Validation of a Previously Developed Geospatial Model That Predictsthe Prevalence of Listeria monocytogenes in New York State ProduceFields

Daniel Wellera Suvash Shiwakotib Peter Bergholzb Yrjo Grohnc Martin Wiedmanna Laura K Strawnd

Department of Food Science Cornell University Ithaca New York USAa Department of Veterinary and Microbiological Sciences North Dakota State University FargoNorth Dakota USAb Department of Population Medicine and Diagnostic Science Cornell University Ithaca New York USAc Department of Food Science andTechnology Eastern Shore Agricultural Research and Extension Center Virginia Polytechnic University Painter Virginia USAd

Technological advancements particularly in the field of geographic information systems (GIS) have made it possible to predictthe likelihood of foodborne pathogen contamination in produce production environments using geospatial models Yet fewstudies have examined the validity and robustness of such models This study was performed to test and refine the rules associ-ated with a previously developed geospatial model that predicts the prevalence of Listeria monocytogenes in produce farms inNew York State (NYS) Produce fields for each of four enrolled produce farms were categorized into areas of high or low pre-dicted L monocytogenes prevalence using rules based on a fieldrsquos available water storage (AWS) and its proximity to water im-pervious cover and pastures Drag swabs (n 1056) were collected from plots assigned to each risk category Logistic regres-sion which tested the ability of each rule to accurately predict the prevalence of L monocytogenes validated the rules based onwater and pasture Samples collected near water (odds ratio [OR] 30) and pasture (OR 29) showed a significantly increasedlikelihood of L monocytogenes isolation compared to that for samples collected far from water and pasture Generalized linearmixed models identified additional land cover factors associated with an increased likelihood of L monocytogenes isolationsuch as proximity to wetlands These findings validated a subset of previously developed rules that predict L monocytogenesprevalence in produce production environments This suggests that GIS and geospatial models can be used to accurately predictL monocytogenes prevalence on farms and can be used prospectively to minimize the risk of preharvest contamination ofproduce

Fresh produce presents a unique food safety challenge due tothe absence of a kill step between harvest and consumption

An increase in recalls and reported outbreaks linked to fresh pro-duce over the past decade (1ndash3) have been associated with con-sumer avoidance of products linked to outbreaks (4 5) This trendcan negatively affect growers and the produce industry (4ndash6) Forexample following a 2011 listeriosis outbreak in the United Statesassociated with fresh cantaloupe (7) cantaloupe consumptiondropped 53 nationwide (6) The prevention of produce contam-ination in production environments is therefore a concern forgrowers the produce industry and public health professionals Todevelop effective prevention strategies it is important to under-stand the ecological processes and environmental factors that af-fect foodborne pathogen prevalence in produce production envi-ronments Technological advancements such as geographicinformation systems (GIS) have the potential to drastically im-prove our ability to examine these processes and to develop noveltools for ensuring the safety of fresh produce

Numerous studies (8ndash21) have examined the ecology of food-borne pathogens in agricultural environments and several (22ndash27) have used GIS and geospatial analysis For example Chapin etal (26) used GIS to organize and extract remotely sensed data toshow that different species of Listeria occupy distinct ecologicalniches in agricultural and natural environments Despite a num-ber of studies that have used GIS to extract or visualize remotelysensed data (22ndash27) only one study (25) has used GIS to predictthe distribution and prevalence of a specific foodborne pathogenin produce production environments This study by Strawn et al(25) used classification tree analysis (CART) to develop a geospa-

tial model that predicts the prevalence of Listeria monocytogenes inNew York State (NYS) produce fields This model consisted of aset of hierarchical rules based on in order the proximity of thefields to surface water temperature the proximity of fields toimpervious cover available water storage (AWS) and the prox-imity of fields to pasture (25) Studies in other disease systems(eg Lyme disease and West Nile virus) have not only developed(28ndash34) but have also validated (35ndash40) geospatial predictive riskmodels These validation studies (35ndash40) demonstrate the utilityof geospatial risk models like the model developed by Strawn et al(25) to accurately and prospectively predict pathogen prevalenceAdditionally these studies (37 39 40) used the output of theirmodels to prioritize and identify risk management strategies sug-gesting that geospatial models can also be integrated with on-farmfood safety plans to develop targeted approaches to disease pre-vention Thus the purpose of this study was to (i) validate the

Received 22 September 2015 Accepted 12 November 2015

Accepted manuscript posted online 20 November 2015

Citation Weller D Shiwakoti S Bergholz P Grohn Y Wiedmann M Strawn LK2016 Validation of a previously developed geospatial model that predicts theprevalence of Listeria monocytogenes in New York State produce fieldsAppl Environ Microbiol 82797ndash 807 doi101128AEM03088-15

Editor D W Schaffner

Address correspondence to Laura K Strawn lstrawnvtedu

Supplemental material for this article may be found at httpdxdoiorg101128AEM03088-15

Copyright copy 2016 American Society for Microbiology All Rights Reserved

crossmark

February 2016 Volume 82 Number 3 aemasmorg 797Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

ability of the model developed by Strawn et al (25) to predicton-farm areas with a significantly higher or lower prevalence of Lmonocytogenes and (ii) identify additional land cover factors thatwere associated with L monocytogenes isolation from produceproduction environments This research also aimed to increaseour understanding of foodborne pathogen ecology and to developtargeted mitigation strategies for risk management in produceproduction environments (eg tailored on-farm food safety ap-proaches) While multiple pathogens can contaminate produce atthe production level we chose L monocytogenes as a model organ-ism to examine contamination at this level due to its high preva-lence in NYS produce production environments (11 22 23 25)We recognize that the model developed by Strawn et al (25) pre-dicts the prevalence of L monocytogenes however since the pres-ence of a Listeria sp is an indicator for L monocytogenes we alsotested the ability of the model to predict Listeria species preva-lence

MATERIALS AND METHODSStudy design A cross-sectional study was conducted over a 6-week periodin July and August of 2014 on four produce farms in NYS The farms werelocated in three regions of NYS western New York (n 2) the HudsonValley (n 1) and the Capital District (n 1) The farms were notselected based on geographic location or management practices and eachfarm was enrolled based on the willingness of the grower to participate

All fields within a farm were classified into four high-risk categoriesand one low-risk category (see Fig 1) based on a set of hierarchical rulesthat were adapted from the study by Strawn et al (25) The rules werebased on a fieldrsquos proximity to water impervious cover and pasture and afieldrsquos AWS (see Fig S1 in the supplemental material ldquoGeographic data

and predicting field riskrdquo for more information) All field areas classifiedinto a given category (eg areas within 375 m of water) were then dividedinto 5-by-5-m plots and a subset of plots was randomly selected fromeach category for sampling One drag swab was collected per plot Themethods used in this study were similar to those of Strawn et al (25) toavoid bias between studies However unlike Strawn et al (25) whose unitof analysis was the field and who collected drag swab composite soilwater and fecal samples we used the plot (ie subfield) as the unit ofanalysis in the study reported here and only drag swabs were collected

