Removal of chronic Mycoplasma ovipneumoniae carrier ewes eliminates pneumonia in a bighorn sheep population

Abstract Chronic pathogen carriage is one mechanism that allows diseases to persist in populations. We hypothesized that persistent or recurrent pneumonia in bighorn sheep (Ovis canadensis) populations may be caused by chronic carriers of Mycoplasma ovipneumoniae (Mo). Our experimental approach allowed us to address a conservation need while investigating the role of chronic carriage in disease persistence. We tested our hypothesis in two bighorn sheep populations in South Dakota, USA. We identified and removed Mo chronic carriers from the Custer State Park (treatment) population. Simultaneously, we identified carriers but did not remove them from the Rapid City population (control). We predicted removal would result in decreased pneumonia, mortality, and Mo prevalence. Both population ranges had similar habitat and predator communities but were sufficiently isolated to preclude intermixing. We classified chronic carriers as adults that consistently tested positive for Mo carriage over a 20‐month sampling period (n = 2 in the treatment population; n = 2 in control population). We failed to detect Mo or pneumonia in the treatment population after chronic carrier removal, while both remained in the control. Mortality hazard for lambs was reduced by 72% in the treatment population relative to the control (CI = 36%, 91%). There was also a 41% reduction in adult mortality hazard attributable to the treatment, although this was not statistically significant (CI = 82% reduction, 34% increase). Synthesis and Applications: These results support the hypothesis that Mo is a primary causative agent of persistent or recurrent respiratory disease in bighorn sheep populations and can be maintained by a few chronic carriers. Our findings provide direction for future research and management actions aimed at controlling pneumonia in wild sheep and may apply to other diseases.


| INTRODUC TI ON
Heterogeneity in infectiousness of individual hosts may dramatically affect pathogen transmission, as illustrated by the role of human superspreaders in the emergence of the severe acute respiratory syndrome-associated coronavirus (SARS-CoV), a zoonotic virus responsible for an acute human epidemic during 2002(Lloyd-Smith, Schreiber, Kopp, & Getz, 2005Woolhouse et al., 1997). Less is known about the importance of individual variation in infectiousness among diseases associated with chronically carried bacterial pathogens. Chronic carriage/shedding is common features of several pathogenic bacteria, so understanding infection risks posed by chronic carrier individuals is relevant to management of infectious diseases across human, domestic animal, and wildlife health sectors (Buhnerkempe et al., 2017;Coyne et al., 2006;Wertheim et al., 2005).

| Study area
We conducted our study in the South Dakota Black Hills from August 2014 to May 2018. This study site includes elevations of 972-2,207 m (Brown & Sieg, 2016), average annual precipitation (Rapid City) of 11.3 cm rainfall and 29.6 cm snow, and temperatures of −30 to 41°C, with an average high of 17°C and an average low of The control population inhabited public and private land in Pennington County and contained three subherds that rarely interacted (Rapid Creek, Spring Creek, and Hill City, Figure 2; 44°04′37″ N, 103°21'16" W). This population utilized canyons for summer range and residential lawns during winter (Smith, Grovenburg, Monteith, & Jenks, 2015). An all-age die-off occurred in 2009, and pneumonia remained a major mortality source . The

