Burrow Dusting or Oral Vaccination Prevents Plague-Associated Prairie Dog Colony Collapse

We established nine treatment plots within a 1375 ha black-tailed prairie dog colony complex (Soapstone–Meadow Springs) located on natural areas in northern Larimer County, Colorado, USA, owned and managed by the City of Fort Collins (Figure 1). Epizootic plague had been last documented in this complex during 2008–2009 (Griffin et al. 2010). We compared three treatments (insecticidal dust, vaccine, placebo) in triplicate. Within each dust–vaccine–placebo block (Bulger Grazing Allotment [BGA], Meadow Springs Ranch [MSR], Soapstone Natural Area [SNA]), we clustered plots closely (Figure 1) to minimize potential confounding effects of spatiotemporal variation in plague activity across the complex. We assigned an entire colony to a single treatment in four cases. In five cases, where we established plots within a larger colony, we buffered boundaries by ≥400 m to minimize spillover between treatments (Table 1; Figure 1). Plot location and size, as well as treatment assignments, were somewhat constrained by other land use priorities, vaccine availability and prior plague mitigation activities (Table 1). Of particular note, vaccine and placebo plots at MSR and SNA were dusted in 2012 as part of a vaccine safety trial or during prior plague management programs (Tripp et al. 2015; Table 1). The BGA and MSR dust plots were spatially adjacent and located within a larger dusted block (Figure 1) but were regarded as independent. We do not believe that these constraints unduly biased overall outcomes.

Insecticidal “dust” plots were established within select prairie dog colonies treated annually during March–April by City of Fort Collins Natural Areas Program staff since 2010. Crews used specialized applicators (Technicide, San Clemente, CA) to deliver 4–5 g of 0.05% deltamethrin powder (DeltaDust^®, Bayer Crop Science, Research Triangle Park, North Carolina; Seery et al. 2003; Biggins et al. 2010; Tripp et al. 2016) into each prairie dog burrow encountered.

Vaccine and placebo treatments described here were also part of a broader multistate effort to evaluate the field efficacy of SPV (Rocke et al. 2017). “Vaccine” plots received baits carrying recombinant raccoonpox virus (RCN-F1/V307; unlicensed Yersinia pestis Vaccine, Live Raccoon Poxvirus Vector, Code 11Y2.R0; Rocke et al. 2014). Each ~4–5 g bait carried about 5 × 10⁷ plaque-forming units of RCN-F1/V307 and consisted of an edible polymer (Food Source Lures, Alabaster, Alabama, USA) and peanut butter, with rhodamine B (0.25%) incorporated as a biomarker (Fernandez and Rocke 2011; Tripp et al. 2014, 2015). We distributed baits by hand along transects spaced 10 m apart, dropping one every ~10 m for an application rate of 99 baits/ha in 2012–2013 or 124 baits/ha in 2014–2015 following methods described by Tripp et al. (2014, 2015). Baits were distributed on vaccine plots each August after juveniles had emerged from natal burrows (and when cattle were absent) to maximize uptake by prairie dogs (Tripp et al. 2014). Vaccine distribution began in 2013, except for an 8 ha subplot in the MSR block that had been treated in 2012 (Table 1; Tripp et al. 2015). “Placebo” plots received baits identical in composition but without RCN-F1/V307; these were applied as described for vaccine. All vaccine and placebo baits were made at NWHC following methods described by Rocke et al. (2017).

We conducted burrow counts, capture, sampling and marking of prairie dogs on all plots within each block using the same schedule and procedures to maintain consistency for comparisons.

Prairie dog capture began 6–20 days after bait distribution to measure and compare bait uptake. The capture schedule followed a robust design method (Pollock et al. 1990; Kendall et al. 1995) to estimate abundance and survival. The robust design specifies that multiple consecutive (or near consecutive) trapping occasions (i.e., trap days) compose a multi-day primary trapping session and that primary sessions are repeated. An interval of approximately 1 year separated trapping sessions in this study, except that two sessions were conducted opportunistically at BGA in 2013 and SNA in 2015 as plague emerged. Survival was estimated for the intervals between trapping sessions (primary period), and abundance and density were estimated from a minimum of four trapping/marking occasions (days) within each session (Otis et al. 1978; Pollock et al. 1990; Gould and Pollock 1997). We calculated density by dividing abundance estimates by effective area trapped, which was the area trapped with a 20 m buffer (half of the estimated home range of a prairie dog) to account for individuals captured on the perimeter. We marked all prairie dogs that were captured during trapping.

Prairie dog capture, handling and marking generally followed methods described elsewhere (Tripp et al. 2009, 2014, 2015, 2016; detailed in Supplemental Material). We captured prairie dogs on at least four consecutive days but anesthetized and sampled individuals only once per trapping session.

We collected fleas from 90 randomly selected burrows at all study plots in May, July and September throughout the study. Prairie dog burrows were classified as active or inactive (Biggins et al. 1993) and swabbed using methods modified from Ecke and Johnson (1952) as described elsewhere (Griffin et al. 2010; Tripp et al. 2015, 2016; detailed in Supplemental Material). We also collected fleas from captured prairie dogs (detailed in Supplemental Material).