Geographic data and prediction of field risk All manipulations ofgeographic data were performed in ArcGIS (version 1022 Environmen-tal Systems Research Institute Redlands CA [41]) AWS data were ob-tained from the US Department of Agriculture (httpdatagatewaynrcsusdagovGDGOrderaspx) Land cover data for NYS for 2006 weredownloaded and extracted from the National Land Cover Database(NLCD) (httpwwwmrlcgovnlcd06_dataphp) Road data were down-loaded from the Cornell University Geographic Information Repository(cugirmannlibcornelledu) Hydrologic data were downloaded fromUS Geological Survey National Hydrography Map (httpviewernationalmapgovviewernhdhtmlpnhd) Maps of each farm were ob-tained from the grower uploaded into ArcGIS and georeferenced If theimage could not be accurately georeferenced a farm map was drawn inArcGIS by identifying field boundaries in satellite images using the origi-nal PDF file of the farm fields as a reference

The predicted field risk for L monocytogenes was based on a hierarchi-cal model developed by Strawn et al (25) using classification tree analysisBriefly we adapted that model by removing the meteorological factors sothe model included only spatial factors (ie proximity to water proximityto impervious cover AWS and proximity to pastures see Fig S1 in thesupplemental material) This adapted model is referred to as the CARTmodel throughout this article The CART model had four splitsrules

FIG 1 Map of predicted prevalence of L monocytogenes on the Homer C Thompson Vegetable Research Farm at Cornell University the expected prevalence ofL monocytogenes is listed in parentheses in the key Note that this map is not based on any of the farms included in this study for confidentiality reasons Mapcreated using ArcGIS software and the base map is from ArcGIS (ESRI [all rights reserved])

Weller et al

798 aemasmorg February 2016 Volume 82 Number 3Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

which in order are the water rule the impervious cover rule the AWSrule and the pasture rule (see Fig S1 in the supplemental material)

Before dividing each farm into areas of high or low predicted L mono-cytogenes prevalence the relevant shapefiles for each farm were generatedusing ArcGIS Hydrology shapefiles were buffered to 395 m road shape-files were buffered to 195 m and pasture shapefiles were buffered to 625m Roads and waterways were buffered by an additional 10 m and 2 mrespectively to give these features a realistic width Additionally the AWSdata were converted from raster to shapefile format The AWS shapefilewas then split into (i) areas with AWS of 42 cm and (ii) areas with AWS42 cm (ie high- and low-AWS areas respectively) The NLCD rasterwas also converted to shapefile format and split creating separate files foreach land cover class (eg pasture grasslands and woody wetlands) TheNLCD shapefiles for developed areas were merged with the road map tocreate an impervious-cover shapefile Similarly all NLCD shapefiles cor-responding to wetlands and forests were merged to create a single wet-lands shapefile and a single forest shapefile respectively

After creation of the relevant shapefiles each farm was categorizedinto areas of high or low predicted L monocytogenes prevalence accordingto the splits in the CART model (see Fig S1 in the supplemental material)For example the buffered hydrology shapefile corresponded to all areaswith a high predicted L monocytogenes prevalence according to the waterrule Similarly all areas that did not have a high predicted prevalenceaccording to the water rule but were included in the impervious-covershapefile corresponded to areas with a high predicted prevalence accord-ing to the impervious cover rule

To assess additional risk factors the distance was calculated from thecenter of each 5-by-5-m sampling plot to land covers of interest (iebarren land grassland forest impervious cover roads scrubland waterand wetlands) The split NLCD shapefiles were used to calculate the dis-tance to barren land grassland and scrubland Similarly the road andhydrology shapefiles were used to calculate the distance to roads and wa-ter Last the merged forest wetlands and impervious-cover shapefileswere used to calculate the distance to those features

Sample collection and preparation The samples were collected andprepared as previously described by Strawn et al (25) Briefly latex gloves(Nasco Fort Atkinson WI) were worn and changed for each sample col-lected For each plot a premoistened drag swab (30 ml of buffered Listeriaenrichment broth [Becton Dickinson Franklin Lakes NJ] in a sterileWhirl-Pak bag) was dragged around the perimeter and diagonals of theplot for 3 to 5 min All samples were transported on ice stored at 4degC andprocessed within 24 h of collection

Bacterial enrichment and isolation Listeria species and L monocyto-genes enrichment and isolation were performed as previously described(25) Briefly each sample was diluted 110 with buffered Listeria enrich-ment broth (Becton Dickinson) and then incubated at 30degC After 4 hListeria selective enrichment supplement (Oxoid Cambridge UnitedKingdom) was added to each enrichment After being incubated for 24and 48 h 50 l of each enrichment was streaked onto L monocytogenesplating medium (LMPM) (Biosynth International Itasca IL) and modi-fied Oxford agar (MOX) (Becton Dickinson) the plates were then incu-bated for 48 h at 35 and 30degC respectively Following incubation up tofour presumptive Listeria colonies were substreaked from MOX to LMPMand incubated at 35degC for 48 h From all LMPM plates up to four pre-sumptive Listeria colonies were then substreaked onto brain heart infu-sion (BHI) plates (Becton Dickinson) and then incubated at 37degC for 24 hThe species and sigB allelic type of one presumptive Listeria colony persample were determined by PCR amplification and sequencing of thepartial sigB gene as previously described (42ndash44)

Positive and negative controls were processed in parallel with the fieldsamples L monocytogenes FSL R3-0001 (45) and uninoculated enrich-ment medium were used as the positive and negative controls respec-tively All isolates were preserved at 80degC and the isolate informationcan be found on the Food Microbe Tracker website

Statistical analysis All statistical analyses were performed in R (ver-sion 31 R Core Team Vienna Austria) The frequency and prevalence ofL monocytogenes were calculated for each predicted risk area for each ruleAlthough the outcome of the CART model was the predicted prevalenceof L monocytogenes all statistical analyses were performed for both (i) Lmonocytogenes and (ii) Listeria spp (including L monocytogenes) sinceListeria spp are more common than L monocytogenes in NYS produceproduction environments and are often used as an index for L monocy-togenes

In order to test the ability of each rule to accurately predict the prev-alence of Listeria spp and L monocytogenes in produce fields and to vali-date the CART model multivariable logistic regression analyses were per-formed using the lme4 package (46) The multivariable model originallycontained all four rules but was reduced using backward selection Theoutcome for the multivariable model was the presence of Listeria spp or Lmonocytogenes Farm was included as a random effect

As the multivariable model used to validate the algorithm adaptedfrom Strawn et al (25) contained only four factors (ie AWS and prox-imity to surface water impervious cover and pasture) univariable gen-eralized linear mixed models (GLMM) (46) were developed to examinethe effect of additional land covers (ie proximity to barren land forestsgrassland roads scrubland and wetlands) on the likelihood of Listeriaspecies and L monocytogenes isolation Since the CART model was basedon a binary interpretation of AWS and proximity to water imperviouscover and pasture univariable GLMMs were also developed to examinethe relationship between these four factors as continuous variables andListeria species and L monocytogenes prevalence In this and all otherGLMMs performed for this study farm was included as a random effectand the outcome was the prevalence of Listeria spp or L monocytogenesAll factors that were significantly associated with the isolation of Listeriaspp or L monocytogenes were tested for correlation with all other factorsthat were found to be significant by univariable analysis

