Kansas witnessed an unprecedented outbreak in Corynebacterium pseudotuberculosis infection among horses, a disease commonly referred to as pigeon fever during fall 2012. Bayesian geostatistical models were developed to identify key environmental and climatic risk factors associated with C. pseudotuberculosis infection in horses. Positive infection status among horses (cases) was determined by positive test results for characteristic abscess formation, positive bacterial culture on purulent material obtained from a lanced abscess (n = 82), or positive serologic evidence of exposure to organism (≥1:512)(n = 11). Horses negative for these tests (n = 172)(controls) were considered free of infection. Information pertaining to horse demographics and stabled location were obtained through review of medical records and/or contact with horse owners via telephone. Covariate information for environmental and climatic determinants were obtained from USDA (soil attributes), USGS (land use/land cover), and NASA MODIS and NASA Prediction of Worldwide Renewable Resources (climate). Candidate covariates were screened using univariate regression models followed by Bayesian geostatistical models with and without covariates. The best performing model indicated a protective effect for higher soil moisture content (OR = 0.53, 95% CrI = 0.25, 0.71), and detrimental effects for higher land surface temperature (≥35°C) (OR = 2.81, 95% CrI = 2.21, 3.85) and habitat fragmentation (OR = 1.31, 95% CrI = 1.27, 2.22) for C. pseudotuberculosis infection status in horses, while age, gender and breed had no effect. Preventative and ecoclimatic significance of these findings are discussed.
Citation: Boysen C, Davis EG, Beard LA, Lubbers BV, Raghavan RK (2015) Bayesian Geostatistical Analysis and Ecoclimatic Determinants of Corynebacterium pseudotuberculosis Infection among Horses. PLoS ONE 10(10): e0140666. https://doi.org/10.1371/journal.pone.0140666
Editor: Glenn F. Browning, The University of Melbourne, AUSTRALIA
Received: January 19, 2015; Accepted: September 29, 2015; Published: October 16, 2015
Copyright: © 2015 Boysen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: Due to ethical restrictions related to protecting horse owner confidentiality, data cannot be made publicly available. Data are available upon request, and interested researchers may submit requests to the corresponding author or directly to the Kansas State University's Institutional Review Board.
Funding: This study was financially supported by the Kansas State Veterinary Diagnostic Laboratory and Veterinary Health Center at Kansas State University. These funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Corynebacterium pseudotuberculosis is a pleomorphic, gram-positive, facultative intracellular bacterium with a worldwide distribution that causes disease in cattle, sheep, goats and horses  . The bacterium is soil-borne, and horses are believed to acquire it from inhaling infected soil or through contamination of abraded skin, and also through biting flies . The bacterium causes three clinical syndromes in horses, the most prevalent of which is the formation of external abscesses that contain abundant purulent material and are characteristically located in the pectoral region, ventral abdomen, sheath, or mammary gland, which is commonly referred to as pigeon fever, pigeon breast or dryland distemper. The second clinical syndrome occurs less frequently (<10% of cases), as an internal infection with development of either single or multiple abscesses in the abdominal or thoracic cavity. Although internal infections are rare, a reported mortality risk of 40% is seen even with appropriate and aggressive antimicrobial treatment, and in the absence of treatment mortality rate increases to 100% . The third and least common manifestation of disease associated with C. pseudotuberculosis is ulcerative lymphangitis.
Infection with C. pseudotuberculosis is traditionally considered to be an arid region disease perhaps due to its first identification among horses in drier areas of California, where the disease is now enzootic. However, outbreaks of this disease outside this region have been recorded periodically , . Potential environmental or latent climatological influences associated with these outbreaks are seldom reported. Large outbreaks in the Midwestern US such as the one Kansas witnessed in 2012 are so far rare. However, the number of cases seen within this region, particularly in Texas has been increasing over the years, indicating a plausible eastward expansion of this disease’s enzootic range. Between the years 2005–2011, the number of confirmed cases tested at Texas Veterinary Medical Diagnostic Laboratory (TVMDL) increased an annual average of 177% , . Most of these cases clustered around the panhandle and Central Texas regions, which experience semi-arid to subtropical climate, unlike the epizootic arid west. The 2012 outbreak in Kansas also occurred in areas that received relatively higher amounts of precipitation but followed an exceptionally warm winter and hot and dry summer seasons that year , .
