Abstract
Management of pathogen transmission is often hindered by uncertainties in the efficacy of and interactions between intervention strategies, sometimes resulting in unintended negative consequences. Yet outbreaks of infectious disease can have serious consequences for wildlife population health, especially species of conservation concern. The endangered Florida panther, for example, experienced an outbreak of feline leukemia virus (FeLV) in 2002-2004, and continues to be affected by this deadly virus. Ongoing management efforts aim to mitigate the effects of FeLV on panthers with limited empirical information about which strategies are most effective and efficient. We used a simulation-based approach to determine optimal FeLV management strategies. We simulated use of proactive management interventions (i.e., proactive vaccination) as well as three reactive strategies, including vaccination in the face of an outbreak, test and removal protocols, and temporary spatial segregation of the panther population. Vaccination strategies included partial vaccine immunity, an understudied type of vaccine imperfection. We compared the effectiveness of different strategies in mitigating FeLV mortalities and duration of outbreaks. Results showed that inadequate proactive vaccination can paradoxically increase the number of disease-induced mortalities in FeLV outbreaks, most likely due to effects of partial vaccine immunity. Combinations of proactive vaccination with reactive test-and-removal or vaccination had a synergistic effect in reducing impacts of FeLV outbreaks. Temporary spatial restrictions were unlikely to be effective under realistic conditions. Our results highlight the importance of integrating management and modeling approaches to identify unexpected consequences and synergies in pathogen management interventions and aid in conservation of at-risk species.
Significance Statement Managing outbreaks of infectious disease is fraught with uncertainties such that seemingly helpful interventions can have unintended negative consequences. We used a simulation approach to determine optimal pathogen management strategies in an endangered carnivore, the Florida panther (Puma concolor coryi), which continues to be affected by outbreaks of feline leukemia virus. We tested interventions including proactive vaccination, reactive vaccination, test-and-removal, and temporary spatial restrictions. We found that inadequate proactive vaccination can counterintuitively increase disease-related mortalities. In contrast, we identified synergies between proactive and reactive strategies that will be key for ongoing conservation of the iconic Florida panther. The results of this study demonstrate the importance of linking modeling and management to optimize pathogen control and avoid unexpected negative consequences.
Introduction
Outbreaks of infectious disease can have significant impacts on the population health of free-ranging wildlife, and are of heightened importance in species of conservation concern (1). For example, the feline retrovirus, feline leukemia virus (FeLV), has been the source of significant outbreaks in two endangered felids: Iberian lynx (Lynx pardinus) and Florida panthers (Puma concolor coryi). In the case of panthers, FeLV caused a deadly outbreak in 2002-2004 (2), spilling over from domestic cats, with subsequent direct transmission among panthers (3). In addition, there is recent evidence of ongoing FeLV spillover to and transmission among panthers (4), necessitating continued management of this deadly pathogen. FeLV inoculation with a domestic cat vaccine has been used previously, but with unknown efficacy in panthers (2). Further, the proportion of the population that must be vaccinated to most efficiently prevent future FeLV outbreaks is unknown, as is how proactive vaccination might interact with other reactive interventions to interrupt an FeLV outbreak. Such uncertainties are common among free-ranging wildlife threatened by infectious disease and hamper efforts to effectively control pathogen transmission.
Mathematical models of infectious disease transmission offer a powerful tool for filling such disease management knowledge gaps (5). Models allow the ethical testing of a wide range of different management approaches and can reveal unexpected consequences of disease control interventions (6). Models have been used to optimize disease management protocols in a variety of free-ranging wildlife species of conservation concern, including Ethiopian wolves (7), chimpanzees (8), and Amur tigers (9). Further, models can serve the important function of balancing the realities of fieldwork with ideal disease control protocols to provide practical, effective guidance for wildlife managers (e.g., 10, 11). Here, we use mathematical models of FeLV transmission in Florida panthers to determine optimal disease prevention and control strategies in this iconic carnivore.