A multivariable GLMM was also developed de novo (ie not based onthe rules reported by Strawn et al [25]) to identify the most importantland cover factors associated with Listeria species and L monocytogenesisolation from drag swab samples Factors that were not correlated andwere significant by univariable analysis were considered candidate factorsfor inclusion in the multivariable model

Predictive models based on the GLMMs for L monocytogenes werethen applied in a GIS platform to generate predictive maps of L monocy-togenes prevalence at the subfield level for comparison with the map thatwas developed using the CART model (Fig 1) Predictive risk maps weredeveloped by inputting the univariable and multivariable GLMMs intoArcGIS The Homer C Thompson Vegetable Research Farm at CornellUniversity Ithaca NY was used to develop these maps to ensure theconfidentiality of the commercial growers enrolled in our study

RESULTS

The overall prevalence of Listeria spp and L monocytogenes fordrag swabs collected from NYS produce farms was 20 and 12respectively Overall Listeria spp (including L monocytogenes)were isolated from 20 (2081056) of the samples L monocyto-genes was isolated from 12 (1281056) of the samples Listeriainnocua was isolated from 40 (421056) of the samples Listeriaseeligeri was isolated from 20 (211056) of the samples andListeria welshimeri was isolated from 16 (171056) of the sam-ples

Overall the prevalence of Listeria spp was greater for all fieldareas with a high predicted prevalence of L monocytogenes isola-tion than that in field areas with a low predicted prevalence ofListeria species (Table 1 and Fig 2) For example the prevalence ofListeria species was 26 (51195) in samples collected from areaswith a high predicted prevalence according to the water rule and

Validation of Models To Predict L monocytogenes Risk

February 2016 Volume 82 Number 3 aemasmorg 799Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

18 (157861) in samples collected from areas with a low pre-dicted prevalence according to the water rule (Table 1 and Fig 2)

The prevalence of L monocytogenes was greater for all fieldareas with a high predicted prevalence of L monocytogenes isola-tion than that in field areas with a low predicted prevalence ac-cording to the water pasture and AWS rules (Table 1 and Fig 3)For example the prevalence of L monocytogenes was 22 (43195) in samples collected from areas with a high predicted preva-lence according to the water rule and 10 (85861) in samplescollected from areas with a low predicted prevalence according tothe water rule (Table 1 and Fig 3)

Rules based on surface water and pasture proximity accu-rately predict L monocytogenes prevalence in environmentalsamples collected from NYS produce production environmentsLogistic regression was performed to test the ability of each rule to

accurately predict L monocytogenes prevalence in NYS produceproduction environments Logistic regression analysis showedthat only the water and pasture rules accurately predicted theprevalence of L monocytogenes in NYS produce production envi-ronments (Table 2) Samples collected from field areas that had ahigh predicted prevalence of L monocytogenes isolation by thewater rule had an increased odds of L monocytogenes isolation(odds ratio [OR] 30 95 confidence interval [CI] 20 46) com-pared to that for samples collected from field areas that had a lowpredicted prevalence Samples collected from field areas that had ahigh predicted prevalence of L monocytogenes by the pasture rulehad an increased odds of L monocytogenes isolation (OR 29 95CI 14 60) compared to that for samples collected from fieldareas that had a low predicted prevalence

While the outcome of the CART model was L monocytogenes

TABLE 1 Frequency and prevalence of Listeria species-positive and L monocytogenes-positive samples for farm fields that had either a high or a lowpredicted risk of L monocytogenes isolation based on land cover factors

Rule

Description by predicted prevalence (no of samples) (n 1056) No of samples positive for (prevalence [])

High Low

Listeria spp (208 [20])a L monocytogenes (128 [12])

High predictedrisk

Low predictedrisk

High predictedrisk

Low predictedrisk

Water 375 m from water (195) 375 m from water (861) 51 (26) 157 (18) 43 (22) 85 (10)Road 9 m from roads (168) 9 m from roads (693) 36 (21) 121 (17) 11 (7) 74 (11)AWSb 42 cm3 AWS (106) 42 cm3 AWS (587) 23 (22) 98 (17) 20 (19) 54 (9)Pasture 625 m from pasture (49) 625 m from pasture (57) 12 (24) 11 (19) 11 (22) 9 (15)a Listeria spp include L monocytogenesb AWS available water storage

FIG 2 Frequency and prevalence of positive Listeria species samples for farm fields that had either a high or a low predicted prevalence of L monocytogenesisolation based on a hierarchical predictive risk model

Weller et al

800 aemasmorg February 2016 Volume 82 Number 3Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

prevalence the ability of the model to predict Listeria speciesprevalence was also validated because Listeria spp are more com-mon than L monocytogenes alone and as a result the findingsbased on Listeria spp are more robust Multivariable logisticregression showed that only the water rule was found to accu-rately predict the prevalence of Listeria spp in NYS produceproduction environments (Table 2) Samples collected from fieldareas that had a high predicted prevalence of L monocytogenes bythe water rule had an increased odds of Listeria species isolation(OR 16 95 CI 11 24) compared to that from samples col-lected from field areas that had a low predicted prevalence of Lis-teria species

Proximity to wetlands and scrublands was associated withaltered likelihood of L monocytogenes isolation from produceproduction environments in NYS As the multivariable modelused to validate the CART model (25) contained only four factorsGLMMs were developed to identify additional land cover factorsthat were associated with the isolation of L monocytogenes fromNYS produce production environments Of the nine land coverfactors that were evaluated six features (ie proximity to forestgrasslands pasture scrublands water and wetlands) were signif-icantly associated with L monocytogenes-positive samples by uni-variable analysis (Table 3) For example for a 100-m increase inthe distance of a sampling site from forests the likelihood of Lmonocytogenes isolation decreased by 14 (OR 086 95 CI074 10) Similarly for a 100-m increase in the distance of asampling site from surface water the likelihood of L monocyto-genes isolation decreased by 23 (OR 077 95 CI 066 090Fig 4)

To identify the most important land cover factors associatedwith L monocytogenes isolation from produce production envi-ronments a multivariable GLMM was developed The six factorsthat were found to be significant by univariable analysis were in-cluded as candidate factors In the final GLMM only three landcover features were retained (see Table S1 in the supplementalmaterial) and no significant interactions (ie P 005) wereobserved between any variables in the model For a 100-m increasein the distance of a sampling site from forests the likelihood of Lmonocytogenes isolation decreased by 13 (OR 087 95 CI076 099) For a 100-m increase in the distance of a sampling sitefrom scrubland the likelihood of L monocytogenes isolation de-creased by 6 (OR 094 95 CI 088 10) Last for a 100-mincrease in the distance of a sampling site from water the likeli-

FIG 3 Frequency and prevalence of positive L monocytogenes samples for farm fields that had either a high or a low predicted prevalence of L monocytogenesisolation based on a hierarchical predictive risk model

TABLE 2 Results of multivariable analyses built using backwardregression (ie only factors with P 005 were retained) that testedpreviously identified rules to accurately predict the effect of differentbinary land cover factors (eg either far away from or close to water) onthe likelihood of Listeria species and L monocytogenes isolation