| Radio-marking, data collection, pathogen detection, and strain typing in adults
We captured bighorn sheep via chemical immobilization (BAM; 0.43 mg/kg butorphanol, 0.29 mg/kg, azaperone, 0.17 mg/kg medetomidine, Wildlife Pharmaceuticals) delivered through dart injection or by aerial net-gunning (Quicksilver Air, Inc. and Hells Canyon Helicopters). We fitted sheep with very high-frequency (VHF) collars (M252OB; Advanced Telemetry Systems) with mortality sensors, which were activated if the collar was motionless for ≥8 hr. After confirming pregnancy with ultrasonography (E.I. Medical Imaging), we fitted ewes with VHF vaginal implant transmitters (VITs; M3900; Advanced Telemetry Systems; Smith, Walsh, et al., 2014).
During capture, we collected information on sex, age, Mo carrier status and antibodies, and presence of other pathogens. We aged individuals up to 3.5 years old using tooth eruption (Valdez & Krausman, 1999) and classified adults >3.5 years old as a single age group. We collected Mo mucosal samples via three swabs, which were consecutively inserted deep into each of the nares and rotated around the cavity wall during removal (Drew et al., 2014). We returned two swabs to their sheath and immersed one in Tryptic Soy Broth with 15% Glycerol (Hardy Diagnostics; Butler et al., 2017).
To sample for other aerobic bacteria that could contribute to respiratory disease, we rotated 3 swabs along each tonsillar crypt and stored them similarly to nasal swabs. We collected blood for serum, held all samples at 4°C, and shipped them on ice to the Washington Animal Disease Diagnostic Lab (WADDL) or Dr. Thomas Besser's lab (Washington State University).
We used real-time polymerase chain reaction (RT-PCR) to detect Mo on nasal swabs (Ziegler et al., 2014). Samples were positive if fluorescence generation exceeded the threshold before the 36th RT-PCR cycle, indeterminate if between the 36th and 40th cycles, and negative if not exceeded through 40 cycles. We determined serum Mo antibody presence by competitive enzyme-linked immunosorbent assay (c-ELISA; Ziegler et al., 2014). We documented other pathogens on swabs through aerobic culture and PCR (Besser et al., 2008).
We determined Mo strain type using genomic DNA extracted from M. ovipneumoniae broth cultures or swabs using DNeasy blood and tissue kits (Qiagen Inc.), following manufacturer's instructions, or PCR-positive DNA extracts obtained directly from WADDL. Mycoplasma ovipneumoniae extracts were genotyped using a multi-locus sequence typing approach that targets four loci.
The targeted loci are partial DNA sequences from the 16S-23S intergenic spacer region (IGS), the small ribosomal subunit (16S), and housekeeping genes encoding RNA polymerase B (rpoB) and gyrase B (gyrB). Protocols and primers for PCR amplification of these loci were described previously ). Commercial service laboratories (Amplicon Express and Eurofins Genomics) conducted bidirectional Sanger DNA sequencing of amplified PCR products using the same primers used in PCR reactions.

| Chronic carrier identification and experimental removal
We began Mo testing in the treatment population in August 2014.
We obtained ≥2 tests from every adult alive in the population before or shortly after chronic carrier removal (Table 1 and Table S1). We classified individuals as chronic carriers (consistently positive), intermittent carriers (negative and positive tests), or noncarriers (all negative tests). We immobilized and relocated all chronic carriers from the treatment population to South Dakota State University (SDSU).
Postremoval, we tested most individuals in both populations annually (Table S2 and S3).
F I G U R E 1 A conceptual depiction of our experiment, where chronic carriers of Mycoplasma ovipneumoniae are identified in two populations, but only removed in one. Bighorn sheep can be classified as chronic carriers, intermittent carriers, and noncarriers; only chronic carriers need to be removed under our operating hypothesis. If chronic carriers are removed, the population should rebound as lamb recruitment improves. Without intervention, the population will continue to decline indefinitely due to low lamb recruitment

| Lamb capture
We monitored VIT radio signals daily using handheld directional radio telemetry (Telonics, Inc.) starting 15 April 2016 and 2017. We located the dam to verify lamb presence when a VIT was expelled. Budgetary constraints prevented VIT implantation in some pregnant ewes in the control population, so we visually monitored females without VITs along with ewes that expelled their VITs prematurely. If we observed a lamb, we attempted to hand capture it. We immobilized lambs we failed to capture as neonates at 2-6 months old. We weighed captured lambs, determined sex, and fitted them with expandable VHF collars (M4200; Advanced Telemetry Systems). We wore latex gloves and minimized handling of neonates to avoid abandonment (Smith, Walsh, et al., 2014). The SDSU Institutional Animal Care and Use Committee approved capture and handling procedures (Approval number 16-00A).