All fleas were identified to the species level using the keys of Stark (1958) and Hubbard (1968). Fleas of a single species from individual prairie dogs and burrows were placed into pools of up to 10 fleas and tested for Y. pestis by PCR (Griffin et al. 2010). Prairie dog and other carcasses were retained for necropsy and plague screening via PCR (Griffin et al. 2010) using the Qiagen supplementary protocol for liver and spleen tissue and the DNeasy blood and tissue kit (Qiagen, Valencia, California, USA); the US Centers for Disease Control and Prevention (Fort Collins, Colorado, USA) confirmed positive carcass results with direct fluorescent antibody (DFA) and mouse inoculation tests. To assess bait uptake, hairs and whiskers from live-trapped prairie dogs were examined under a fluorescence microscope (Fernandez and Rocke 2011).

We compared flea abundance between plots using Wilcoxon rank-sum tests with P-values adjusted by the Holm (1979) procedure. The R statistical base package (pairwise.wilcox.test) was used for these analyses (R Development Core Team, Version 2.15.2 2012). We compared burrow activity between plots using Chi-square tests (prop.test).

We used program MARK (Burnham and Anderson 1998) to estimate abundance and probabilities of apparent survival, capture and recapture. We used the Huggins robust design model (Pollock et al. 1990; Huggins 1991) to estimate parameters of interest in the BGA and MSR blocks. The SNA block had sufficient data to support a multistate robust design model (Nichols and Coffman 1999). We fit models with the following variables: sex, estimated age (animals ≤15 months old vs. >15 months old in multistate robust design models; animals first marked at ≤15 mo old vs. first marked >15 mo old in robust design only models), primary period, treatment (dust, vaccine or placebo), proportion of colony trapped, and whether the plot had experienced epizootic plague as defined by the observed decline of the colony (>90% reduction in prairie dog numbers).

Fitting all possible models and combinations for both survival and recapture would have resulted in a set of over 1000 models; therefore, model selection was conducted according to methods detailed in Supplemental Material. ΔAICc (Akaike Information Criteria; Burnham and Anderson 2002) is the difference between the lowest-AICc model and the model referenced; AICc weight is the relative weight or belief assigned to each model. We show models with ΔAICc ≤ 10.

Epizootic plague affected all three blocks during the study period but emerged asynchronously among and within blocks (Table 1; Figure 1). We detected plague in fleas and prairie dog carcasses in the southeastern portion of the complex as our field experiment began in August 2013 (Table 1; Figure 1). The epizootic subsequently spread in a northwesterly direction, reaching the northern portion of the complex in 2015 (Table 1; Figure 1). Burrow activity and prairie dog density declined sharply in placebo plots at all three blocks when epizootic plague emerged (Figures 2, 3; Table S1).

Patterns in the corresponding dust and vaccine plots were less consistent and appeared strongly influenced by the timing of treatment applications relative to plague emergence (Table 1; Figure 2). Temporal asynchrony in epizootic plague emergence and its strong effects on prairie dog dynamics rendered most planned pooled comparisons uninformative. Consequently, we report analyses and patterns for each block separately in chronological order of plague emergence (BGA, MSR, SNA).

We first detected plague at the BGA block in August and October 2013 on the vaccine and placebo plots, respectively (Table 1; Figure 2). Plague detection portended rapid declines in burrow activity on both of these plots (Figure 2). Placebo plot burrow activity collapsed (= 0%) by May 2014 and remained low through October 2015 (range 1–32%; Figure 2). Burrow activity on the vaccine plot also collapsed by March 2014 and remained low (0–40%; Figure 2). In contrast, we recovered no plague-positive fleas or carcasses on the BGA dust plot and prairie dog burrow activity remained high (64–100%; Figure 2). In March 2014, burrows in the dust plot were ≥71% more active (Chi-square test; P < 0.0001) than in the vaccine or placebo plots (which did not differ; P = 0.4751).

The pattern at the MSR block was similar. We first detected plague in August 2013 on the vaccine plot and in June 2014 on the placebo (Table 1). Rapid declines in burrow activity followed (Figure 2). Burrow activity on the placebo plot collapsed by September 2014 and remained low (0–11%; Figure 2). Burrow activity on the vaccine plot decreased to 16% by April 2014 but rebounded to between 32 and 62% thereafter (Figure 2). Like BGA, we recovered no plague-positive fleas or carcasses on the MSR dust plot, and burrow activity remained high (68–98%; Figure 2). In September 2014, burrow activity in the MSR dust plot was 56% greater (P < 0.0001) than in the vaccine or placebo plots; activity in the vaccine plot (32%) was greater than in the placebo plot (0%; P < 0.0001).

Burrow activity remained relatively high at the SNA block until 2015. We first detected plague on the placebo plot in July 2014 and on the dust and vaccine plots in March 2015 (Table 1). Declines in burrow activity preceded or followed plague detection (Figure 2). Placebo plot burrow activity decreased to 6% by June 2015 and collapsed by September. Dust plot burrow activity decreased to 12% by September 2015. Burrow activity on the vaccine plot declined less dramatically, with 60% remaining active in June 2015. In September 2015, 50% fewer burrows were active in the dust plot (P < 0.0001) than in the vaccine plot but 12% more burrows were active (P = 0.0019) than in the placebo plot; 62% more burrows were active on the vaccine plot than on the placebo plot (P < 0.0001).

