Global Risk Assessment of the Occurrence of Bovine Lumpy Skin Disease: Based on an Ecological Niche Model

Lumpy skin disease (LSD) is a highly contagious disease in bovine animals. An outbreak of LSD can cause devastating economic losses to the cattle industry. To investigate the distribution characteristics of historical LSD epidemics, LSD was divided into four phases for directional distribution analysis based on trends in epidemic prevalence. Ecological niche models were developed for LSD as well as for two vectors ( Stomoxys calcitrans and Aedes aegypti ), and global predictive maps were generated for the probability of LSD occurrence and the potential distribution of the two LSD vectors. Te models had good predictive performance (the AUC values were 0.894 for the LSD model, 0.911 for the S. calcitrans model, and 0.950 for the A. aegypti model). Te LSD combined vector prediction map was generated by combining the distribution maps of Stomoxys calcitrans and Aedes aegypti with fuzzy overlay tool in ArcGIS. Te LSD combined vector prediction map was combined with the LSD prediction map to generate the LSD vector transmission risk map. Te eastern and northwestern regions of North America, the eastern and northern regions of South America, the central and southern regions of Africa, the southern region of Europe, the northwestern and southeastern regions of Asia, and the eastern region of Australia were predicted to provide suitable environmental conditions for the occurrence of LSD. Cattle density, bufalo density, and bio2 (mean diurnal range) were identifed as key variables for the occurrence of LSD. Te fndings of this study can be useful to policymakers in developing and implementing preventive measures of LSD for the health of cattle and the cattle industry.


Introduction
Bovine lumpy skin disease (LSD) is an infectious viral disease caused by the LSD virus (LSDV, genus goat pox virus; family Poxviridae) and is listed as a notifable disease by the World Organization for Animal Health (WOAH) [1]. Te LSDV shares antigenic similarities with sheep and goat pox viruses. However, it cannot be distinguished by routine serological testing [2]. Te LSDV is highly host-specifc and only infects and causes disease in bovids (mainly cattle and bufalo) under natural conditions, and no human infections have been reported [3]. Te initial signs of infection in cattle include fever, depression, loss of appetite, excessive salivation, and markedly enlarged lymph nodes on the body surface [4]. Soon after the onset of fever, extensive skin nodules appear on the head, neck, limbs, udder, genitalia, and perineum of the afected cattle, gradually covering the entire body as the disease progresses. Te skin nodules appear as frm, restricted, and rounded elevations, sometimes with lesions in the subcutaneous tissue and even in the muscle [5]. Te morbidity and mortality in cattle caused by LSD are highly correlated with the strain of the virus, the breed of cattle, and immunization. Typically, the incidence of LSD ranges from 3-85%, but the mortality is relatively low (1-5%) [4]. LSD causes a reduction in the quality and quantity of milk production, damage to hides, weakness and weight loss, infertility, and abortions resulting in signifcant economic losses to the cattle industry [6].
LSD was frst discovered in Zambia in 1929 and has shown endemicity mainly in Africa, including the Sahara and Madagascar [7]. Subsequently, the disease spread to most African countries, followed by the Middle East in 1986, and to southeastern and northeastern Europe since 2014, afecting the Caucasus countries, Kazakhstan, Russia, Greece, Armenia, Bulgaria, and the Republic of Macedonia [8]. Te disease expanded to Asia since 2019, with outbreaks in India, Bangladesh, Nepal, Bhutan, Vietnam, and Myanmar [9]. As the epidemic spreads, the disease gradually threatens other European and Asian countries.
Te primary source of infection for LSD is infected cattle. However, the transmission of the LSD virus from infected cattle is not fully understood. Direct contact with infected animals and transmission by blood-sucking insects (e.g., mosquitoes, fies, and ticks) have long been suggested as important routes for LSDV transmission [10]. Although direct contact is considered a less efective source of infection [11], studies assessing the risk of LSDV transmission by blood-sucking insects have shown that stable fies (Stomoxys calcitrans) and mosquitoes (Aedes aegypti) are the most likely vectors of LSD [12]. Indeed, the LSD-causing virus was detected in S. calcitrans caught on LSD-infected animals in the feld [13]. S. calcitrans can also transmit LSDV from experimentally infected animals to naive cattle [14]. Ae. aegypti has been shown to transmit LSDV from infected to susceptible cattle and can be a more efcient vector than other mosquitoes [15,16]. Terefore, this study considered two important vectors, S. calcitrans and Ae. aegypti, and explored their potential global distribution.
Spatiotemporal studies that focus on exploring the transmission and distribution characteristics of infectious diseases contribute to a better understanding of the patterns and risk factors of disease epidemics [17]. In addition, a historical spatiotemporal distribution of epidemics of infectious diseases may complement clinical and molecular studies of the diseases [18]. Terefore, spatial epidemiological tools are increasingly being applied to monitor and allocate resources in the face of emerging threats posed by infectious diseases to animals [19][20][21]. Te ecological niche model (ENM) has widely been used to predict the potential distribution of disease and for disease risk mapping [22,23]. Te ENM explores the relationships between specifc environmental variables and disease occurrence and helps to predict the potential distributions of a disease, which are important to respond to the spread and transmission of diseases [24]. In addition, predicting the spatial distribution of vectors by using ENM can inform decision-making to address control measures in targeted areas with the presence of vectors [25,26]. In this study, a directional distribution analysis was performed using the standard deviation ellipse method to explore the historical spread of the LSD epidemic globally. A maximum entropy modeling was applied to LSD and LSD vector occurrence data to identify potential areas suitable for LSD amplifcation and vector transmission. Te fndings of this study will provide useful information for policymakers to develop tools and allocate resources for LSD pathogen detection and vector surveillance.