| Survival monitoring
We monitored adult collars for mortality signals 2 times/week from 13 March 2016 to 1 May 2018 and lamb collars daily from capture until 20 November; thereafter, we monitored lambs similarly to adults. When a collar indicated mortality, we located the carcass and examined the site for predator presence (scat, tracks, scrapes) and the cadaver for caching, hemorrhaging, and skeletal disarticulation (Stonehouse, Anderson, Peterson, & Collins, 2016). We performed necropsies on adults in the field and shipped swabs to WADDL for pathogen testing. When possible, we shipped lamb cadavers to WADDL for necropsy.
Cause of death was sometimes uncertain; therefore, we used the knowledge gained from necropsies, evidence at the site, and behavioral observations to assign the likelihood of each individual's cause of death. Cause-specific mortality categories for lambs and adults were "predation," "pneumonia," and "other," and, in addition for adults only, "human-caused." We created a probability for each category, and the resulting vector of prior predictive probabilities summed to one for each individual. If cause of death was certain, the vector contained a single, nonzero entry for the appropriate causeof-death category (Walsh, Norton, Storm, Van Deelen, & Heisey, 2018). As an example of this process, a carcass might be scavenged shortly after dying, and movement of the collar could delay detection of the mortality by a week. The lack of disarticulation might rule F I G U R E 2 Ranges of study populations of bighorn sheep in the Black Hills, South Dakota, USA, 2014-2018 out predation, but the carcass's desiccated state would make it difficult to tell whether pneumonia or another internal malady caused death. If we had previous behavioral evidence that indicated pneumonia, in this example we would assign a 70% predictive probability to "pneumonia," 30% to "other," and 0% to all other categories.

| Adult a priori models
Our adult global model calculated log unit cumulative hazard as the baseline log unit cumulative hazard rate. For the i th individual during the j th week, we denoted β treatment as the treatment effect, β positive test as the positive test effect, β strain as the strain type effect, β year as the year effect, β age as the individual's age effect, and β sex as an individual's sex effect. We signified the week effect with ρ j .

| Lamb covariates
We modeled daily survival rates, incorporating treatment (i.e., population identity), year (Gaillard et al., 2000), strain type exposure (Cassirer TA B L E 1 Summary of bighorn sheep pathogen testing results and survival estimates in the treatment and control populations, We coded treatment and individual sex effects as with adults. We designated year as a binary variable (2016 = 0, 2017 = 1) and birth timing consisted of three groups: lambs born within ± 3 days of the median date of lamb births in a given year, those born >3 days before the peak period, and those born >3 days after the peak period . We coded strain type exposure as categorical (exposure to introduced strain = 2, exposure to resident strain = 1, exposure to neither = 0). Based on strain typing data, lambs in the Spring Creek subherd were exposed to the introduced strain in 2017, and the resident strain in 2016. All lambs in Rapid Creek and Hill City subherds were classified as only exposed to the resident strain for the duration of the study. Since we removed all carriers of Mo in the treatment population in January 2016 and there were no Mo-positive tests postremoval, we classified treatment lambs as being exposed to neither strain for the entire study. Birth weight was measured to the nearest 0.10 kg. For late-caught lambs, we imputed missing birth weight and birth timing values using the empirical distributions of these variables for lambs caught as neonates (Gelman et al., 2014).

| Lamb a priori models
Our global model calculated daily log unit cumulative hazard as ln(Λ i,j ) = γ + β treatment × treatment i + β strain [strain type i ] + β year × year i + β sex × sex i + β birth timing [birth timing i ] + β birth weight × birth weight i + ρ j .
The γ, β treatment , and β sex parameters were included in the model in the same way as in the adult model. For the i th individual during the j th day, we denoted the year effect as β year , β strain as the strain type exposure effect, β birth timing as the birth timing effect, and β birth weight as the birth weight effect. We signified the day effect with ρ j . We then calculated cause-specific mortality while incorporating observer uncertainty into parameter estimation (Walsh et al., 2018).

