Central questions in ecology center around species coexistence: what regulates the observed diversity of biological communities? How do species interact within a shared environment to stabilize their abundance? Similarly, questions in epidemiology center around the coexistence of pathogen communities, bringing community ecology to the forefront of medical science. Understanding the coexistence of viral communities is particularly important to predict the effects of vaccines, which perturb the underlying community dynamics. The advent of effective, multivalent vaccines against human papillomavirus (HPV), a community of 200 distinct viral types, highlights this principle. Biological interactions between HPV types, which may be synergistic, neutral, or competitive, have vast implications for the effects of HPV vaccines. Particularly, if HPV types compete, then the eradication of vaccine-targeted types could lead to “type replacement”, a phenomenon that has been observed in other pathogens (e.g. Weinberger et al. 2011).

A major challenge to understanding community assembly lies in quantifying the relative contributions of inter-species interactions and environmental drivers of co-occurrence. Ecologists have long used statistical network models of to infer these mechanisms from ecological survey data. However, snap-shot observations of ecological communities can lead to biased interpretations of species interactions. For example, two competing species may transiently aggregate if they share environmental requirements (i.e., positive correlations in co-occurrence over space), even if competitive exclusion would ultimately result (i.e., unobserved negative correlations in co-occurrence over time). This paradox is illustrated in Fig. 1. In this paper, we address this challenge by leveraging longitudinal co-occurrence data to develop a dynamical statistical regression model. We separately estimate spatial and temporal correlations in species persistence and colonization rates, distinguishing fixed effects, markers of direct species interactions, from random individual-level and environmental effects.

We fit our models to a large database tracking infection with 10 different HPV types in men over time. We detect only sparse, weak pairwise fixed effects of HPV types on persistence and colonization, suggesting that type interactions are limited (Fig. 2d, 2e). This implies that HPV vaccines are unlikely to cause type replacement. Conversely, we detect substantial positive correlations in among-patient random effects, suggesting that HPV types respond similarly to common host traits (Fig. 2b). Otherwise put, our results suggest that HPV type diversity depends on shared environmental drivers, or high-risk sub-populations, in agreement with a previous mechanistic model of HPV dynamics fit to the same data (Ranjeva et al. 2017).

Ultimately, we hope that our model provides a computationally tractable way to develop hypotheses about species interactions in ecological communities, taking advantage of increasingly available longitudinal data. This approach can guide resource-intensive mechanistic modeling and manipulative experiments. 

Written by: Sylvia Ranjeva and Joseph Mihaljevic

 References:

Ranjeva, Sylvia L., et al. "Recurring infection with ecologically distinct HPV types can explain high prevalence and diversity." Proceedings of the National Academy of Sciences 114.51 (2017): 13573-13578.

Weinberger, Daniel M., Richard Malley, and Marc Lipsitch. "Serotype replacement in disease after pneumococcal vaccination." The Lancet 378.9807 (2011): 1962-1973.