<%BANNER%>
PRIVATE ITEM
Digitization of this item is currently in progress.
Ecological Niche Modelling of the Bacillus anthracis A1.a sub-lineage in Kazakhstan
CITATION PDF VIEWER
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/AA00009675/00001
 Material Information
Title: Ecological Niche Modelling of the Bacillus anthracis A1.a sub-lineage in Kazakhstan
Series Title: BMC Ecology
Physical Description: Archival
Language: English
Creator: Mullins, Jocelyn
Lukhnova, Larissa
Aikimbayev, Alim
Pazilov, Yerlan
Van Ert, Matthew
Blackburn, Jason K.
Publisher: BioMed Central
Publication Date: 2011
 Notes
Abstract: Background: Bacillus anthracis, the causative agent of anthrax, is a globally distributed zoonotic pathogen that continues to be a veterinary and human health problem in Central Asia. We used a database of anthrax outbreak locations in Kazakhstan and a subset of genotyped isolates to model the geographic distribution and ecological associations of B. anthracis in Kazakhstan. The aims of the study were to test the influence of soil variables on a previous ecological niche based prediction of B. anthracis in Kazakhstan and to determine if a single sub-lineage of B. anthracis occupies a unique ecological niche. Results: The addition of soil variables to the previously developed ecological niche model did not appreciably alter the limits of the predicted geographic or ecological distribution of B. anthracis in Kazakhstan. The A1.a experiment predicted the sub-lineage to be present over a larger geographic area than did the outbreak based experiment containing multiple lineages. Within the geographic area predicted to be suitable for B. anthracis by all ten best subset models, the A1.a sub-lineage was associated with a wider range of ecological tolerances than the outbreaksoil experiment. Analysis of rule types showed that logit rules predominate in the outbreak-soil experiment and range rules in the A1.a sub-lineage experiment. Random sub-setting of locality points suggests that models of B. anthracis distribution may be sensitive to sample size. Conclusions: Our analysis supports careful consideration of the taxonomic resolution of data used to create ecological niche models. Further investigations into the environmental affinities of individual lineages and sublineages of B. anthracis will be useful in understanding the ecology of the disease at large and small scales. With model based predictions serving as approximations of disease risk, these efforts will improve the efficacy of public health interventions for anthrax prevention and control.
 Record Information
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution.
System ID: AA00009675:00001

Downloads

This item is only available as the following downloads:

( PDF )


Full Text

PAGE 1

RESEARCHARTICLE OpenAccessEcologicalNicheModellingofthe Bacillus anthracis A1.asub-lineageinKazakhstanJocelynMullins1,2,LarissaLukhnova3,AlimAikimbayev4,YerlanPazilov3,MatthewVanErt2and JasonKBlackburn1,2*AbstractBackground: Bacillusanthracis ,thecausativeagentofanthrax,isagloballydistributedzoonoticpathogenthat continuestobeaveterinaryandhumanhealthprobleminCentralAsia.Weusedadatabaseofanthraxoutbreak locationsinKazakhstanandasubsetofgenotypedisolatestomodelthegeographicdistributionandecological associationsof B.anthracis inKazakhstan.Theaimsofthestudyweretotesttheinfluenceofsoilvariablesona previousecologicalnichebasedpredictionof B.anthracis inKazakhstanandtodetermineifasinglesub-lineageof B.anthracis occupiesauniqueecologicalniche. Results: Theadditionofsoilvariablestothepreviouslydevelopedecologicalnichemodeldidnotappreciablyalter thelimitsofthepredictedgeographicorecologicaldistributionof B.anthracis inKazakhstan.TheA1.aexperiment predictedthesub-lineagetobepresentoveralargergeographicareathandidtheoutbreakbasedexperiment containingmultiplelineages.Withinthegeographicareapredictedtobesuitablefor B.anthracis byalltenbest subsetmodels,theA1.asub-lineagewasassociatedwithawiderrangeofecologicaltolerancesthantheoutbreaksoilexperiment.Analysisofruletypesshowedthatlogitrulespredominateintheoutbreak-soilexperimentand rangerulesintheA1.asub-lineageexperiment.Randomsub-settingoflocalitypointssuggeststhatmodelsof B. anthracis distributionmaybesensitivetosamplesize. Conclusions: Ouranalysissupportscarefulconsiderationofthetaxonomicresolutionofdatausedtocreate ecologicalnichemodels.Furtherinvestigationsintotheenvironmentalaffinitiesofindividuallineagesandsublineagesof B.anthracis willbeusefulinunderstandingtheecologyofthediseaseatlargeandsmallscales.With modelbasedpredictionsservingasapproximationsofdiseaserisk,theseeffortswillimprovetheefficacyofpublic healthinterventionsforanthraxpreventionandcontrol.BackgroundAnthraxisadiseaseofwildlife,livestockandhumans thatremainsapublichealthproblemthroughoutthe world. Bacillusanthracis ,thecausativeagentofanthrax, isasoil-borne,spore-formi ngbacteriumwhichpersists insoilforlongperiodsoftimeunderappropriateconditions[1].Certainsoilparameters,includingpH,organic contentandcalcium,maybeassociatedwithsporesurvival[1-5].Anthraxoutbreaksamonglivestockand wildliferesultfromexposuretothesesporesandare possiblyinfluencedbyclimaticandphysiologicalevents [6,7].Inendemicareas,humancasesofanthraxprimarilyresultfromcontactwithinfectedlivestockduring slaughterorbutchering[8,9]andcontroloflivestock diseasethroughvaccinationandactivesurveillanceof livestockandwildlifeisessentialforpreventinghuman disease[10].However,widespreadactivesurveillanceis costlyandvaccinationofeveryanimalisnotfeasible.It isfarmorepracticaltofocustheseeffortsonareasof highrisk.Toidentifytheseitisnecessarytoimprove ourunderstandingoftheecologyof B.anthracis throughwhichanimalinfectionoccurs. Theecologyofapathogensuchas B.anthracis canbe exploredusingsimilartoolsasthoseusedforspecies distributionmodellingandconservationplanning.For example,ecologicalnichemodelling(ENM)hasbeen usedtopredictthepotentialecologicalandgeographic distributionofpathogensbasedonoutbreaklocations [10-14],presenceofdiseasevectors[15-17]anddisease *Correspondence:jkblackburn@ufl.edu1DepartmentofGeography,UniversityofFlorida,Gainesville,FL,USA FulllistofauthorinformationisavailableattheendofthearticleMullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 2011Mullinsetal;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreativeCommons AttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,andreproductionin anymedium,providedtheoriginalworkisproperlycited.