Species by rule

Odds ratio forListeria species orL monocytogenes detection 95 CIa P value

Listeria sppb

Waterc 16 11 24 0008L monocytogenes

Pastured 29 14 60 0005Waterc 30 20 46 0001

a CI confidence intervalb Listeria spp include L monocytogenesc The water rule predicts a high prevalence of L monocytogenes for areas within 375 mof surface water and a low prevalence for areas 375 m from surface waterd The pasture rule predicts a high prevalence of L monocytogenes for areas within 625m of pasture and a low prevalence for areas 625 m from surface water

Validation of Models To Predict L monocytogenes Risk

February 2016 Volume 82 Number 3 aemasmorg 801Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

hood of L monocytogenes isolation decreased by 15 (OR 08595 CI 076 095)

Predictive risk maps (Fig 5) were then developed using theunivariable and multivariable GLMMs for L monocytogenes de-scribed above (Table 3 see also Table S1 in the supplemental ma-terial) The maps were developed to allow for comparisons withthe map based on the CART model (Fig 1) and as a proof of aconcept to assess if the multivariable GLMM for L monocytogenescould be used to predict L monocytogenes prevalence at the sub-field level This map shows that multivariable GLMM can be usedto generate a map of L monocytogenes prevalence and that thismap is at a finer scale than that of maps based on CART analyses

Proximity to forests and scrublands was associated with anincreased likelihood of Listeria species isolation from produceproduction environments in NYS Similar to L monocytogenesGLMMs were also developed to identify additional land coverfactors that were associated with the isolation of Listeria spp fromNYS produce production environments Of the nine land coverfactors that were evaluated five features (ie proximity to forestpasture scrublands water and wetlands) were significantly asso-

TABLE 3 Results of univariable analyses that tested the effect ofdifferent land cover factors treated as continuous variables on thelikelihood of Listeria species and L monocytogenes isolation

Proximity byland cover factor Odds ratioa 95 CIb P value

Listeria sppc

Forest 084 074 095 0009Pasture 092 083 10 0117Scrubland 093 088 099 0044Water 085 076 095 0005Wetlands 093 086 10 0058

L monocytogenesForest 086 074 10 0060Grassland 104 099 11 0104Pasture 092 081 10 0148Scrubland 088 081 095 0002Water 077 066 090 0001Wetlands 092 084 10 0088

a For a 100-m increase in the distance of a given sampling point from the given landcover factorsb CI confidence intervalc Listeria spp include L monocytogenes

FIG 4 True prevalence (bars) and predicted prevalence of Listeria species-positive samples (A) and L monocytogenes-positive samples (B) (line) based on mixedmodels that included proximity to water as a risk factor True prevalence was calculated for 50-m bins (eg all samples that were between 0 and 50 m from waterwent into the first bin) the sample size for each bin is noted at the bottom of each column Among five samples collected 650 m away from water two wereListeria species positive and none were L monocytogenes positive Prevalence is reported as a decimal

Weller et al

802 aemasmorg February 2016 Volume 82 Number 3Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

ciated with Listeria-positive samples by univariable analysis (Ta-ble 3) For example for a 100-m increase in the distance of asampling site from forests the likelihood of Listeria species isola-tion decreased by 16 (OR 084 95 CI 074 095) Similarlyfor a 100-m increase in the distance of a sampling site from surfacewater the likelihood of Listeria species isolation decreased by 15(OR 085 95 CI 076 095 Fig 4) No strong correlations (iecorrelation coefficient of 05) were observed between any of thesignificant factors by univariable analysis

To identify the most important land cover factors associatedwith Listeria species isolation from produce production environ-ments a multivariable GLMM was developed The five factors thatwere found to be significant by univariable analysis were includedas candidate factors In the final GLMM only three land cover

factors were retained (see Table S1 in the supplemental material)and no significant interactions were observed between the vari-ables in the final model For a 100-m increase in the distance of asampling site from scrubland the likelihood of Listeria speciesisolation decreased by 14 (OR 086 95 CI 079 093) For a100-m increase in the distance of a sampling site from water thelikelihood of Listeria species isolation decreased by 24 (OR 07695 CI 065 089) Last for a 100-m increase in the distance of asampling site from wetlands the likelihood of Listeria species iso-lation decreased by 9 (OR 091 95 CI 083 099)

DISCUSSION

The primary objectives of this study were (i) to validate previouslydeveloped geospatial rules that predicted areas of significantly

FIG 5 Map of predicted prevalence of L monocytogenes for the Homer C Thompson Vegetable Research Farm at Cornell University based on the results of (i)univariable generalized linear mixed models in which proximities to scrubland (A) water (B) and wetlands (C) were included as risk factors and (ii) amultivariable generalized linear mixed model in which proximities to scrubland water and wetlands were included as risk factors (D) Note that this map is notbased on any of the farms included in this study for confidentiality reasons Maps were created using ArcGIS software and base maps are from ArcGIS (ESRI [allrights reserved])

Validation of Models To Predict L monocytogenes Risk

February 2016 Volume 82 Number 3 aemasmorg 803Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

higher or lower prevalence of L monocytogenes and (ii) to identifyadditional land cover factors that may be associated with an in-creased or decreased likelihood of L monocytogenes isolation inproduce production environments Our study validated two ofthe four rules (ie the water and pasture rules) that comprised theCART model (25) Additionally among land cover factors thatwere not included in the original CART model but were testedhere proximity to scrubland and proximity to wetlands werefound to be significantly associated with an increased likelihood ofL monocytogenes isolation These findings suggest that on-farmproduce safety is complicated by the ecological context unique toeach field and by the scale (eg the farm field and subfield levels)at which prevalence is assessed Thus it is essential to have toolsthat allow growers to account for both ecological context and scalewhen developing on-farm produce safety plans The validation ofthe water and pasture rules in this study demonstrates the appli-cation of geospatial models for prospective and accurate predic-tion of pathogen prevalence on produce farms suggesting thatGIS is a promising tool for food safety

Geospatial models have the ability to accurately predict thelikelihood of L monocytogenes isolation from produce produc-tion environments In this study proximity to surface water andproximity to pasture were significantly associated with L mono-cytogenes isolation from produce production environments by lo-gistic regression These findings validated two of the four rulesfrom the CART model adapted from the study by Strawn et al(25) These findings were also consistent with other studies con-ducted on L monocytogenes in NYS agricultural environments(22 23 26) and on L monocytogenes and other foodborne patho-gens in agricultural and nonagricultural environments (19 47ndash50) For example in a Canadian study Lyautey et al (47) foundthat proximity to dairy operations was one of the most importantpredictors of L monocytogenes-positive surface water samplesThe repeated identification of an association between L monocy-togenes isolation and proximity to water pasture and other live-stock-associated areas suggests that our findings are translatableto other farms in NYS In our study reported here proximity towater and proximity to pasture were significantly associated withL monocytogenes isolation by GLMM and logistic regression fur-ther supporting the robustness of this association By validatingtwo of the rules adapted from the CART model our study dem-onstrates that geospatial models can be used to accurately andprospectively predict the prevalence of L monocytogenes in pro-duce production environments