While there is a chance for new infections to occur outside the epizootic areas through contact with horses that were infected elsewhere, it is more or at least equally likely for the infections to originate locally from C. pseudotuberculosis that may be present in soils and/or other hosts. This is often suspected to be the source of infection for horses that have had no apparent contact with travelling or infected horses. Information on the ecology and spatial distribution of the pathogen is however lacking. Environmental and climatological influences such as a prolonged drought could trigger infections in new areas , . Understanding such influences may have important preventative and management implications, and as well as help recognize the eco-climatological influences in the ecology and evolution of this disease. Doherr et al.  reported that young adult horses and those that had contact with other horses on summer pastures were at significant risk for pigeon fever in California. Besides this study, evaluations of environmental or climatological risk factors for C. pseudotuberculosis infection, which could vary over geographic regions cannot be found. Soil types and conditions such as moisture levels, and pH could affect the survival and C. pseudotuberculosis infectivity, while a horse’s exposure to extreme climatic events and different ground cover or landscape patterns that support reservoir hosts could increase risk of infection.
In this study, we evaluated associations of such potential risk factors for C. pseudotuberculosis infection among horses that were seen between 2005–2013 period in Kansas and neighboring states. For this, we used medical records of horses that had a positive or negative diagnosis for C. pseudotuberculosis infection that were received at the Kansas State University’s Veterinary Health Center (VHC) and Kansas State Veterinary Diagnostic Laboratory (KSVDL), and publicly available environmental and climatological covariate data. Following univariate screening of candidate variables, models were constructed in a Bayesian hierarchical framework, which has advantages over traditional methods when there is a potential for spatial influence on disease data , .
Materials and Methods
2.1 Study region
Most cases reported to VHC and KSVDL during the 2012 outbreak originated from Kansas, and few were from neighboring Nebraska and Missouri (Fig 1). With a mid-latitude and continental location, temperature extremes in Kansas occur in July (26°C) and January (−1.6°C) (30 year averages). Over 70% of the precipitation occurs during the warm season, with a peak in June (105 mm) and the minimum in January (18 mm) (30 year averages) . There are large variations in the yearly precipitation received across the state of Kansas, with eastern Kansas receiving up to three times more rainfall than west . As a result, climate and vegetation are transitional between the humid east and semi-arid western portions of Kansas.
2.2.1 Ethics statement.
This study did not involve human subjects. No patient owner names were recorded or used during the completion of this study to maintain patient owner confidentiality and to adhere to the International Ethical Guidelines for Biomedical Research Involving Human Subjects. All patient owner records/information was anonymized and de-identified prior to analysis.
2.2.2 Epidemiological data.
All medical records entered between January 2005 and April 2013 were searched from the VHC and KSVDL databases that had keywords typically associated with pigeon fever infection in horses (Equine; Corynebacterium; pseudotuberculosis; pigeon fever; pectoral; abscess). A total of 317 unique records were retrieved and the owner’s address at the time infection was diagnosed were digitized, out of which 32 (10.0%) had business addresses provided and were therefore removed. On further spatial evaluation of the remaining records using land cover/land use data  31 (8.6%) records revealed that the land within 2.5 km area surrounding their addresses contained built environment, suggesting that the horses had lived elsewhere and were therefore removed. From the remaining 254 records all information pertaining to the diagnosis (presence or absence of C. pseudotuberculosis infection in horses), horse demographics (age, sex, and breed), and season of case presentation (spring = March–May, summer = June–August, Fall = September–November, Winter = December–February) was digitized. Horses were grouped into eight categories by 5-year ages groups (<1, 1 to 6, 6 to 11, 12 to 17, 18 to 23, 24 to 29, >29, and Unknown) and by sex into four groups (Male, Female, Male neutered, and Unknown) (Table 1).