PAGE 2

reservoirs[18].Theecologi calnicheofapathogen,as forothertypesofspecies,isconceptualizedastheNdimensionalhypervolumeofecologicalparameters withinwhichthespeciescanbemaintainedwithout immigration[19,20].Vario usapproachestoENMidentifynon-randomassociationsbetweenaspecies locality dataandenvironmentalparameters.Ecologicalniche modellingexperimentsof B.anthracis areparticularly usefulconsideringthepotentialassociationsbetween sporesurvivalandecologicalconditions[1,5].Results canbeusedasaproxyfordiseaseriskandintegrated intofocusedsurveillancestrategiesforwildlifeandlivestockinendemicareasandintovaccinationstrategies thattargetatriskherdsbeforeandduringoutbreak events[10]. Recently,studiesofdiseaseecologyhavecombined moleculargenotypingtechniquesandecologicalniche modellingtoprovideevidencethatgeneticlineagesofa pathogencanhavedifferentenvironmentalassociations andpotentialgeographicdistributions[12,21].Ingeneral B.anthracis hasrelativelylimitedglobaldiversity.However,multiplelocusvariablenumbertandemrepeatanalysis(MLVA)systemsfor B.anthracis candifferentiate strainsintodistinctlineages andsub-lineages[22-24]. Analysesofaglobalcollectionof B.anthracis isolates suggeststhattheAlineageisgloballydistributed,while otherlineages(BandC)aregeographicallyrestricted. Thesefindingsmaybeexplainedbyadaptivedifferences, someofwhichcarryfitnesscoststhatlimitabundance anddistributionofcertainlineagesorsub-lineages [3,24].Theecologicalnicheof B.anthracis hasbeen modelledintheUnitedStatesandKazakhstanusing locationsofreportedoutbreaks[10,11,25,26].Astated limitationoftheseexperimentswasthattheoutbreak datapotentiallyincludedmultiplestrainsof B.anthracis [10,11].Iflineagesof B.anthracis doexhibitnichespecializationanduniquegeographicdistributions,thenit isplausiblethatcurrentoutbreakbasedecologicalniche modelsarebiasedtowardadominantstraininaparticularlandscape.Itwouldthenfollowthatsinglelineage modelsmaybetterpredictpresenceofthepathogenat localscalesandincreasethevalueofpublichealthmeasures[10]. KazakhstanissituatedinCentralAsia,aregionwith someofthehighestreportedhumananthraxincidence andmortalityratesintheworld[27,28].Themajorityof humananthraxcasesinKazakhstanarerelatedtoexposuretoinfectedlivestockorhandlingofproducts derivedfrominfectedlivestock[9].Inruralareasof Kazakhstan,veterinarycareandsurveillanceprograms arelimitedbythecountry slargelandmassandwidely distributedruralpopulations.Vaccinationoflivestock occursmainlyinresponsetodetectedoutbreaks.In countriessuchasKazakhstan,prioritizingareasfor vaccinationandsurveillancearenecessaryfordisease control.Ourgrouprecentlycreatedamulti-variateecologicalnichemodeltochara cterizethebroadenvironmentalconditionsthatsupport B.anthracis across Kazakhstan[11,26].Inaparalleleffort,Aikimbayevetal. usedaneightmarkerMLVAtypingsystem(MLVA-8) todescribethediversityof B.anthracis withinKazakhstanfrom88archivalstrains[22,29]. Inthisstudy,wefirstexpandedonthepreviouslypublishedoutbreakbasedmodellingexperimentbyadding foursoilvariables(pH,calciumlevels,organiccontent andbaselinewatersaturation)totheoriginalsetof environmentalvariables.Despiteliteraturesuggestinga strongrelationshipbetweensoilcharacteristicssuchas highcalciumlevelsandalkalinepHandsporepersistence[1,4,5],theinfluenceofavailablesoilvariableson B.anthracis ENMpredictionshasnotbeencomparativelyexamined[10].WenextusedthesetwelveenvironmentalvariablesandthecollectionofMLVA-8 genotypedsamplestocreateanA1.asub-lineagespecific ecologicalnichemodelforKazakhstan.ResultsAccuracyMetricsEcologicalnichemodellingwasperformedusingthe GeneticAlgorithmforRule-SetPrediction(GARP).Four experimentswererun(outbr eak-soil,A1.asub-lineage, smallsouthernoutbreakandlargesouthernoutbreak) andaresummarizedinTable1.Allmodellingprocesses reachedconvergenceofaccuracy(0.01)priortoreaching themaximumiterationsetting(=1,000).TheoutbreaksoilmodelhadanAreaUndertheCurve(AUC)of 0.7188andwassignificantlydifferentfromarandom model.Totalomissionoftheoutbreak-soilmodelwas 2.6%andaverageomissionwas9.9%,indicatingthat 97.4%ofthetestingpointswerepredictedbyatleast onebestsubsetmodeland89.1%werepredictedbyall models.TheAUCoftheA1.asub-lineagemodelwas 0.6964andwasalsosignificantlydifferentfromarandommodelandhadatotalandaverageomissionof0 and13.1,respectively.Boththelargeandsmalloutbreak modelshadAUCssignificantlydifferentthanrandom. AccuracymetricsforallmodelsareshowninTable2.PredictedDistributionsof B.anthracisLocationsusedforinputintoGARPareshowninFigure 1.Basedonareasofagreementofaminimumofsixof thebestsubsetmodels,theoutbreak-soilexperiment predicted B.anthracis acrossmuchofnorthernKazakhstanandinanarrowbandofthesoutheast.Theinterior ofthecountry,whichisprimarilyarid,wasnotpredictedtobesuitableforthepathogen.Theresultsare similartothoseoftheexperimentwithoutthesoilvariableswithrespecttothegeographicextentofareasofMullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page2of14

PAGE 3

sixormorebestsubsetmodelagreement(Figure2). Theoutbreak-soilexperimentexpandedtwoareasinthe northwhichhadlowermodelagreement.TheA1.asublineageexperimentpredictsamoreextensivegeographic distributionthanthatoftheoutbreakexperiment, includingareasinthenortherninteriorandwestern portionsofthecountry(Figure3).Thenorthernpockets oflesssuitablegeographicareasseenintheoutbreaksoilexperimentwerepredictedtobeunsuitablebased onagreementofsixormorebestsubsetmodels.The overallextentsofthegeographicpredictionsofthetwo experimentsweremoresimilarinthesouththaninthe north.Thelargeandsmallsouthernoutbreakexperimentsbothpredictedsimilargeographicextentsasthe outbreak-soilexperiment(Figure4).Allthreeprojected experiments(A1.asub-linea ge,largesouthernoutbreak andsmallsouthernoutbreak)wereruntenadditional timesusingrandomexternaldatasplits.Thesubsetsof A1.asub-lineageandsmallsouthernoutbreakexperimentsshowedgreaterdegreesofspatialheterogeneity thandidthelargesouthernoutbreakexperimentset. (seeAdditionalFile1:RandomSubsetsforillustration, availableasaPDFfile,andAdditionalFile2:Accuracy metricsforrandomsubsets,availableasaPDFfile). EachGARPmodeliscomposedof50if-thentyperules (logic,range,negatedrangeandatomic)whichpredictthe presenceorabsenceofthespeciesforeachpixel.Rule typesforthetenbestsubsetmodelsfromtheoutbreaksoilandA1.asub-lineageexperimentswereextractedand aresummarizedinTable3.Justoverhalfoftheoutbreaksoilexperimentruleswerelogitandnoatomicruleswere included,whereasrangerulesmadeupover60%ofthe A1.asub-lineageexperimentruletypesandthisexperimentincludedfouratomic rulesinthebestsubsets. Between6and13rulesdefinedgreaterthan90%ofareas predictedtobesuitablefor B.anthracis foreachofthe bestsubsets.Ofthe95ruleswhichpredictedthemajority ofthelandscapeintheoutbreak-soilexperiment,the majority(83%)werepresencerulesandofthese62%were rangerules.TheA1.asub-lineageexperimenthad99total rulespredictthemajorityofthelandscape;allbutoneof thesewasapresenceruleand73%ofthepresencerules wererangerules.Theenvironmentaltolerancesdescribed bythedominantrulessuggestthatmeanNDVI,altitude, meantemperature,minimumsoilcalciumandminimum soilorganiccontentarelimitingvariablesfor B.anthracis inKazakhstan(Figure5).Medianminimumvaluesof meanNDVI,NDVIamplitude,annualprecipitation,dry monthprecipitation,wetmonthprecipitation,meantemperature,altitudeandsoilorganiccontentaresignificantly differentbetweentheA1.asub-lineageandtheoutbreaksoilexperimentusingtheWilcox-Mann-Whitneytestata 95%significancelevel.MedianmaximumvaluesofNDVI amplitude,meantemperature ,drymonthprecipitation, altitude,soilbasesaturationandsoilorganiccontentdiffer betweenthetwoexperiments. Table1SummaryofexperimentsExperimentExternalDataSplit(%Training/% Testing) AreausedforModel Building LocalityData Outbreak-Soil85/15AllKazakhstanAllspatiallyuniquelivestockoutbreaks A1.aSub-lineage80/20SouthernPolygonSpatiallyuniqueA1.aisolatesinsouthernpolygon SmallSouthern Outbreak 85/20SouthernPolygonRandomsub-setofspatiallyuniquelivestockoutbreaksin southernpolygon LargeSouthern Outbreak 80/15SouthernPolygonAllspatiallyuniquelivestockoutbreaksinsouthernpolygon Table2SamplesizesandaccuracymetricsforGARPmodelbuildingandevaluationTable3Model MetricOutbreak-SoilA1.asub-lineageLargeSouthernOutbreakSmallSouthernOutbreak Ntobuildmodels*2182611326 Ntotestmodels3913145147 TotalOmission2.60.00.00.0 AverageOmission9.913.119.115.5 TotalCommission32.719.1812.7417 AverageCommission58.466.1149.2656.35 AUC 0.71880.69640.74010.7386 SE0.04660.08170.24100.04 Z90.944.444916.328416.6241*Nwasdividedinto50%training/50%testingforeachmodeliteration AUC=areaunderthecurveMullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page3of14