| Analysis framework
Specifically, we treated the true cause of death for each individual as a latent, unknown variable with an assigned vector of prior predictive probabilities. As described above, these priors specified the observer's belief that each cause of death was the true cause of death given their assessment of the available evidence (Table S4).
We imputed the true cause of death, using a data augmentation approach that generated a cause of death at each Markov chain Monte Carlo (MCMC) iteration (Gelman et al., 2014), based on a categorical distribution with a parameter vector equal to the prior predictive probability vector specified for that individual. Using random starting values, we ran three MCMC chains for 100,000 iterations and removed the first 10,000 repetitions for burn-in. We looked for evidence of nonconvergence of the chains via graphical checks, and none was observed for any of the following analyses.
We calculated Watanabe-Akaike Information Criteria (WAIC) from each model to identify models that best described the evidence in the data (Gelman et al., 2014). We considered models differing by ≤2 WAIC as alternatives to the top ranked model but preferred the simplest model (Burnham & Anderson, 2002). We based our conclusions on parameter estimates from the best model.

| Collaring effort
We radio-collared 86 adult bighorn sheep between 1 August 2014 and 1 May 2018:29 in the treatment (10 rams, 19 ewes) and 57 (19 rams, 38 ewes) in the control population. This constituted 100% and ~90% of adults in the respective populations by the study's end.

| Chronic carrier identification and removal in the treatment population
We tested all treatment population adults that survived long enough to be tested (7 rams, 17 ewes) for Mo over 60 sampling events (Table 1). Fifty-one (85%) swabs tested negative with RT-PCR, 1 (2%) tested indeterminate, and 8 (13%) tested positive. Two females were classified as chronic carriers of the BH-058 strain (n = 7 positive tests). The other adult that tested positive subsequently tested negative and died before experimental manipulation. We classified all other individuals tested twice as noncarriers (n = 19 individuals). We removed the two chronic carriers on 13 March 2016.

| Postremoval pathogen testing in the treatment population
After removal of chronic carriers, we did not detect Mo in 35 samples (26 negative and 9 indeterminate) collected from 26 individuals (9 males, 17 females) in the treatment population (Table 1 and   Table S3). We collected serum from 20 individuals, and antibodies were detected in 10 (50%); 2 (10%) tested indeterminate, and 8

| Pathogen testing and strain typing in the control population
Between 1 January 2016 and 1 May 2018, we collected 80 samples from 57 control population bighorn sheep (Table 1 and Table S2).
We obtained 38 Mo-positive samples (48%), 9 indeterminate samples (11%) and 32 negative samples (41%). We were able to strain type 36 of the 38 positive samples. We found that 24 samples were the resident BHS-058 strain also detected in the treatment population, but 11 samples were typed as BHS-043 (introduced and 2 (4%) tested negative. We lacked serum from 6 individuals. We
Hazard of pneumonia-induced mortality for adults in the treatment population was significantly lower than in the control population (Probability difference = −52%, CI = −78%, −15%; Figure 3c).

