Optimizing noninvasive sampling of a zoonotic bat virus

Abstract

Outbreaks of infectious viruses resulting from spillover events from bats have brought much attention to bat-borne zoonoses, which has motivated increased ecological and epidemiological studies on bat populations. Field sampling methods often collect pooled samples of bat excreta from plastic sheets placed under-roosts. However, positive bias is introduced because multiple individuals may contribute to pooled samples, making studies of viral dynamics difficult. Here, we explore the general issue of bias in spatial sample pooling using Hendra virus in Australian bats as a case study. We assessed the accuracy of different under-roost sampling designs using generalized additive models and field data from individually captured bats and pooled urine samples. We then used theoretical simulation models of bat density and under-roost sampling to understand the mechanistic drivers of bias. The most commonly used sampling design estimated viral prevalence 3.2 times higher than individual-level data, with positive bias 5-7 times higher than other designs due to spatial autocorrelation among sampling sheets and clustering of bats in roosts. Simulation results indicate using a stratified random design to collect 30-40 pooled urine samples from 80 to 100 sheets, each with an area of 0.75-1 m², and would allow estimation of true prevalence with minimum sampling bias and false negatives. These results show that widely used under-roost sampling techniques are highly sensitive to viral presence, but lack specificity, providing limited information regarding viral dynamics. Improved estimation of true prevalence can be attained with minor changes to existing designs such as reducing sheet size, increasing sheet number, and spreading sheets out within the roost area. Our findings provide insight into how spatial sample pooling is vulnerable to bias for a wide range of systems in disease ecology, where optimal sampling design is influenced by pathogen prevalence, host population density, and patterns of aggregation.

Keywords: bat virus; sampling bias; under roost sampling; viral prevalence.

FIGURE 1

Conceptual drawing of sampling techniques commonly used to estimate viral prevalence at the roost level. Individual‐level sampling is shown in (a) where individual bats are captured and each provides a sample that is used to calculate prevalence. Both (b) and (c) show under‐roost sampling techniques that collect urine droplets from plastic sheets laid beneath roosts. The pooled quadrant technique (b) pools urine droplets that fall within each of the four quadrants of a plastic sheet. The pooled sheet technique (c) pools urine droplets within each plastic sheet. The examples of prevalence calculation show how overestimation of prevalence at the roost level can occur due to multiple bats contributing to a sample. Note that this toy example assumes all bats are captured and the assay used to test samples has perfect sensitivity and specificity

FIGURE 2

Changes in Hendra virus prevalence over time at a roost in Boonah, Queensland, from May 2013 to June 2014. Solid lines show viral prevalence estimated by generalized additive models (GAMs) fitted to observe field data collected from individually captured bats (red), and under‐roost sampling techniques that aggregate urine samples at the pooled quadrant level (blue) and at the pooled sheet level (green). Note that the GAM for individually captured bats begins in later because the study period for this level of sampling begins in June 2013. See Figure 1 for conceptual drawing of sampling types. Shaded regions indicate the standard error of fitted GAMs. [Correction added on 17 September 2021, after first online publication: Figure 2 caption has been updated in this version.]

FIGURE 3

Illustration of one simulation of a kernel density estimation of bat density within a roost. The top row shows pixel images, and the bottom row shows perspective plots of: the density of roosting positions and individual‐level movement around them (left), an isometric Gompertz probability density function centered on the roost to model roost‐level movement (middle), and the final estimated intensity function used to model bat density (right). [Correction added on 17 September 2021, after first online publication: Figure 3 caption has been updated in this version.]

FIGURE 4

Examples of one simulation of each of the four under‐roost sheet sampling designs explored in this study generated for a roost with a 30 m radius. The quadrant design (top left), which follows methods found in previously published studies (Edson, Field, McMichael, Jordan, et al., 2015; Field et al., 2011, 2015), is comprised of 10 3.6 × 2.6 m sheets, each divided into 1.8 × 1.6 m quadrants for pooling urine samples. The other three designs (uniform, stratified, and random) are all “small‐sheet” designs that reduce sheet area, increase sheet number, and disperse sheets about the roost area. The small‐sheet designs plotted above each contain 100 one‐m2 sheets. The stratified design is generated using a sequential inhibition process with and inhibitory radius of 2 m. [Correction added on 17 September 2021, after first online publication: Figure 4 caption has been updated in this version.]

FIGURE 5

Results of 1,000 simulations performed over all possible values of true prevalence for four different under‐roost sheet sampling designs (see scenario 2 in Table 1). The dashed red line indicates $\widehat{p}=p$, and mean estimation bias for all simulations is printed in the lower right corner of each plot

FIGURE 6

Results of the global sensitivity analysis performed in scenario 3, where the quadrant (blue points) and stratified (orange points) designs are compared to determine what drives differences in estimation bias between the two designs. Table 1 shows the parameters used in the simulation. The barplot (a) shows the relative influence of each parameter determined by a boosted regression tree emulator. Plots e and f show the value of estimation bias fitted by the emulator as a function of five influential parameters (blue: quadrant, orange: stratified sampling design)

FIGURE 7

Global sensitivity analysis of scenario 4, where the influence of sheet dimension parameters is explored to determine optimal application of the stratified sheet sampling design. The plots display results from two boosted regression tree emulators: one for estimation bias (top row) and the other for the probability of false negatives (bottom row). Each response is plotted against sheet dimension parameters (from left to right): sheet area s, number of sheets h, minimum distance between sheets d_s , and number of samples collected n_s . The red lines indicate the trend of the points given by smooth spline regression (sreg function in the fields R package; Nychka et al. (2015))