PAGE 4

Valuesofthelimitingvaria bleswereextractedfrom areasoftenbestsubsetmodelagreementandplottedin twodimensionalvariablespace.TheA1.asub-lineage experimentshowedabroaderecologicalenvelopethan theoutbreak-soilexperimentbasedonareasoftenbest subsetmodelagreement,despitethesmallergeographic areapredictedbyagreementofalltenmodels(Figure 6).ThetwoA4locations,whicharedistantfromeach othergeographically,arefoundwithinanarrowrangeof meanNDVIandmeantemperature,butoccupynearly oppositeendsoftherangeofprecipitationvalues. Finally,theA3.blocationwasassociatedwithecological conditionstowardstheouterboundariesoftheecologicalenvelopepredictedbytheoutbreak-soilexperiment.DiscussionThisstudyassessestheadditionofsoilvariablestoa previouslydevelopedecologicalnichemodelfor Bacillus anthracis andisthefirstknowntomodeltheecological andgeographicdistributionofasinglesub-lineageof B. anthracis .Inclusionofavailablesoilvariablesintoour anthraxoutbreakmodelresu ltedinsubtlechangesin thelikelihoodofthepathogeninareasofnorthern Kazakhstan,butdidnotsubstantiallychangetheextent ofgeographicpredictionsorr esultsofrulesetanalyses [26].Theareaspredictedaslesssuitablebytheoutbreak-soilmodelcorrespondtoregionsoflocallydifferentvaluesforallfoursoilvariables(seeAdditionalFile 3:SoilVariables,availableasaPDFdocument).However,itisnotknownwhethertheseareasrepresenta uniqueecologicalregionorifmeasurementinthese areaswasaffectedbyerrororbias.Thelowminimum soilcalciumassociationfo undintherulesetanalyses contrastswithpreviousliteraturesuggestingthat B. anthracis sporepersistenceisassociatedwithhighsoil calciumlevels[3,30].Theresultsofourrulesetanalysis, however,arenotdirectlycomparabletopreviouswork inthatdifferentunitsofmeasurementandsampling techniqueswereused.Inaddition,thesoilsdataavailableforthisstudyhadarelativelycoarseresolutionof1 kmandwerefurtheraggregatedto8kmtomatchother climaticdataformodeldevelopment.Asaresult,fine Figure1 MapofKazakhstanwithanthraxlocalitydata .(A)Locationsofgenotypedisolates.Insetshowslocationsofoutbreaksusedforthe fulloutbreakmodel.GreenpointsindicatetrainingdataforinputintoGARPandyellowpointsindicateindependentpointsusedtoevaluate modelaccuracy.Shadedareaindicatessouthernpolygonusedforcreatingprojectedmodels. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page4of14

PAGE 5

resolutionrelationshipsbetweensoilandanthraxoccurrenceswouldlikelybemissedinthisexperiment. Improvedresolutionofsoilandoutbreakdata,suchas exactcarcasslocations,arelikelynecessarytocharacterizetheroleofsoilparametersinpromotinganthrax sporepersistence[3,31]andforbetterunderstanding thespatio-temporaldynamicsandecologyoflocaloutbreaks[32]. TheA1.asub-lineageexperimentpredictedamore extensivegeographicareaofanthraxpresencethandid Figure2 Predicteddistributionof Bacillusanthracis inKazakhstan .Predicteddistributionof Bacillusanthracis inKazakhstanbasedon outbreakdatawithandwithoutsoilvariables.(A)Outbreakexperiment(excludingsoilvariables)[26],(B)Outbreak-soilexperiment(includingsoil variables),(C)Differencesbetweendistributionspredictedbythetwoexperiments. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page5of14

PAGE 6

theoutbreakexperiment.Thisismostpronouncedin thenorthernandcentralportionsofthecountry.The medianminimumvaluesofmostvariablesdefinedby thedominantrulesetsweres ignificantlydifferent. Whenecologicalvaluesoflimitingvariableswere extractedfromgeographicareasofbestsubset agreementandplottedintwodimensionalvariablespace,theA1.asub-lineagewasassociatedwithalarger ecologicalenvelopethantheoutbreak-soildata.This findingillustratesthatanalysisofdominantrulesets aloneshouldbeinterpretedwithsomecaution.The variablerangesderivedfromthedominantrulesets Figure3 Predictedgeographicdistributionofthe Bacillusanthracis A1.asub-lineage .Comparisonofpredictedgeographicdistributionsof B.anthracis .(A)distributionof B.anthracis predictedbythesub-lineageexperiment,(B)differencebetweenpredicteddistributionsofthesublineageandtheoutbreak-soilexperiments. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page6of14

PAGE 7

Figure4 Predictedgeographicdistributionof B.anthracis basedonthelargesouthernoutbreakexperimentandsmallsouthern outbreakexperiment .Predictedgeographicdistributionof B.anthracis basedon(A)largesouthernoutbreakexperimentand(B)small southernoutbreakexperiment.(C)Differencebetweenpredictedgeographicdistributions.Greenpointsindicatetrainingdataforinputinto GARPandyellowpointsindicateindependentpointsusedtoevaluatemodelaccuracy. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page7of14

PAGE 8

Table3Rulestypesfromtenbestmodelsoftheoutbreak-soilandA1.asub-lineageexperiments.Valuesshownare numberofruletypesintheruleset(column%)Outbreak-SoilRuleSet RuleType261321242529394851Total Logit17(34)32(64)34(68)25(50)22(44)26(52)36(72)35(70)25(50)27(54)279(55.8) NegatedRange2(4)1(2)0(0)0(0)2(4)2(4)0(0)1(2)7(14)7(14)22(4.4) Range31(62)17(34)16(32)25(50)26(52)22(44)14(28)14(28)18(36)16(32)199(39.8) A1.aSub-lineageRuleSet RuleType1102140495154769193Total Atomic1(2)2(4)1(2)0(0)0(0)0(0)0(0)0(0)0(0)0(0)4(0.8) Logit15(30)5(10)22(44)20(40)30(60)2(4)7(14)24(44)22(44)9(18)156(31.2) NegatedRange10(20)0(0)4(8)2(4)0(0)0(0)0(0)7(14)0(0)0(0)23(5.6) Range24(48)43(86)23(46)28(56)20(40)48(96)43(86)19(38)28(56)41(82)317(63.4) Figure5 Medianrangesofenvironmentalvariablespredicting B.anthracis presencebytheoutbreak-soilexperiment .O=Outbreak-Soil Experiment;A=A1.asub-lineageexperiment. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page8of14