Occurrence Data Collection.
Geographical coordinates of the historical LSD outbreak and LSD cases were obtained from the Food and Agriculture Organization of the United Nations (FAO) records (https://empres-i.apps.fao.org/). Te records included 4316 global cases of LSD from 1/1/2006 to 18/09/2022. Occurrence records for Ae. aegypti and S. calcitrans from 2006 to 2022 were obtained from the Global Biodiversity Information Facility (GBIF) (https:// www.gbif.org/) database. Te GBIF database was searched for Aedes aegypti and Stomoxys calcitrans as keywords and clicked on "OCCURRENCES" in the results to display the distribution records of the species. Te time period selected was 2006-2022 and then downloaded. After cleaning and sorting, only the occurrence records with specifc latitude and longitude coordinates were retained and the rest were excluded from the analysis.
To reduce duplication and spatial autocorrelation in the data, the "trim duplicate occurrences" function in the ENMTool software was used to flter the occurrence records to ensure that there was only one record in a grid [27]. In the end, a total of 2196 occurrence records for LSD, 610 for Ae. aegypti, and 1485 for S. calcitrans were used to run the models (Table S1).

Directional Distribution Analysis.
Directional distribution analysis is most often used in epidemiology to determine the directionality of disease distribution and transmission. A trend in the directional distribution of LSD was determined using a standard deviation ellipse, a method widely applied to study epidemics of infectious diseases in animals and humans. Te standard deviation ellipse generates ellipsoidal polygons. Te attributed values for these polygons include the X and Y coordinates of the mean center, long and short axes, and the orientation of the ellipse. Analysis of these features and elliptical attribution values helped us to determine whether the distribution of the LSD epidemic points was in a specifc direction. Based on trends in the epidemic prevalence, we identifed the epidemic progression patterns in four phases: 2006-2009, 2010-2013, 2014-2017, and 2018-September 2022. Te standard deviation ellipse for each stage was calculated by the directional distribution tool within the spatial statistics tool of ArcGIS 10.2. Te ellipse size was set to one standard deviation, incorporating approximately 68% of the number of cases in a phase.