The inclusion criteria employed for selecting cases and controls that were enrolled in this study is as following. Medical records that indicated positive bacterial culture for C. pseudotuberculosis on purulent material obtained from abscess and/or a positive titer (≥ 512) on synergistic hemolysis-inhibition titer (SHI) test  were considered to confirm positive infection status in horses. Among horses that were presented at VHC, those that had initially displayed characteristic abscess formation (n = 25) were tentatively determined to be positive for C. pseudotuberculosis infection. Positive bacterial culture on purulent material obtained from abscess further confirmed infection status among these horses. Horses received at VHC that show symptoms of respiratory disease or systemic inflammatory disease that are consistent with pigeon fever but without visible external abscess are routinely referred for screening using SHI test. During the study period, horses that met these criteria (n = 21) were referred for SHI testing, and 14 were later confirmed positive based on an SHI titer ≥ 1:512. In two of these cases horses had both a positive culture and an SHI titer ≥ 1:512. Among horses that were referred for diagnostic testing at KSVDL, case records that indicated positive bacterial culture for C. pseudotuberculosis from purulent material obtained from abscess (n = 26) and SHI titer ≥ 1:512 (n = 17) confirmed pigeon fever status in horses. The above inclusion criteria resulted in a total of 82 cases.
The inclusion criteria for controls used medical records, queried using the same keywords as above that indicated negative bacterial culture for C. pseduotuberculosis from purulent material obtained from abscess (n = 45), SHI titers ≤ 512 (n = 7, VHC; n = 29, KSVDL), and determination of other etiological causes during necropsy (n = 45) and cytologic examinations (n = 9), routine biopsy (n = 16) and clinical evaluations (n = 19, VHC). This resulted in the inclusion of 172 controls for this study.
2.2.3 Covariate data.
Environmental covariates from three broad thematic groups were obtained from publicly available sources using Geographic Information Systems (GIS) and remote-sensing methods; soil attributes from USDA State Soil Geographic Survey (SSURGO) , land use/land cover attributes from the USGS National Land Cover Dataset (NLCD) , and climate data from NASA Moderate Resolution Imaging Spectroradiometer (MODIS)  and Prediction of Worldwide Renewable Resources (POWER) . Covariate values from these geographic datasets were obtained by constructing 2.5 km buffer areas surrounding locations where horses had resided prior to their diagnosis. A list of all variables evaluated in this study is present in Table 2. Two landscape metrics were derived from the NLCD. The total edge contrast index (TECI) were estimated using Fragstats 4.0  by, where eik is the total length of edge between patch types i and k, E* is the total length of edge in landscape, dik is the dissimilarity (edge contrast weight) between patches i and k; and, largest patch index (LPI) was calculated by,
Where aij is the area of patch i, j and A being total landscape area.
2.3 Statistical analysis
2.3.1 Univariate covariate selection.
Candidate covariates were screened in order to avoid convergence and mixing issues that result when numerous covariates are included in a model. Separate univariate logistic regression models evaluated each covariate independently and those that were significant at p ≤ 0.2 were kept. A logistic regression takes the form,
Where pi is presence of C. pseudotuberculosis infection, β0 the intercept coefficient, and βk the coefficient for the explanatory variable vki(k = 1,..,n). Non-linearity in covariate association with infection status in horses was checked and covariates with non-linear relationship were categorized based on diagnostic scatter plots. The variables, daytime land surface temperature and night time land surface temperature were categorized in to four and three categories, respectively for ease of interpretation despite their relationship with case status being linear. All other variables were kept in their original, continuous format. Multicollinearity among screened variables was tested by estimating the variance inflation factor (VIF) and all variables with a VIF ≥ 10 were considered to indicate multicollinearity , in which case, the least biologically relevant variable was dropped at a time until multicollinearity was absent. Two-way interaction terms were evaluated, and care was taken not to remove candidate variables that were deemed clinically relevant .
2.3.2 Bayesian model specification.
We assumed that infection status Yij of a horse i at location Sj, which takes the value of 1 if the horse has positive diagnosis and 0 otherwise, followed a Bernoulli distribution, Yij ∼ Ber(pij). The method is an extension of a GLM using the logit link function as follows:
The stochastic spatial component Si was modeled as a zero mean Gaussian process with variance σ2 and autocorrelation function, ρ(dij, θ), where dij = xi − xj measures the Euclidean distance between xi and xj, and θ = [ϕ,σ2,τ2], where ϕ and τ are the decay and nugget parameters in a resulting semivariogram. The spatial component helps to account for the spatial variation in the residuals after model fit using covariates alone. The model parameters were estimated using a Bayesian framework with a Markov chain Monte Carlo (MCMC) algorithm in OpenBugs program  in a Beocat cluster computing environment . Distributions of covariate effects on C. pseudotuberculosis prevalence are seldom available for the region; therefore non-informative, uniform priors were selected for the regression parameters, βk and their variance component, . This allows the observed data to have the greatest influence on posterior distributions without being constrained by the choice of prior and can also improve MCMC convergence .
