Pteropus lylei primarily forages in residential areas in Kandal, Cambodia

Abstract Bats are the second most species‐rich Mammalian order and provide a wide range of ecologically important and economically significant ecosystem services. Nipah virus is a zoonotic emerging infectious disease for which pteropodid bats have been identified as a natural reservoir. In Cambodia, Nipah virus circulation has been reported in Pteropus lylei, but little is known about the spatial distribution of the species and the associated implications for conservation and public health. We deployed Global Positioning System (GPS) collars on 14 P. lylei to study their movements and foraging behavior in Cambodia in 2016. All of the flying foxes were captured from the same roost, and GPS locations were collected for 1 month. The habitats used by each bat were characterized through ground‐truthing, and a spatial distribution model was developed of foraging sites. A total of 13,643 valid locations were collected during the study. Our study bats flew approximately 20 km from the roost each night to forage. The maximum distance traveled per night ranged from 6.88–105 km and averaged 28.3 km. Six of the 14 bats visited another roost for at least one night during the study, including one roost located 105 km away. Most foraging locations were in residential areas (53.7%) followed by plantations (26.6%). Our spatial distribution model confirmed that residential areas were the preferred foraging habitat for P. lylei, although our results should be interpreted with caution due to the limited number of individuals studied. Synthesis and applications: Our findings suggest that the use of residential and agricultural habitats by P. lylei may create opportunities for bats to interact with humans and livestock. They also suggest the importance of anthropogenic habitats for conservation of this vulnerable and ecologically important group in Cambodia. Our mapping of the probability of occurrence of foraging sites will help identification of areas where public awareness should be promoted regarding the ecosystem services provided by flying foxes and potential for disease transmission through indirect contact.


| INTRODUC TI ON
Bats are the second most species-rich Mammalian order with over 1,300 species worldwide (Voigt & Kingston, 2016) and provide a wide range of ecologically important and economically significant ecosystem services (Kunz, Torrez, Bauer, Lobova, & Fleming, 2011).