PAGE 9

summarizeapproximatelyonl yone-fifth(orfewer)of thetotalnumberofrulesgeneratedbyGARPintheten bestsubsets.Thevaluesextractedfrompredictedareas onthelandscapearederivedfromall500rulescontainedinthe10bestsubsestsandrepresentthespectrumofcomplexinteractionsbetweenvariablesandthe landscape.Becausegeographicareasofmodelagreement canbethoughtofasrepresentativeofallsub-sampled regionsorpopulations[33],thefindingofalargerecologicalenvelopefortheA1.asub-lineageexperiment lendssupporttothehypothesisthattheA1.asublineageof B.anthracis mayhavebroadenvironmental tolerancesthatinfluenceitsbroadgeographicdistribution[3,24]. TheAlineageismorewidelydistributedgloballythan othersubtypes,perhapsreflectingagreaterleveloffitnessascomparedtootherlineages[24].Thisfinding hasbeenshownonalocalscaleaswell.Isolatesofthe AlineageinKrugerNationalPark,SouthAfrica,as definedbyMLVA-8typing,weremorediffuselydistributedandshowedadistinctlydifferentspatialclusterpatternthanthoseoftheBlineage.Furthermore,theB 0.0 0.1 0.2 0.3 0.4 200 400 600 800 Mean NDVIAnnual Precipitation (mm) A1a sublineage OutbreakSoil Background A3.b A4 KZ Genotype 9 0.0 0.1 0.2 0.3 0.4 5051 01 5 Mean NDVIMean Temperature (C) A1a sublineage OutbreakSoil Background 0.0 0.1 0.2 0.3 0.4 200 400 600 800 Mean NDVIAnnual Precipitation (mm) Outbreak Soil Large S. Outbreak Small S. Outbreak Background 0.0 0.1 0.2 0.3 0.4 5051 01 5 Mean NDVIMean Temperature (C) Outbreak Soil Large S. Outbreak Small S. Outbreak Background Figure6 Distributionof B.anthracis inecologicalspace .Predicteddistributionof B.anthracis inecologicalspacebasedonareasoftenbest subsetmodelagreement.Redpoints=outbreaklocations,bluepoints=A1.aisolatelocations,blacktriangles=A4locations,blackcircle=A3.b location. Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page9of14

PAGE 10

lineageisolatesoccupiedanarrowrangeofavailable ecologicalconditionswith inthoseoccupiedbytheA lineageisolates[3].Intheexperimentsreportedhere, theA1.asub-lineagelocationswereassociatedwith lowerpHvaluesthantheoutbreaklocationsandthis couldprovisionallysupportthefindingsfromKruger NationalPark[3].Itfollowsthatthata B.anthracis lineageand/orsub-lineageotherthanA1.amaypredominateinthenorthernregionsofKazakhstanandis drivingthenarrowergeographicandecologicalpredictionoftheoutbreak-soilexperiment.Genotypingof additionalisolatesfromnorthernKazakhstanisnecessarytoevaluatethishypothes is.Locationsofoutbreaks oftheA4lineageingeographicalandecologicalspace suggeststhatthisgenotypemayalsohavearelatively broaddistribution,whichisconsistentwiththeA4sublineagebeingfoundacros stheMiddleEastandChina [29,34].Althoughwecannotmakeinferencesregarding theenvironmentalaffinities ofthesingleA3.bisolate, wenoteitislocatedonthefareasternborderof Kazakhstaninanareapredictedtobeunsuitablefor anthraxbytheoutbreak-soilexperiment.TheA3.bsublineagehasbeenisolatedfromgeographicallylimited areas,mostnotablynorthernChinaandTexas,andthe presenceofthisgenotypeislikelyaresultofhistorical traderoutes[24,34,35]. Thegeneticdiversityof B.anthracis isolatesinsouthernKazakhstanisnotsurprisinggiventhelocationof thisareaalongthehistoricSilkRoad[29],butthis diversityalsoimpliesthatthisregionissupportiveof sporepersistence.Associationsbetweengenotypesof B. anthracis ,environment,virulenceorhostspecieshave notyetbeenfullyexploredanditisunknownwhether genotypeinfluencesepidemiologicalcharacteristicsof outbreaks.Understandingtheserelationshipswill improveourunderstandingofanthraxdiseaseecology, helpfocussurveillanceandefficientlydirectproactive vaccination.Furthermore,thisknowledgecanhelpdistinguishbetweennaturallyoccurringoutbreaks,contaminationandpotentialbiote rrorism,andgreatlyenhance epidemiologicaltraceback(tracingoutbreakstosource) effortsduringoutbreaks. FewstudiesusingGARPhavequantifiedandexaminedruletypes.GARPbeginscreatingrulesetsby choosingthefirstruletypeatrandomandsuccessful rulesarecarriedforwardintosubsequentrulesets. Thus,thefirstruletypechosenatrandomwilloften predominateinthefinalrulesets.BlackburnandJoyner et.al.bothpresentedsummariesanddistributionofthe dominantruletypespredicting B.anthracis inthecontinentalUnitedStatesandKazakhstan,respectively [26,32].Blackburnnotedapredominanceofrangerules amongatotalof63rulespredictinggreaterthan90%of thelandscape[32].JoyneretaldividedKazakhstaninto northernandsouthernhalvesandmodelledthetwosectionsseparately[11].Agreaterpercentageofrangerules describedthenorthernhalfofthecountry,whereaslogit rulespredominantlydescribedthesouthernhalfofthe country.Here,logitrulesdominatedintheout-break soilexperimentandrangerulesintheA1.asub-lineage experiment.Furtherworkisrequiredtoteaseout whetherdominantruletypesresultfromthestochastic natureofGARPorarerelatedtocomplexinteractions betweentheorganismandenvironmentalvariables. Additionalfutureworkshouldalsoexplorehowrule setsandecologicalvaluesfoundoverpredictedareasof thelandscapecanbeusedtoenhanceourunderstandingoftheecologyofanorganism. SeveralauthorshavenotedthatENM-basedpredictionsofspecieswithwidespreaddistributionsshow reducedmodelaccuracywhichcanbeimprovedby dividingspeciesortherangeintosub-units[33,36-39]. Oneexplanationforthea pparentpooraccuracyof modelsofwidespreadspeciesistheuseofAUC.The AUCissensitivetotheareapredictedtobesuitablefor aspeciesrelativetothetotallandareaanalyzed[33,40]. Otherconsiderationsincludenon-uniformityofpresencelocations(geographicalbias),andbiologicalfactorssuchaslocalecologicaladaptationsandgenetic diversity[33,37-39].Forexample,modellingof Francisellatularensis genotypesintheUSyieldedoverlapping, yetdifferent,geographicpredictionsandecologicalassociations[12].Interestingly,thisdifferencewasapparent atintermediate,asopposedtocoarse,phylogenetic levels.Similarly,Fisheret.al.showedthatthreegenotypecategoriesofthebroadl ydistributedpathogenic fungus Penicilliummarneffei correlatedwithenvironmentalheterogeneityacrossVietnam[21].UsingGARP, thegenotypesclasseswerepredictedtooccupythree non-overlappinggeographicareas.Asaconsequenceof B.anthracis beingawidelydistributedspecies,models ofanthraxoutbreaksaresubjecttosimilarlimitationsin accuracy.Genotypespecificmodelsmaythereforehave improvedaccuracyandpredictivepower,andshouldbe exploredatmultiplephylogeneticlevels. Itisworthwhiletoevaluatemodellinglimitations whichcouldpotentiallyexplaindifferencesbetween experiments.DespitestudiesshowingGARPtobe robustwhenpredictingnewlandscapes[14,25,41],our useofaprojectedmodellingstrategymaywronglypredictgeographicandecologicaldistributiongiventhe largegeographicareainquestion.However,projected modelscreatedusingsouthernoutbreakpointsshow similargeographicpredictionsastheoutbreak-soil experiment,supportingthatthebroadergeographicA1. asub-lineagepredictionisnotsimplyanartefactofthe modellingtechnique.Thatthelargeandsmallsouthern outbreakexperimentsshowedalesserdegreeofmodelMullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page10of14