2.3.3 Bayesian geostatistical modeling.
Two Bayesian geostatistical models were fit to describe C. pseudotuberculosis infection status among horses. The first model included only the intercept and stochastic spatial component, Si, and a second model included the covariates screened in the univariate procedure as fixed effects along with Si. The reason for excluding/including the covariates was to determine if this had an effect on the model fit and to determine whether the covariates accounted for part or all of the spatial autocorrelation in pigeon fever data. For both models an initial burn-in of 10,000 iterations was allowed (later discarded), followed by 100,000 simulations after which the intercept (α), smoothing parameter (κ), decay parameter (ϕ) and covariate parameter estimate values were recorded. In order to determine whether sufficient iterations had been conducted to fully describe the posterior distributions, Monte Carlo error was calculated for each variable and if was ≤ 0.05, it was decided that sufficient iterations had been conducted . For the second model, a full model with all covariates was fitted followed by removal of covariates one at a time. If the removal of a covariate resulted in higher DIC value (≥5) units we considered that to indicate poor performance and that variable was retained in the model. Previously removed covariates did not re-enter the final model. The median covariate estimates from the posterior distribution were calculated and exponentiated to provide odds ratios (ORs) along with their corresponding uncertainty measures (95% Bayes credible intervals). Potential confounders (age, sex, and breed) were inserted in the final Bayesian model one at a time and changes to DIC values and parameter estimates were recorded. If these additions resulted in ≥10% change to the posterior median estimates then those variables were retained in the model and adjusted ORs recorded.
The observed spatial distribution of C. pseudotuberculosis infection in Kansas based on VHC and KSVDL medical records is present in Fig 1. Most cases recorded during the study period were reported during the fall of 2012 (Sep–Nov) (Fig 2A and 2B), indicating that the infection had occurred during the preceding summer months or late spring in the region. From the univariate, deterministic, non-spatial logistic models, five variables retained their statistical significance and others were discarded. Higher soil moisture levels had a protective effect (OR = 1.22, 95% CI =1.08, 1.30) on C. pseudotuberculosis infection status in horses; and, mixed forests (OR = 1.18, 95% CI = 1.04, 1.12), grassland/herbaceous cover (OR = 1.22, 95% CI = 1.08, 1.30), total edge contrast index (OR = 1.33, 95% CI = 1.17, 2.21), and day time land surface temperature (≥35°C) (OR = 2.28 95% CI = 2.14, 3.31) had significantly increased the odds of infection status in horses. Of these variables, mixed forest and grassland/herbaceous cover did not retain significance in the Bayesian geostatistical model with covariates, and, individual host factor covariates (age, sex and breed) were not significantly associated with case status when they were inserted one at a time. Box plots showing differences in statistical distribution of significant univariate variables and those retained in the Bayesian geostatistical model surrounding case and control locations are present in Figs 3 and 4, respectively.
Soil moisture readings pertain to the soil moisture profile for a referenced map unit in a 2.5 km buffer area surrounding case control locations, recorded every month and averaged annually and describe the representative soil moisture situation for the map unit throughout the year . Total edge contrast index, a measure of landscape fragmentation was measured by estimating the sum of lengths of each edge segment in the landscape multiplied by the corresponding contrast weight, divided by the total length of edge in the landscape and multiplied by 100 for percentages, for a 2.5 km buffer area surrounding case control locations .