Variable Collection.
Environmental and topographic variables afect habitat conditions for arthropod survival and have been widely used in ecological niche modeling [28][29][30][31][32]. Environmental and topographic variables assembled for this study included 19 bioclimatic variables, solar radiation, wind speed, elevation, and normalized diference vegetation index (NDVI) ( Table 1). Data for the 19 bioclimatic variables, solar radiation, wind speed, and elevation were downloaded from the WorldClim 2.1 database for 1970-2000. NDVI data were downloaded from the National Tibetan Plateau Data Center [33]. Tis dataset is the most recent version of the long series (1981-2015) normalized diference vegetation index product of the NOAA Global Inventory Monitoring and Modeling System (GIMMS). We used the maximum value compositing method to obtain the annual mean NDVI for 1981-2015 [34,35]. Te global density data for cattle and bufalo were obtained from the FAO livestock systems database [36,37]. All raster data were at a resolution of 5 arc minutes. Before the model development, we performed multicollinearity and correlation analyses between 19 bioclimatic variables using the basic functions and the USDM package of R software. Te variables with an absolute value of Pearson correlation coefcient >0.7 and a VIF value >5 were excluded from the model development. Te fnal remaining bioclimatic variables were used with solar radiation, wind speed, altitude, and NDVI to construct the ecological niche model.

Ecological Niche
Modeling. MaxEnt is a popular ecological niche modeling method with strong predictive performance [38,39]. MaxEnt 3.4.1 software was employed to predict the current distribution of LSD, Ae. aegypti, and S. calcitrans. Te model was set up as follows: 80% of the occurrence records are used to train the model and the remaining 20% are used to evaluate the model performance, the output format was logistic, the features were automatically selected, the regularization multiplier � 1, the maximum number of background points � 10000, the replicated run type was Bootstrap, and random seed was ticked. Te model was repeated 10 times, and the average of the 10 results was taken as the fnal prediction. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) criteria, with AUC values ranging from 0-1. Te higher the AUC, the better the performance of the model. Furthermore, an AUC greater than 0.5 indicates that the predictive capacity of a model is higher than a model using random prediction.

LSD Transmission Risk Assessment.
Te model outputs were converted to raster fles in ArcGIS 10.2 to produce an LSD occurrence suitability map and habitat suitability maps for Ae. aegypti and S. calcitrans. Te fuzzy overlay tool in the spatial analysis tools of ArcGIS 10.2 allows for the analysis of the possibility that a phenomenon belongs to multiple ensembles during the multicriteria overlay analysis. Te LSD combined vector prediction map was generated by combining Ae. aegypti and S. calcitrans habitat suitability maps with the fuzzy overlay tool. Te overlay type "OR" was selected, meaning that each raster of the output layer was the maximum of the input layer. Te LSD combined vector prediction map refects the global distribution of LSD vectors. Ten, the LSD combined vector prediction map was combined with the LSD occurrence suitability map by selecting "AND" as the overlay type in the fuzzy overlay tool. Selecting "AND" as the overlay type ensured that each raster of the output layer was the minimum of the input layer, kept the likelihood of LSD and the presence of the vector to a minimum, and produced a map that showed potential LSD vector transmission risk [40].

Model Evaluation and Variable Contributions.
Te models were evaluated based on AUC values, and evaluation results are shown in Figure S1. Te mean AUC values for the LSD, S. calcitrans, and Ae. aegypti models were 0.894, 0.911, and 0.950, respectively, demonstrating a strong performance of all three models. Te variable sieving procedure was employed to select important variables contributing to each model. Te results of selected variables and their contribution to the models are shown in Table 2. Variables that contributed more than 10% to a model were considered important. Cattle (cattle density), bio2 (mean diurnal range), and bufalo (bufalo density) had a contribution of more than 10% and were considered important variables in the LSD model. In the Ae. aegypti model, bio3 (isothermality), bio19 (mean precipitation of coldest quarter), srad (solar radiation), and NDVI were considered essential for the survival of Ae. aegypti. Te important variables in the S. calcitrans model were bio19, bio6 (min temperature of the coldest month), and cattle.

Response Curves for Signifcant Variables in the Model.
Te response curves for the signifcant variables in the LSD model are shown in Figure 2. Although cattle and bufalo densities contributed more to the model, the probability of LSD occurrence remained relatively stable as both variables increased. Te response curve for bio2 showed that the environmental suitability for LSD was higher at a mean diurnal range of approximately 7-13°C. Te response curves for the LSD vectors are shown in Figures S2 and S3. Ae. aegypti occurred in areas with high levels of isothermality and solar radiation, appropriate winter precipitation, and suitable vegetation cover. Areas with mild winter temperatures, suitable precipitation, and the presence of cattle were suitable for S. calcitrans.