PAGE 11

agreementthantheoutbreak-soilexperimentandthat thelargeoutbreaksubsettingprocedurehadimproved spatialhomogeneityoverthesmallsouthernoutbreaks subsettingindicatesthatprojectedmodelsmaybesensitivetoissuesofsamplesizeandclustering.Issuesrelatingtosamplesizeappearto beimportantinourstudy. Thepredictedgeographicdistributionof B.anthracis variedamongthe10randomdatasplitsforboththeAl. asub-lineageandsmallsouthernoutbreakexperiments. Incontrast,thesamerandomdatasplittingprocedure performedwiththelargesouthernoutbreakandthe outbreak-soilexperimentsshowedspatialhomogeneity amongmodels[26].AlthoughStockwellandPeterson andHernandezfoundthatasfewas10presencepoints areadequateforaccurateGARPmodels[36,39],additionalworkhasshownthatcertainspeciesorgeographic scenariosaremoresensitivetosamplesizethanothers andthatmodelaccuracycanbesensitivetobothsample sizeandextentofthespecies range[42].Here,any effectofsamplesizeislikelyexaggeratedbyprojection ontoalargegeographicareaandtherelativelylimited resolutionthedata[36,42]. Someadditionallimitationsapplytoourfindings.The A1.asub-lineageisolatecollectionwasderivedfrommultiplespecies,includinglivestockandhumans,andfrom soilsamples.Theoutbreakdata,however,werederived fromlivestockonly.Theimpactofthisonmodelcomparisonsisunknownsinceassociationsbetweenhost, genotypeandenvironmentareasyetunexplored.Weare limitingourmodelledareab ypoliticalboundariesas opposedtobiogeographiclimits.Finally,thegenotyped isolatecollectionisgeographicallybiasedtowardsthe southeastportionofKazakhstanandspansarelatively longperiodoftime[29].Samplingbiasiscommonin nichemodelsusinghistoric alcollectionsandcancreate artificialpatternsinthedata,althoughGARPisarguably lesssensitivethanothermodellingalgorithmstospatial bias[10,37,43,44].Futuregenotypingofadditionalisolatesfromunder-sampledportionsofthecountry,particularlythenorthernoblasts,willbeessentialtobetter characterizethegeneticdiversityandecologyofanthrax inKazakhstan,allowingtheconstructionofmorerefined predictivemodels.ConclusionsTheinclusionofavailablesoilvariablesresultedin subtlechangesinthepredictedgeographicdistribution ofanthraxinKazakhstan,buttheexperimentislimited bythenatureofavailablesoilvariables.Standardized soilvariablesandfinerresolutiondatawillbeessential tocharacterizingtheimportanceofsoilparametersin B. anthracis persistence.TheA1.as ub-lineageexperiment showedalargergeographicandecologicaldistribution thantheoutbreakbasedexperiment.Understanding genetic-environmentalassociationswillbeessentialto accuratemodellingofanthraxforuseindiseasepreventionandcontrolinKazakhstanMethodsAnthraxoccurrencedataPresencepointsforoutbreak-basedmodelsweretaken fromthedatabasecreatedbyJoyneret.al.[26].Briefly, historicalrecordswereusedtoconstructadatabaseof 3,947anthraxoutbreaksreportedinKazakhstanbetween 1937and2006.Thedataweresequentiallyfilteredto createadatasetcontainingthelatitudeandlongitudeof outbreaksincattle,sheepandgoatswhichoccurred between1960and2000.Thistimeperiodreflectsthe implementationofmassvaccinationandcorrespondsto theaverageddatafromboththeWorldClimandsoils datasets[45,46].Thefinaldatasetcontained258spatiallyuniquepoints,meaningthatonlyoneoutbreak pointoccurredineach8km2pixel.An8km2resolution waschosenbecauseoutbreaksweremappedtothenearestvillageandsomeoutbreaksoccurredgreaterthan1 kmfromthevillagecoordinates.Thisdatasetishereafterreferredtoasthefulloutbreakdataset. AseconddatasetwasconstructedusingoutbreakisolatesgenotypedbyAikimbayevet.al.[29].Isolateswere groupedintotheA1.a(n=78),A3.b(n=6)andA4(n =4)sub-lineagesusingunweightedpairgroupmethod witharithmeticmean(UPGMA)clusteranalysisandthe 89 B.anthracis genotypesidentifiedbyKeimet.al.[22]. Thiswasfilteredtocontainonlyspatiallyuniquepoints ataresolutionof8km2resultingin42spatiallyunique points,ofwhich39wereA1.a,twowereA4and1was A3b.LocalitydataaremappedinFigure1.Onlythe A1.asub-lineagehadanadequatenumberofspatially uniquepointsformodelling .TheA1.alocationswere geographicallybiasedtowardsthesouth-easternportion ofthecountry.Toreducetheproblemoflargely unsampledareasbeingconsideredasabsencepoints,we createdapolygonencompassi ngsoutheastKazakhstan usinglatitude48Nand60Easthenorthernandwesternboundaries,respectively,andthecountry spolitical boundariesinthesouthandeast.Theboundariesofthis southernpolygonwerederivedfromexaminationofthe locationsoftheA1.aisolatesandfromthedifferent northernandsouthernecologicalassociationsnotedby Joyner[26].Thesouthernpolygonwasusedtoclipthe A1.alocalitypointsinArcMapandthissetofsouthern A1.alocationswasusedforthesub-lineageexperiment. Thesameprocedurewasusedtocreateasouthernoutbreakdataset.EnvironmentalDataWeusedsixenvironmentalcoveragesdownloadedfrom theWorldClimwebsitehttp://www.worldclim.org[46].Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page11of14