The Bayesian geostatistical model without any covariates had a DIC value of 740 and the value of the decay parameter for spatial correlation (ϕ) was 0.3, indicating that a high degree of spatial correlation of C. pseudotuberculosis infection prevalence was evident between locations with relatively large distances separating them. From the Bayesian geostatistical model which included all covariates, the removal of mixed forest and grassland/herbaceous cover one at a time did not significantly affect the model performance, but removing the remaining three covariates one at a time resulted in ≥10% increase to the DIC value and were therefore retained (Table 3). The final covariate model had a lower DIC value (681) relative to the model without any covariates, indicating that the presence of covariates accounted for additional spatial autocorrelation in the infection status among horses. The decay parameter for spatial correlation (ϕ), was still 0.3. Inclusion of host factors (age, gender and breed) did not change the posterior parameter estimates by ≥10%. The posterior distributions for all parameters followed a normal distribution pattern, and values for individual covariates revealed no convergence issues within the burn-in period and no mixing issues were noted.
Physical environment and climatological factors influenced C. pseudotuberculosis infections among horses in the study region; and, based on the spatial correlation estimate, it appears highly likely that pigeon fever infections in the study region were influenced by common factors. As in previous studies , , seasonality in case arrivals was noted in this study, most occurring during late summer and fall months and are likely due to infections acquired in the spring and early summer. Infection with C. pseudotuberculosis has been noted to be common among young adult horses  but like in the present study gender and breed were not significantly associated with case status in prior studies. The lack of significant pigeon fever association with a horse’s age in this study is similar to Hall et al.,  in their epidemiological investigation of an outbreak in 1997 in Western Colorado. It is likely that horses of all ages are equally susceptible to C. pseudotuberculosis under outbreak conditions in naïve populations.
Non-spatial models used for analyzing data with inherent spatial structure (as in this study) can lead to biased regression parameters, underestimated standard errors, narrow confidence intervals and overestimation of parameter coefficients , . We therefore used a generalized linear geostatistical model in a Bayesian framework to account for spatial autocorrelation and to incorporate uncertainties in the input data and model parameters. This approach likely provided more accurate parameter and significance estimates and increased predictive accuracy , , . Cases of pigeon fever recorded over several years (2005–2013) were modeled in this study; however, most of the cases occurred during the 2012–2013 period spread over approximately four month period. Despite this, temporal or spatio-temporal interaction effects was not noted in the disease data during the initial screening of variables. This could be because of single mode in the outbreak data and the relatively wide-spread occurrence over a four month period.
Soil is a reservoir for C. pseudotuberculosis and at least one of its biovars, equi has recently been shown to survive in different soil types under wide ranging environmental (humidity, ambient temperature) conditions . In the present study, areas with soil that can hold higher moisture levels had a protective effect from C. pseudotuberculosis infection for horses, and conceivably this effect is similar for horses in other regions as well. Soil types that retain higher levels of moisture typically have higher clay content, and are perhaps covered by vegetation and permit less run-off . Soil-borne bacterial ecology is highly regulated by soil physiochemical properties such as organic matter content, moisture and pH levels, the significance for many of which are currently not known for C. pseudotuberculosis. Future evaluations of such effects are worth considering.
Horses that had positive diagnosis for C. pseudotuberculosis infection in the present study had experienced weekly average land surface temperatures (LST) greater than 35°C(95°F) for at least a seven day period or more, eight to twelve weeks prior to diagnosis. Higher LSTs for prolonged periods of time (variable on different ground cover types) is an indicator used by meteorologists for determining drought conditions, which has long been suspected to be associated with C. pseudotuberculosis infection. The Great Plains region in general and Kansas in particular experienced prolonged drought conditions during the spring and summer months in 2012 , and recorded historically high temperatures combined with low rainfall , , , circumstances that may have favored an outbreak. More proximal causes for drought related disease incidence may include multiple factors. Soil is highly prone to wind erosion under drought conditions , consequently favoring inhalation of C. pseudotuberculosis containing airborne dust. Weakened immune system in horses exposed to persistently high temperatures or drought stress and any soil biochemical processes that boost C. pseudotuberculosis infectivity could also be proximal factors. Soil warming and drought-like conditions have resulted in altered microbial community characteristics and soil enzymatic activity under field conditions [36–38], and improving our understanding on any such effects regarding C. pseudotuberculosis may help devising preventive methods for this disease.