Nipah virus was first identified in pigs and people in Malaysia in
1998 (Chua, 2000) and has reemerged annually in Bangladesh since 2001 (Luby et al., 2009). NiV causes lethal encephalitis in people, and bats in the Pteropus genus are the reservoir (Epstein, Field, Luby, Pulliam, & Daszak, 2006). Transmission of the virus in Malaysia is presumed to have occurred as a result of pigs consuming bat-contaminated fruits, followed by contamination of humans working with pigs (Chua, 2003). In Bangladesh, direct bat-to-human transmission of the virus occurs through the consumption of date palm sap . NiV has been isolated or detected in several Pteropus species in Southeast Asia, including P. medius in Bangladesh, P. lylei in Thailand and Cambodia, and P. vampyrus and P. hypomelanus in Malaysia. However, despite its detection in P. hypomelanus, a serological study on Tioman Island did not find the virus in any of the local people (Chong, Tan, Goh, Lam, & Bing, 2003) that the bats live among and regularly interact with (Aziz, Clements, Giam, Forget, & Campos-Arceiz, 2017). Seasonal NiV shedding patterns have been suggested for P. lylei in Thailand, with peak shedding occurring in May (Cappelle, Hul, Duong, Tarantola, & Buchy, 2014;Wacharapluesadee et al., 2010).
Understanding the capacity of a reservoir to spread the virus at local and regional levels to humans and domestic animals is fundamental to surveillance and prevention initiatives. Knowledge about the distribution and movement patterns of these bat species is thus required, and telemetry (measurement and transmission of data from remote sources) is a valuable tool to monitor the drivers and characteristics of fruit bat movements (Smith et al., 2011). This can be used to develop appropriate host management strategies that maximize the conservation of bat populations and minimize the risk of disease outbreaks in domestic animals and humans.
Telemetry studies have been undertaken on several Pteropus species in Asia and Australia. In Australia, tracking of fourteen P. poliocephalus males revealed that these are highly mobile between roosts and regularly travel long distances (Roberts, Catterall, Eby, & Kanowski, 2012). For instance, one P. alecto was tracked between Papua New Guinea and Australia and traveled more than 3,000 km over 11 months (Breed, Field, Smith, Edmonston, & Meers, 2010). In Southeast Asia, the movements of seven P. vampyrus males encompassed Malaysia, Indonesia, and Thailand, indicating the need for regional management plans for such species (Epstein et al., 2009).
These studies highlight the difference between migratory and nomadic flying foxes and the need to adapt management strategies to relevant geographic scales.
At a local scale, telemetry studies indicate that Pteropus bats make foraging flights on a nightly basis, with distances from the roost ranging from a few kilometers to 20-30 km. Depending on species, foraging sites range from apparently intact forest to cultivated areas.
In Bangladesh, the roosting ecology of P. giganteus is associated with forest fragmentation, likely because fragmented forests offers more foraging options to the bats, including fruit species cultivated by humans (Hahn et al., 2014). Conversely, in the Philippines, most foraging locations of eight Acerodon jubatus were situated in closed forest remote from areas of evident human activity (de Jong et al., 2013). Another study on A. jubatus and P. vampyrus in the Philippines suggested these species prefer undisturbed forest types and select against disturbed and agricultural areas (Mildenstein, Stier, Nuevo-Diego, & Mills, 2005). Foraging also repeatedly occurred 15-30 km from the roost on average. Similarly, movements of P. alecto were very similar between nights with most foraging sites located less than 6 km from roost sites. In Thailand, P. lylei also undertakes relatively short foraging movements (2.2-23.6 km) on a nightly basis to fields, plantations, backyards, and mangroves (Weber et al., 2015).
In Cambodia, three flying fox species are thought to occur, large flying fox P. vampyrus which is listed as "near threatened" by IUCN, Lyle's flying fox P. lylei which is listed as "vulnerable," and island flying fox P. hypomelanus, which is listed as "least concern" (IUCN, 2008; Kingsada et al., 2011). Most of the known flying fox roost sites in Cambodia are located on the grounds of pagodas, where hunting is limited by the presence of the monks (Ravon, Furey, Hul, & Cappelle, 2014). Consequently, these are often located in the middle of villages close to human and domestic animal populations, and available foraging areas mostly comprise anthropogenic landscapes. Flying foxes in Cambodia are likely to interact frequently with humans and to depend on human activities for their subsistence. As a consequence, understanding of their preferred foraging areas is important to inform public health and conservation actions.
The objective of our study was to use telemetry data to determine and characterize foraging locations visited by flying foxes ecosystem services provided by flying foxes and potential for disease transmission through indirect contact.

| Study site
The P. lylei roost selected for this study was located at Wat Pi Chey Saa Kor (11.200 N,105.058 E), Kom Poung Kor village, Koh Thom District, Kandal Province ( Figure 1). The site comprises a grove of trees on the grounds of a Buddhist pagoda which encompasses 21 roost trees with an estimated population of 4,000 flying foxes (Ravon et al., 2014). The village is bisected by a road with houses on either side and is characterized by a mosaic of agriculture that lacks significant areas of natural vegetation/forest. Land uses in the region include cultivation of rice and other crops, backyards, plantations, and various backyard animal farming activities.

| Study period
Our study was conducted from April 18, 2016 to May 17, 2016 when shedding of the NiV by P. lylei is believed to peak in Cambodia , similar to Thailand (Wacharapluesadee et al., 2010
Weight, forearm length, sex, age, and reproductive status were documented for each bat. Animals were selected for collaring based on weight. Adult males and females without pups weighing at least 400 g were selected so that collars, weighting 20 g, would comprise <5% of body mass (Brigham, 1988). Pregnant and lactating female bats were excluded to avoid adding extra burdens.
Selected bats were anesthetized by injecting medetomidine into the pectoral muscle (Epstein, Zambriski, Rostal, Heard, & Daszak, 2011). GPS devices (FLR V, Telemetry Solutions ™ , www.telemetrysolutions.com) attached to nylon bands were secured around the neck of each bat using catgut suture (1.0) and three surgical knots (Figure 2), which were presumed to last for at least 30 days.
Following collar attachment, atipamezol was injected intramuscularly. Each collared bat was kept in a separate cage during recovery from anesthesia and offered pieces of mango ad libitum prior to release.
We deployed 14 GPS collars on 13 adult males and one adult female (Table 1). Collars 1-5 were programed to transmit one location every 30 min from 5 p.m. to 6 a.m. while collars 6-14 were programed to transmit one location every 30 min for the first night only and one location every 5 min from 5 p.m. to 6 a.m. on following nights. As a consequence, collars 1-5 were expected to last for 1 month and allow observations of foraging behavior across the expected annual excretion peak of NiV. Collars 6-14 were expected to last for 10 days and provide detailed information on P. lylei foraging sites, including night roosts. Data were remotely downloaded each morning from active collars with a base station, which automatically connected to the GPS collars when within reading distance (10-20 m).