PAGE 12

TheWorldClimvariablesarecalculatedfrominterpolationofmonthlytemperatureandprecipitationmeasurementsrecordedatstationsl ocatedworldwidebetween 1961and2000.Monthlyvaluesaretransformedby WorldCliminto19bioclim aticvariablegridsthat describeannualtrends,seasonalityandpotentiallylimitingecologicalparameterssuchastemperatureofthe coldestandwarmestmonths.Twosatellite-derived environmentalvariablesdescribingtemperatureand vegetationmeasureswereobtainedfromtheTrypanosomiasisandLandUseinAfrica(TALA)researchgroup (Oxford,UnitedKingdom)[47]. Weaddedfoursoilvariablestothesetofeightenvironmentallayersusedinthepreviousstudytotestthe influenceofsoilparametersontheoutbreakmodel.Soil variableswerederivedfromtheHarmonizedWorldSoil Databaseandwereavailableat1km2resolution[45]. Allcoverageswerere-sampledto8km2andclippedto theboundariesofKazakhstaninArcView3.3.(EnvironmentalSystemsResearchinstitute,Redlands,CA).An identicalsetofcoverageswasclippedtothesouthern polygon.ThefinalsetofcoveragesisgiveninTable4.EcologicalNicheModellingThisstudyusedtheGeneticAlgorithmforRule-SetPrediction(GARP)toperformtheecologicalnichemodelling[48].Modelsweredev elopedinDesktopGARP v.1.1.3,whichgivestheusertheoptiontowriteoutthe rulesetsforeachmodel.Briefly,GARPisapresence onlymodellingtechniquethatdetectsnon-randomassociationsbetweenspecieslocalitiesandspecificenvironmentalvariables.Throughaniterativeprocess, relationshipsareexpresse dasaseriesoflogicstatements,orrules,ofwhichtherearefourtypes:(1)logitbasedonlogisticregression;(2)atomic-singlevaluefor agivenvariablethatpredictspresence;(3)range-a rangeofvaluesofagivenvariablethatpredictspresence;and(4)negatedrange-arangeofvaluesoutside ofwhichpresenceispredicted.EachindividualGARP modelisasetof50rulesthatarerandomlygenerated, testedandmodified.Theusersetsamaximumnumber ofmodelstobecreatedinasingleexperiment.Abest subsetsprocedurewithinGARPthenselectsasetof optimalmodelsbasedonuserdefinedomissionand commissioncriteria[48].T healgorithmisatwo-step process,wherefirstrelationshipsaredefinedinvariable spacethrougharandomwalkandthenappliedtothe geographiclandscapewherethoseconditionsaremet [25].GARPthereforehasthebenefitofbeingableto projectrulesetsontotheenvironmentallayersofadifferentlandscapeandhasbeenshowntoberobustin thisapplication[25,41].ModelbuildingandevaluationTotesttheeffectofsoilsontheoutbreakexperiment weusedthefulloutbreakdatasetaslocalitypointsand thetwelveenvironmentalvariablesdescribedinTable3. The258spatiallyuniquepoin tswererandomlydivided intoan85%(n=218)trainingsetusedformodelbuildinganda15%(n=39)testingsetformodelevaluation. The32southernA1.apointsweredividedintoan80% (n=26)trainingsetanda20%(n=6)testingsetin ordertomaximizepointsavailablefortesting[33,39,42]. TheA1.atrainingsetwasinputintoGARPwiththeset ofenvironmentalcoveragesclippedtothesouthern polygonformodeldevelopment.RulesfromthissouthernA1.aexperimentwereprojectedontotheentire landscapeofKazakhstan.Inordertotesttherobustness oftheA1.amodelprojectionsgiventherelativelysmall samplesizeandissuesoftransferability,twoexperimentsusingthesouthernoutbreakdatawereperformed.Thefirstutilizedall142southernoutbreak pointsandisreferredtoasthelargesouthernoutbreak experiment.Forthesecon d,32pointswererandomly selectedfromthesouthernoutbreakdataset(small southernoutbreakexperiment).Fortheseexperiments an85%/15%and80%/20%,respectively,externaldata splitwasperformedandtheexperimentsconductedas forthesub-lineage.ThefourexperimentsaresummarizedinTable4. Forallnichemodellingexperiments,wespecified200 modelswithamaximumof1,000iterationsandaconvergencelimitof0.01.Thetrainingdatawereinputinto GARPwitha50%training/50%testinginternaldatapartition.Thebestsubsetprocedureselectedthebest20 modelsundera10%hardomissionthresholdanda50% commissionthreshold.Theresultingtenbestsubset modelswereimportedinArcGISandsummatedusing therastercalculatorfunctionoftheSpatialAnalyst Table4EnvironmentalcoveragesusedforGARPmodelsEnvironmentalVariable(unit)NameSource Elevation(m)AltitudeWorldClim* AnnualTemperatureRange(C)BIO7WorldClim AnnualMeanTemperature(C)BIO1WorldClim PrecipitationofDriestMonth(mm)BIO14WorldClim PrecipitationofWettestMonth(mm)BIO13WorldClim AnnualPrecipitation(mm)BIO12WorldClim NDVIAmplitude(nounits)wd1014a1TALA MeanNDVI(nounits)wd1014a0TALA SoilpH(-log(H ))HWSD TopsoilCalcium(%weight)HWSD TopsoilOrganicContent(%weight)HWSD SubsoilBaseSaturation(%)HWSD*http://www.worldclim.org[46] TrypanosomiasisandLandUseinAfrica(TALA)researchgroup[47] HarmonizedWorldSoilDatabase[45]Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page12of14