Pathogen management strategies can be preventive (hereafter, proactive) or reactive. Among proactive and reactive strategies, vaccination is a cornerstone in veterinary medicine (1, 12), and functions by reducing the availability of susceptible individuals. Vaccination can be applied randomly, or targeted to high risk individuals (e.g., 13); to likely superspreaders (e.g., in chimpanzees: 8); seasonally (e.g., 11, 14); or spatially, as with vaccine barriers or firewalls to prevent spread of a pathogen to new areas or subsets of populations (e.g., 15–17). In panthers, vaccination has been used for active management of FeLV (2, 4), but with uncertainties regarding optimal distribution, individual efficacy, and level of population protection needed.
Among reactive pathogen management strategies, test-and-removal (or test-and-cull) is commonly used in domestic species to functionally reduce the infectious period of infected individuals (18). However, this approach is rarely used in wildlife due to the generally low availability of field diagnostic tests and/or difficulty in recapturing individuals after a positive diagnosis (18). Rather, pathogen control in wildlife often relies on non-selective culling in an attempt to reduce density and—in theory—transmission (19). However, non-selective culling can have a host of negative consequences for disease management including removal of immune individuals (18, 20) or culling-induced perturbation (e.g., 21). Thanks to ready availability of a field diagnostic test, FeLV management in domestic cats and Iberian lynx has made use of test-and-removal or isolation of infected individuals (22, 23), and is part of future FeLV mitigation plans in panthers. However, uncertainties remain regarding how test-and-removal might interact with proactive vaccination efforts to more effectively reduce pathogen impacts.
Larger-scale isolation or quarantine measures may also be used for disease management in free-ranging wildlife, in the form of physical or behavioral barriers between affected and unaffected subsets of a population. For example, fencing has been used to prevent transmission of foot-and-mouth disease between wildlife and cattle in South Africa (24), and winter feeding grounds may help prevent transmission of brucellosis between elk and cattle in the Greater Yellowstone Ecosystem through behavioral separation (25). For panthers, while physical barriers to prevent spillover from domestic cats are impractical, it may be possible to use temporary spatial barriers to reduce the spatial spread of FeLV among panthers after a spillover event. Specifically, the major I-75 freeway is fenced throughout Florida panther habitat to reduce vehicle strikes, with regular wildlife underpasses the main means for wildlife to traverse this barrier. While never before attempted, it may be possible to physically block these underpasses under emergency conditions to prevent the spread of FeLV from the northern to southern subsets of the panther population, or vice versa.
Given the ongoing risks of FeLV to panther conservation and the uncertainties regarding application of proactive and reactive FeLV management strategies, the objectives of this study were to test singly, and in combination, the effectiveness of: (1) proactive vaccination, (2) reactive vaccination, (3) reactive test-and-removal, and (4) reactive temporary spatial restrictions for reducing the population level impacts of FeLV in Florida panthers. We used a mathematical modeling approach to address these objectives in order to efficiently and ethically test a wide range of intervention protocols for the protection of this endangered carnivore.
Methods
Simulation pipeline
To examine the effect of different disease management regimes on FeLV control, we used spatially-explicit, network simulations adapted from our research determining drivers of retrovirus transmission in panthers (26). This approach involves two steps: (1) simulation of a contact network among panthers, and (2) simulation of FeLV transmission on this network. In brief, we simulated panther populations of 150 individuals (27) and used our previously described exponential random graph model for retrovirus transmission in panthers (26) to simulate contact networks constrained by network density among simulated populations (see supplementary methods and Table S1 for further details).