Interestingly while our findings were generally consistent withthe previously reported CART model (25) neither the AWS northe impervious cover rules were validated by our findings Thismay be the result of small differences in sampling protocols be-tween the study by Strawn et al (25) and the study reported hereStrawn et al (25) used drag swab composite soil fecal and watersamples in their analyses while in the study reported here onlydrag swab samples were collected As each sample type likely rep-resents a unique L monocytogenes population from a distinct eco-logical niche (eg water versus soil) it seems plausible that differ-ent factors would be associated with the isolation of L monocytogenesin each study Therefore the fact that the AWS and imperviouscover rules were not validated may indicate that these rules areassociated with L monocytogenes isolation from one of the sampletypes that were collected by Strawn et al (25) but not in the studyhere (eg water samples) Future studies that investigate geospa-

tial factors associated with contamination risk for actual produce(ie not environmental samples) are thus needed to increase theaccuracy of predictive models and allow growers to maximize sur-veillance efforts However these studies will require considerablylarger sample sizes as pathogen prevalence on produce tends to besignificantly lower than that in environmental samples (22) Alsoin the study reported here more samples were collected fromareas at low predicted risk than from areas at high predicted riskthis was due to the fact that samples were collected in commercialsettings Future studies should aim to collect comparable samplesizes from high- and low-risk areas as well

Identification of additional factors (eg proximity to wet-lands) that were not included in the original CART model butwere found to be associated with the prevalence of L monocyto-genes in produce production environments may aid in the refine-ment of prediction models Importantly these same factors havealso been identified as risk factors for Listeria and L monocytogenescontamination in past studies of natural (26) and agricultural (2326) environments However while the study reported here did notfind any significant interactions between the different landscapefactors studied a previous report did find that interactions be-tween landscape and meteorological factors significantly affectedthe probability of isolating Listeria spp from soil vegetation andwater (24) Similarly previous studies (9 11 19 49 51ndash53) foundthat management practices were significantly associated with thelikelihood of isolating L monocytogenes from on-farm environ-ments Management practices and meteorological factors whichwere not considered in the study reported here may thus affect therelationships between L monocytogenes prevalence and landscapefactors Further improvement of geospatial models may thereforebe achieved by integration of additional environmental (bothlandscape and meteorological) and management practice dataWhile the development of such models would require larger datasets these models might account for temporal (eg changes inmanagement practices or meteorological factors over time) andspatial variation and would thus facilitate the identification ofadditional risk factors and control strategies

Issues of scale need to be considered when developing andvalidating geospatial models for preharvest produce safety as-sessment Despite the fact that the pasture rule was validated bylogistic regression proximity to pasture was not retained in thefinal multivariable GLMM This difference may be a function ofscale which is defined by the resolution (ie grain) and extent ofthe available spatial data Numerous studies (54ndash62) have foundthat changing the study scale changes the strength of associationsand interactions For example in a study on habitat use by Eleodeshispilabris McIntyre (62) found that E hispilabris avoided shrubsat small scales but selectively occupied shrubland at larger scaleswhich may be due to different mechanisms influencing habitatselection at the different scales Thus studies that look at similaroutcomes (eg L monocytogenes prevalence) at different scales(eg field and subfield levels) may identify different predictorvariables The issue of scale is complicated by the grain and accu-racy of the remotely sensed data available particularly if the scaleof the input data differs from the scale of the model (63) Forexample while the 2006 NLCD has a national accuracy of 78(64) the odds of misclassification increase as landscape heteroge-neity increases (65) Therefore in highly mosaic environmentssuch as produce farms NLCD accuracy is lower This may alsoexplain why proximity to pasture was not retained in the final

Weller et al

804 aemasmorg February 2016 Volume 82 Number 3Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

GLMM particularly since misclassification of grass-dominatedlandscapes such as pasture accounted for 26 of all inaccuracies(64) It is therefore important that researchers are cognizant of thelimitations associated with the use of remotely sensed data to de-velop geospatial predictive risk models On the other hand theselimitations can be minimized by carefully designing studies andusing appropriate analyses that account for scale (54 63 66) Inaddition improved data collection strategies (eg using drones)could be used to address these issues in the future Despite differ-ences in study scale it is important to note that proximity to pas-ture was significantly associated with L monocytogenes prevalenceby univariable GLMM which does support the validation of thepasture rule by logistic regression

Ecological and food safety implications of edge interactionson farm landscapes In the present work edge interactions be-tween produce farms and four other land cover types (ie forestscrubland water and wetland) were observed The elevated prev-alence of L monocytogenes in ecotones (ie the transitional areawhere two ecological communities meet) is consistent with pat-terns observed in other disease systems (eg Lyme disease [67ndash70]) This is also consistent with our current understanding ofinfectious disease emergence as they frequently arise at the inter-face between human habitats and other ecosystems (67ndash71) Eco-tones are most abundant in fragmented landscapes and theirpresence intensifies ecological processes For example ecotonesare often more diverse than surrounding communities (69 72 73)and provide an ideal habitat for ldquoedge speciesrdquo (eg ticks androdents [69]) Additionally ecotones and the associated habitatfragmentation affect the nature and rate of species interaction(eg intensifying competition [69 74]) In this context our re-sults suggest that food grown within short distances of ecotonesspecifically the boundaries between farm fields and forests waterscrublands or wetlands is at an increased risk for L monocyto-genes contamination Thus risk management plans need to con-sider the potential for increased preharvest food safety risks asso-ciated with produce grown in or near ecotones For examplegrowers could create buffer zones of unharvested product near theedges of fields increase surveillance andor decontamination ofproduce grown near field edges or stage harvesting and process-ing so that higher-risk material (ie produce grown near fieldedges) is harvested and processed last These concerns are partic-ularly pertinent for small farms that have a larger ratio of ecotoneto field area thus future studies should account for farm sizewhen developing and validating on-farm intervention strategies

Predictive risk maps based on GIS-enabled models allow forthe visualization of preharvest food safety risk at multiplescales The CART model predicted prevalence at the field levelwhile the GLMMs developed in the study reported here predictedL monocytogenes prevalence at every point within a field (ie atthe subfield level) Thus the CART model generated a map ofdiscrete areas of high and low predicted prevalence (Fig 1) whilethe GLMMs produced a risk gradient map (Fig 5) As previouslymentioned different mechanisms drive ecological processes atdifferent scales so the factors that are significantly associated withL monocytogenes isolation at the field and subfield levels may dif-fer Therefore the model and predictive map that are most appro-priate for use by the grower depend on the scale of their risk man-agement plan (ie farm field or subfield level) In general mapsbased on the GLMM are more appropriate as those maps offergreater resolution than CART models which allows for the devel-

opment of more targeted mitigation strategies However the abil-ity to develop both map types demonstrates the flexibility ofgeospatial tools and the utility of GIS for visualizing the output ofdifferent model types Overall GIS offers a unique opportunity tolook at variation across scales and to account for cross-scale dif-ferences in predictive models by allowing for the integration andvisualization of remotely sensed and field-collected data

