Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Ecological Niche Modeling of Bacillus anthracis on Three Continents: Evidence for Genetic-Ecological Divergence?

  • Jocelyn C. Mullins,

    Affiliation Spatial Epidemiology & Ecology Research Laboratory, Department of Geography, University of Florida, Gainesville, Florida, United States of America

  • Giuliano Garofolo,

    Affiliations Istituto Zooprofilattico Sperimentale dell’Abruzzo e del Molise “G. Caporale”, Teramo, Italy, Anthrax Reference Institute of Italy, Istituto Zooprofilattico Sperimentale della Puglia e della Basilicata, Foggia, Italy

  • Matthew Van Ert,

    Affiliations Spatial Epidemiology & Ecology Research Laboratory, Department of Geography, University of Florida, Gainesville, Florida, United States of America, Emerging Pathogens Institute, University of Florida, Gainesville, Florida, United States of America

  • Antonio Fasanella,

    Affiliation Anthrax Reference Institute of Italy, Istituto Zooprofilattico Sperimentale della Puglia e della Basilicata, Foggia, Italy

  • Larisa Lukhnova,

    Affiliation Anthrax Laboratory, Kazakh Science Center for Quarantine and Zoonotic Diseases, Almaty, Kazakhstan

  • Martin E. Hugh-Jones,

    Affiliation Department of Environmental Sciences, School of the Coast and Environment, Louisiana State University, Baton Rouge, Louisiana, United States of America

  • Jason K. Blackburn

    Affiliations Spatial Epidemiology & Ecology Research Laboratory, Department of Geography, University of Florida, Gainesville, Florida, United States of America, Emerging Pathogens Institute, University of Florida, Gainesville, Florida, United States of America


We modeled the ecological niche of a globally successful Bacillus anthracis sublineage in the United States, Italy and Kazakhstan to better understand the geographic distribution of anthrax and potential associations between regional populations and ecology. Country-specific ecological-niche models were developed and reciprocally transferred to the other countries to determine if pathogen presence could be accurately predicted on novel landscapes. Native models accurately predicted endemic areas within each country, but transferred models failed to predict known occurrences in the outside countries. While the effects of variable selection and limitations of the genetic data should be considered, results suggest differing ecological associations for the B. anthracis populations within each country and may reflect niche specialization within the sublineage. Our findings provide guidance for developing accurate ecological niche models for this pathogen; models should be developed regionally, on the native landscape, and with consideration to population genetics. Further genomic analysis will improve our understanding of the genetic-ecological dynamics of B. anthracis across these countries and may lead to more refined predictive models for surveillance and proactive vaccination programs. Further studies should evaluate the impact of variable selection of native and transferred models.


Bacillus anthracis is a soil-borne, spore forming bacteria and the causative agent of anthrax in wildlife, livestock and humans worldwide. Metabolically dormant B. anthracis spores can persist in landscapes with suitable soil and ecological characteristics for long periods of time [1] such that years may pass between outbreaks. More recent evidence suggests B. anthracis may have an active soil life cycle [2]. Despite a long recorded history of anthrax [3], the environmental and epidemiological catalysts for epizootics are poorly understood. Control of the disease in livestock and humans is best achieved through annual vaccination of livestock and adequate surveillance to identify outbreaks early in the epidemic course [4]. In wildlife populations, control is limited to active surveillance and proper disposal of carcasses [5]. The economic conditions in many anthrax endemic areas, however, combined with expansive rural geographies are such that early recognition of outbreaks and proactive distribution of vaccine is challenging, but could be facilitated by active, targeted efforts in areas of high likelihood of occurrence. Ecological niche models of B. anthracis identify geographic areas suitable for pathogen persistence, and these areas should be considered priorities for surveillance and vaccination. In particular, areas predicted to be supportive of the pathogen which also have a history of outbreak clusters should be targeted for active control [6].

Much of the earlier literature on anthrax ecology describes soil conditions that are favorable for B. anthracis persistence, including higher calcium levels and pH [79]. In addition to soil, there is evidence that endemic anthrax areas are associated with warmer temperatures, higher soil moisture content and topography [9,10]. These relationships between potential pathogen persistence and these environmental variables make ecological niche modeling valuable for predicting the spatial distribution of anthrax. Ecological niche models (ENMs) are frequently used to predict the potential distributions of species in ecologic and geographic space. A variety of correlative algorithms are available, all of which aim to identify non-random relationships between known location data for the species and environmental variables and then identify geographic areas of predicted presence. Models developed on a native landscape and determined to accurately predict the species’ presence in the native range can be projected, or transferred, onto novel geographical areas. Transferred models have been used to predict suitable areas for invasive species [11,12], to predict the effects of climate change on species’ range [13,14], and to test niche conservatism hypotheses [15,16]. Ecological niche modeling has been incorporated into phylogeographic studies by evaluating whether genetically defined sub-populations are associated with divergence in ecological niche and geographic distribution [1721]. Combining ENM and phylogeography is particularly informative for studies of globally distributed pathogens which have intricate, often little understood, interactions with spatially heterogeneous abiotic and biotic factors which may be linked to genetic variation. From a methodological perspective, widely distributed species tend to result in less accurate ecological niche models [22], and models for such species can be improved by dividing the larger population into biologically meaningful sub-populations such as those defined by genetic analysis [17,23].

