skip to main content


Title: How to make more from exposure data? An integrated machine learning pipeline to predict pathogen exposure
Abstract

Predicting infectious disease dynamics is a central challenge in disease ecology. Models that can assess which individuals are most at risk of being exposed to a pathogen not only provide valuable insights into disease transmission and dynamics but can also guide management interventions. Constructing such models for wild animal populations, however, is particularly challenging; often only serological data are available on a subset of individuals and nonlinear relationships between variables are common.

Here we provide a guide to the latest advances in statistical machine learning to construct pathogen‐risk models that automatically incorporate complex nonlinear relationships with minimal statistical assumptions from ecological data with missing data. Our approach compares multiple machine learning algorithms in a unified environment to find the model with the best predictive performance and uses game theory to better interpret results. We apply this framework on two major pathogens that infect African lions: canine distemper virus (CDV) and feline parvovirus.

Our modelling approach provided enhanced predictive performance compared to more traditional approaches, as well as new insights into disease risks in a wild population. We were able to efficiently capture and visualize strong nonlinear patterns, as well as model complex interactions between variables in shaping exposure risk from CDV and feline parvovirus. For example, we found that lions were more likely to be exposed to CDV at a young age but only in low rainfall years.

When combined with our data calibration approach, our framework helped us to answer questions about risk of pathogen exposure that are difficult to address with previous methods. Our framework not only has the potential to aid in predicting disease risk in animal populations, but also can be used to build robust predictive models suitable for other ecological applications such as modelling species distribution or diversity patterns.

 
more » « less
NSF-PAR ID:
10460095
Author(s) / Creator(s):
 ;  ;  ;  ;  ;  ;
Publisher / Repository:
Wiley-Blackwell
Date Published:
Journal Name:
Journal of Animal Ecology
Volume:
88
Issue:
10
ISSN:
0021-8790
Page Range / eLocation ID:
p. 1447-1461
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The outcome of pathogen spillover from a reservoir to a novel host population can range from a “dead‐end” when there is no onward transmission in the recipient population, to epidemic spread and even establishment in new hosts. Understanding the evolutionary epidemiology of spillover events leading to discrete outcomes in novel hosts is key to predicting risk and can lead to a better understanding of the mechanisms of emergence. Here we use a Bayesian phylodynamic approach to examine cross‐species transmission and evolutionary dynamics during a canine distemper virus (CDV) spillover event causing clinical disease and population decline in an African lion population (Panthera leo) in the Serengeti Ecological Region between 1993 and 1994. Using 21 near‐complete viral genomes from four species we found that this large‐scale outbreak was likely  ignited by a single cross‐species spillover event from a canid reservoir to noncanid hosts <1 year before disease detection and explosive spread of CDV in lions. Cross‐species transmission from other noncanid species probably fuelled the high prevalence of CDV across spatially structured lion prides. Multiple lines of evidence suggest that spotted hyenas (Crocuta crocuta) could have acted as the proximate source of CDV exposure in lions. We report 13 nucleotide substitutions segregating CDV strains found in canids and noncanids. Our results are consistent with the hypothesis that virus evolution played a role in CDV emergence in noncanid hosts following spillover during the outbreak, suggest that host barriers to clinical infection can limit outcomes of CDV spillover in novel host species.

     
    more » « less
  2. Abstract

    Determining parameters that govern pathogen transmission (such as the force of infection, FOI), and pathogen impacts on morbidity and mortality, is exceptionally challenging for wildlife. Vital parameters can vary, for example across host populations, between sexes and within an individual's lifetime.

    Feline immunodeficiency virus (FIV) is a lentivirus affecting domestic and wild cat species, forming species‐specific viral–host associations. FIV infection is common in populations of puma (Puma concolor), yet uncertainty remains over transmission parameters and the significance of FIV infection for puma mortality. In this study, the age‐specific FOI of FIV in pumas was estimated from prevalence data, and the evidence for disease‐associated mortality was assessed.

    We fitted candidate models to FIV prevalence data and adopted a maximum likelihood method to estimate parameter values in each model. The models with the best fit were determined to infer the most likely FOI curves. We applied this strategy for female and male pumas from California, Colorado, and Florida.

    When splitting the data by sex and area, our FOI modeling revealed no evidence of disease‐associated mortality in any population. Both sex and location were found to influence the FOI, which was generally higher for male pumas than for females. For female pumas at all sites, and male pumas from California and Colorado, the FOI did not vary with puma age, implying FIV transmission can happen throughout life; this result supports the idea that transmission can occur from mothers to cubs and also throughout adult life. For Florida males, the FOI was a decreasing function of puma age, indicating an increased risk of infection in the early years, and a decreased risk at older ages.

    This research provides critical insight into pathogen transmission and impact in a secretive and solitary carnivore. Our findings shed light on the debate on whether FIV causes mortality in wild felids like puma, and our approach may be adopted for other diseases and species. The methodology we present can be used for identifying likely transmission routes of a pathogen and also estimating any disease‐associated mortality, both of which can be difficult to establish for wildlife diseases in particular.

     
    more » « less
  3. Abstract

    Hosts and parasites are embedded in communities where species richness and composition can influence disease outcomes (diversity–disease relationships). The direction and magnitude of diversity–disease relationships are influenced by variation in competence (ability to support and transmit infections) of hosts in a community. However, host susceptibility to parasites, which mediates host competence, is not static and is influenced by environmental factors, including pollutants. Despite the role that pollutants can play in augmenting host susceptibility, how pollutants influence diversity–disease dynamics is not well understood.

