Inference of Infectious Disease Transmission through a Relaxed Bottleneck Using Multiple Genomes Per Host

Abstract

In recent times, pathogen genome sequencing has become increasingly used to investigate infectious disease outbreaks. When genomic data is sampled densely enough amongst infected individuals, it can help resolve who infected whom. However, transmission analysis cannot rely solely on a phylogeny of the genomes but must account for the within-host evolution of the pathogen, which blurs the relationship between phylogenetic and transmission trees. When only a single genome is sampled for each host, the uncertainty about who infected whom can be quite high. Consequently, transmission analysis based on multiple genomes of the same pathogen per host has a clear potential for delivering more precise results, even though it is more laborious to achieve. Here, we present a new methodology that can use any number of genomes sampled from a set of individuals to reconstruct their transmission network. Furthermore, we remove the need for the assumption of a complete transmission bottleneck. We use simulated data to show that our method becomes more accurate as more genomes per host are provided, and that it can infer key infectious disease parameters such as the size of the transmission bottleneck, within-host growth rate, basic reproduction number, and sampling fraction. We demonstrate the usefulness of our method in applications to real datasets from an outbreak of Pseudomonas aeruginosa amongst cystic fibrosis patients and a nosocomial outbreak of Klebsiella pneumoniae.

Keywords: genomic epidemiology; infectious disease outbreak; transmission analysis; within-host diversity and evolution.

Fig. 1.

Difference in posterior probability estimates of transmission between a dataset with 1 observation per host and a dataset with 5 observations per host. The underlying transmission network remains the same; it is defined by the black squares, which show the true transmissions in the simulated dataset. The gray squares show the reverse relationship, switching the true infector and infected hosts. Black squares containing red demonstrate higher posterior probabilities being assigned to the true transmission links as a result of including more observations. Elsewhere, blue indicates lower posterior probabilities being assigned to incorrect transmission links.

Fig. 2.

Distribution of posterior link probabilities inferred in the simulation studies with 1 (left) and 5 (right) observations per host. The top plots show bidirectional link probabilities in which the roles of infector and infected host may switch, the bottom plots show the directional link probabilities in which the infector and infected host must be correctly inferred. The red lines relate to pairs of individuals for which a transmission link exists, and the blue lines relate to pairs of individuals that are not linked.

Fig. 3.

Varying the 4 key simulation parameters. Vertical bars show 95% central credible intervals, while solid circles show posterior means. Horizontal or diagonal lines show true values for simulations. a) Varying π. b) Varying R. c) Varying κ. d) Varying λ.

Fig. 4.

Transmission analysis of P. aeruginosa. a) Dated phylogeny colored by host according to the iteration with highest posterior probability. b) Matrix of transmission probabilities from each host (row) to any other (column).

Fig. 5.

Transmission analysis of K. pneumoniae. a) Dated phylogeny colored by host according to the iteration with highest posterior probability. b) Matrix of transmission probabilities from each host (row) to any other (column).

Fig. 6.

a) Example transmission tree with 6 hosts. Points indicate the infected times of each host. Filled circles show observed hosts, and empty circles show unobserved hosts. b) Example phylogenetic tree with 7 leaves from 5 observed hosts. Leaf labels indicate the host, followed by the sample number for that host. Each coalescence node is given a label. c) Example colored phylogenetic host with 7 leaves from 5 observed hosts, and 6 hosts overall. The branch color indicates the host, and the asterisks indicate transmissions. Here, host 3 is infected with 2 lineages.