The phylogeography of B. anthracis has been described at the global level [24] and in multiple regions [7,2527] using a combination of genetic markers including single nucleotide polymorphisms (SNPs) [24] and multiple locus variable number tandem repeats (MLVA) [24,28,29]. Although the geographic distribution of genetic lineages as defined by SNP and MLVA analysis exhibits heterogeneity, globally successful lineages can dominate strain collections in countries on separate continents. For example, the MLVA defined A1.a is widely distributed throughout North America and Eurasia [24]; this sublineage was further defined using SNP analysis into the trans-Eurasian ancestor (TEA) and the related Western North American (WNA) group [24].

The possibility that genetic variation in bacterial pathogens is associated with spatial and ecological divergence is supported by studies demonstrating unique epidemiologic characteristics or ecological affinities among genetic groups of pathogens within a geographic region [17,18,30]. In the case of B. anthracis, a study in Kruger National Park, South Africa [7] revealed intriguing genetic-ecological associations for the pathogen and a differential distribution of genotypes based on soil characteristics. In this study, Smith et al. [7] detected spatial and ecological differences between genotypes representing the evolutionarily distinct B. anthracis A and B branches within the relatively narrow geographic boundaries of Kruger National Park. More recently, work by Mullins et al. [21] suggested that B. anthracis genotypes belonging to A1.a sublineage in Kazakhstan were associated with a broader ecological space than the larger population containing multiple A cluster sublineages, further supporting the importance of genetic information in building relevant ENMs for the species.

In the countries of United States, Italy and Kazakhstan, strains belonging to the A1.a sublineage are ecologically established and dominate strain collections [2426,28,29], providing a unique opportunity to study the dynamics of this successful group across diverse landscapes. In the United States, ecological niche modeling of anthrax outbreak locations predicted pathogen persistence primarily along a narrow corridor running from southwest Texas northward through the Dakotas [4]. In that landscape the A1.a sublineage (SNP group WNA) is dominant, although small areas appear to support other sublineages, including A3.b and A4. The genetic diversity of B. anthracis and geographic distribution of anthrax outbreaks in Italy was described by Fasanella et al. [26], Lista et al. [29] and Van Ert et al. [24], and confirmed by Garofolo et al. [31]. The dominant genotypes in Italy fell into the MLVA A1.a group and TEA SNP group. Isolates belonging to the A1.a/TEA sublineage also predominate in the southern portion of Kazakhstan [25], where the ecological niche and predicted geographic distribution has been modeled [21,32]. We developed ecological niche models of this globally successful B. anthracis sublineage in the United States, Italy and Kazakhstan. Models were reciprocally transferred to determine if pathogen presence could be accurately predicted on novel landscapes.


Anthrax occurrence data and environmental data

Geographic information system (GIS) databases of B. anthracis isolates belonging to the MLVA-defined A1.a sublineage and SNP defined TEA/WNA from the United States (U.S.), Italy, and Kazakhstan were used to support our analyses. All isolates were derived from collections previously genotyped using MLVA typing systems containing the set of eight markers (MLVA-8) described in Keim et al. [28] and SNP analysis [24]. The U.S. isolates were derived from Kenefic et al. [33], Blackburn et al. [4] and new strains from recent field collections in western Montana and Texas (Blackburn, unpublished data). The isolates from Italy were first reported in Fasanella et al. [26] and mapped for this study. Several isolates from Italy were mapped only to the nearest province and, therefore, we removed the isolates from the analysis. Kazakh data were derived from Mullins et al. [21]. Isolates were independently grouped into genetic lineages using the unweighted pair group method with arithmetic mean (UPGMA) cluster analysis or canSNP group assignments [2426]. Remaining occurrence data were reduced to spatially unique points at an 8 km2 resolution to accommodate the spatial uncertainty of the data [13]. Each occurrence dataset was randomly divided into an 80% training set for model building and a 20% dataset for testing the native projection. Figure 1 shows the occurrence point distributions for each of the countries used in this analysis.

Figure 1. Geographic distribution of the training and testing points used for ecological niche model building and evaluation.

Occurrences are shown for (A) the United States, (B) Italy and (C) Kazakhstan. Model training points are illustrated in green and independent data for model evaluation are yellow.