| Spatial data and site characterization
Global Positioning System data were transferred each morning to a computer, converted into KML format (QGIS, version 2.14), and mapped to identify foraging locations visited by bats the previous night (Google Earth, version 7.1). Foraging sites were identified based on clusters of two or more locations obtained from individual bats and as many as possible were visited depending on accessibility.
Tree species visited by bats and evidence of foraging such as partially F I G U R E 1 Location of the study area and other flying fox roost sites known in Cambodia eaten fruits were recorded to facilitate identification of roosting and feeding trees. Nonfruiting trees were also recorded.

| Habitat use
All locations were classified in three major categories: roost locations (all points less than 30 m from the roost site), foraging locations (a cluster of ≥2 two points separated by <30 m where the bat spent at least 10 min at night (i.e., from 6 p.m. to 6 a.m.)), and commuting locations (isolated points connecting the roost and foraging sites located >30 m from a foraging or roost location). Based on patterns visible in Google Earth, five habitat types were recognized for foraging locations: plantations (including fruit trees within the plantation and trees around the plantation), residential areas (locations within 50 m of human settlements, including pagodas, backyards, roads), agricultural lands (any cultivated land not included in "plantations" and "residential areas"), rivers, and uncultivated areas (all locations not included in the preceding categories).

| Spatial analysis
The home range of an animal illustrates spatial and temporal use of an area and is defined as the area commonly used for normal activities such as foraging for food, breeding, and caring for young (Burt, 1943). We used the kernelUD() function of the Adehabitat package in R software (Version 3.2.3) to estimate the home range for all bats, using the kernel density method (Calenge, 2006). The function computes the different percentage levels of home range estimation, for example the 50% home range identifies the areas where an individual is likely to occur 50% of the time.
We used QGIS to analyze the trajectories of each bat and to generate heatmaps based on kernel density estimation. The density was calculated based on the number of points in a location, with larger numbers of clustered points resulting in larger values. We also used the sp package in R software to calculate the maximum linear distance traveled from the roost per night.
The spatial distribution of foraging sites in the study area was modeled using the GPS data collected, a set of generated background data and land cover data. We created a map which classified habitats according to their expected influence on foraging site selection by the bats. This map was the product of a classification procedure based on Landsat images (30 m spatial resolution) acquired in 2015 and ground-truthing. Details of the classification are provided as Appendix (Supporting information Appendix S1: Table S1), and the result is illustrated by (Supporting information Appendix S1: Figure   S1). The eight different habitats identified in this classification were F I G U R E 2 Collared Pteropus lylei, southern Cambodia TA B L E 1 Characteristics of Pteropus lylei studied and GPS device performance, southern Cambodia. The proportion of valid data corresponds to the proportion of locations recorded with valid geographic coordinates speculated to have the following impacts on the distribution of foraging sites. Plantations were expected to be highly attractive to bats because of the high density of fruit available. Tree vegetation was expected to be attractive because of the potential presence of fruit consumed by bats. Water bodies such as rivers were also expected to attract the bats due to the presence of fruit trees on their banks.
Residential areas were expected to have mixed effects as a source of disturbance for the bats and a potential source of fruit in backyards.
The four remaining habitats in the classification (rice field, bare soil, flooded vegetation, and shrubland) were not expected to attract the bats.
To train and validate the model, we used all GPS locations of foraging sites and generated an equivalent number of background locations in the study area which were used as pseudoabsences by the model. Half of the data were randomly assigned to a training dataset and the other half to a validation dataset. We used a generalized linear model with the training dataset as the response variable with a binomial distribution (1 for presence and 0 for pseudoabsence) and habitat type as an explanatory qualitative variable.
To deal with the discrepancy between the spatial resolution of our classification (30 m) and GPS points (1-5 m), we calculated the distances of all data points to the closest habitats with an expected influence on bat habitat selection: plantations, tree vegetation, water bodies, and residential areas. Because of this discrepancy and landscape fragmentation in the study area, GPS locations of bats foraging in attractive habitats could be recorded in an adjacent nonattractive habitat. We therefore generated four explanatory variables (dPlant, dTree, dWater, and dResid) to allow us to capture the spatial structure of the study area. Using the distance to these attractive habitats as explanatory variables in the model would then help take into account the limited spatial resolution of our habitat classification as well as spatial autocorrelation.

