Anthony O'Hare

Whole Genome Sequencing Reveals Local Transmission Patterns of Mycobacterium bovis in Sympatric Cattle and Badger Populations

Whole genome sequencing (WGS) technology holds great promise as a tool for the forensic epidemiology of bacterial pathogens. It is likely to be particularly useful for studying the transmission dynamics of an observed epidemic involving a largely unsampled ‘reservoir’ host, as for bovine tuberculosis (bTB) in British and Irish cattle and badgers. BTB is caused by Mycobacterium bovis, a member of the M. tuberculosis complex that also includes the aetiological agent for human TB. In this study, we identified a spatio-temporally linked group of 26 cattle and 4 badgers infected with the same Variable Number Tandem Repeat (VNTR) type of M. bovis. Single-nucleotide polymorphisms (SNPs) between sequences identified differences that were consistent with bacterial lineages being persistent on or near farms for several years, despite multiple clear whole herd tests in the interim. Comparing WGS data to mathematical models showed good correlations between genetic divergence and spatial distance, but poor correspondence to the network of cattle movements or within-herd contacts. Badger isolates showed between zero and four SNP differences from the nearest cattle isolate, providing evidence for recent transmissions between the two hosts. This is the first direct genetic evidence of M. bovis persistence on farms over multiple outbreaks with a continued, ongoing interaction with local badgers. However, despite unprecedented resolution, directionality of transmission cannot be inferred at this stage. Despite the often notoriously long timescales between time of infection and time of sampling for TB, our results suggest that WGS data alone can provide insights into TB epidemiology even where detailed contact data are not available, and that more extensive sampling and analysis will allow for quantification of the extent and direction of transmission between cattle and badgers.

Estimating epidemiological parameters for bovine tuberculosis in British cattle using a Bayesian partial-likelihood approach.

Fitting models with Bayesian likelihood-based parameter inference is becoming increasingly important in infectious disease epidemiology. Detailed datasets present the opportunity to identify subsets of these data that capture important characteristics of the underlying epidemiology. One such dataset describes the epidemic of bovine tuberculosis (bTB) in British cattle, which is also an important exemplar of a disease with a wildlife reservoir (the Eurasian badger). Here, we evaluate a set of nested dynamic models of bTB transmission, including individual- and herd-level transmission heterogeneity and assuming minimal prior knowledge of the transmission and diagnostic test parameters. We performed a likelihood-based bootstrapping operation on the model to infer parameters based only on the recorded numbers of cattle testing positive for bTB at the start of each herd outbreak considering high- and low-risk areas separately. Models without herd heterogeneity are preferred in both areas though there is some evidence for super-spreading cattle. Similar to previous studies, we found low test sensitivities and high within-herd basic reproduction numbers (R0), suggesting that there may be many unobserved infections in cattle, even though the current testing regime is sufficient to control within-herd epidemics in most cases. Compared with other, more data-heavy approaches, the summary data used in our approach are easily collected, making our approach attractive for other systems.

Developing a framework for risk-based surveillance of tuberculosis in cattle: a case study of its application in Scotland

SUMMARY Due to its substantially lower prevalence of bovine tuberculosis (bTB) relative to other areas of Great Britain, Scotland was designated as an officially (bovine) TB-free region in 2009. This paper investigates resultant possibilities for reducing surveillance by developing risk-based alternatives to current 4-year testing of eligible herds. A model of freedom of infection was used to develop strategies that specifically tested herds that are at risk of infection but would probably not be identified by slaughterhouse meat inspection. The performance of current testing is mimicked by testing all herds that slaughter fewer than 25% of their total stock per year and regularly import animals from high-incidence areas of England and Wales or from Ireland. This system offers a cost reduction by requiring 25% fewer herd and animal tests and 25% fewer false positives.

Risk of Foot-and-Mouth Disease Spread Due to Sole Occupancy Authorities and Linked Cattle Holdings

Livestock movements in Great Britain are well recorded, have been extensively analysed with respect to their role in disease spread, and have been used in real time to advise governments on the control of infectious diseases. Typically, livestock holdings are treated as distinct entities that must observe movement standstills upon receipt of livestock, and must report livestock movements. However, there are currently two dispensations that can exempt holdings from either observing standstills or reporting movements, namely the Sole Occupancy Authority (SOA) and Cattle Tracing System (CTS) Links, respectively. In this report we have used a combination of data analyses and computational modelling to investigate the usage and potential impact of such linked holdings on the size of a Foot-and-Mouth Disease (FMD) epidemic. Our analyses show that although SOAs are abundant, their dynamics appear relatively stagnant. The number of CTS Links is also abundant, and increasing rapidly. Although most linked holdings are only involved in a single CTS Link, some holdings are involved in numerous links that can be amalgamated to form “CTS Chains” which can be both large and geographically dispersed. Our model predicts that under a worst case scenario of “one infected – all infected”, SOAs do pose a risk of increasing the size (in terms of number of infected holdings) of a FMD epidemic, but this increase is mainly due to intra-SOA infection spread events. Furthermore, although SOAs do increase the geographic spread of an epidemic, this increase is predominantly local. Whereas, CTS Chains pose a risk of increasing both the size and the geographical spread of the disease substantially, under a worse case scenario. Our results highlight the need for further investigations into whether CTS Chains are transmission chains, and also investigations into intra-SOA movements and livestock distributions due to the lack of current data.

Broadwick: a framework for computational epidemiology

BackgroundModelling disease outbreaks often involves integrating the wealth of data that are gathered during modern outbreaks into complex mathematical or computational models of transmission. Incorporating these data into simple compartmental epidemiological models is often challenging, requiring the use of more complex but also more efficient computational models. In this paper we introduce a new framework that allows for a more systematic and user-friendly way of building and running epidemiological models that efficiently handles disease data and reduces much of the boilerplate code that usually associated to these models. We introduce the framework by developing an SIR model on a simple network as an example.ResultsWe develop Broadwick, a modular, object-oriented epidemiological framework that efficiently handles large epidemiological datasets and provides packages for stochastic simulations, parameter inference using Approximate Bayesian Computation (ABC) and Markov Chain Monte Carlo (MCMC) methods. Each algorithm used is fully customisable with sensible defaults that are easily overridden by custom algorithms as required.ConclusionBroadwick is an epidemiological modelling framework developed to increase the productivity of researchers by providing a common framework with which to develop and share complex models. It will appeal to research team leaders as it allows for models to be created prior to a disease outbreak and has the ability to handle large datasets commonly found in epidemiological modelling.

Inference in High-Dimensional Parameter Space

Model parameter inference has become increasingly popular in recent years in the field of computational epidemiology, especially for models with a large number of parameters. Techniques such as Approximate Bayesian Computation (ABC) or maximum/partial likelihoods are commonly used to infer parameters in phenomenological models that best describe some set of data. These techniques rely on efficient exploration of the underlying parameter space, which is difficult in high dimensions, especially if there are correlations between the parameters in the model that may not be known a priori. The aim of this article is to demonstrate the use of the recently invented Adaptive Metropolis algorithm for exploring parameter space in a practical way through the use of a simple epidemiological model.