Grids representing six bioclimatic variables (Table 1) and altitude were downloaded from WorldClim ( and two satellite-derived environmental variables describing measures of vegetation were obtained from the Trypanosomiasis and Land Use in Africa (TALA) Research Group (Oxford, United Kingdom; Table 1) [34,35]. WorldClim bioclimatic grids (Bioclim) may be more biologically meaningful than annual mean, maximum and minimum values because the manipulation of monthly data results in variables that represent annual trends and seasonality as well as extremes in environmental conditions that limit a species’ range. All grids were resampled to 8 km2 and clipped to country boundaries using ArcView 3.3 with the GARP datasets extension (Environmental Systems Research institute, Redlands, California, USA). The eight variables in the environmental dataset were chosen based on parameters reported to reflect persistence of B. anthracis in soils and used in previous studies [4,13,32,36]. The background extent of each country was determined by political boundaries. These extents were chosen because reporting and control of anthrax is conducted within political boundaries.

Environmental Variable (unit)NameSource
Elevation (m)AltitudeWorldClim
Annual Temperature Range (°C)BIO7WorldClim
Annual Mean Temperature (°C)BIO1WorldClim
Precipitation of Driest Month (mm)BIO14WorldClim
Precipitation of Wettest Month (mm)BIO13WorldClim
Annual Precipitation (mm)BIO12WorldClim
NDVI Amplitude (no units)wd1014a1TALA
Mean NDVI (no units)wd1014a0TALA

Table 1. Environmental variables used to develop ecological niche models.

Trypanosomiasis and Land Use in Africa (TALA) Research Group (Oxford, United Kingdom) [57]
Download CSV

Model development

This study employed the Genetic Algorithm for Rule-Set Prediction (GARP) [37] with a best subset procedure to perform the ecological niche modelling experiments [38]. Briefly, the GARP approach uses a two-step procedure in which sets of rules are developed iteratively to predict presence or absence in variable space using presence only input data and background data. These rule sets are combinations of if/then statements derived from either variable ranges or logistic regression functions. This process is a random walk and develops multiple models; the best subset procedure aids in the selection of optimal models based on user-defined thresholds of omission and commission. Optimal rule sets are then projected onto the landscape. Training data were input into GARP with a 50% training/50% testing internal data partition. For all experiments, we specified 200 models with a maximum of 1,000 iterations and a convergence limit of 0.01. The 10 best subset models were selected using a 10% hard omission threshold and a 50% commission threshold. These output models were imported into ArcMap 10 (Environmental Systems Research institute, Redlands, California, USA) and summated to generate a single cumulative raster file of model agreement for B. anthracis presence. Grid cell values thus ranged from 0 (all models predict absence) to 10 (all models predict presence). Native models were trained in each of the three countries, then rule sets were projected onto the two other landscapes (for example, the model trained in the U.S. was projected onto Kazakhstan and Italy). As described in Mullins, et al. [21], Kazakh models were trained with a subset of southern A1.a isolates and projected onto the entire Kazakh landscape, the U.S. and Italy.

Model evaluation

Predictive performance of the best model subset was evaluated with an area under the curve (AUC) in a receiver operating characteristic (ROC) analysis using withheld independent test data for native models and all data for projected models. Values of AUC approaching one indicate a well-performing model while an AUC equal to 0.5 indicates the model performs no better than random and are tested statistically with a z-score (Z) and standard error (SE) estimates. AUCs were interpreted in conjunction with measures of omission and commission calculated using the summated 10 best models [39,40]. Total commission is the percent of pixels which are predicted as presence by areas of 10 model agreement. Average commission is the average area predicted as presence by all subset models. The greater the difference between the 2 measures of commission, the greater the spatial heterogeneity among the 10 best subset models [22]. Two measure of omission were also calculated. Total omission was calculated as the total number of test points falling into areas predicted as absence by all 10 models. Summed area omission (SAO) was calculated as the omission error of areas of 10 model agreement. Models with small SAO values are desirable because complete model agreement represents the most conservative threshold with which to predict areas of presence. Models which most robustly predict areas with a high likelihood of pathogen persistence facilitate implementation of cost effective, focused public health measures such as surveillance and pre-emptive vaccination.


All experiments reached convergence of accuracy prior to the maximum 1,000 iterations. The AUC scores for native projections all performed significantly better than random, and native models each had zero total omission and low average omission (Table 2). The native U.S. model predicted B. anthracis distributed in a north–south corridor in the center of the country (Figure 2). This band widens as it moves from southwest Texas northward into South and North Dakota and eastern Montana. Areas of the interior northwest are also predicted. The native Italian model predicted large sections of the southeastern mainland as well as the islands of Sardinia and Sicily. Areas of high likelihood also include coastal regions in central Italy and two relatively isolated regions of the northeastern and central north portions of the country. The Italian model accurately predicted areas of provinces in the northeast where A1.a strains have been documented, but were excluded from the analysis because of imprecise GIS data. In Kazakhstan, the native projection strongly predicted presence along the mountainous region of the southern portion of the country and a broad area in the north, while predicting the interior of the country with low model agreement.

