Industry-Wide Surveillance of Marek's Disease Virus on Commercial Poultry Farms

Abstract

Marek's disease virus is a herpesvirus of chickens that costs the worldwide poultry industry more than US$1 billion annually. Two generations of Marek's disease vaccines have shown reduced efficacy over the last half century due to evolution of the virus. Understanding where the virus is present may give insight into whether continued reductions in efficacy are likely. We conducted a 3-yr surveillance study to assess the prevalence of Marek's disease virus on commercial poultry farms, determine the effect of various factors on virus prevalence, and document virus dynamics in broiler chicken houses over short (weeks) and long (years) timescales. We extracted DNA from dust samples collected from commercial chicken and egg production facilities in Pennsylvania, USA. Quantitative PCR was used to assess wild-type virus detectability and concentration. Using data from 1018 dust samples with Bayesian generalized linear mixed effects models, we determined the factors that correlated with virus prevalence across farms. Maximum likelihood and autocorrelation function estimation on 3727 additional dust samples were used to document and characterize virus concentrations within houses over time. Overall, wild-type virus was detectable at least once on 36 of 104 farms at rates that varied substantially between farms. Virus was detected in one of three broiler-breeder operations (companies), four of five broiler operations, and three of five egg layer operations. Marek's disease virus detectability differed by production type, bird age, day of the year, operation (company), farm, house, flock, and sample. Operation (company) was the most important factor, accounting for between 12% and 63.4% of the variation in virus detectability. Within individual houses, virus concentration often dropped below detectable levels and reemerged later. These data characterize Marek's disease virus dynamics, which are potentially important to the evolution of the virus.

Keywords: Marek's disease virus; epidemiology; surveillance; vaccine escape; virulence evolution.

Figure 1:

The structure of the data in our study. Left panel: a schematic example of a sampling hierarchy generated by the structure of the poultry industry. Reading from the bottom up, multiple samples were collected from a single flock, multiple flocks were reared in a single house over time, multiple houses were located on a single farm, multiple farms were associated with a single operation (company), and multiple operations were rearing chickens that typically belonged to a single production type. This created a nested hierarchical structure in the data. One example of such a hierarchy is shown here. Right panel: the actual number of unique levels are given by “C” for the cross-sectional data, “L” for the longitudinal data, “A” for the air tube data, and “F” for the feather tip data.

Figure 2:

Summary plots of the cross-sectional data depicting the number of assays that were performed as a function of production type (A), operation (B), farm (C), sex (D), month of the year (E), bird age (F), and flock size (G). For example, in panel B, 520 assays were run for samples collected from operation 4. Also depicted are the approximate locations of origin of each sample (H) and each farm (I). Note that to maintain farm location anonymity, normal random variables with mean 0 and standard deviation 0.1 were added to the points when plotting latitude and longitudes in H and I. In all plots, black color depicts breeder facilities, red color depicts broiler facilities, and blue color depicts layer facilities.

Figure 3:

Fraction of tests with detectable virus. Each point shows the mean for a different house with grey bars depicting 95% confidence intervals on the mean (Supplement A.15). Confidence intervals vary between houses because of variable sample sizes. Different rows depict different production types (top–breeders, middle–broilers, bottom–layers). Solid black lines separate different operations (companies). Dashed red lines separate different farms. Note that prevalence estimates are from the raw data, not corrected to account for potential confounding effects such as bird age, collection date, or flock.

Figure 4:

Fraction of variance on the latent scale attributable to each model factor. Points are median values and lines are 95% credible intervals. Marginal and conditional R² values represent the variance explainable by all fixed effects, and all fixed plus random effects respectively. Note that only the values for the best model (Table 1) are shown.

Figure 5:

Effect sizes for fixed effects. The top panel shows the median and 95% credible interval for the three production types. The middle panel shows the median and 95% credible interval for the effect of bird age on the probability of detecting virus in a dust sample. The bottom panel shows the median and 95% credible interval for the effect of collection date on the probability of detecting virus.

Figure 6:

Longitudinal surveillance data for three broiler farms in Pennsylvania. Each panel is labelled “X-Y”, where “X” gives a unique farm identification, and “Y” gives a house number on that farm such that each two character label is unique. Each of the three farms shown in this figure had two houses. All of these farms began associated with the same operation, but farm “C” changed operations in the middle of our surveillance. The timing of this change is denoted by an asterisk in the plot. All farms followed an “all-in, all-out” policy meaning that houses had discrete periods of rearing and down time. To represent the presence or absence of birds, white intervals cover periods when birds were present, grey intervals cover periods when birds were absent, and blue intervals cover unknown periods. Each point represents the log-mean virus concentration (VCN) for that set of dust samples. Error bars are 95% confidence intervals calculated as explained in Supplement A.15. The dotted horizontal line shows the approximate qPCR limit of detection for a single test.

Figure 7:

Longitudinal surveillance data for two additional broiler farms in Pennsylvania. Symbols, colors and layout as in fig. 6. Both of these farms had four houses. Farm “D” was associated with the same operation as the farms in fig. 6, but farm “E” was not. Note also that farm “E” changed operations during our surveillance period, the timing of which is marked with an asterisk.

Figure 8:

Air tube data (left column) and feather tip data (right column) for two broiler farms in Pennsylvania. Symbols, colors and layout as in fig. 6. Note that the dynamics in the air tube data and feather tip data are highly similar to one another, and are highly similar to that of the corresponding houses in the cross-sectional data (fig. 6). As in fig. 6, a change in operation on farm C is denoted by an asterisk.