Plague’s emergence in the BGA block preceded or coincided with the beginning of our study, confounding survival analyses. The lowest-AICc model (weight ~0.9; Table 2) indicated differences in apparent survival (S, accounts for survival and emigration), capture (c) and recapture (p) probabilities among primary periods with c and p being unequal. In the nearest ranked model (ΔAICc = 5.5; weight ~0.06; Table 2), S changed differently among age groups across primary periods. Estimated monthly S was 0.55 (SE = 0.24) between August and October 2013, then ≥0.82 (0.04) for the remainder of the study (Table S2). Except for the 2-day October 2013 primary period in which no recaptures occurred, c ≥ p (Table S2). Estimated densities on the BGA dust plot were generally higher and more stable than either vaccine or placebo plots, both of which collapsed by 2014 (Figure 3, Table S1). Prairie dog densities were higher throughout the experiment on the placebo colony than on the vaccine colony (Figure 3). This is likely because the former remained occupied to some extent, whereas the latter was temporarily devoid of prairie dogs after the epizootic. Thus, recovery occurred more rapidly on the placebo colony.

Plague management showed stronger effects in the MSR and SNA blocks. The low-AICc model for each suggested that S varied by treatment and primary period and that p varied by age and primary period, with p = c (weights ~0.57 and ~0.82, respectively; Table 2).

At MSR, all five lowest-scoring AICc models (ΔAICc ≤ 10; combined weight ~0.998; Table 2) included an effect of dust or vaccine on survival. The MSR block had sufficient data to fit multistate robust design models. We did not trap on the dust plot prior to 2013, but monthly S was estimated at ≥0.81 (0.03) throughout the study (Table S2; Figure 4). Monthly S in the vaccine plot was estimated at 0.89 (0.02) between September 2012 and September 2013 for animals marked by Tripp et al. (2015), then declined to 0.79 (0.02) before rebounding (0.92 ± 0.02; Figure 4; Table S2). Estimated monthly S in the placebo plot was 0.91 (0.02) between September 2012 and September 2013 for animals marked by Tripp et al. (2015), but 0.0 (0.0) after epizootic plague emerged (Figure 4; Table S2). Capture rates for adults were slightly higher than for young (Table S2); capture and recapture rates were equal.

The three lowest-scoring AICc models for SNA (combined weight ~0.997; Table 2) all included an effect of treatment on S. Estimated monthly S in the dust plot ranged from 0.74 (0.03) to 0.83 (0.04) (Figure 4; Table S2). In the vaccine plot, estimated monthly S remained ≥0.81 (0.05) throughout the study (Figure 4; Table S2). In contrast, estimated monthly S of prairie dogs in the SNA placebo plot was 0.79 (0.04) between August 2013 and August 2014, but 0.0 (0.0) thereafter (Figure 4; Table S2). As at MSR, p and c were equal but varied across primary periods, with young having higher capture probabilities than adults (Table S2).

Yersinia pestis DNA-positive fleas were most frequent when flea abundance in burrows and on prairie dogs tended to be highest (Figure 2). When flea data from all blocks and years were pooled by treatment, dust plots had fewer fleas per burrow (\( \bar{x} \) = 0.1; P < 0.0001; Figure 2) than vaccine (\( \bar{x} \) = 2.5) or placebo plots (\( \bar{x} \) = 2.5). Similarly, prairie dogs captured on dust plots had fewer fleas (\( \bar{x} \) = 0.4; P < 0.0001) than on vaccine (\( \bar{x} \) = 15.8) or placebo plots (\( \bar{x} \) = 14). Burrows and prairie dogs on dusted plots had fewer fleas (P < 0.05) than vaccine or placebo plots on all blocks and during all years (2013–2015). Differences in flea abundance between vaccine and placebo plots were more variable. At BGA, there was no difference in fleas per burrow on vaccine (\( \bar{x} \) = 1.9) and placebo plots (\( \bar{x} \) = 0.73; P = 0.0860), while fleas were more abundant on prairie dogs on vaccine (\( \bar{x} \) = 12.7) than on placebo plots (\( \bar{x} \) = 6.3; P = 0.0317). At MSR, there were fewer fleas per burrow on vaccine (\( \bar{x} \) = 3.7) than on placebo plots (\( \bar{x} \) = 5.0; P = 0.036) but fleas on prairie dogs were more abundant on vaccine plots (\( \bar{x} \) = 19.0) when compared to placebo (\( \bar{x} \) = 14.7; P = 0.0317). At SNA, there was no difference in fleas per burrow on vaccine (\( \bar{x} \) = 2.0) and placebo plots (\( \bar{x} \) = 1.8; P = 0.980), while fleas were less abundant on prairie dogs on vaccine (\( \bar{x} \) = 10.7) than on placebo plots (\( \bar{x} \) = 16.3; P = 0.0442). See Supplemental Materials for additional flea data.