We used a transmission model developed in Gilbertson et al. (26). Briefly, this model used a susceptible-infectious-recovered compartmental framework, where individuals could be progressively, regressively, or abortively infected (2). Progressive infections always resulted in death, while regressive infections eventually recovered with immunity, and abortive infections were always considered immune. Importantly, based on our previous work, we allowed both progressives and regressives to be infectious, though regressives were less likely to transmit (Table S1). Duration of infectiousness is long for FeLV (estimated as a mean of 18 weeks, 2; but see 26 and Table S1). As in Gilbertson et al. (26), we included only disease-induced mortality, and, in order to preserve key network structure, allowed territories vacated by deaths to be reoccupied by new susceptible individuals (hereafter, respawning). Outbreaks were initiated by a single, randomly selected non-isolate individual in the population (see supplement discussion), and proceeded in weekly time steps for up to five years.
Our primary objective was to examine the effect of different FeLV management regimes on epidemic outcomes, so in our primary simulations, we held network generation and transmission parameters constant at previously supported values (see supplementary methods and Table S1 for further details). We evaluated the consistency of our results to the choice of parameter values with a sensitivity analysis (see below). Hereafter, a parameter set represents the unique set of network, transmission, and management parameters for any given set of simulations. A full simulation includes simulation of a single contact network and FeLV transmission on that network (with or without management interventions). For each parameter set, we performed 100 full simulations.
For a baseline scenario, we recorded key epidemic outcomes in the absence of interventions. For management scenarios we recorded key outcomes in the presence of interventions (Figure 1). These key outcomes were (1) the number of mortalities, (2) the duration of an epidemic, and (3) the proportion of epidemics that “failed” per 100 successful epidemics. A failed epidemic (i.e. stochastic fadeout, 28) was one in which fewer than 5 individuals acquired progressive or regressive infections. The outcomes of mortalities and epidemic durations were summarized as median values per parameter set, as results were often skewed; all outcomes were compared between baseline and management scenarios. All simulations were performed in R version 3.6.3 (29).
Proactive vaccination
In the first management scenario, we examined the effect of different levels of population proactive vaccination (proportion of the population vaccinated prior to an outbreak), and different ratios of single versus boosted vaccination. We simulated from 10-80% (in 10% increments) of the population having some degree of vaccine-induced immunity to FeLV at the onset of an outbreak. These vaccinations were distributed randomly in the population. Among the vaccinated individuals, 0, 50, or 100% received a second boosting inoculation. Actual vaccine efficacies are unknown for panthers, but based on efficacy studies in domestic cats (30, 31), we conservatively assumed that boosted vaccination would prevent 80% of infections and single vaccination would prevent 40% of infections. We modeled this efficacy as a binomial probability to represent partial vaccine immunity (versus binary immunity; 32). We conducted proactive vaccination scenarios in a full factorial design, for a total of 24 proactive vaccination parameter sets, and 2,400 full simulations (100 full simulations per parameter set).
Reactive vaccination
During simulated periods of reactive vaccination administration, panthers were selected at a rate of one panther per week for vaccination. We assumed that managers would not know the disease status of an individual selected for vaccination and that vaccination would be ineffective in infectious or recovered individuals. In the case of previously vaccinated individuals, a re-vaccination changed the vaccine efficacy of singly vaccinated individuals (efficacy of 40%) to the efficacy for boosted individuals (efficacy of 80%). We further assumed that managers would know which individuals had received two vaccinations, so boosted individuals were not selected for additional vaccination attempts.
We varied the timing of the onset of reactive vaccination after the initiation of an FeLV outbreak to reflect the difficulty of epidemic detection in this elusive carnivore. We therefore began reactive vaccination at an optimistic, but difficult-to-attain time point of 26 weeks, and a more realistic time point of 52 weeks. In addition, we varied the distribution of reactive vaccination. While proactive vaccines were always distributed randomly, reactive vaccination was either randomly distributed or spatially distributed in an attempted vaccine barrier based on proximity to the I-75 freeway (see supplement for further details).