Total Edge Contrast Index (TECI), a measure of landscape fragmentation  was significantly associated with C. pseudotuberculosis infection status among horses in the present study. Horses that were exposed to highly fragmented landscapes may have received infection through biting insects inhabiting such landscapes, potentially also involving a transmission cycle that could include wildlife reservoirs for C. pseudotuberculosis. Landscape fragmentation leads to more and smaller landscape patches, increased patch isolation, decreased complexity of patch shape, and higher proportions of edge habitats , conditions that lead to loss of biodiversity and “dilution effect” , that subsequently favor arthropod-borne disease transmission to wildlife and humans , . Insects have been incriminated as mechanical vectors for C. pseudotuberculosis , , and increased activity of flying-insects was observed by horse owners weeks prior to their horses received infection . It is possible for insects to act as vectors for this pathogen from infected to susceptible hosts, some of which could also include other ungulates that inhabit fragmented landscapes. Kelly et al., 2012  reported the isolation of C. pseudotuberculosis from abscess from captive elks, and this bacterium has been suggested to infect multiple hosts , including deer that may inhabit fragmented habitats. Therefore, a possibility for a transmission cycle involving insects and wild ruminant hosts seems likely but this premise needs further evaluation.
It is important to discuss a limitation of this study. Our study population originated from medical records maintained at two facilities, Kansas State University’s VHC and KSVDL, and not private practices or other sources in the region where horses may have been treated for C. pseudotuberculosis infection. Despite these two veterinary medical facilities being the largest in the state with a broad clientele in the region, there does exists a statistical possibility for some of our estimates and risk factors to change if horses from a larger population were to be included; a problem typical of many retrospective case-control studies that can only be fully mitigated by conducting carefully designed prospective sampling. Future, more robust studies are therefore necessary to better understand pigeon fever outbreaks and ecoclimatic drivers of this disease.
Infection with C. pseudotuberculosis occurs in horses of all ages, gender and breed in the study region, and this disease is driven by environmental (soil moisture, landscape fragmentation) and hot and dry climatic conditions (LST ≥ 35°C). The identification of these risk factors for pigeon fever is important in the context of disease prevention. Also, it is noteworthy that soil moisture and land surface temperature are climate-change indexes that are projected to change under different IPCC (Intergovernmental Panel on Climate Change) scenarios as soon as next decade. Studies on the implication of such changes for the ecology and evolution C. pseudotuberculosis and similar soil-borne bacteria that have animal and public health significance are needed.
This study was financially supported by the Kansas State Veterinary Diagnostic Laboratory and Veterinary Health Center at Kansas State University. These funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank two anonymous reviewers for their comments that helped improve an earlier version of this manuscript, and Shana Barnett, Kansas State University, for excellent technical support.
Conceived and designed the experiments: CB RKR EGD LAB. Performed the experiments: CB RKR EGD. Analyzed the data: CB RKR. Contributed reagents/materials/analysis tools: CB RKR EGD LAB BVL. Wrote the paper: CB RKR EGD LAB.
- 1. Aleman M, Spier SJ, Wilson WD, Doherr M (1996) Corynebacterium pseudotuberculosis infection in horses: 538 cases (1982–1993). Journal of the American Veterinary Medical Association 209: 804–809. pmid:8756884
- 2. Pratt SM, Spier SJ, Carroll SP, Vaughan B, Whitcomb MB, Whitcomb B, et al. (2005) Evaluation of clinical characteristics, diagnostic test results, and outcome in horses with internal infection caused by Corynebacterium pseudotuberculosis: 30 cases (1995–2003). Journal of the American Veterinary Medical Association 227: 441–448. pmid:16121612
- 3. Kilcoyne I, Spier SJ, Carter CN, Smith JL, Swinford AK, Cohen VMD (2014) Frequency of Corynebacterium pseudotuberculosis infection in horses across the United States during a 10-year period. Journal of the American Veterinary Medical Association 245: 309–314. pmid:25029310
- 4. Pigeon Fever: Confirmed cases in Texas horses have soared in six years, TVMDL data shows. Available: http://tvmdl.tamu.edu/wp-content/uploads/2012/07/Pigeon-Fever-handout-10.12.pdf. Accessed 30 July 2015.
- 5. Szonyi B, Swinford A, Clavijo A, Ivanek R (2014) Re-emergence of Pigeon Fever (Corynebacterium pseudotuberculosis) Infection in Texas Horses: Epidemiologic Investigation of Laboratory-Diagnosed Cases. Journal of Equine Veterinary Science 34: 281–287.