    Using an amphibian–trematode model, we tested how NaCl influences diversity–disease dynamics. We predicted that NaCl exposure can alter relative susceptibility of host species to trematodes, leading to cascading effects on the diversity–disease relationship. To test these predictions, we exposed hosts to benign or NaCl environments and generated communities that differed in number and composition of host species. We exposed these communities to trematodes and measured disease outcomes at the community (total infections across all hosts within a community) and species levels (average number of infections per host species within a community).

    Host species differed in their relative susceptibility to trematodes when exposed to NaCl. Consequently, at the community level (total infections across all hosts within a community), we only detected diversity–disease relationships (dilution effects) in communities where hosts were exposed to NaCl. At the species level, disease outcomes (average number of infections/species) and whether multi‐species communities supported lower number of infections relative to single‐species communities depended on community composition. Notably, however, as with overall community infection, diversity–disease relationships only emerged when hosts were exposed to NaCl.

    Synthesis.Pollutants are ubiquitous in nature and can influence disease dynamics across a number of host–parasite systems. Here, we show that NaCl exposure can alter the relative susceptibility of host species to parasites, influencing the relationship between biodiversity and disease at both community and species levels. Collectively, our study contributes to the limited knowledge surrounding environmental mediators of host susceptibility and their influence on diversity–disease dynamics.

     
    more » « less
  4. Abstract

    The spatial organization of a population can influence the spread of information, behaviour and pathogens. Group territory size and territory overlap and components of spatial organization, provide key information as these metrics may be indicators of habitat quality, resource dispersion, contact rates and environmental risk (e.g. indirectly transmitted pathogens). Furthermore, sociality and behaviour can also shape space use, and subsequently, how space use and habitat quality together impact demography.

    Our study aims to identify factors shaping the spatial organization of wildlife populations and assess the impact of epizootics on space use. We further aim to explore the mechanisms by which disease perturbations could cause changes in spatial organization.

    Here we assessed the seasonal spatial organization of Serengeti lions and Yellowstone wolves at the group level. We use network analysis to describe spatial organization and connectivity of social groups. We then examine the factors predicting mean territory size and mean territory overlap for each population using generalized additive models.

    We demonstrate that lions and wolves were similar in that group‐level factors, such as number of groups and shaped spatial organization more than population‐level factors, such as population density. Factors shaping territory size were slightly different than factors shaping territory overlap; for example, wolf pack size was an important predictor of territory overlap, but not territory size. Lion spatial networks were more highly connected, while wolf spatial networks varied seasonally. We found that resource dispersion may be more important for driving territory size and overlap for wolves than for lions. Additionally, canine distemper epizootics may have altered lion spatial organization, highlighting the importance of including infectious disease epizootics in studies of behavioural and movement ecology.

    We provide insight about when we might expect to observe the impacts of resource dispersion, disease perturbations, and other ecological factors on spatial organization. Our work highlights the importance of monitoring and managing social carnivore populations at the group level. Future research should elucidate the complex relationships between demographics, social and spatial structure, abiotic and biotic conditions and pathogen infections.

     
    more » « less
  5. Abstract

    Models of host–pathogen interactions help to explain infection dynamics in wildlife populations and to predict and mitigate the risk of zoonotic spillover. Insights from models inherently depend on the way contacts between hosts are modelled, and crucially, how transmission scales with animal density.

    Bats are important reservoirs of zoonotic disease and are among the most gregarious of all mammals. Their population structures can be highly heterogeneous, underpinned by ecological processes across different scales, complicating assumptions regarding the nature of contacts and transmission. Although models commonly parameterise transmission using metrics of total abundance, whether this is an ecologically representative approximation of host–pathogen interactions is not routinely evaluated.

    We collected a 13‐month dataset of tree‐roostingPteropusspp. from 2,522 spatially referenced trees across eight roosts to empirically evaluate the relationship between total roost abundance and tree‐level measures of abundance and density—the scale most likely to be relevant for virus transmission. We also evaluate whether roost features at different scales (roost level, subplot level, tree level) are predictive of these local density dynamics.

    Roost‐level features were not representative of tree‐level abundance (bats per tree) or tree‐level density (bats per m2or m3), with roost‐level models explaining minimal variation in tree‐level measures. Total roost abundance itself was either not a significant predictor (tree‐level 3D density) or only weakly predictive (tree‐level abundance).

    This indicates that basic measures, such as total abundance of bats in a roost, may not provide adequate approximations for population dynamics at scales relevant for transmission, and that alternative measures are needed to compare transmission potential between roosts. From the best candidate models, the strongest predictor of local population structure was tree density within roosts, where roosts with low tree density had a higher abundance but lower density of bats (more spacing between bats) per tree.

    Together, these data highlight unpredictable and counterintuitive relationships between total abundance and local density. More nuanced modelling of transmission, spread and spillover from bats likely requires alternative approaches to integrating contact structure in host–pathogen models, rather than simply modifying the transmission function.

     
    more » « less