Because vaccination is resource and time intensive, we evaluated the effect of reactive vaccination for 6 months per year versus year-round. Among reactive vaccination scenarios, we also included scenarios with proactive vaccination, specifically where 0-60% of the population was proactively vaccinated (in 20% increments) at the initiation of an outbreak. Because it is highly unlikely that 100% of the proactively vaccinated population would have received boosted vaccination, we used a more conservative—yet still challenging to attain—ratio of 50% of the proactively vaccinated population being boosted (i.e. efficacy of 80%). We held this ratio of efficacy constant for proactive vaccination in all reactive scenarios. We again used a factorial design across all variations, resulting in 32 parameter sets and a total of 3,200 full simulations under reactive vaccination scenarios.
Reactive test-and-removal
Test-and-removal scenarios were built around a protocol in which panthers infectious at capture are removed from the population through humane euthanasia or temporary removal until recovery. For simplicity, we assumed that all progressively infected individuals would show clinical signs and be humanely euthanized at capture, while infectious regressive individuals were temporarily removed from the population until their recovery and re-released into any open territory. A maximum of five individuals were allowed to be temporarily removed in this way at one time.
We expected that managers would be able to capture and test one panther per week at most, and that captures occurred during a 17 week (about 4 month) capture season, in accordance with current panther capture protocols. We assumed that managers would not know the disease status of a target individual until the capture occurred, so captures were not targeted by infection state. We varied the onset of test-and-removal, such that the intervention began 26 or 52 weeks after the initiation of an epidemic. Captures were also random or spatially targeted. If spatially targeted, captures (and consequent removals) only occurred on the same side of the I-75 freeway as the initial FeLV infection.
As in the reactive vaccination scenarios, we again also included varying degrees of proactive vaccination (0-60% in 20% increments). Reactive test-and-removal scenarios were simulated in a full factorial design across variations, for a total of 16 parameter sets and 1,600 full simulations.
Reactive underpass closures
To determine the potential utility of closing I-75 wildlife underpasses in a FeLV outbreak, we considered a “best case” scenario in which underpass closures were completely effective at preventing transmission across the freeway. Underpasses were closed instantaneously either 26 or 52 weeks after the initiation of an FeLV outbreak, and remained closed for 4, 13, 26, or 52 weeks. We again included variations in proactive vaccination (0-60% in 20% increments), as in other reactive management scenarios. The factorial design here resulted in 32 parameter sets for underpass closure scenarios, for a total of 3,200 full simulations.
Sensitivity analysis
We used a latin hypercube sampling (LHS) approach to generate 50 sensitivity analysis parameter sets across our 8 network and transmission parameters using the lhs package in R (33). We repeated our baseline scenario simulations across these 50 parameter sets and completed 50 simulations per parameter set for all sensitivity analyses.
Due to the high computational effort required to perform sensitivity simulations across all management scenarios, we focused only on the proactive vaccination scenarios. Specifically, we evaluated a subset of 12 LHS parameter sets across a subset of the proactive vaccination conditions in a factorial design (see supplementary methods for further details), resulting in 108 parameter sets with 50 full simulations per parameter set (5,400 full simulations). This approach allowed us to examine sensitivity of our proactive vaccination results across different outbreak sizes and network and transmission parameters, while mitigating computational effort associated with exploring such a wide range of parameters and scenario variations. Sensitivity analysis simulation results were evaluated using scatterplots and Partial Rank Correlation Coefficients (PRCC; 34, 35); proactive vaccination scenarios were evaluated for alignment with our qualitative result that low proactive vaccination can increase disease-induced mortalities.
Results
Progressive infections are expected to result in death in panthers, and are therefore of key concern for management efforts. For simplicity, hereafter, we refer to the number of progressive infections in simulations as the number of mortalities. In the baseline, nointervention scenario, the median number of mortalities was 34 (range: 1-54); median duration of epidemics was 119.5 weeks, and 34 epidemics failed (fewer than 5 progressive or regressive infections) per 100 successful epidemics.