| D ISCUSS I ON
We found that pneumonia can be maintained in bighorn sheep populations by a few individuals chronically carrying Mo. After removing these individuals, we detected no deaths attributable to pneumonia and 76% of lambs survived to 6 months of age, similar to other healthy populations in our study region (37.5%-90%; Parr et al., 2018;Zimmerman, 2008). In contrast, we detected pneumonia-induced mortality in adults and juveniles in a control population where Mo carriers remained. Average lamb survival to 6 months in the control population was 35%, which is similar to other unhealthy populations (0%-60% of collared lambs surviving; Cassirer et al., 2018;Grigg et al., 2017;. We detected other pneumonia-associated pathogens in both populations, but their presence failed to induce pneumonia in adults or lambs in the absence of Mo (Table 1). This finding makes physiological sense: while other pathogens contribute to disease , Mo appears necessary, by disrupting mucociliary clearance, for bighorn sheep to establish lung infections (Cassirer et al., 2018;Niang et al., 1998). Although leukotoxigenic Pasteurella was detected in the treatment population before removal and not afterward, the three individuals harboring it were not removed (Table S1 and Table S3). Our results indicate that Mo is warranted as a focal pathogen in efforts to eliminate pneumonia in wild sheep populations.
We found 11 intermittent carriers in our control population but none in our treatment population (Table 1). Strain typing of samples from intermittent carriers in the control population revealed that eight changed from noncarriers to carriers when infected with a strain previously found in bighorn sheep in Deadwood, South Dakota and western Nebraska (Kamath et al., 2019). These individuals therefore were not intermittently carrying a single strain of Mo. More likely, they lacked strain-specific immunity and were acutely carrying the introduced strain of Mo .
Importantly, they were not contributing to the persistence of the original strain. The role of intermittent carriers in population-level dynamics of Mo is not known ; however, our assumption is that chronic carriers are necessary for persistence.
Our study indicates that in some cases intermittent carriage can be attributed to the introduction of a new strain type. Future studies could consider invasion of new strains as a potential factor influencing Mo carriage patterns.
Although our study demonstrates improved lamb survival resulting from removing chronic carriers, the impact on adult survival is not as pronounced. We detected a statistically significant reduction in pneumonia-induced adult mortality in the treatment population relative to the control, but other sources of mortality in the treatment population offset this effect. For adults, our best survival model included a treatment effect and was significantly better (i.e., ≥2 ΔWAIC) than models lacking this effect; however, the effect was not statistically significant. This finding may indicate that the treatment effect was biologically important, but a larger sample size is needed to assess statistical significance given the overall high survival rates of both the treatment and control populations. Previous studies found that adult survival generally rebounded to or above previous levels in the years following all-age die-offs (Manlove, Cassirer, Cross, Plowright, & Hudson, 2016;Plowright et al., 2013).
Clarifying It remains unclear which host factors contribute to chronic carriage and how to minimize the number of tests necessary to identify chronic carriers. Plowright et al. (2017) found that age and homozygosity at a specific locus were associated with persistent carriage of Mo in bighorn sheep, suggesting possible host factors that might contribute to variation in infectious period and carriage rates among populations. Coinfections, underlying disease, and high exposure frequency and/or dose may also shape host resistance to infection and increase likelihood of chronic or intermittent carriage (Fox et al., 2015). Chronic carriers might also be individuals that adopt a strategy of tolerance (limiting the harm caused by a given parasite burden) rather than resistance (limiting parasite burden) to combat Mo.
This predisposition for tolerance has a genetic basis and is variable among individuals within a species (Råberg, Graham, & Reed, 2009).
Identifying factors associated with carriage status would provide insight into temporal variation in disease dynamics and population response to infection. The ability to determine the likelihood of intermittent versus chronic carriage at an individual or population level would also help minimize the sampling effort required to efficiently identify chronic carriers. Our results indicate the minimum required sampling intensity will vary across populations.
Given the cost of identifying chronic carriers, wildlife managers might question whether complete depopulation followed by repopulation is preferable to selective removal. However, complete depopulation of wildlife is not always feasible, especially in rugged or remote terrain (Courchamp, Chapuis, & Pascal, 2003). Furthermore, while depopulation removes the pathogen along with its host, ensuing reintroductions may fail (Griffith, Scott, Carpenter, & Reed, 1989) depends on how the disease is spread and maintained within and between populations. As infectious disease outbreaks increase globally (Jones et al., 2008), broadening our understanding of various transmission patterns will be integral to mitigating disease outbreaks and conserving at-risk wildlife populations. Wieseler, J. Jenson., and B. Felts were integral to completing field work. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

CO N FLI C T O F I NTE R E S T
None declared.