Conclusions This study yielded quantitative data that showedthat L monocytogenes contamination on produce farms is depen-dent on the specific ecological context of a produce farm and thatgeospatial predictive risk maps can be used to accurately and pro-spectively predict L monocytogenes prevalence in NYS produceproduction environments Additionally other land cover factorswere identified that should be examined in future studies to de-velop higher-resolution models The implementation of geospa-tial predictive models by the produce industry may increase theunderstanding of risk factors that promote foodborne pathogenprevalence and persistence in produce fields and it will assistgrowers in focusing their food safety efforts Geospatial modelsallow for the development of individualized preventive measureson produce farms as they enable growers to proactively assess andaddress environmental factors that may increase the risk of con-tamination events on their specific farms For example predictiverisk maps can identify areas of high predicted pathogen prevalencewithin farms and enable growers to make more informed deci-sions about the management of crops in these areas includingtargeted pathogen surveillance programs and altered manage-ment practices Thus geospatial predictive risk models and mapshave a promising future in preharvest food safety as they can beapplied to any location and utilize the unique combination oflandscape characteristics (eg proximity to domestic animal op-erations) soil properties (eg available water storage) and cli-mate (eg precipitation) of a farm in the prediction process

ACKNOWLEDGMENTS

This work was supported by the Center for Produce SafetyWe thank Maureen Gunderson and Sherry Roof for their technical

assistance and Erika Mudrak David Kent Saurabh Mehta Julia Finkel-stein Francoise Vermeylen and Sadie Ryan for their statistical supportWe also thank Randy Worobo and Betsy Bihn for helping us enroll grow-ers in the study

FUNDING INFORMATIONCenter for Produce Safety (CPS) provided funding to Peter Bergholz andMartin Wiedmann under grant number 201400970-01

REFERENCES1 DeWaal C Glassman M 2014 Outbreak alert 2014 a review of food-

borne illness in America from 2002ndash2011 Center for Science in the PublicInterest Washington DC httpcspinetorgreportsoutbreakalert2014pdf

2 Painter JA Hoekstra RM Ayers T Tauxe RV Braden CR Angulo FJGriffin PM 2013 Attribution of foodborne illnesses hospitalizationsand deaths to food commodities by using outbreak data United States1998 ndash2008 Emerg Infect Dis 19407ndash 415 httpdxdoiorg103201eid1903111866

3 Bean N Griffin P Goulding J Ivey C 1990 Foodborne disease out-breaks 5-year summary 1983ndash1987 MMWR Surveill Summ 39(SS-01)15ndash23

4 Dillaway R Messer K Bernard J Kaiser H 2011 Do consumer re-sponses to media food safety information last Appl Econ Perspect Policy33363ndash383 httpdxdoiorg101093aeppppr019

5 Peake WO Detre JD Carlson CC 2014 One bad apple spoils the bunch

Validation of Models To Predict L monocytogenes Risk

February 2016 Volume 82 Number 3 aemasmorg 805Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

An exploration of broad consumption changes in response to food recallsFood Policy 4913ndash22 httpdxdoiorg101016jfoodpol201406006

6 Bailin D 2013 Killer cantaloupes ignoring the science behind foodsafety Union of Concerned Scientists Cambridge MA httpwwwucsusaorgsitesdefaultfileslegacyassetsdocumentscenter-for-science-and-democracykiller-cantaloupes-study-2013-fullpdf

7 US Food and Drug Administration 2013 Environmental assessmentfactors potentially contributing to the contamination of fresh whole can-taloupe implicated in a multi-state outbreak of salmonellosis US Foodand Drug Administration Silver Spring MD httpwwwfdagovFoodRecallsOutbreaksEmergenciesOutbreaksucm341476htm

8 Park S Szonyi B Gautam R Nightingale K Anciso J Ivanek R 2012Risk factors for microbial contamination in fruits and vegetables at thepreharvest level a systematic review J Food Prot 752055ndash2081 httpdxdoiorg1043150362-028XJFP-12-160

9 Park S Navratil S Gregory A Bauer A Srinath I Szonyi B NightingaleK Anciso J Jun M Han D Lawhon S Ivanek R 2014 Farm manage-ment environment and weather factors jointly affect the probability ofspinach contamination by generic Escherichia coli at the preharvest stageAppl Environ Microbiol 802504 ndash2515 httpdxdoiorg101128AEM03643-13

10 Nightingale KK Schukken YH Nightingale CR Fortes ED Ho AJ HerZ Grohn YT McDonough PL Wiedmann M 2004 Ecology and trans-mission of Listeria monocytogenes infecting ruminants and in the farmenvironment Appl Environ Microbiol 704458 ndash 4467 httpdxdoiorg101128AEM7084458-44672004

11 Strawn LK Groumlhn YT Warchocki S Worobo RW Bihn EA Wied-mann M 2013 Risk factors associated with Salmonella and Listeria mono-cytogenes contamination of produce fields Appl Environ Microbiol 797618 ndash7627 httpdxdoiorg101128AEM02831-13

12 Liu C Hofstra N Franz E 2013 Impacts of climate change on themicrobial safety of pre-harvest leafy green vegetables as indicated by Esch-erichia coli O157 and Salmonella spp Int J Food Microbiol 163119 ndash128httpdxdoiorg101016jijfoodmicro201302026

13 Geacuteneacutereux M Grenier M Cocircteacute C 2015 Persistence of Escherichia colifollowing irrigation of strawberry grown under four production systemsfield experiment Food Control 47103ndash107 httpdxdoiorg101016jfoodcont201406037

14 Chitarra W Decastelli L Garibaldi A Gullino ML 2014 Potentialuptake of Escherichia coli O157H7 and Listeria monocytogenes fromgrowth substrate into leaves of salad plants and basil grown in soil irrigatedwith contaminated water Int J Food Microbiol 189139 ndash145 httpdxdoiorg101016jijfoodmicro201408003

15 Vivant A-L Garmyn D Maron P-A Nowak V Piveteau P 2013Microbial diversity and structure are drivers of the biological barrier effectagainst Listeria monocytogenes in soil PLoS One 8e76991 httpdxdoiorg101371journalpone0076991

16 Omac B Moreira RG Castillo A Castell-Perez ME 2015 Growth ofListeria monocytogenes and Listeria innocua on fresh baby spinach leaveseffect of storage temperature and natural microflora Postharvest BiolTechnol 10041ndash51 httpdxdoiorg101016jpostharvbio201409007

17 Castro-Ibaacutentildeez I Gil MI Tudela JA Allende A 2014 Microbial safetyconsiderations of flooding in primary production of leafy greens a casestudy Food Res Int 6862ndash 69 httpdxdoiorg101016jfoodres201405065

18 Kilonzo C Li X Vivas EJ Jay-Russell MT Fernandez KL Atwill ER2013 Fecal shedding of zoonotic food-borne pathogens by wild rodents ina major agricultural region of the central California coast Appl EnvironMicrobiol 796337ndash 6344 httpdxdoiorg101128AEM01503-13

19 Benjamin L Atwill ER Jay-Russell M Cooley M Carychao D GorskiL Mandrell RE 2013 Occurrence of generic Escherichia coli E coli O157and Salmonella spp in water and sediment from leafy green produce farmsand streams on the central California coast Int J Food Microbiol 16565ndash76 httpdxdoiorg101016jijfoodmicro201304003