Proactive vaccination alone
Proactive vaccination paradoxically increased the number of mortalities across a range of conditions, especially without vaccine boosting (Figures 2, S2). Even with 50% of vaccinates receiving a booster, proactive vaccination only reduced mortalities from the baseline scenario at high levels of population vaccination (i.e., 60-80%). With 100% boosting, proactive vaccination increased mortalities at low levels of population vaccination (10-20%), had marginal effects at 30-40% population vaccination, and was strongly effective at about 50% population vaccination levels and higher (at 80% population vaccination, median 17.5 mortalities versus 34 with no interventions). Proactive vaccination consistently lengthened the duration of epidemics, relative to the baseline scenario (up to a median duration of 143.5 weeks; Figures S3-4). When all vaccinated individuals received a booster, vaccination reduced the probability of a successful outbreak even at 40% population vaccination (52 failures versus 34 failures per 100 successful epidemics; Figure S13). In contrast, when no vaccinated individuals received a booster, proactive vaccination was largely only effective at reducing the probability of epidemics at very high levels of population vaccination.
Proactive and reactive vaccination
Reactive vaccination alone did not reduce mortalities (Figure 3). However, reactive vaccination appeared to work synergistically with proactive vaccination, particularly at moderate to high levels of proactive vaccination (i.e., at least 40-60% of the population proactively vaccinated). This largely held true regardless of the timing of intervention onset and the strategy for reactive vaccination distribution (i.e., random versus spatial; Figures S5-6). The mortalityreducing effects of reactive vaccination were, however, slightly reduced if reactive vaccination occurred for only 6 months out of the year (Figure 3). A ratio of greater than 1.5 inoculations per vaccinated individual appeared to promote the largest reductions in mortalities (e.g., as few as a median of 25 mortalties with year-round reactive vaccination and 60% proactive vaccination versus 34 mortalities with no interventions; Figure 3). Adding reactive vaccination largely did not affect the durations of simulated epidemics (Figures S7-8), and had little impact on the probability of epidemics failing, particularly in comparison to proactive vaccination alone (Figure S13).
Proactive vaccination with test-and-removal
Test-and-removal alone did not reduce mortalities (Figure 4). Like reactive vaccination, however, test-and-removal appeared to work synergistically with proactive vaccination, especially at moderate to high levels of proactive vaccination (i.e., at least 40-60% of the population proactively vaccinated; as few as a median of 26 mortalities). This largely held true regardless of the timing of the onset of the intervention or the targeting of captures (i.e., random versus spatial). Notably, simulated captures were only conducted for about 4 months per simulation year, in contrast to at least 6 months of reactive vaccination per year. Captures were marginally more likely to successfully identify actively infectious individuals when initiated earlier in an outbreak (Figure S9). The addition of test-and-removal largely did not affect the durations of epidemics (Figure S10). When coupled with proactive vaccination, test-and-removal had a modest effect in reducing the probability of a successful epidemic (e.g., maximum of 50 failed epidemics per 100 successful versus 34 failed epidemics per 100 successes with no interventions; Figure S13).
Reactive underpass closures
Reactive underpass closures were ineffective in the absence of proactive vaccination (Figure S11). The most effective impact of underpass closures on reducing FeLV mortalities occurred when onset of closure was early (26 weeks after epidemic initiation), lasted for at least 13 weeks, and occurred in conjunction with at least 40-60% of the population being proactively vaccinated (though the clearest effects occurred when at least 60% were proactively vaccinated). Under these conditions, underpass closures synergistically reduced mortalities from baseline scenarios (as few as a median of 22 mortalities with underpass closures) and increased the probability that an epidemic would fail (maximum of 75 failed epidemics per 100 successes with underpass closures; Figure S13). In some cases, simulations showed a marked decrease in transmission during the period of underpass closures, but reopening often resulted in a subsequent resurgence of infections (Figure 5). Underpass closures had no clear impact on the duration of outbreaks (Figure S12).
Sensitivity analyses
Simulated epidemic sizes were variable across the full 50 sensitivity analysis parameter sets in the absence of FeLV interventions (range: median 7.5 mortalities-median 47.5 mortalities; Figure S14). Based on PRCC, the parameters for network density, transmission potential from regressives, weekly contact rates, and baseline transmission potential were positively associated with median mortalities in the absence of interventions; the parameter for the infection-induced mortality rate was negatively associated with median mortalities (see supplementary results, Figure S15).
