Potential Risk Levels of Invasive Fire Blight in Apple Orchards Around the World

Fire blight, a disease of apple trees caused by Erwinia amylovora, occurs worldwide except in South America, South-Central Africa, and most of Oceania. Ecological niche models can determine the potential distribution of species and measure the risk of pest invasion. This study aims to develop global climate suitability models using MaxEnt software for E. amylovora and to determine the regions in which apple cultivation and the bacterium are most likely to co-occur. Most occurrence data for E. amylovora (93%) are from the northern hemisphere, distributed between 63.90 to 14.56 degree days in regions of Africa, Central America, North America, Asia, and Europe. The only country in the southern hemisphere that this bacteria has been detected is New Zealand (Oceania). Apples are cultivated on every continent except Antarctica, between 61.55 to 44.41 degree days. We find that regions of South-Eastern Africa, Argentina, Australia, Southern Brazil, Bolivia, Chile, China, the United States, Madagascar, Morocco, Mexico, New Zealand, Peru, Tunisia, Uruguay, and the majority of Europe are suitable for both E. amylovora and apple cultivation. These results provide information on the potential worldwide distribution of E. amylovora in apple production area.

. MaxEnt is a model that predicts the distribution of probabilities that a species will be present in a particular locality, based on the principle of maximum entropy (Phillips et al., 2006). This software requires data on locations in which the pest and its host are known to occur Phillips & Dudík, 2008;Merow et al., 2013).
Despite the severe impact that E. amylovora would have if it were introduced into new countries, at present, no studies have analyzed the invasion risk of this species. This study, therefore, aims to use MaxEnt modelling to predict suitable areas for E. amylovora and open-field cultivation of apples.

Distribution
We used species occurrence data from the Centre for Agriculture and Biosciences International-CABI (2019) We then performed spatial filtering using the spThin package in the R software (Ripley, 2001). Data handling maintains the most significant number of occurrence records and checks all possible filter combinations, eliminating outliers and using a minimum distance of 10 km (Boria et al., 2014;Team, 2014), ensuring that each cell has only a single occurrence record.

Climatic Data
Our analysis considers nineteen bioclimatic variables (Tables S1 and S2) from the WorldClim version 1.4 dataset (Hijmans et al., 2017) and a spatial resolution of 2.5 min (about 5 km). This resolution is considered high-quality for analyses at the global scale (Elith & Leathwick, 2009). WorldClim uses global climate records from 1960 to 1990 to estimate air temperature and rainfall (mean, maximum, and minimum), as well as other parameters such as seasonal variables and extreme climate indices (Hijmans & Elith, 2013).
We used the SDMtools package in ArcGIS to remove highly correlated variables with a Pearson correlation coefficient of r ≥ 0.75 as a cutoff, following Kumar et al. (2014). We kept one strongly correlated pair of variables in the dataset.

Determination of Risk Levels
The maximum test sensitivity plus specificity (MTSPS) was chosen to determine the distribution of suitability classes of E. amylovora in apple crops that are at risk of pest invasion. We used a cutoff for both species, with values above the cutoff considered unsuitable. The cutoff for E. amylovora and M. domestica was set at 0.3575 and 0.3457, respectively.