20 Jay-Russell MT Hake AF Bengson Y Thiptara A Nguyen T 2014Prevalence and characterization of Escherichia coli and Salmonella strainsisolated from stray dog and coyote feces in a major leafy greens productionregion at the United States-Mexico border PLoS One 9e113433 httpdxdoiorg101371journalpone0113433

21 Linke K Ruumlckerl I Brugger K Karpiskova R Walland J Muri-KlingerS Tichy A Wagner M Stessl B 2014 Reservoirs of Listeria species inthree environmental ecosystems Appl Environ Microbiol 805583ndash5592httpdxdoiorg101128AEM01018-14

22 Weller D Wiedmann M Strawn L 2015 Spatial and temporal factorsassociated with an increased prevalence of Listeria monocytogenes in spin-ach fields in New York State Appl Environ Microbiol 816059 ndash 6069 httpdxdoiorg101128AEM01286-15

23 Weller D Wiedmann M Strawn LK 2015 Irrigation is significantlyassociated with an increased prevalence of Listeria monocytogenes in pro-duce production environments in New York State J Food Prot 781132ndash1141 httpdxdoiorg1043150362-028XJFP-14-584

24 Ivanek R Groumlhn YT Wells MT Lembo AJ Jr Sauders BD WiedmannM 2009 Modeling of spatially referenced environmental and meteoro-logical factors influencing the probability of Listeria species isolation fromnatural environments Appl Environ Microbiol 755893ndash5909 httpdxdoiorg101128AEM02757-08

25 Strawn LK Fortes ED Bihn EA Nightingale KK Groumlhn YT WoroboRW Wiedmann M Bergholz PW 2013 Landscape and meteorologicalfactors affecting prevalence of three food-borne pathogens in fruit andvegetable farms Appl Environ Microbiol 79588 ndash 600 httpdxdoiorg101128AEM02491-12

26 Chapin TK Nightingale KK Worobo RW Wiedmann M Strawn LK2014 Geographical and meteorological factors associated with isolation ofListeria species in New York State produce production and natural envi-ronments J Food Prot 771919 ndash1928 httpdxdoiorg1043150362-028XJFP-14-132

27 Sauders BD Overdevest J Fortes E Windham K Schukken Y LemboA Wiedmann M 2012 Diversity of Listeria species in urban and naturalenvironments Appl Environ Microbiol 784420 ndash 4433 httpdxdoiorg101128AEM00282-12

28 Eisen L Lozano-Fuentes S 2009 Use of mapping and spatial and space-time modeling approaches in operational control of Aedes aegypti anddengue PLoS Negl Trop Dis 3e411 httpdxdoiorg101371journalpntd0000411

29 Sabesan S Raju HKK Srividya A Das PK 2006 Delimitation of lym-phatic filariasis transmission risk areas a geo-environmental approachFilaria J 512 httpdxdoiorg1011861475-2883-5-12

30 Machado-Machado EA 2012 Empirical mapping of suitability to denguefever in Mexico using species distribution modeling Appl Geogr 3382ndash93 httpdxdoiorg101016japgeog201106011

31 Lobitz B Beck L Huq A Wood B Fuchs G Faruque AS Colwell R2000 Climate and infectious disease use of remote sensing for detectionof Vibrio cholerae by indirect measurement Proc Natl Acad Sci U S A971438 ndash1443 httpdxdoiorg101073pnas9741438

32 Kolivras KN 2006 Mosquito habitat and dengue risk potential in Hawaiia conceptual framework and GIS application Prof Geogr 58139 ndash154httpdxdoiorg101111j1467-9272200600521x

33 Ekpo UF Mafiana CF Adeofun CO Solarin ART Idowu AB 2008Geographical information system and predictive risk maps of urinaryschistosomiasis in Ogun State Nigeria BMC Infect Dis 874 httpdxdoiorg1011861471-2334-8-74

34 Raso G Matthys B N=Goran EK Tanner M Vounatsou P Utzinger J2005 Spatial risk prediction and mapping of Schistosoma mansoni infec-tions among schoolchildren living in western Cocircte drsquoIvoire Parasitology13197ndash108 httpdxdoiorg101017S0031182005007432

35 Larson SR DeGroote JP Bartholomay LC Sugumaran R 2010 Eco-logical niche modeling of potential West Nile virus vector mosquito spe-cies in Iowa J Insect Sci 10110 httpdxdoiorg10167303101011001

36 Slater H Michael E 2013 Mapping Bayesian geostatistical analysis andspatial prediction of lymphatic filariasis prevalence in Africa PLoS One8e71574 httpdxdoiorg101371journalpone0071574

37 Sabesan S Raju KH Subramanian S Srivastava PK Jambulingam P2013 Lymphatic filariasis transmission risk map of India Vector BorneZoonotic Dis 13657ndash 665 httpdxdoiorg101089vbz20121238

38 Diuk-Wasser MA Vourcrsquoh G Cislo P Hoen AG Melton F Hamer SARowland M Cortinas R Hickling GJ Tsao JI Barbour AG Kitron UPiesman J Fish D 2010 Field and climate-based model for predicting thedensity of host-seeking nymphal Ixodes scapularis an important vector oftick-borne disease agents in the eastern United States Glob Ecol Biogeogr19504 ndash514httpdxdoiorg101111j1466-8238201000526x

39 Pullan RL Gething PW Smith JL Mwandawiro CS Sturrock HJGitonga CW Hay SI Brooker S 2011 Spatial modelling of soil-transmitted helminth infections in Kenya a disease control planning toolPLoS Negl Trop Dis 5e958 httpdxdoiorg101371journalpntd0000958

40 Bhunia GS Chatterjee N Kumar V Siddiqui NA Mandal R Das P

Weller et al

806 aemasmorg February 2016 Volume 82 Number 3Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from

Kesari S 2012 Delimitation of kala-azar risk areas in the district ofVaishali in Bihar (India) using a geo-environmental approach Mem InstOswaldo Cruz 107609 ndash 620 httpdxdoiorg101590S0074-02762012000500007

41 ESRI 2014 ArcGIS Desktop release 1022 Environmental Systems Re-search Institute Redlands CA

42 Bundrant BN Hutchins T den Bakker HC Fortes E Wiedmann M2011 Listeriosis outbreak in dairy cattle caused by an unusual Listeriamonocytogenes serotype 4b strain J Vet Diagn Invest 23155ndash158 httpdxdoiorg101177104063871102300130

43 den Bakker HC Bundrant BN Fortes ED Orsi RH Wiedmann M2010 A population genetics-based and phylogenetic approach to under-standing the evolution of virulence in the genus Listeria Appl EnvironMicrobiol 766085ndash 6100 httpdxdoiorg101128AEM00447-10

44 Nightingale KK Windham K Wiedmann M 2005 Evolution and mo-lecular phylogeny of Listeria monocytogenes isolated from human and an-imal listeriosis cases and foods J Bacteriol 1875537ndash5551 httpdxdoiorg101128JB187165537-55512005

45 Roberts AJ Wiedmann M 2006 Allelic exchange and site-directed mu-tagenesis probe the contribution of ActA amino-acid variability to phos-phorylation and virulence-associated phenotypes among Listeria monocy-togenes strains FEMS Microbiol Lett 254300 ndash307 httpdxdoiorg101111j1574-6968200500041x