Training LandscapeUnited StatesItalyKazakhstan
Total Omission062.113.2086.896.40000
Average Omission6.771511.474.882.2019.21017.7
Total Commission8.3800.8427.860019.7713.179.0428.67
Average Commission22.887.7143.5450.210.053.1746.354.2890.983.35

Table 2. Sample sizes and accuracy metrics for all native models and projections.

Native projection are under the curve (AUC) scores are shown in bold for comparison. SE = standard error, Z = z-score, SAO = summed area omission.
Native training and testing are for southern training area only
statistically significant value
Download CSV
Figure 2. Predicted distribution of Bacillus anthracis by native and transferred projections.

Native models were built for (A) the United States, (B) Italy and (C) Kazakhstan. Color ramp indicates the level of model agreement from zero (no models predict presence) to ten (all models in the best subset predict presence).

Models projected to novel landscapes performed poorly. Transferred models which performed better than random were poor, and some transferred models were more dispersed than random; measures of omission and commission were consistent with under-prediction by the Italian and U.S. models and over-prediction by the Kazakh model (Figure 2, Table 2). The native U.S. model when transferred to Italy predicted similar geographic areas of Italy as the native Italian model, but with very low model agreement. The transferred model failed to predict the area in southern Basilicata which has experienced anthrax outbreaks and gave low agreement for known endemic areas. The native U.S. model transferred to Kazakhstan broadly predicted, with low model agreement, almost the entire landscape. Small areas in the southern and northern regions were predicted with higher model agreement. These areas were also predicted by the native Kazakh model, whereas in the northeast corner of the country, the U.S. and Kazakh models were very different.

The model trained in Italy transferred to Kazakhstan predicted a highly endemic area of southern Kazakhstan, although with very low model agreement. Projected onto the U.S., the model trained in Italy did highlight some areas predicted by the native trained model, namely western Texas, western Oklahoma and Colorado, but again with low model agreement. When transferred, the models trained in Kazakhstan over-predicted all of Italy and the majority of the landscape in the U.S.


Evidence increasingly suggests phylogenetic analysis provides a meaningful way to subdivide B. anthracis, and other species, for more accurate niche modeling [7,17,19,21,23]. Furthermore, an ability to apply models developed in a landscape with known anthrax locations to one in which anthrax is not reported or is poorly recognized would be of great value in predicting potential areas of emerging disease. This study tested whether ecological niche models of the B. anthracis MLVA-8 defined A1.a sublineage can be used to predict the distribution of the same sublineage on novel landscapes. Our native models reasonably predicted the occurrence points in our experiments, and furthermore the native Italian model predicted a region in Italy where the A1.a sublineage has been isolated, but due to lack of spatial data was not included in the occurrence dataset. Transferred models, however, failed to accurately predict documented anthrax occurrence points and tended to over-predict or under-predict presence on the non-native landscapes. Where transferred models were successful in predicting areas of known persistence, model agreement tended to be low. Both over-prediction and under-prediction of transferred models are considered failures, although these failures have different practical consequences for public health applications. When areas suitable for pathogen presence are not predicted by a model, the failure to incorporate this area into surveillance programs will result in cases of disease being overlooked. Over-prediction by a model, on the other hand, hinders epidemiologic investigations of cases by misallocating resources to monitor regions that are erroneously predicted to support the pathogen.

The failure of transferred models to accurately and consistently predict known occurrences presents an interpretive challenge. The lack of transferability may have resulted from methodological shortcomings of transferring models, or could reflect genetic-ecological divergence of the pathogen [15,17,19,41,42]. Evidence supporting genetic-ecological divergence was demonstrated in Kruger National Park, South Africa, where the A lineage had broader ecological tolerances than the B lineage within the park [7]. Our results suggest that this ecological divergence in B. anthracis strains may also occur in a widely distributed sublineage. It is possible that successful genetic groups with broader tolerances are more likely to become established across ecological extremes, and the local population would then differentiate to develop a unique genetic signature associated with the local ecology [43]. In contrast, B. anthracis lineages with more limited tolerances, such as the B lineage and, although its ecological associations have not been characterized, the geographically limited C lineage, would be restricted ecologically and geographically. This process of regional-scale differentiation leading to spatially structured genetic variation has been described among other introduced species or pathogens [15,17,18,33,44]. That Italian and U.S. models under predicted pathogen occurrence in Kazakhstan, whereas Kazakh models overpredicted a large extent of Italy and the U.S., may indicate that Kazakh strains have adapted to a broader ecological envelope. Although more complete genomic data are required to substantiate this hypothesis, this may reflect an earlier introduction of the sublineage into Kazakhstan than in Italy or North America.