- 6. NOAA National Climatic Data Center. BAMS State of the Climate 2012. Available: http://www.ncdc.noaa.gov/bams-state-of-the-climate/2012.php. Accessed 1 October 2015.
- 7. Hoerling M, Eischeid J, Kumar A, Leung R, Mariotti A, Mo K, et al. (2014) Causes and predictability of the 2012 Great Plains drought. Bulletin of the American Meteorological Society 95: 269–282.
- 8. Hepworth-Warren K, Sponseller BT, Wong DM, Kinyon JM (2014). Isolation of Corynebacterium pseudotuberculosis biovar equi from a horse in central Iowa. Case Reports in Veterinary Medicine Article ID 436287, 3 pages, https://doi.org/10.1155/2014/436287
- 9. Doherr MG, Carpenter TE, Hanson KMP, Wilson WD, Gardner IA (1998) Risk factors associated with Corynebacterium pseudotuberculosis infection in California horses. Preventive Veterinary Medicine 35: 229–239. pmid:9689656
- 10. Dunson DB (2001) Commentary: Practical advantages of Bayesian analysis of epidemiologic data. American Journal of Epidemiology 153: 1222–1226. pmid:11415958
- 11. Banerjee S, Carlin BP Gelfand AE (2014). Hierarchical modeling and analysis for spatial data. CRC Press, 2014.
- 12. Raghavan RK, Harrington JA Jr (2013) Communicating Kansas climate and climate change: Initial development of a web-based resource. Papers of the Applied Geography Conference 36:11–16.
- 13. Goodin DG, Mitchell JE, Knapp MC, Bivens RE (2004) Climate and weather atlas of Kansas. An Introduction. Available: http://www.k-state.edu/ksclimate/documents/kgsed.pdf. Accessed: 1 October 2015.
- 14. Fry JA, Xian G, Jin SM, Dewitz JA, Homer CG, Limin Y, et al. (2011) Completion of the 2006 National Land Cover Database for the conterminous United States. PE&RS, Photogrammetric Engineering & Remote Sensing 77: 858–864.
- 15. Jeske JM, Spier SJ, Whitcomb MB, Pusterla N, Gardner IA (2013) Use of antibody titers measured via serum synergistic hemolysis inhibition testing to predict internal Corynebacterium pseudotuberculosis infection in horses. Journal of the American Veterinary Medical Association 242: 86–92. pmid:23234286
- 16. Natural Resources Conservation Service. Soil Geography. Available: http://www.nrcs.usda.gov/wps/portal/nrcs/main/soils/survey/geo/. Accessed 1 October 2015.
- 17. Land Process Distributed Active Archive Center (LP DAAC). MODIS Overview. Available: https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table. Accessed 1 October 2015.
- 18. White JW, Hoogenboom G, Stackhouse PW Jr, Hoell JM (2008) Evaluation of NASA satellite- and assimilation model-derived long-term daily temperature data over the continental US. Agricultural and Forest Meteorology 148: 1574–1584.
- 19. McGarigal K, Cushman SA, Ene E (2012) FRAGSTATS v4: Spatial Pattern Analysis Program for Categorical and Continuous Maps. Available: http://www.umass.edu/landeco/research/fragstats/fragstats.html. Accessed 1 October 2015.
- 20. Hosmer DW, Lemeshow. S. Model‐Building Strategies and Methods for Logistic Regression. Applied Logistic Regression 1990: 91–142.
- 21. OpenBugs. Available: http://www.openbugs.net/w/FrontPage. Accessed 1 October 2015.
- 22. Beocat Documentation. Available: https://www.cis.ksu.edu/beocat/documentation. Accessed 1 October 2015.
- 23. Gelman A, Carlin JB, Stern HS, Rubin DB (2004) Posterior simulation: In: Bayesian Data Analysis. Boca Raton, FL: Chapman and Hall/CRC. pp 283–310.
- 24. Zacarias OP, Andersson M (2011) Spatial and temporal patterns of malaria incidence in Mozambique. Malaria Journal 10.