46 Bates D Maechler M Bolker B Walker S 2015 Fitting linear mixed-effects models using lme4 J Stat Softw 671ndash 48

47 Lyautey E Lapen DR Wilkes G McCleary K Pagotto F Tyler KHartmann A Piveteau P Rieu A Robertson WJ Medeiros DT EdgeTA Gannon V Topp E 2007 Distribution and characteristics of Listeriamonocytogenes isolates from surface waters of the South Nation River wa-tershed Ontario Canada Appl Environ Microbiol 735401ndash5410 httpdxdoiorg101128AEM00354-07

48 Haley BJ Cole DJ Lipp EK 2009 Distribution diversity and seasonalityof waterborne salmonellae in a rural watershed Appl Environ Microbiol751248 ndash1255 httpdxdoiorg101128AEM01648-08

49 Park S Navratil S Gregory A Bauer A Srinath I Jun M Szonyi BNightingale K Anciso J Ivanek R 2013 Generic Escherichia coli con-tamination of spinach at the preharvest stage effects of farm managementand environmental factors Appl Environ Microbiol 794347ndash 4358 httpdxdoiorg101128AEM00474-13

50 Wilkes G Edge TA Gannon VP Jokinen C Lyautey E Neumann NFRuecker N Scott A Sunohara M Topp E Lapen DR 2011 Associationsamong pathogenic bacteria parasites and environmental and land usefactors in multiple mixed-use watersheds Water Res 455807ndash5825 httpdxdoiorg101016jwatres201106021

51 Beuchat LR Ryu JH 1997 Produce handling and processing practicesEmerg Infect Dis 3459 ndash 465 httpdxdoiorg103201eid0304970407

52 Johannessen GS Loncarevic S Kruse H 2002 Bacteriological analysis offresh produce in Norway Int J Food Microbiol 77199 ndash204 httpdxdoiorg101016S0168-1605(02)00051-X

53 Oliveira M Usall J Vintildeas I Solsona C Abadias M 2011 Transfer ofListeria innocua from contaminated compost and irrigation water to let-tuce leaves Food Microbiol 28590 ndash596 httpdxdoiorg101016jfm201011004

54 Levin SA 1992 The problem of pattern and scale in ecology Ecology731943ndash1967 httpdxdoiorg1023071941447

55 Krummel JR Gardner RH Sugihara G OrsquoNeill RV Coleman PR 1987Landscape patterns in a disturbed environment Oikos 48321ndash324 httpdxdoiorg1023073565520

56 Openshaw S Taylor P 1979 A million or so correlation coefficients p127ndash144 In Wrigley N (ed) Statistical methods in the spatial sciencesPion London United Kingdom

57 Thompson CM McGarigal K 2002 The influence of research scale on

bald eagle habitat selection along the lower Hudson River New York(USA) Landsc Ecol 17569 ndash586 httpdxdoiorg101023A1021501231182

58 Wu J Shen W Sun W Tueller PT 2002 Empirical patterns of the effectsof changing scale on landscape metrics Landsc Ecol 17761ndash782 httpdxdoiorg101023A1022995922992

59 Gehlke CE Biehl K 1934 Certain effects of grouping upon the size of thecorrelation coefficient in census tract material J Am Stat Assoc 29169 ndash170 httpdxdoiorg10108001621459193410506247

60 Holland JD Bert DG Fahrig L 2004 Determining the spatial scale ofspeciesrsquo response to habitat Bioscience 54227 httpdxdoiorg1016410006-3568(2004)054[0227DTSSOS]20CO2

61 Sherry TW Holmes RT 1988 Habitat selection by breeding Americanredstarts in response to a dominant competitor the least flycatcher Auk105350 ndash364 httpdxdoiorg1023074087501

62 McIntyre NE 1997 Scale-dependent habitat selection by the darklingbeetle Eleodes hispilabris (Coleoptera Tenebrionidae) Am Midl Nat 138230 ndash235 httpdxdoiorg1023072426671

63 OrsquoSullivan D Unwin D 2002 Geographic information analysis 2nd edJohn Wiley amp Sons Hoboken NJ

64 Wickham JD Stehman SV Gass L Dewitz J Fry JA Wade TG 2013Accuracy assessment of NLCD 2006 land cover and impervious surfaceRemote Sens Environ 130294 ndash304 httpdxdoiorg101016jrse201212001

65 Smith JH Stehman SV Wickham JD Yang L 2003 Effects of landscapecharacteristics on land-cover class accuracy Remote Sens Environ 84342ndash349 httpdxdoiorg101016S0034-4257(02)00126-8

66 MacEachren AM Robinson A Hopper S Gardner S Murray R Gahe-gan M Hetzler E 2005 Visualizing geospatial information uncertaintywhat we know and what we need to know Cartogr Geogr Infect Sci 32139 ndash160 httpdxdoiorg1015591523040054738936

67 Despommier D Ellis BR Wilcox BA 2006 The role of ecotones inemerging infectious diseases Ecohealth 3281ndash289 httpdxdoiorg101007s10393-006-0063-3

68 Halos L Bord S Cotteacute V Gasqui P Abrial D Barnouin J Boulouis HJVayssier-Taussat M Vourcrsquoh G 2010 Ecological factors characterizingthe prevalence of bacterial tick-borne pathogens in Ixodes ricinus ticks inpastures and woodlands Appl Environ Microbiol 764413ndash 4420 httpdxdoiorg101128AEM00610-10

69 Lambin EF Tran A Vanwambeke SO Linard C Soti V 2010 Patho-genic landscapes interactions between land people disease vectors andtheir animal hosts Int J Health Geogr 954 httpdxdoiorg1011861476-072X-9-54

70 McFarlane RA Sleigh AC McMichael AJ 2013 Land-use change andemerging infectious disease on an island continent Int J Environ ResPublic Health 102699 ndash2719 httpdxdoiorg103390ijerph10072699

71 Jones BA Grace D Kock R Alonso S Rushton J Said MY McKeeverD Mutua F Young J McDermott J Pfeiffer DU 2013 Zoonosisemergence linked to agricultural intensification and environmentalchange Proc Natl Acad Sci U S A 1108399 ndash 8404 httpdxdoiorg101073pnas1208059110

72 Smith TB Wayne RK Girman D Bruford MW 2005 Evaluating thedivergence-with-gene-flow model in natural populations the importanceof ecotones in rainforest speciation p 148 ndash165 In Bermingham E DickCW Moritz C (ed) Tropical rainforests past present and future TheUniversity of Chicago Press Chicago IL

73 Brooks RA Bell SS 2001 Mobile corridors in marine landscapes en-hancement of faunal exchange at seagrasssand ecotones J Exp Mar BioEcol 26467ndash 84 httpdxdoiorg101016S0022-0981(01)00310-0

74 Roland J 1993 Large-scale forest fragmentation increases the duration oftent caterpillar outbreak Oecologia 9325ndash30 httpdxdoiorg101007BF00321186

Validation of Models To Predict L monocytogenes Risk

February 2016 Volume 82 Number 3 aemasmorg 807Applied and Environmental Microbiology

on Septem

ber 22 2020 by guesthttpaem

asmorg

Dow

nloaded from