LSD and LSD Vectors Suitability Maps.
Te eastern and northwestern regions of North America, the eastern and northern regions of South America, the central and southern regions of Africa, the southern region of Europe, the northwestern and southeastern regions of Asia, and the eastern region of Australia were identifed as suitable for the prevalence of LSD (Figure 3). Te high-risk areas for LSD were concentrated in the southern part of Europe and the northwestern and southeastern parts of Asia. Southern North America, southeastern South America, southeastern Africa, southern Asia, and eastern Australia were identifed as environmentally suitable for the subsistence of Ae. aegypti ( Figure S4). Te southern region of North America, the southeastern and northwestern regions of South America, the western region of Europe, the eastern region of Asia, and the eastern region of Australia may be the current ecological range of S. calcitrans ( Figure S5).

Mapping LSD Transmission Risk Areas.
Te LSD transmission risk map (Figure 4) was generated by combining the LSD occurrence suitability map ( Figure 3) and the LSD combined vector prediction map ( Figure S6). Results showed that the eastern region of North America, eastern South America, southern Europe, southern Asia, and eastern Australia might be at high risk of LSD transmission.

Discussion
Te spread of the LSD epidemic is of international concern as it is an important statutory notifable infectious disease. Despite the low mortality rate, LSD causes signifcant economic losses to the cattle industry. In addition, since LSD is considered a transboundary and trade band disease, it signifcantly impedes the international trade of livestock products. Using directional distribution analysis and ecological niche modeling, we explored the directional distribution of historical epidemics and areas of environmental suitability for the LSD globally. Due to logistical issues (e.g., delays in data reporting by the WOAH and FAO), September 2022 LSD outbreak data from India were not included in this study. Our results showed that the directional distribution trend in LSD from 2006-2009 was not apparent (Figure 1) [41,42]. Compared with other fying insects, S. calcitrans and Ae. aegypti may be efective vectors [12], and both may play important roles in cross-border transmission of LSD. Our results showed that the high-risk areas for LSD were concentrated in the countries surrounding the Black Sea, north of the Mediterranean Sea, northwest and southeast Asia, and parts of Africa and the Americas (Figure 3). Te environmental suitability for LSD disease across a large part of India reported in this study is supported by a severe LSD outbreak in India in September 2022 [43]. Tese fndings also support the robustness and reliability of our modeling approach. Te risk map suggested that the countries with no history of LSD or with a few cases, such as Colombia, Italy, Spain, France, Portugal, Slovenia, Slovakia, Hungary, Croatia, Bosnia and Herzegovina, Romania, Moldova, Georgia, Armenia, Azerbaijan, Tajikistan, Pakistan, China, and Myanmar, were at high-risk regions of LSD occurrence. Tese countries should be on alert for the importation and potential disease outbreak.
Te response curves for the environmental variables showed that changes in each variable afect the probability of LSD incidence. Te response curves for cattle density and bufalo density were generally smooth, with little efect of both variables on the incidence of LSD ( Figure 2). However, the contribution of these variables to the model was reasonably high (cattle density: 44.2% and bufalo density: 10%). Previous mathematical modeling studies also reported similar results, with no clear relationships between cattle density and LSD infection rates [44]. Te transmission patterns can be explained by indirect transmission [44]. Tis may be because direct animal contact is less rapid and effcient in transmitting the virus and does not play a significant role in transmission, with LSD more likely to be transmitted by blood-sucking insects [45,46]. Te response curve for bio2 showed that the temperature range of 7-13°C was suitable for LSDV. Lower latitudes experience greater daily variation in solar height and mean diurnal range, implying that the LSD may be more prevalent in the low and mid-latitudes with warm climatic conditions. Te warmer and humid climates in lower latitudes can be conducive to the reproduction and survival of LSD vectors such as mosquitoes, fies, and ticks and transmitting LSD virus (ESM [4]).