When focusing on a subset of parameter sets for proactive vaccination sensitivity analysis, low levels of proactive vaccination (e.g., 20% population proactive vaccination) were sometimes effective in reducing the number of mortalities, in contrast to our primary results. PRCC results from proactive scenario sensitivity analysis suggested that the parameters for network density, transmission potential from regressives, and weekly contact rates were positively correlated with increased mortalities at low levels of vaccination (Figure S16). However, network density did not have a clearly monotonic relationship with the difference between mortalities with and without proactive vaccination (Figure S17). We therefore performed additional post-hoc sensitivity analyses to further interrogate the relationship between network density and our qualitative outcome of increased mortalities at low levels of proactive vaccination (see supplementary results for detailed discussion of post-hoc analysis). These additional analyses found some evidence that low levels of proactive vaccination were least effective at reducing mortalities at intermediate values of parameters governing network connectivity (network density and proportion adults), especially when coupled with increased transmission potential (e.g. higher infectiousness of regressive individuals and/or increased weekly contact rates; Figures S19-20).
Discussion
In this study we found unexpected consequences and impacts of several epidemic management strategies for disease control in small populations of conservation concern. Although our simulation results provide guidelines for FeLV management in Florida panthers, they also demonstrate the power of partnering modeling approaches and population management questions to test and optimize disease control strategies in free-ranging wildlife (36). Furthermore, the principles of transmission and available methods of disease control underlying our findings provide insights for pathogen control in other host-pathogen systems, including humans and livestock.
Proactive vaccination alone may worsen epidemic outcomes under some conditions
Our simulation results showed a paradoxical increase in FeLV mortalities with low levels of proactive vaccination. This counterintuitive finding is likely due, at least in part, to partial vaccine immunity, a type of vaccine imperfection often overlooked in studies of wildlife disease (32). Under partial vaccine immunity, vaccinates can act as a semi-protected susceptible pool, contracting infection later in the course of an epidemic, ultimately prolonging epidemics and increasing the total number of mortalities. Only with adequate herd immunity can these effects be avoided. This counterintuitive result is consistent with findings by Rees et al. (37), who found reduced vaccine effectiveness with heterogeneity in the spatial distribution of hosts. Importantly, our results suggest that studies assuming full protection from vaccination may significantly underestimate the level of vaccination needed for population protection—underestimates which may even make epidemics worse if only partial vaccine immunity can be achieved.
Our sensitivity analysis suggests that populations with intermediate levels of network connectivity and/or high transmission potential may be most vulnerable to these paradoxical effects. As in spatially structured populations (38), high or low connectivity causes infection to fade out quickly; with intermediate connectivity, vaccine failures provide a steady supply of new susceptibles. Alternatively, in cases of high transmission potential, vaccination may shift a rapid fade-out epidemic to a sustained epidemic scenario, as in Rees et al. (37). While our sensitivity analysis was limited by computational demands and use of some discrete parameters (which may affect PRCC inference; 34), our findings are consistent with this broader body of literature.
FeLV vaccine efficacy may operate differently in reality from our simulation structure: for example, some vaccinates may have zero vaccine-induced immunity, while others have 100% protection (binary immunity; 32). Alternatively, vaccination may not protect from infection but could reduce viral shedding or increase survival of infected individuals (32). In the case of binary immunity, vaccine efficacy would be unlikely to prolong and worsen epidemics as we saw here. In contrast, increased survival of infected individuals without changes to shedding potential could extend or worsen outbreaks, and even favor the evolution of virulence (32). Future research should assess vaccinated panthers’ immune response to FeLV infection in vitro to help refine understanding of vaccine imperfection in panthers and how imperfect immunity may further alter vaccination guidelines.
