Integrating serological and genetic data to quantify cross-species transmission: brucellosis as a case study

Abstract

Epidemiological data are often fragmented, partial, and/or ambiguous and unable to yield the desired level of understanding of infectious disease dynamics to adequately inform control measures. Here, we show how the information contained in widely available serology data can be enhanced by integration with less common type-specific data, to improve the understanding of the transmission dynamics of complex multi-species pathogens and host communities. Using brucellosis in northern Tanzania as a case study, we developed a latent process model based on serology data obtained from the field, to reconstruct Brucella transmission dynamics. We were able to identify sheep and goats as a more likely source of human and animal infection than cattle; however, the highly cross-reactive nature of Brucella spp. meant that it was not possible to determine which Brucella species (B. abortus or B. melitensis) is responsible for human infection. We extended our model to integrate simulated serology and typing data, and show that although serology alone can identify the host source of human infection under certain restrictive conditions, the integration of even small amounts (5%) of typing data can improve understanding of complex epidemiological dynamics. We show that data integration will often be essential when more than one pathogen is present and when the distinction between exposed and infectious individuals is not clear from serology data. With increasing epidemiological complexity, serology data become less informative. However, we show how this weakness can be mitigated by integrating such data with typing data, thereby enhancing the inference from these data and improving understanding of the underlying dynamics.

Keywords: Bayesian modelling; brucellosis; data integration; epidemiological modelling; genetics; serology; state-space models.

Fig. 1.

Plausible alternative epidemiological scenarios for inter-species transmission and sources of human Brucella infection. Arrows indicate the direction and magnitude of transmission. Question marks indicate that transmission occurs with an unknown magnitude (which will be estimated by our models). In Scenario 1, humans can be infected by caprids with B. melitensis and cattle with B. abortus; in Scenario 2 caprids with B. melitensis can transmit to humans and cattle but only caprids can transmit infection to humans; and in Scenario 3 both caprids and cattle with B. melitensis can infect humans.

Fig. 2.

Population structures used in simulations. In population structure 1 there is a positive correlation between cattle and caprid numbers in each household (HH). In population structure 2 there is clear segregation and each household has mostly cattle or mostly caprids. Population structure 3 shows an intermediate relationship with weak correlation of cattle and caprid numbers and represents the structure of the real sampled population from northern Tanzania.

Fig. 3.

Serology sampling strategy. The continuous bold line shows the relationship between the number of animals present at each household and the number sampled. The dotted line shows the number of animals that would be sampled if all animals present were sampled.

Fig. 4.

Results from the serology model on the Brucella field survey data. The left panel shows the raw mean seroprevalence per household, per species (with associated s.d.; black) and the equivalent model estimated means (with associated 95% credible intervals; blue). The right panel shows the posterior distributions of the coefficients governing the contribution of cattle (β_1,h in red) and caprids (β_2,h in blue) to the probability of human infection in northern Tanzania.

Fig. 5.

Model estimates for the influence of animal population size on infection probability of cattle (β_1,c & β_2,c, left panel) and caprids (β_1,s & β_2,s, right panel).

Fig. 6.

Posterior distributions of the coefficients governing the effect of Brucella-seropositive cattle (β_1,h in red) and caprids (β_2,h in blue) on the probability of human infection. These posteriors were obtained from the serology only model applied to each combination of the epidemiological scenarios and population structures used for simulations. Small vertical lines on the x-axes correspond to the coefficient values used for simulation.

Fig. 7.

Posterior distributions for the coefficients describing the contributions of different infected animal populations to the probability of human infection from the model integrating genetic and serology data (α_1,h in green, α_3,h in red and α_4,h in grey), with decreasing levels of genetic-typing data (50% in top row, 10% in middle row and 5% in bottom row within each epidemiological scenario). Small vertical lines on the x-axes correspond to the coefficient values used for simulation.