Evidence for genetic differentiation within the MLVA-8 defined A1.a sublineage includes the separation of this sublineage into distinct TEA and WNA SNP groups [24]. More recent SNP analysis of the WNA lineage [33] suggests a significant evolutionary divergence between the North American (U.S.) and Eurasian (Italy and Kazakhstan) strains. The evolutionary divergence between Eurasian strains in Kazakhstan and Italy is not as well defined. However, a 15-marker VNTR analysis of the Eurasian group reported by Van Ert et al. [24] suggests a high level of genetic diversity exists within the ‘TEA’ SNP group (A. Br.008/009). More recent VNTR based analyses based on 25 markers indicates that the Kazakh ‘A1.a’ strains, as previously defined using the 8-marker system, exhibit a considerable degree of genetic divergence (Sytnik, unpublished data) from European strains. Although this VNTR based diversity is intriguing, more comprehensive analysis is required to more accurately measure the evolutionary relationships between these populations, particularly considering that discovery bias inherently limits resolution in ‘canonical’ SNP data [45]. There is clearly a need to genotype existing collections of B anthracis isolates with the highest resolution systems available, such as the Lista 25 marker system [29] or the 31 marker system [46]. Such an effort will enhance our understanding of the phylogenetics and the ecology of the pathogen. Existing niche models should then be reconstructed according to emerging phylogenetic evidence. Despite the limitations of the genetic data used in the present analysis, the finding that native models were successful while transferred models failed to predict anthrax occurrence points suggests that niche specialization may have occurred within this broadly distributed sublineage and this suggestion of genetic-ecological associations warrants additional investigation.

From a modeling perspective, however, the effect of variable selection [41,42,47,48] and background [4951] on transferred projections must be considered as potential methodological limitations when interpreting the results of this study. The variable set used in the current experiments was chosen based on variables considered to be limiting factors in the persistence of B. anthracis based on previous niche modeling efforts performed within single landscapes [4,21,32]. In studies exploring the effects of variable selection on transferred ecological niche modeling experiments, changes in the dimensionality [41,52] and source [41] of environmental datasets resulted in different geographic predictions, and these difference were more pronounced in the novel landscapes than in the native ones. Variable selection can limit transferability because limiting factors for the species vary geographically [42,49], and, in addition, interactions among ecological variables may differ across landscapes, which would alter the relationship of more distal variables with the pathogen [51]. Ecological variables limiting the spatial distribution of B. anthracis may also vary with genetic lineage, reflecting niche specialization suggested by the work of Smith et al. [7] in Kruger National Park, South Africa. The impact of variable selection, as well as that of different techniques for selection of environmental variable sets, on these results will be evaluated in additional experiments.

A second methodological consideration is that of the extent used for background. We have defined the background in this study according to political boundaries instead of using other suggested techniques such as minimum convex polygons [42], global ranges [53], or the accessible extent [50,51] because B. anthracis dispersal to novel areas is primarily anthropogenic and therefore not constrained by natural physical boundaries or biological limits to movement. Therefore the potential area of dispersal, as currently understood, is limited by control and prevention measures which follow political boundaries. Similarly, the nature of surveillance for, diagnosis of and reporting of anthrax outbreaks is dependent on policies which fall within such limits. Despite this practical consideration, however, we recognize that within the political boundaries used for defining the extent for these models, considering the large geographic extents and latitudinal differences between the countries, the ranges of environmental variables is likely differ between landscapes. Hence, concerns about extrapolation are significant and should be addressed in additional experiments.

While we suggest the results presented here may reflect niche differentiation within this sublineage of B. anthracis, while considering the potential effects of methodological problems, it is important to also consider the source of isolate/occurrence data and the effects of control efforts on the disease in each country [4]. It is possible that the broad native predictions within Kazakhstan reflect an insufficient capacity to maintain widespread vaccination to reduce the overall burden of anthrax [54]. In the case of the U.S., models likely reflect areas where the disease persists after decades of widespread vaccination efforts have resulted in a clear reduction in overall outbreak numbers and an accompanying contraction in the spatial extent of disease [55]. Differences in historical control, alone or in synergy with genetic and ecological factors, could explain some portion of the over- and under- prediction observed in the transference of models.

In this study we have used measures of omission, commission and AUC as accuracy metrics to evaluate native models and their transferability. The AUC measure has limitations when evaluating ecological niche models [39,40]. Two described limitations are that AUC is influenced by the geographic extent of the landscape being studied and the AUC uses the entire ROC plot which would render comparisons between modeling platforms unreliable. Here, however, we evaluate models over the same extent (ie, a native U.S. model and a projection transferred to the U.S.) and all using the GARP modeling platform. The optimum weighting of omission and commission, although equally weighted in the AUC, varies depending on the purpose of the experiment. In this study ENM was used to predict the potential presence of a pathogen and infer subsequent disease risk to inform public health policy. Errors of commission, or over-prediction, will result in excess expenditures for surveillance and interventions and mislead trace-back efforts, whereas errors of omission, or under-prediction, could result in sustained transmission [56]. Viewed in this context, optimal weighting of omission and commission will depend on the relative costs of surveillance and disease. The summed area omission (SAO) allows for refinement of models with the goal being that areas of total model agreement can be used as a conservative threshold for predicted presence. By calculating omission only in areas of complete model agreement, this conservative evaluation of model performance maximizes predictive value without incorporating potentially large geographic areas of low model agreement and therefore lower risk. Expensive surveillance and prevention programs can then be effectively targeted to highest risk areas.