Based on our results, we argue that wildlife managers should continue vaccinating available panthers and prioritize boosting vaccinated individuals to develop a core population of high-immunity individuals, rather than a broadly distributed low-immunity population. This recommendation should be most effective at increasing the probability of epidemic failure, and aligns with other wildlife studies which have emphasized vaccination of core populations (39) or risk-based sub-populations (13).
Temporary spatial restrictions are unlikely to be effective under realistic scenarios
Here, we examined the effect of wildlife underpass closures as a novel method to restrict connectivity of the panther population under emergency disease control conditions. Such temporary spatial restrictions could increase other types of panther mortality (e.g. vehicle strikes or intraspecific conflict), so such an intervention would need to reduce FeLV mortalities at a rate greater than these other types of panther mortality in order to be a viable disease control strategy. Unfortunately, our simulations found that temporary underpass closures were generally no more effective at reducing FeLV mortalities than less risky reactive interventions. Further, underpass closures were less effective when occurring after the peak of simulated epidemics, such that spatial restrictions would need to occur early in an epidemic to be effective. However, it should be noted that spatial restrictions may be more effective if used in combination with other reactive FeLV management strategies (e.g., reactive vaccination in concert with temporary spatial restrictions to prevent transmission resurgence after restrictions are removed).
Extensive research has considered how landscape barriers and fragmentation may affect pathogen transmission and control in wildlife (e.g., 37, 40–42). By closing underpasses, we artificially fragmented panther habitat, assuming complete efficacy of closures in preventing transmission across the I-75 freeway. In reality, some individuals would likely successfully traverse this barrier. Because habitat fragmentation can promote pathogen persistence (42), a “semipermeable” freeway barrier may, in fact, ultimately worsen epidemic dynamics. Alternatively, landscape restrictions to animal movement can facilitate more efficient disease control, as in Haydon et al. (7), where vaccinating to control pathogen spread along habitat corridors reduced the level of vaccine coverage needed to protect endangered Ethiopian wolves. While panthers are generally well-connected spatially, expansion of the population north of its current range may favor similar metapopulation-oriented disease control strategies. However, given uncertainties in the effect of underpass closures on other sources of mortalities and their limited effectiveness in our simulations, we suggest temporary underpass closures should currently be considered a low priority for FeLV management in panthers.
Reactive and proactive strategies can work synergistically to reduce epidemic impacts
Our simulations showed that both reactive vaccination and test-and-removal strategies reduced FeLV mortalities in panthers when used in combination with moderate levels of proactive vaccination. Test-and-removal had more consistent effects with arguably less effort than reactive vaccination (four months of captures compared to year-round reactive vaccination). We therefore suggest that test-and-removal be prioritized over reactive vaccination, especially if identification of actively infectious individuals can be improved.
In our simulations, captures that were most aligned with the initial wave of infectious individuals (i.e., with earlier onset) were more likely to identify actively infectious individuals for removal. This finding highlights the importance of targeting captures to individuals likely to be actively infectious. However, determining infection status in cryptic wildlife is difficult, and consequently supports the increased use of remote tracking technologies that may be able to (1) identify behavior changes associated with sickness, and (2) detect the onset of an epidemic more quickly. This conclusion is consistent with similar findings in Channel Island foxes, where increasing the number and frequency of tracking of sentinel individuals was important for early identification of epidemics (16).
A key component of the success of test-and-removal here is the selective removal we simulated, which avoids removing immune individuals that contribute to overall herd immunity (18, 20). However, we have simplified the field testing process in our simulations. The common field-available FeLV diagnostic test identifies antigenemia, which is key for identifying actively infectious individuals. However, the duration of antigenemia—and even degree of infectiousness—in regressively infected individuals is unclear in panthers. We may therefore overestimate the effect of removing regressive individuals, but given their reduced infectiousness in our simulations, we still expect test-and-removal to be a key strategy for mitigating FeLV impacts in panthers.
