample can (in theory) be used to back-calculate an individual’s TOI. Unfortunately, there are multiple caveats to this approach, one of which is the requirement to model the antibody response. Ideally this is done using experimental data where individuals are sampled repeatedly after a known TOI, but this is often impossible or highly demanding.
Cross-sectional data from field studies are much more readily available, and sometimes contain repeated measurements for single individuals. Here we investigate the potential to infer within-host antibody dynamics and TOI from such sparse longitudinal measurements. We test different approaches for estimating antibody dynamics using data from wild-caught Channel Island foxes (Urocyon littoralis). Between 2001 and 2016, 330 foxes were sampled between 2 and 8 times at uneven intervals including gaps larger than 1 year, resulting in 1,150 samples. Antibody titers against Leptospira interrogans serovar Pomona were measured using microscopic agglutination tests.
Results/Conclusions
A Bayesian model fitting approach is used in order to incorporate multiple sources of uncertainty arising from individual variation in antibody response, time gaps between sampling sessions, and limited numbers of samples. Model fitting performance mostly depends on the interplay between initial antibody decay rate and length of the seroconversion interval (i.e. time from last negative test to first positive test), where steeper initial decay rates help counter the loss of information resulting from longer seroconversion intervals.
This work provides a case study for the model-based estimation of antibody dynamics using only field data collected unevenly in a wildlife species. Antibody decay models can consequently be combined with additional sources of information, such as pathogen presence, age, sex, outbreak seasonality or population density, to improve estimates of TOI from field data. Using a Bayesian approach ensures that all available data are integrated effectively, and that uncertainties are propagated accurately, despite the many challenges faced by wildlife disease ecologists.