Previous ecologic niche models described conditions favorable for B. anthracis outbreaks based on collections of isolates from multiple genetic lineages that were likely biased towards a dominant subset of genotypes [4,32] and the potential for using genetic analysis to improve models was subsequently demonstrated [21]. Our findings suggest genetic-ecological divergence exists among geographically dispersed populations of B. anthracis from the MLVA-8 defined A1.a sublineage. Some caveats apply, however. More comprehensive and higher resolution genomic data is required to better characterize the genetic differences between these populations. In addition, the methodological problems discussed here must be explored in order to refine native models and to evaluate whether variable selection methods will enhance the transferability of models. Until our understanding of the genetic-ecological dynamics of B anthracis is better developed, we suggest that B. anthracis is best modeled on a country or regional level and with consideration of the genetic diversity of the population. Moving forward, understanding B. anthracis genetic-ecological associations on the landscape will result in construction of ecological niche models that are sensitive to genotype and region and are more successful in predicting outbreaks.

Author Contributions

Conceived and designed the experiments: JCM JKB MVE. Performed the experiments: JCM JKB. Analyzed the data: JCM GG MVE JKB. Contributed reagents/materials/analysis tools: JKB GG AF LL MEHJ. Wrote the manuscript: JCM JKB MVE GG.


  1. 1. Hugh-Jones M, Blackburn J (2009) The ecology of Bacillus anthracis. Mol Aspects Med 30: 356-367. doi: PubMed: 19720074.
  2. 2. Schuch R, Fischetti VA (2009) The secret life of the anthrax agent Bacillus anthracis: bacteriophage-mediated ecological adaptations. PLOS ONE 4: e6532. doi: PubMed: 19672290.
  3. 3. Sternbach G (2003) The history of anthrax. J Emerg Med 24: 463-467. doi: PubMed: 12745053.
  4. 4. Blackburn JK, McNyset KM, Curtis A, Hugh-Jones ME (2007) Modeling the geographic distribution of Bacillus anthracis, the causative agent of anthrax disease, for the contiguous United States using predictive ecologic niche modeling. Am J Trop Med Hyg 77: 1103-1110. PubMed: 18165531.
  5. 5. Blackburn JK, Curtis A, Hadfield TL, O’Shea B, Mitchell MA et al. (2010) Confirmation of Bacillus anthracis from flesh-eating flies collected during a West Texas anthrax season. J Wildl Dis 46: 918-922. PubMed: 20688697.
  6. 6. Kracalik IT, Blackburn JK, Lukhnova L, Pazilov Y, Hugh-Jones ME et al. (2012) Analysing the spatial patterns of livestock anthrax in Kazakhstan in relation to environmental factors: a comparison of local (Gi*) and morphology cluster statistics. Geospatial Health 7: 111-126. PubMed: 23242686.
  7. 7. Smith KL, DeVos V, Bryden H, Price LB, Hugh-Jones ME et al. (2000) Bacillus anthracis diversity in Kruger National Park. J Clin Microbiol 38: 3780-3784. PubMed: 11015402.
  8. 8. Van Ness G, Stein CD (1956) Soils of the United States favorable for Anthrax. J Am Vet Med Assoc 128: 7-9. PubMed: 13278269.
  9. 9. Van Ness GB (1971) Ecology of Anthrax. Science 172: 1303-1307. doi: PubMed: 4996306.
  10. 10. Hugh-Jones ME, de Vos V (2002) Anthrax and wildlife. Rev Sci Tech 21: 359-383. PubMed: 11974621.
  11. 11. Mukherjee A, Diaz R, Thom M, Overholt W, Cuda J (2012) Niche-based prediction of establishment of biocontrol agents: an example with Gratiana boliviana and tropical soda apple. Biocontrol Sci Technol 22: 447-461. doi:
  12. 12. Peterson AT (2003) Predicting the geography of species’ invasions via ecological niche modeling. Q Rev Biol 78: 419-433. doi: PubMed: 14737826.
  13. 13. Joyner TA, Lukhnova L, Pazilov Y, Temiralyeva G, Hugh-Jones ME et al. (2010) Modeling the potential distribution of Bacillus anthracis under multiple climate change scenarios for Kazakhstan. PLOS ONE 5: e9596. doi: PubMed: 20231894.
  14. 14. Nakazawa Y, Williams R, Peterson AT, Mead P, Staples E et al. (2007) Climate change effects on plague and tularemia in the United States. Vector-Borne Zoonotic Dis 7: 529-540. doi: PubMed: 18047395.