Notably, reactive vaccination, in concert with proactive vaccination, is also a viable alternative strategy to test-and-removal. Our simulations showed reactive vaccination to have the strongest effects for reducing FeLV mortalities when at least 50% of the population was vaccinated, and with a ratio of about 1.5 vaccines per vaccinated individual (i.e., 50% of vaccinated individuals received a booster). We therefore suggest that managers should prioritize boosting at least half of vaccinated individuals in a reactive vaccination response scenario. We did not see a strong effect of attempting a vaccine barrier, in contrast to Sanchez et al (16), where a simulated vaccine barrier could effectively halt spread of a pathogen in Channel Island foxes. This difference is likely due to the differences in home range size and movement capacity between the two species, with foxes ranging less widely than panthers. In reality, if an empirical outbreak of FeLV in panthers exhibited a stronger spatial signal than was featured in our simulations, spatially targeted reactive vaccination may yet be a worthwhile intervention strategy.
Importantly—particularly if levels of population protection from proactive vaccination are unknown—both reactive vaccination and test-and-removal strategies mitigated the negative effects of low levels of population protection seen with inadequate proactive vaccination. It is therefore vital for managers to incorporate these reactive strategies in the event of future FeLV outbreaks.
Limitations and future directions
In this study, we considered the effects of partial vaccine immunity, but we made the simplifying assumption of no waning vaccine or infection-induced immunity over time. However, our findings with regard to imperfect efficacy are at least partially representative of the likely consequences of waning immunity, in that the loss of immunity supplies new susceptibles to the population. Should immunity not outlast the course of an FeLV outbreak, this process could prolong outbreaks and result in increased mortalities. Future research should therefore examine the effects of waning vaccine immunity, particularly considering the value of revaccinating individuals which may be experiencing loss of vaccine protection.
Our simulation results found that relatively high levels of vaccination were required to reduce impacts of FeLV in panthers, in contrast to studies in other endangered species (7, 9). However, here we are investigating a pathogen with a long duration of infectiousness and lack the advantages of distinct corridors between panther sub-populations for reducing required levels of vaccination. It is therefore unsurprising that panthers would require higher levels of FeLV population vaccination than was found, for example, for rabies vaccination in Ethiopian wolves (7) or canine distemper virus vaccination in Amur tigers (9). However, we also did not assume the presence of preexisting population immunity prior to proactive vaccination, and some degree of population immunity likely already exists in panthers, given ongoing exposures (4). This would reduce necessary vaccination levels in panthers, as would higher vaccine efficacy than we conservatively assumed here (39). Future research could prioritize better understanding the realized individual immunity after single and boosted FeLV vaccination in panthers to determine if inoculations are more protective than we conservatively assumed here.
Conclusions
Our simulation results highlight the risks of inadequate proactive vaccination, particularly with partial vaccine immunity. We recommend prioritizing boosted vaccination in panthers, and joint use of proactive vaccination and reactive strategies to mitigate the risks of imperfect vaccination and most effectively reduce the impacts of FeLV in this iconic carnivore. This research highlights the value of linking modeling and management priorities to identify unexpected consequences of interventions and determine optimal pathogen management strategies in free-ranging wildlife.
Authors’ Contributions
All authors conceived the ideas and designed methodology; MLJG performed simulations, analyzed data, and led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.
Data Availability
Full R code for simulations is available on GitHub (https://github.com/mjones029/FeLV_Management_Simulations) and upon acceptance will be archived at Zenodo.
Acknowledgements
This research was supported by the National Science Foundation (DEB-1413925, 1654609, and 2030509). MLJG was supported by the Office of the Director, National Institutes of Health (NIH T32OD010993), the University of Minnesota Informatics Institute MnDRIVE program, and the Van Sloun Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Florida panther data collected by the Florida Fish and Wildlife Conservation Commission is fully supported by donations to the Florida Panther Research and Management Trust Fund via the registration of “Protect the Panther’’ license plates. We acknowledge the efforts of National Park Service staff in the collection of Florida panther data utilized in this study.