Validation
The global distributions of E. amylovora and apple crops were obtained from the maximum entropy-based model using MaxEnt v. 3.3.3k (Phillips et al., 2006). The suitability index generated by MaxEnt ranges from 0 for not suitable to 1 for suitable. A total of 50,000 points were randomly selected for each species, representing areas of current occurrence. Sampling bias was generated in data collected without sampling from external sources. This procedure was made using a kernel density estimate in SDMToolbox (Brown, 2014;Jarnevich et al., 2015). The polarization surface compensates for sampling intensity and possible sampling bias (Jarnevich et al., 2015).
The settings used in the MaxEnt models for E. amylovora and M. domestica were based on specific resource-type combinations and the regularization multiplier (RM) (Jarnevich et al., 2015;Merow et al., 2013). Combined sets of linear (L), quadratic (Q), product (P), threshold (T), and hinge (H) features were used to control the number of parameters, and thus the model complexity for both species.
The MaxEnt fade-by-clamping option was used to eliminate extrapolations outside the environmental range (Owens et al., 2013). The contribution of environmental variables was estimated using the jackknife method. Of the response curves generated by MaxEnt, only those representing relationships between the probabilities of presence for each species and each environmental predictor were chosen. All response curves were evaluated based on biological logic, and those that failed this test were eliminated.
The omission rate (OR) and the area under the curve (AUCcv) were used to evaluate the models Liu et al., 2013). The OR measures the extent to which the model omits the existence of localities, where the target species occurs. The AUCcv is obtained from the integration of the receiver operating characteristic (ROC) curve, which is the relationship of sensitivity with the complement of specificity (1-specificity). Sensitivity is defined as the proportion of real presence concerning the total occurrences predicted by the model, Characteristics of the selected model Contribution Importance of permutation Annual mean temperature (bio1;°C) 11.6 (-5.6-25.2) 80.9 80.9 Annual rainfall ( Concerning M. domestica, the mean annual temperature was the most important variable (71.6% contribution) ( Table 2). It was followed by the mean diurnal range (10.3% contribution), annual temperature range (9.3% contribution), annual precipitation (7.1% contribution), precipitation of the driest month (1.0% contribution), and precipitation seasonality (0.7% contribution) (  Annual temperature variation (bio7; ° C) 28.1 (9.5-49. Of a total of 12 models that use different combinations, the linear, quadratic and hinge (LQH) model showed the best performance for E. amylovora, considering six environmental variables (bio1, bio2, bio7, bio12, bio14, and bio15). The LQH model resulted in an RM value of 1.5, AUCcv of over 0.9, and the lowest OR values (10% of 0.1098 and 0% of 0.004) (  Note. * Selected model; † The names of the variables are described in Table 1; § L = linear component, Q = quadratic component, P = product, T = threshold, and H = hinge; ∞ RM = regularization multiplier. Out of the various combinations of the 12 models analyzed, the linear and hinge (LH) model had the best performance for M. domestica, with six environmental variables (bio1, bio2, bio7, bio12, bio14, and bio15). The LH model resulted in an RM value of 1.0, AUCcv of over 0.9, and the lowest OR values (10% of 0.1199, and 0% of 0.0048) ( Table 4).  Figure 2B). The entire European continent, except for Norway, presented suitability for crops, besides Tunisia, Algeria, Morocco, and some regions of southeastern Africa. In Asia, the suitability was detected mainly in areas of Japan, South Korea, and North Korea, besides some border regions of southeastern Asia. In Oceania, New Zealand and some areas of Australia also showed suitability for M. domestica. The model also indicated suitability in some areas of Madagascar and northeastern Brazil, in which M. domestica is not yet cultivated ( Figure 2B).
The model detected areas that are suitable for both species in all continents except Antarctica. In the Americas, appropriate areas were found in southern Brazil, Uruguay, mainly southeastern Argentina, southern Chile, south-central Bolivia, southwestern Peru, central Ecuador, some regions of Colombia, western Mexico, northeastern United States, and some points in extreme southern Canada ( Figure 2C).  Figure 2C).
In the Asian continent, the suitability for both species was found in Japan, South Korea, southeastern China, and border regions with Bhutan, Nepal, India, and Pakistan. Some points were also detected in Indonesia, Afghanistan, Iran, Turkmenistan, Kazakhstan, Syria, Yemen, Kyrgyzstan, and Tajikistan ( Figure 2C).
In Africa, suitability was higher in the southern region of the continent, including South Africa and some areas of Namibia, Angola, Zimbabwe, Tanzania, Kenya, Ethiopia, Madagascar, Tunisia, Algeria, and Morocco ( Figure  2C).
In Oceania, New Zealand and the southern and southwestern regions of Australia were the most suitable regions for E. amylovora and M. domestica ( Figure 2C).

Discuss
The ntina, occo, m by matic ating material from fruit trees and ornamental plants is an important dispersion factor of E. amylovora over long distances. Therefore, the source regions of this bacterium should be monitored since even asymptomatic plants can spread the bacteria if ideal conditions occur among the host, pathogen, and environment (Cambra et al., 2002).
According to Rodoni et al. (1998), E. amylovora was detected in Melbourne, Australia, in 1997, an area that our study indicates as suitable for the propagation of this pathogen. In the following years, the Australian authorities carried out quarantine actions, leading to the country to become the only one in the world to eradicate this species (Rondoni et al. 2001), in 1999. However, since our model indicated that the range of M. domestica in the country overlapped with the potential range of E. amylovora, defence actions are imperative in order to avoid a possible reentry of the disease.
In the United States, the distribution area of E. amylovora is larger than the cultivation areas of M. domestica.
Since other host species (pears and quinces) are cultivated in the United States, measures to contain and suppress the pest are necessary to minimize its damage. For the cultivation of apples in these regions, the use of resistant varieties, healthy propagating material, biological control, monitoring, and use of disease prediction models is recommended (Aćimović et al., 2015).  , 2018). Because it is a highly aggressive, rapidly-spreading disease, and since there are currently no effective chemical treatments to control it, rapid diagnostic tests for early detection in areas that are free of the disease allow the destruction of infested material, which is essential for the control and eradication of the species (DGADR, 2011;Powney et al., 2011). Accordingly, it is important to carry out more detailed studies on the entry risk of this bacterium in Brazil, mainly because the southern region, where the production areas of M. domestica are concentrated, were demonstrated to be suitable for E. amylovora.
In contrast, the countries belonging to Mercosur that have shown suitability for E. amylovora in some regions, such as Argentina, Chile, Uruguay, Bolivia, Ecuador, Peru, and Colombia, have restrictions on the entry of fruits that are hosts to E. amylovora, as described in Mercosur Resolution No. 50/05 GMC (Mercosul, 2019). We observed in this study that all countries in South America that have suitability for the pest also have phytosanitary restrictions to prevent its entry, mainly through commercial relations with countries currently infested by E. amylovora. This work considered temperature and precipitation to determine suitability for E. amylovora and M. domestica. Thus, further studies considering other variables are necessary, such as the presence of other host species (pear and quince), the presence of natural enemies and antagonistic microorganisms, and the resistance of cultivars, among others.
Suitability maps are important tools for pest risk analyses, quarantine strategies, and to support phytosanitary actions. These results can help to develop strategies to prevent the introduction, dispersion, and establishment of E. amylovora, in addition to supporting future research and supporting biosafety practices.
This study presents relevant information about the potential risk of the worldwide distribution of E. amylovora in apple crops and the suitability of both species using the MaxEnt model. The maps can serve to support monitoring programs in countries where the species already occurs and to determine guidelines and measures to prevent the risk of invasion of E. amylovora in other regions.