| Collar performance
A total of 84 bats were caught, 14 of which were selected for col-

| Habitat use
Tree species identified during visits to foraging sites are listed in Table 2. Partially eaten mango (Mangifera indica, n = 15) and sapodilla (Manilkara zapota, n = 3) were found at exact GPS foraging locations ( Figure 3). It was not possible to detect whether leaves or flowers were also consumed.  (Table 3). (Supporting information Appendix S1: Figure S2) shows the spatial distribution of the foraging sites in the study area.

| Movement patterns and flight distances
The maximum distance traveled per bat/night ranged from 6.88 to 105.14 km and averaged 28.3 km ( for bat #8 and 50 km during one night (site C) for bat #10 (Figure 4).

| Spatial analysis
The complete results of the home range estimations for all bats are shown in (Supporting information Table S2). The estimated home ranges were maximal for bats #08 and #10 which went to distant roosts, with 95% home range of respectively 5,984 and 1,158 km 2 .
For the eight bats that did not join another roost, the 95% home range ranged from 29.5 to 316.8 km 2 with an average 95% home range of 104.5 km 2 (SD = 115.5 km 2 ). The 50% home range of these same eight bats ranged from 4.3 to 41.1 km 2 with an average 95% home range of 14.9 km 2 (SD = 13.4 km 2 ). Our heatmap of GPS locations showed that most foraging sites and night roosts were located <15 km from the roost ( Figure 5). The spatial distribution model showed that foraging locations were significantly negatively correlated with the distance to the roost, residential areas, and water bodies. Conversely, foraging locations were significantly and positively correlated with distance to plantations. Residential areas, trees, and plantations were the main foraging habitats used by the bats (Table 4). Our map of the probability of P. lylei foraging sites highlights areas close to the roost but also helps to identify further areas where bat-human interfaces could occur ( Figure 6). Model performance was very good with a cross-validated AUC of 0.93.