- 25. Hall K, McCluskey BJ, Cunningham W (2001) Corynebacterium pseudotuberculosis infections (pigeon fever) in horses in western Colorado: An epidemiological investigation. Journal of Equine Veterinary Science 21: 284–286.
- 26. Lawson AB (2013) Bayesian disease mapping: hierarchical modeling in spatial epidemiology. Vol. 2. CRC Press.
- 27. Dormann CF, McPherson JM, Araujo MB, Bivand R, Bolliger J, Carl G, et al. (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30: 609–628.
- 28. Cliff A, Ord K (1972) Testing for spatial autocorrelation among regression residuals. Geographical Analysis 4:267–284.
- 29. Schabenberger O, Gotway CA (2004) Statistical methods for spatial data analysis. CRC Press.
- 30. Spier SJ, Toth B, Edman J, Quave A, Habasha F, Garrick M, et al. (2012) Survival of Corynebacterium pseudotuberculosis biovar equi in soil. Veterinary Record 170.
- 31. Munson SM, Belnap J, Okin GS (2011) Responses of wind erosion to climate-induced vegetation changes on the Colorado Plateau. Proceedings of the National Academy of Sciences of the United States of America 108: 3854–3859. pmid:21368143
- 32. Kansas Office of the Governor, Declaration of federal disaster areas. Available: http://www.kwo.org/Drought.html. Accessed 1 October 2015.
- 33. Blunden J, Arndt DS, editors. State of the Climate in 2012. Bulletin of the American Meteorological Society: (2013) 94:S1–S238.
- 34. NOAA National Climatic Data Center. National Overview–Annual 2012. Available: http://www.ncdc.noaa.gov/sotc/national/2012/13. Accessed 1 October 2015.
- 35. Kuhn NJ (2014) Soil Loss 3. The Soil Underfoot: Infinite Possibilities for a Finite Resource (2014): 37.
- 36. Hueso S, Garcia C, Hernandez T (2012) Severe drought conditions modify the microbial community structure, size and activity in amended and unamended soils. Soil Biology & Biochemistry 50: 167–173.
- 37. Acosta-Martinez V, Moore-Kucera J, Cotton J, Gardner T, Wester D (2014) Soil enzyme activities during the 2011 Texas record drought/heat wave and implications to biogeochemical cycling and organic matter dynamics. Applied Soil Ecology 75: 43–51.
- 38. Hugh AL Henry (2012) Soil extracellular enzyme dynamics in a changing climate. Soil Biology and Biogeochemistry 47: 53–59.
- 39. Fahrig L (2003) Effects of habitat fragmentation on biodiversity. Annual review of ecology, evolution, and systematics 34: 487–515.
- 40. Schmidt KA, Ostfeld RS (2001) Biodiversity and the dilution effect in disease ecology. Ecology 82: 609–619.
- 41. Halos L, Bord S, Cotte V, Gasqui P, Abrial D, Boulouis HJ, et al. (2010) Ecological Factors Characterizing the Prevalence of Bacterial Tick-Borne Pathogens in Ixodes ricinus Ticks in Pastures and Woodlands. Applied and Environmental Microbiology 76: 4413–4420. pmid:20453131
- 42. Robinson GR, Holt RD, Gaines MS, Hamburg SP, Johnson ML, Fitch HS, et al. (1992) Diverse and contrasting effects of habitat fragmentation. Science 257: 524–526. pmid:17778686
- 43. Spier SJ, Leutenegger CM, Carroll SP, Loye JE, Pusterla JB, Carpenter TE, et al. (2004) Use of a real-time polymerase chain reaction-based fluorogenic 5 ' nuclease assay to evaluate insect vectors of Corynebacterium pseudotuberculosis infections in horses. American Journal of Veterinary Research 65: 829–834. pmid:15198224
- 44. Kelly EJ, Rood KA, Skirpstunas R (2012) Abscesses in Captive Elk Associated with Corynebacterium pseudotuberculosis, Utah, USA. Journal of Wildlife Diseases 48: 803–805. pmid:22740551
- 45. Lipsky BA, Goldberger AC, Tompkins LS, Plorde JJ (1982) Infections caused by nondiptheria corynebacteria. Reviews of Infectious Diseases 4: 1220–1235. pmid:6760340