PAGE 13

extension.Thiscreatedasinglecumulativerasterfileof modelagreementfor B.anthracis presenceranging from0(allmodelspredictabsence)to10(allmodels predictpresence).Themoremodelsthatpredictpresenceforagivenpixel,thehigherthelikelihoodthat thepixelcansupport B.anthracis RuletypesfromthetenbestmodelsoftheoutbreaksoilexperimentandtheA1.asub-lineageexperiment wereextractedwithapythonscript(K.M.McNyset,US NOAA)andsummarizedtoillu stratetherelativenumbersofeachruletype.Dominantrules,orthesubsetof rulesthattogetherpredictover90%ofthelandscape, foreachmodelwereidentif ied.Weextractedtheminimumandmaximumvaluesofrangerulesusingthe pythonscript.Whenlogitruleswereidentified,we extractedtherangeofvaluesacrossthepixelspredicted bythatruleusingthe ExtractValuestoPoints routine oftheSpatialAnalystextensioninArcMap.Median minimumandmaximumvaluesforeachvariablewere calculatedinSAS(SAS9.2,Cary,N.C.)andplottedasa bargraph.Differencesinmedianandmaximumvalues betweenexperimentswereassessedusingWilcoxonMann_WhitneytestinSAS. Predictiveperformanceofthebestsubsetmodelswas evaluatedwithanareaunderthecurve(AUC)ina receiveroperatingcharacteristic(ROC)analysisusing theindependenttestdatawithheldfromtheoriginal datasets[40].Fortheprojectedmodels,testingpoints includedpresencepointsoutsidethesouthernpolygon inadditiontopointswithheldfromwithinthesouthern polygon.ValuesofAUC,whichrangefrom0.5(nodifferentfromrandom)to1(aperfectmodel),arederived frommeasuresofsensitivity(absenceofomissionerror) andspecificity(absenceof commissionerror).Thecalculatedvalueiscomparedtothatofarandommodel usingaz-test.Inaddition, measuresofomissionand commissionwerecalculatedusingthesummedtenbest subsetmodels.TotalandaverageomissionvaluesevaluatehowwellGARPpredictsthepresenceofknown localitypointsnotincludedinthemodelbuildingdata. Totalandaveragecommissionisthepercentofpixels predictedaspresencebythesummatedmodelandthe averageofthisvalueforalltenbestsubsetmodels, respectively.Largevariationbetweenthetwomeasures ofcommissionsuggestssubstantialvariationbetween theproportionsofthelandscapepredictedpresentby eachofthetenbestsubsetmodels[33]. Environmentalvaluesof5,000randomlychosenpoints fromareaspredictedbyall10ofthebestsubsetmodels wereextractedusingthe ExtractValuestoPoints routineoftheSpatialAnalystextensioninArcMap.Values ofeachenvironmentalvariableateachpresencepoint andat5,000randompointsrepresentingthetotalavailableenvironmentalspace(background)weresimilarly extracted.Specificenvironmentalvaluesappearingtobe limitingfactorsforpredictionof B.anthracis werechosenbasedontherulesetevaluation(Figure5)and visualizedin2-dimensionalecologicalspaceagainstthe backgroundofavailableenvironmentalconditionsusing R2.1.1http://www.R-project.org.AdditionalmaterialAdditionalfile1:RandomSubsets .Predictedgeographicdistribution ofB.anthracisbasedon10randomsubsetsofinputlocalitypointsfor theAa.1sub-lineage,largesouthernoutbreakandsmallsouthern outbreakexperiments. Additionalfile2:Accuracymetricsforrandomsubsets .Accuracy metricsof10randomsubsetsofinputlocalitypointsfortheAa.1sublineage,largesouthernoutbreakandsmallsouthernoutbreak experiments Additionalfile3:SoilVariables .Mappedvaluesofthefoursoil variables(minimumsoilpH,minimumsoilorganiccontent,minimumsoil calciumandminimumsoilbasesaturation). Acknowledgements ThisworkwasfundedbytheUSDefenseThreatReductionAgencythrough theCooperativeBiologicalEngagementPrograminKazakhstanandthe EmergingPathogensInstituteattheUniversityofFlorida.JKBandJCMare supportedthroughtheDTRAJointUniversityPartnershipadministered throughtheUniversityofNewMexico.MVEissupportedbytheEmerging PathogensInstituteattheUniversityofFlorida.G.Temiraliyeva,Y. Sansyzbayev,T.A.Joyner,A.J.Curtis,M.E.Hugh-Jonesassistedwiththe developmentandmaintenanceoftheanthraxGISdatabase. Authordetails1DepartmentofGeography,UniversityofFlorida,Gainesville,FL,USA.2Spatial EpidemiologyandEcologyResearchLaboratory,EmergingPathogens Institute,UniversityofFlorida,Gainesville,FL,USA.3KazakhScienceCentre forQuarantineandZoonoticDiseases,MinistryofHealthoftheRepublicof Kazakhstan,Almaty,Kazakhstan.4ScientificandPracticalCentreofSanitary andEpidemiologicalExpertiseandMonitoring,MinistryofHealthofthe RepublicofKazakhstan,Almaty,Kazakhstan. Authors contributions JCMplannedthestudy,rantheexperimentsandwrotethemanuscript.JKB contributedtoplanningandrunningtheexperimentsanddraftingthe manuscript.LL,YP,JKBconstructedtheGISdatabaseforoutbreaksand genotypedstrains.LL,YP,AAcollectedandmanagedthe Bacillusanthracis strains.MVE,LL,YP,AAgenotypedthestrains.Allauthorsreviewedthefinal draftofthemanuscript. Received:9October2011Accepted:12December2011 Published:12December2011 References1.Hugh-JonesM,BlackburnJ: TheecologyofBacillusanthracis. Molecular AspectsofMedicine 2009, 30 :356-367. 2.DragonDC,RennieRP: Theecologyofanthraxspores:toughbutnot invincible. TheCanadianVeterinaryJournal 1995, 36 :295. 3.SmithKL,DeVosV,BrydenH,PriceLB,Hugh-JonesME,KeimP: Bacillus anthracisdiversityinKrugerNationalPark. JClinMicrobiol 2000, 38 :3780-3784. 4.VanNessG,SteinCD: SoilsoftheUnitedStatesfavorableforAnthrax. J AmVetMedAssoc 1956, 128 :7-9. 5.VanNessG: EcologyofAnthrax. Science 1971, 172 :1303-1307. 6.EppT,WaldnerC,ArgueCK: Case-controlstudyinvestigatingananthrax outbreakinSaskatchewan,Canada Summer2006. CanVetJ 2010, 51 :973-978.Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page13of14