| D ISCUSS I ON
Our study yielded two main results. First, our study bats mostly foraged in residential areas (53.7% of foraging locations) rather than in plantations (25.6%) and our spatial model indicated that residential areas were the preferred foraging habitat (Table 4). While other studies have shown that P. lylei and P. giganteus can primarily forage in anthropogenic landscapes (Hahn et al., 2014;Luskin, 2010;Weber et al., 2015), our data indicate a particularly strong interface through residential backyards where the potential for contact between bats and humans would be higher due to continuous human presence. This could potentially facilitate NiV transmission to humans and domestic animals and two transmission routes have been documented in previous outbreaks of NiV. The first is directly from bats to humans due to consumption of raw palm sap contaminated by flying foxes, which has led to recurrent outbreaks in Bangladesh (Luby et al., 2009). The second route was suggested for the Malaysian outbreak where pigs were likely infected after consuming fruit contaminated by flying foxes (Chua, 2003) and supported by isolation of the virus from fruit partially eaten by bats in Malaysia (Chua et al., 2002). Consistent with this second route, a direct bat-to-human transmission via ingestion of fruit has been suggested for another paramyxovirus in Malaysia (Yaiw et al., 2007). Thus, by frequently foraging in residential areas, P. lylei could contaminate fruit where humans and domestic animals live, increasing the chance of indirect contact. As such, further information on the use by local residents of fruit partially eaten by bats would help to characterize transmission risks and inform preventative actions including promotion of public awareness. Similarly, palm sap collectors in the study area reported seeing flying foxes on palm trees and urine and feces on collection containers. As our data also indicate that P. lylei visits these trees (Table 2), research on palm sap collection in the area is needed to assess the risk associated with this potential transmission route.
F I G U R E 3 Partially consumed mangoes at a GPS foraging location of Pteropus lylei, Kandal Province, southern Cambodia Our finding that P. lylei mostly forages in residential areaswhich mostly correspond to backyards-rather than in plantations was unexpected because human disturbance would likely be higher in the former and food availability greater in the latter. and all individuals showed fidelity to foraging areas, indicating repeated utilization once a food resource was located. This is presumably more energy-efficient than random foraging and is consistent with studies of A. jubatus in the Philippines (de Jong et al., 2013) and P. alecto in Australia (Palmer & Woinarski, 1999  From a conservation perspective, the apparent preference for backyards and plantations suggest that our P. lylei population is highly dependent on human activities for foraging. As such, understanding of community knowledge, attitudes, and practices regarding bats will be important to develop appropriate conservation and public awareness strategies and is now underway. Nevertheless, that residential backyards were the most strongly selected foraging habitat suggests that conflict with humans may be limited in our study area. This is consistent with the fact that other patches of trees were also attractive to our study bats ("Tree vegetation" in Table 4), albeit less than backyards and plantations. Were major bat-human conflicts to occur in our study area, the few attractive non-human-dominated habitats present could possibly become overselected by the bats.
However, our results must of course be interpreted with caution as only 14 bats in the same population were studied.
Second, because six of our 14 study bats visited at least one other roost during our 28-day study, it would appear that movements to other roost sites are relatively frequent. However, these movements were limited in time and the fidelity shown to the day roost by all of our study bats is consistent with the non-nomadic ecology attributed to P. lylei. Similar to observations for P. vampyrus (Epstein et al., 2009) and P. medius (Epstein, unpublished), visits to four other roosts including one 105 km from the study site were observed. These frequent exchanges between roosts are consistent with a regional circulation of different NiV strains in Southeast Asia suggested in previous studies (Epstein, 2017;Wacharapluesadee et al., 2016). From a conservation perspective, they also suggest that P. lylei in Cambodia is likely a metapopulation and that conservation strategies should be planned on a regional scale. This is consistent with the results of another telemetry study on the migratory P. vampyrus, calling for a comprehensive protection by regional management plans across their international range (Epstein et al., 2009 found in people on the island (Chong et al., 2003). As such, close and frequent interfaces between bats and humans, including bats roosting in the middle of villages and feeding on cultivated fruit in residential backyards and orchards (Aziz, Clements, Giam et al., 2017) may not be sufficient to lead to an emergence. Other factors such as cultural and agricultural practices must be taken into account.
Different agricultural practices may lead to different levels of exposure in the countries of Southeast and South Asia. Conditions specific to intensive pig farming in Malaysia or palm sap collection in Bangladesh may explain why the virus emerged in these countries.
Nevertheless, understanding the ecology of P. lylei may significantly improve our ability to target limited resources for interventions, and educational campaigns that discuss the risks of NiV to people and their domestic animals (Nahar et al., 2014;Parveen et al., 2016). In particular, while based on only 14 individuals, our mapping of the probability of occurrence of foraging sites for the P. lylei will help targeting prevention measures to areas where contact between flying foxes and humans can be expected.

ACK N OWLED G M ENTS
For their help in the field, we thank Pring Long and Yen Sroy.  (Epstein).

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
JC and AT conceived the study and designed methodology; JC and TH coordinated the capture of the bats with the help of KC, SR, NF, VH, and CN; JHE coordinated the deployment of the GPS collars; AT, MG, and AJ collected environmental data and produced the land cover map of the study area; KC and SR collected the GPS data from the collars; KC and SR analyzed the data and led the writing of the manuscript. KC, SR, and JC drafted the first version of the manuscript and all authors contributed critically to the drafts and gave final approval for publication.

DATA ACCE SS I B I LIT Y
The data used in this study are available on Movebank ( Annelise Tran https://orcid.org/0000-0001-5463-332X