  15. 15. Broennimann O, Treier UA, Müller-Schärer H, Thuiller W, Peterson AT et al. (2007) Evidence of climatic niche shift during biological invasion. Ecol Lett 10: 701-709. doi: PubMed: 17594425.
  16. 16. Mukherjee A, Williams D, Wheeler G, Cuda J, Pal S et al. (2011) Brazilian peppertree (Schinus terebinthifolius) in Florida and South America: evidence of a possible niche shift driven by hybridization. Biol Invasions: 1-16.
  17. 17. Fisher MC, Hanage WP, De Hoog S, Johnson E, Smith MD et al. (2005) Low effective dispersal of asexual genotypes in heterogeneous landscapes by the endemic pathogen Penicillium marneffei. PLOS Pathog 1: 1986-1990. PubMed: 16254598.
  18. 18. Nakazawa Y, Williams RA, Peterson AT, Mead PS, Kugeler KJ et al. (2010) Ecological niche modeling of Francisella tularensis subspecies and clades in the United States. Am J Trop Med Hyg 82: 912-918. doi: PubMed: 20439975.
  19. 19. Rissler LJ, Apodaca JJ (2007) Adding more ecology into species delimitation: ecological niche models and phylogeography help define cryptic species in the black salamander (Aneides flavipunctatus). Syst Biol 56: 924-942. doi: PubMed: 18066928.
  20. 20. Rice NH, Martinez-Meyer E, Peterson AT (2003) Ecological niche differentiation in the Aphelocoma jays: a phylogenetic perspective. Biol J Linn Soc 80: 369-383. doi:
  21. 21. Mullins J, Lukhnova L, Aikimbayev A, Pazilov Y, Van Ert M et al. (2011) Ecological Niche Modelling of the Bacillus anthracis A1.a sub-lineage in Kazakhstan. BMC Ecol 11: 32. doi: PubMed: 22152056.
  22. 22. McNyset KM (2005) Use of ecological niche modelling to predict distributions of freshwater fish species in Kansas. Ecol Freshw Fish 14: 243-255. doi:
  23. 23. Gonzalez SC, Soto-Centeno JA, Reed DL (2011) Population distribution models: Species distributions are better modeled using biologically relevant data partitions. BMC Ecol 11: 20. doi: PubMed: 21929792.
  24. 24. Van Ert MN, Easterday WR, Huynh LY, Okinaka RT, Hugh-Jones ME et al. (2007) Global Genetic Population Structure of Bacillus anthracis. PLOS ONE 2: e461. PubMed: 17520020.
  25. 25. Aikembayev AM, Lukhnova L, Temiraliyeva G, Meka-Mechenko T, Pazylov Y et al. (2010) Historical distribution and molecular diversity of Bacillus anthracis, Kazakhstan. Emerg Infect Dis 16: 789-796. doi: PubMed: 20409368.
  26. 26. Fasanella A, Van Ert M, Altamura SA, Garofolo G, Buonavoglia C et al. (2005) Molecular diversity of Bacillus anthracis in Italy. J Clin Microbiol 43: 3398–3401. doi: PubMed: 16000465.
  27. 27. Simonson TS, Okinaka RT, Wang B, Easterday WR, Huynh L et al. (2009) Bacillus anthracis in China and its relationship to worldwide lineages. BMC Microbiol 9: 71. doi: PubMed: 19368722.
  28. 28. Keim P, Price LB, Klevytska AM, Smith KL, Schupp JM et al. (2000) Multiple-locus variable-number tandem repeat analysis reveals genetic relationships within Bacillus anthracis. J Bacteriol 182: 2928–2936. doi: PubMed: 10781564.
  29. 29. Lista F, Faggioni G, Valjevac S, Ciammaruconi A, Vaissaire J et al. (2006) Genotyping of Bacillus anthracis strains based on automated capillary 25-loci multiple locus variable-number tandem repeats analysis. BMC Microbiol 6: 33. doi: PubMed: 16600037.
  30. 30. Kugeler KJ, Mead PS, Janusz AM, Staples JE, Kubota KA et al. (2009) Molecular epidemiology of Francisella tularensis in the United States. Clin Infect Dis 48: 863-870. doi: PubMed: 19245342.
  31. 31. Garofolo G, Serrecchia L, Corrò M, Fasanella A (2011) Anthrax phylogenetic structure in Northern Italy. BMC Res Notes 4: 273. doi: PubMed: 21801397.
  32. 32. Joyner TA (2010) Ecological niche modeling of a zoonosis: A case study using anthrax outbreaks and climate change in Kazakhstan. University of Florida.
  33. 33. Kenefic LJ, Pearson T, Okinaka RT, Schupp JM, Wagner DM et al. (2009) Pre-columbian origins for north american anthrax. PLOS ONE 4: e4813. doi: PubMed: 19283072.
  34. 34. Hay SI, Tatem AJ, Graham AJ, Goetz SJ, Rogers DJ (2006) Global environmental data for mapping infectious disease distribution. Adv Parasitol 62: 37-77. PubMed: 16647967.
  35. 35. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) Very high resolution interpolated climate surfaces for global land areas. Int J Climatol 25: 1965-1978. doi:
  36. 36. Mullins J, Lukhnova L, Aikimbayev A, Pazilov Y, Van Ert M et al. (2011) Ecological Niche Modelling of the Bacillus anthracis A1.a sub-lineage in Kazakhstan. BMC Ecology 11: 32.
  37. 37. Stockwell D, Peters D (1999) The GARP modelling system: problems and solutions to automated spatial prediction. Int J Geogr Inf Sci 13: 143-158. doi:
  38. 38. Anderson RP, Lew D, Peterson AT (2003) Evaluating predictive models of species’ distributions: criteria for selecting optimal models. Ecol Modell 162: 211-232. doi:
  39. 39. Lobo JM, Jiménez-Valverde A, Real R (2008) AUC: a misleading measure of the performance of predictive distribution models. Glob Ecol Biogeogr 17: 145-151. doi:
  40. 40. Peterson AT, Papes M, Soberón J (2008) Rethinking receiver operating characteristic analysis applications in ecological niche modeling. Ecol Modell 213: 63-72. doi:
  41. 41. Peterson A, Nakazawa Y (2008) Environmental data sets matter in ecological niche modelling: an example with Solenopsis invicta and Solenopsis richteri. Glob Ecol Biogeogr 17: 135-144.
  42. 42. Rödder D, Lötters S (2010) Explanative power of variables used in species distribution modelling: an issue of general model transferability or niche shift in the invasive Greenhouse frog (Eleutherodactylus planirostris). Naturwissenschaften 97: 781-796. doi: PubMed: 20617298.
  43. 43. Holt R, Barfield M, Gomulkiewicz R (2005) Theories of niche conservatism and evolution: could exotic species be potential tests? In: DF Sax. Species invasions: insights into ecology, evolution, and biogeography. Sinauer Associates. pp. 259-290.
  44. 44. Carrel MA, Emch M, Jobe RT, Moody A, Wan XF (2010) Spatiotemporal Structure of Molecular Evolution of H5N1 Highly Pathogenic Avian Influenza Viruses in Vietnam. PLOS ONE 5: e8631. doi: PubMed: 20072619.
  45. 45. Pearson T, Busch JD, Ravel J, Read TD, Rhoton SD et al. (2004) Phylogenetic discovery bias in Bacillus anthracis using single-nucleotide polymorphisms from whole-genome sequencing. Proc Natl Acad Sci U S A 101: 13536-13541. doi: PubMed: 15347815.
  46. 46. Beyer W, Bellan S, Eberle G, Ganz HH, Getz WM et al. (2012) Distribution and molecular evolution of Bacillus anthracis genotypes in Namibia. PLOS Neglected Trop Dis 6: e1534. doi: PubMed: 22413024.
  47. 47. Jiménez-Valverde A, Nakazawa Y, Lira-Noriega A, Peterson AT (2009) Environmental correlation structure and ecological niche model projections. Biodivers Informatics 6.
  48. 48. Araujo MB, Guisan A (2006) Five (or so) challenges for species distribution modelling. J Biogeogr 33: 1677-1688. doi:
  49. 49. Austin M (2007) Species distribution models and ecological theory: a critical assessment and some possible new approaches. Ecol Modell 200: 1-19. doi:
  50. 50. Barve N, Barve V, Jiménez-Valverde A, Lira-Noriega A, Maher SP et al. (2011) The crucial role of the accessible area in ecological niche modeling and species distribution modeling. Ecol Modell 222: 1810-1819. doi:
  51. 51. Elith J, Kearney M, Phillips S (2010) The art of modelling range-shifting species. Methods Ecol Evolution 1: 330-342. doi:
  52. 52. Rödder D, Schmidtlein S, Veith M, Lötters S (2009) Alien invasive slider turtle in unpredicted habitat: a matter of niche shift or of predictors studied? PLOS ONE 4: e7843. doi: PubMed: 19956684.
  53. 53. Broennimann O, Guisan A (2008) Predicting current and future biological invasions: both native and invaded ranges matter. Biol Lett 4: 585-589. doi: PubMed: 18664415.
  54. 54. Woods CW, Ospanov K, Myrzabekov A, Favorov M, Plikaytis B et al. (2004) Risk factors for human anthrax among contacts of anthrax-infected livestock in Kazakhstan. Am J Trop Med Hyg 71: 48–52. PubMed: 15238688.
  55. 55. Blackburn JK (2006) Evaluating the spatial ecology of anthrax in North America: Examining epidemiological components across multiple geographic scales using a GIS-based approach. Louisiana State University.
  56. 56. Peterson AT, Soberon J, Pearson RG, Anderson RP, Martinez-Meyer E et al. (2011) Ecological Niches and Geographic Distributions (MPB-49). Princeton University Press.
  57. 57. Hay SI, Tatem AJ, Graham AJ, Goetz SJ, Rogers DJ (2006) Global environmental data for mapping infectious disease distribution. Adv Parasitol 62: 37-77. doi: PubMed: 16647967.