PAGE 14

7.ParkinsonR,RajicA,JensonC: Investigationofananthraxoutbreakin Albertain1999usingageographicinformationsystem. CanVetJ 2003, 44 :315-318. 8.BalesME,DannenbergAL,BrachmanPS,KaufmannAF,KlatskyPC, AshfordDA: EpidemiologicResponsestoAnthraxOutbreaks:AReviewof FieldInvestigations,1950-2001. EmergingInfectiousDiseases 2002, 8 :1163. 9.WoodsCW,OspanovK,MyrzabekovA,FavorovM,PlikaytisB,AshfordDA: Riskfactorsforhumananthraxamongcontactsofanthrax-infected livestockinKazakhstan. TheAmericanJournalofTropicalMedicineand Hygiene 2004, 71 :48. 10.BlackburnJK,McNysetKM,CurtisA,Hugh-JonesME: Modelingthe geographicdistributionofBacillusanthracis,thecausativeagentof anthraxdisease,forthecontiguousUnitedStatesusingpredictive ecologicnichemodeling. AmericanJournalofTropicalMedicineand Hygiene 2007, 77 :1103-1110. 11.JoynerTA,LukhnovaL,PazilovY,TemiralyevaG,Hugh-JonesME, AikimbayevA,BlackburnJK: Modelingthepotentialdistributionof Bacillusanthracisundermultipleclimatechangescenariosfor Kazakhstan. PLoSOne 2010, 5 :e9596. 12.NakazawaY,WilliamsRA,PetersonAT,MeadPS,KugelerKJ,PetersenJM: EcologicalnichemodelingofFrancisellatularensissubspeciesand cladesintheUnitedStates. AmJTropMedHyg 2010, 82 :912-918. 13.PetersonAT,BauerJT,MillsJN: Ecologicandgeographicdistributionof filovirusdisease. EmergingInfectiousDiseases 2004, 10 :40-47. 14.RonSR: PredictingtheDistributionoftheAmphibianPathogen BatrachochytriumdendrobatidisintheNewWorld1. Biotropica 2005, 37 :209-221. 15.AdjemianJCZ,GirvetzEH,BeckettL,FoleyJE: AnalysisofGenetic AlgorithmforRule-SetProduction(GARP)modelingapproachfor predictingdistributionsoffleasimplicatedasvectorsofplague,Yersinia pestis,inCalifornia. JMedEntomol 2006, 43 :93-103. 16.CostaJ,PetersonAT,BeardCB: Ecologicnichemodelingand differentiationofpopulationsofTriatomabrasiliensisneiva,1911,the mostimportantChagas diseasevectorinnortheasternBrazil (hemiptera,reduviidae,triatominae). AmericanJournalofTropicalMedicine andHygiene 2002, 67 :516-520. 17.PetersonAT,ShawJ: Lutzomyiavectorsforcutaneousleishmaniasisin SouthernBrazil:ecologicalnichemodels,predictedgeographic distributions,andclimatechangeeffects. IntJParasitol 2003, 33 :919-931. 18.PetersonAT,Sanchez-CorderoV,BenBeardC,RamseyJM: Ecologicniche modelingandpotentialreservoirsforChagasdisease,Mexico. Emerging InfectiousDiseases 2002, 8 :662-667. 19.GrinnellJ: Theniche-relationshipsoftheCaliforniaThrasher. Auk 1917, 34 :427-433. 20.HutchinsonGE: PopulationStudies-AnimalEcologyandDemographyConcludingRemarks. ColdSpringHarbSym 1957, 22 :415-427. 21.FisherMC,HanageWP,DeHoogS,JohnsonE,SmithMD,WhiteNJ, VanittanakomN: Loweffectivedispersalofasexualgenotypesin heterogeneouslandscapesbytheendemicpathogenPenicillium marneffei. PLoSPathogens 2005, 1 :1986-1990. 22.KeimP,PriceL,KlevytskaA,SmithK,SchuppJ,OkinakaR,JacksonP,HughJonesM: Multiple-locusvariable-numbertandemrepeatanalysisreveals geneticrelationshipswithinBacillusanthracis. JournalofBacteriology 2000, 182 :2928. 23.ListaF,FaggioniG,ValjevacS,CiammaruconiA,VaissaireJ,LeDoujetC, GorgO,DeSantisR,CarattoliA,CiervoA: GenotypingofBacillus anthracisstrainsbasedonautomatedcapillary25-locimultiplelocus variable-numbertandemrepeatsanalysis. BMCMicrobiology 2006, 6 :33. 24.VanErtMN,EasterdayWR,HuynhLY,OkinakaRT,Hugh-JonesME,RavelJ, ZaneckiSR,PearsonT,SimonsonTS,U RenJM, etal : GlobalGenetic PopulationStructureofBacillusanthracis. PLoSOne 2007, 2. 25.BlackburnJ: IntegratingGeographicInformationSystemsandEcological NicheModelingintoDiseaseEcology:ACaseStudyofBacillusanthracis intheUnitedStatesandMexico. In EmergingandEndemicPathogens. Editedby:O ConnelK,etal.SpringerScience;2010:59-88. 26.Joyner, etal : Ecologicalnichemodelingofazoonosis:acasestudy usinganthraxoutbreaksandclimatechangeinKazakhstan. Universityof Florida;2010. 27.CherkasskiyBL: Anationalregisterofhistoricandcontemporaryanthrax foci. JApplMicrobiol 1999, 87 :192-195. 28.Hugh-JonesM: 1996-97GlobalAnthraxReport. JApplMicrobiol 1999, 87 :189-191. 29.AikembayevAM,LukhnovaL,TemiraliyevaG,Meka-MechenkoT,PazylovY, ZakaryanS,DenissovG,EasterdayWR,VanErtMN,KeimP, etal : Historical distributionandmoleculardiversityofBacillusanthracis,Kazakhstan. EmergInfectDis 2010, 16 :789-796. 30.HimsworthCG: Thedangeroflimeuseinagriculturalanthrax disinfectionprocedures:thepotentialroleofcalciuminthepreservation ofanthraxspores. CanVetJ 2008, 49 :1208-1210. 31.VanNessG: GeologicImplicationsofAnthrax. TheGeologicalSocietyof AmericaSpecialPaper 1967, 90 :61-64. 32.BlackburnJK: EvaluatingthespatialecologyofanthraxinNorthAmerica: Examiningepidemiologicalcomponentsacrossmultiplegeographic scalesusingaGIS-basedapproach. LouisianaStateUniversity;2006. 33.McNysetKM: Useofecologicalnichemodellingtopredictdistributions offreshwaterfishspeciesinKansas. EcolFreshwFish 2005, 14 :243-255. 34.SimonsonTS,OkinakaRT,WangB,EasterdayWR,HuynhL,U RenJM, DukerichM,ZaneckiSR,KeneficLJ,BeaudryJ: BacillusanthracisinChina anditsrelationshiptoworldwidelineages. BMCMicrobiology 2009, 9 :71. 35.KeneficLJ,PearsonT,OkinakaRT,ChungWK,MaxT,TrimCP,BeaudryJA, SchuppJM,VanErtMN,MarstonCK, etal : Texasisolatescloselyrelatedto BacillusanthracisAmes. EmergInfectDis 2008, 14 :1494-1496. 36.HernandezPA,GrahamCH,MasterLL,AlbertDL: Theeffectofsamplesize andspeciescharacteristicsonperformanceofdifferentspecies distributionmodelingmethods. Ecography 2006, 29 :773-785. 37.PetersonAT: Predictingspecies geographicdistributionsbasedon ecologicalnichemodeling. TheCondor 2001, 103 :599-605. 38.RiceNH,Martinez-MeyerE,PetersonAT: Ecologicalnichedifferentiationin theAphelocomajays:aphylogeneticperspective. BiolJLinnSoc 2003, 80 :369-383. 39.StockwellDRB,PetersonAT: Effectsofsamplesizeonaccuracyofspecies distributionmodels. EcologicalModelling 2002, 148 :1-13. 40.WileyE,McNysetKM,PetersonAT,RobinsCR,StewartAM: Nichemodeling andgeographicrangepredictionsinthemarineenvironmentusinga machine-learningalgorithm. Oceanography 2003, 16 :120-127. 41.PetersonAT,PapeM,EatonM: Transferabilityandmodelevaluationin ecologicalnichemodeling:acomparisonofGARPandMaxent. Ecography 2007, 30 :550-560. 42.PetersonAT,BallLG,CohoonKP: PredictingdistributionsofMexicanbirds usingecologicalnichemodellingmethods. Ibis-JournaloftheBritish OrnithologistsUnion 2002, 144 :27. 43.AraujoMB,GuisanA: Five(orso)challengesforspeciesdistribution modelling. JBiogeogr 2006, 33 :1677-1688. 44.StockwellD,PetersD: TheGARPmodellingsystem:problemsand solutionstoautomatedspatialprediction. IntJGeogrInfSci 1999, 13 :143-158. 45.FAO/IIASA/ISRIC/ISSCAS/JRC: HarmonizedWorldSoilDatabase(version1.1) Rome,ItalyandIIASA,Laxenburg,Austria;2009. 46.HijmansRJ,CameronSE,ParraJL,JonesPG,JarvisA: Veryhighresolution interpolatedclimatesurfacesforgloballandareas. IntJClimatol 2005, 25 :1965-1978. 47.HaySI,TatemAJ,GrahamAJ,GoetzSJ,RogersDJ: Globalenvironmental dataformappinginfectiousdiseasedistribution. AdvParasit 2006, 62 :37-77. 48.AndersonRP,LewD,PetersonAT: Evaluatingpredictivemodelsof species distributions:criteriaforselectingoptimalmodels. Ecological Modelling 2003, 162 :211-232.doi:10.1186/1472-6785-11-32 Citethisarticleas: Mullins etal .: EcologicalNicheModellingofthe Bacillusanthracis A1.asub-lineageinKazakhstan. BMCEcology 2011 11 :32.Mullins etal BMCEcology 2011, 11 :32 http://www.biomedcentral.com/1472-6785/11/32 Page14of14

PAGE 15

nnrrr !"!

PAGE 16

n nrrr !!

PAGE 17

n rrr !""

PAGE 18

nnnrn !"# !$%& n n r $ (')" *) $ +* +* ++ + $* ,+ -* ", $ )" $"" + )$ +$ ,$ *,, (-"* "$++ )$ "$ +,$ *,,$ "+), *,+) *",, *$ .-"*) +$) "*$ +*+ "++ +$") "" *,), "** /0, ) ") +) *, *) "" "" 1n) )) $+) )"" ))$,) )$ )$"" )) $$,*+ $+"+ nnnrn2 3 !"# !)+%&n n r ,) $*, $ $ (')$ $) $,* "+ )" $, $ -" ") $"$ "$ $++ ",* "$ *,, (-)" )$, $"$,*, """ *$ $,$+ )"+ *,* .-*,)" "+, +*)""+* *+ "$$ +) "** +$ ")$ /0"" ") +") "" *, " ") 1n$ )$)$+ )+*""$,, ))*$ *"" $,*+ *,$ )"+

PAGE 19

$nnnrn2 3** !$# !)*%&n n r '+ + + ) ('*"* $ ," *" -+)+ +), $+ + $" **$ ,*, (-*$$**$ ** ) "*) *$) **" "$$ *)* *+ .-+$"+$" ", +$+ + ++ +$ ++ + + /0)+) ), ) )* )* )* )* )" )" 1n*+$*"$* *$$ ")$ +"" *,*$) "$$ $"$" "$" "))

PAGE 20

nr r nr n!r "#$ n! %#&'$(&#)#*'&+,$ *$&,*#