One of the best ways to prevent and control infectious diseases is to cut of the transmission. Te vectors of LSD are not well understood, but Ae. aegypti and S. calcitrans have    been shown to play an important role in the transmission of LSD [12]. We collected historical records of the presence of S. calcitrans and Ae. aegypti and developed separate models using topography, climate, and vegetation to identify the distribution of LSD vectors and thus assess the risk of vectorborne LSD globally. Our results showed that appropriate temperatures and precipitation were necessary for S. calcitrans to survive the winter ( Figure S3). S. calcitrans overwinter as eggs and larvae and are sensitive to changes in temperature and precipitation [47]. Excessive cold and dry conditions can prolong larval development and reduce the survival rate of the larvae [48]. Cattle density was also a signifcant predictor of S. calcitrans distribution, suggesting that cattle provide a blood meal for the survival of S. calcitrans. Ae. aegypti distribution was also infuenced by temperature and precipitation in addition to vegetation cover. Results showed that Ae. aegypti performed well under high temperatures, sufcient moisture conditions, and enough vegetation cover ( Figure S2). Temperature afects the survival, growth, and reproductive rate of Ae. aegypti [49,50]. Precipitation and vegetation cover provide resting habitats and stimulate egg incubation [51,52]. Our results also implied that the potential distribution ranges of Ae. aegypti and S. calcitrans overlapped little. We generated the LSD combined vector suitability map by overlaying the individual map of each species using the fuzzy overlay tool ( Figure S6). A comparison of the LSD environmental suitability map with the LSD vectors suitability map suggested that most of the high-risk areas for LSD were also suitable for the survival of its vectors. Te eastern United States, eastern and northwestern South America, Black Sea coast, northern Mediterranean countries, Pakistan, southern, eastern and northern of India, Bangladesh, Cambodia, Vietnam, southeastern China, and eastern Australia were considered at high risk of LSD vector transmission ( Figure 4). In these countries and regions, the environmental suitability of LSD and the habitat suitability of the LSD vectors are both high and extremely suitable for the occurrence and spread of LSD. To reduce the risk of potential LSD outbreak and spread in these countries and regions, cattle should be vaccinated regularly, the movement of cattle should be controlled, vector surveillance should be increased, and regular disease surveillance should be conducted.
Tere are some limitations to our study. We did not have some relevant data to better predict the long-distance transmission of LSD. For instance, we could not consider the data related to long-distance cattle migration, which may be related to the cross-border spread of LSD [41]. Because of the paucity of information on LSD vectors and transmission mechanisms, we could not include several other potential insect vectors but only considered S. calcitrans and Ae. aegypti. Terefore, the LSD transmission risk map may be underestimated.

Conclusion
Te directional distribution trend in the LSD epidemic was not obvious during the frst phase (from 2006-2009).
However, the LSD epidemic from 2010-2013 and 2014-2017 showed northeast-southwest distributional trends. From 2018 to September 2022, the disease showed a northwestsoutheast distribution. Our model identifed areas suitable for LSD occurrence and vector transmission. Cattle density, bufalo density, and bio2 (mean diurnal range) were the key factors infuencing the occurrence of LSD. Tis study could be used for risk-based surveillance patterns that selectively target high-risk areas of LSD occurrence and vector transmission.

Data Availability
Te datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Ethical Approval
Te authors confrm that the ethical policies of the journal, as noted on the journal's author guidelines page, have been adhered to.

Conflicts of Interest
Te authors declare that there are no conficts of interest.  Figure S4. Habitat suitability map for Aedes aegypti. Te warmer colors depict areas of high habitat suitability while cooler colors depict areas of low habitat suitability. Figure S5: Habitat suitability map for Stomoxys calcitrans. Te warmer colors depict areas of high habitat suitability while cooler colors depict areas of low habitat suitability. Figure S6: Habitat suitability of LSD combined vector. Te warmer colors depict areas of high habitat suitability while cooler colors Transboundary and Emerging Diseases 7 depict areas of low habitat suitability. Table S1: Filtering of occurrence records.