Insectivorous bats selectively source moths and eat mostly pest insects on dryland and irrigated cotton farms

Abstract Insectivorous bats are efficient predators of pest arthropods in agroecosystems. This pest control service has been estimated to be worth billions of dollars to agriculture globally. However, few studies have explicitly investigated the composition and abundance of dietary prey items consumed or assessed the ratio of pest and beneficial arthropods, making it difficult to evaluate the quality of the pest control service provided. In this study, we used metabarcoding to identify the prey items eaten by insectivorous bats over the cotton‐growing season in an intensive cropping region in northern New South Wales, Australia. We found that seven species of insectivorous bat (n = 58) consumed 728 prey species, 13 of which represented around 50% of total prey abundance consumed. Importantly, the identified prey items included major arthropod pests, comprising 65% of prey relative abundance and 13% of prey species recorded. Significant cotton pests such as Helicoverpa punctigera (Australian bollworm) and Achyra affinitalis (cotton webspinner) were detected in at least 76% of bat fecal samples, with Teleogryllus oceanicus (field crickets), Helicoverpa armigera (cotton bollworm), and Crocidosema plebejana (cotton tipworm) detected in 55% of bat fecal samples. Our results indicate that insectivorous bats are selective predators that exploit a narrow selection of preferred pest taxa and potentially play an important role in controlling lepidopteran pests on cotton farms. Our study provides crucial information for farmers to determine the service or disservice provided by insectivorous bats in relation to crops, for on‐farm decision making.

productivity by suppressing pest arthropods is high. This pest control service has been estimated to be worth billions of dollars to agriculture globally by decreasing insect crop damage and increasing yield (Boyles, Cryan, McCracken, & Kunz, 2011;Cleveland et al., 2006;Maine & Boyles, 2015;Naylor & Ehrlich, 1997). However, few studies have explicitly investigated the composition and abundance of dietary prey items or assessed the ratio of pest and beneficial arthropods consumed, making it difficult to assess the quality of the pest control service provided by bats.
Advances in molecular methods have facilitated the identification of cryptic dietary items and enabled predator-prey interactions to be revealed to a fine taxonomic level (Pompanon et al., 2012).
Metabarcoding is one method that has been used to examine the impact of bats as control agents of selected arthropod pests (Bohmann et al., 2011;Brown, Braun de Torrez, & McCracken, 2015;Burgar et al., 2014;Krauel et al., 2018) and monitor fluctuations in insect pests through bat-scat assays (Maslo et al., 2017). Metabarcoding can identify prey species that are overlooked via traditional microscopic dietary analysis methods, such as partially digested prey items (Pompanon et al., 2012). Qualitative and semiquantitative applications of metabarcoding can illustrate the efficiency of insectivorous bats as pest control agents as they enable identification of prey species and semiquantitative estimation of relative abundances (based on "relative read" abundance) in diets (Deagle et al., 2018;Thomas, Deagle, Eveson, Harsch, & Trites, 2016). Although semiquantitative estimates of relative abundance are not without bias (Pompanon et al., 2012;Thomas et al., 2016), they may provide a more accurate view of population-level diet variation than traditional methods (Deagle et al., 2018).
Insectivorous bats exploit major lepidopteran cotton pests in the genus Helicoverpa (cotton bollworm or corn earworm moth) in agricultural systems worldwide (Brown et al., 2015;Krauel et al., 2018;Lee & McCracken, 2005;McCracken et al., 2012). However, the rise of transgenic cotton has significantly reduced the number of Helicoverpa and other lepidopteran pests in intensive cropping landscapes (Whitehouse, Wilson, & Fitt, 2005;Williams, Wilson, & Vogel, 2011). Transgenic cotton has also modified arthropod communities (Whitehouse et al., 2005;Zhao, Ho, & Azadi, 2011) and the arthropod prey available to insectivorous bats (Federico et al., 2008). Understanding which prey items insectivorous bats consume in transgenic cotton is a significant knowledge gap, given that transgenic cotton accounts for the majority of cotton grown globally, including 4.58 million ha in the United States alone (ISAAA, 2017).
In Australia,99.5% of cotton grown is transgenic. Cotton is an important agricultural export in Australia and integral to the social fabric and viability of cotton-growing rural communities, contributing A$2 billion dollars annually (Cotton Australia, 2017). Thus, identification of prey to the lowest taxonomic level is vital to determine the extent to which insectivorous bats provide a pest control service or disservice to high-value commodity crops such as transgenic cotton and corn.
This study investigated insectivorous bat diets to determine the range of prey items consumed and therefore evaluate the effectiveness of bats as pest regulators. Insectivorous bats are assumed to provide a pest regulation service, yet major knowledge gaps remain in relation to the diversity of insectivorous bats present, and information on what they eat in transgenic cotton crops.
We asked three questions: (a) How diverse is the insectivorous bat community foraging in cotton crops over the cotton-growing season (November-March)? (b) Which prey items are consumed by insectivorous bats, and what are the relative quantities of pest and beneficial arthropods consumed? (c) Do the arthropods in insectivorous bat diets reflect arthropod prey abundance across the summer-growing season? This study enabled species-level dietary exploration of an insectivorous bat community to evaluate the contribution to natural pest control in transgenic cotton-growing landscapes. 150°04′52.5″E). These farms were located in the "humid subtropical"

| Study sites and sample collection
Köppen-Geiger climate zone (Kottek, Grieser, Beck, Rudolf, & Rubel, 2006) in the Brigalow Belt South Bioregion (Environment Australia, 2000). Bollgard II ® cotton (Bt-cotton), containing two genes derived from the common soil bacterium, Bacillus thuringiensis, was grown on both farms, with 10% of the cropping area dedicated to unsprayed cotton refuge crops (conventional cotton and pigeon pea).
Trapping for insectivorous bats occurred over a total of 48 nights, including eight nights in December 2014 at Site 1 and eight nights each month during the cotton-growing season (November 2015-March 2016) at Site 2. Bats were captured using mist nets (Ecotone) and harp traps (Faunatech) placed in cotton crops or on the edge of crops (Figure 1). Mist nets and harp traps were moved to a different sampling location ( Figure 1) each night at each site (Marques et al., 2013). On each occasion, we used three mist nets (12 × 2.5 m, 15 × 2.5 m, and 18 × 2.5 m; 5 shelves, Denier 110, 16 mm mesh size) and three harp traps (standard 4.2-m 2 two-bank traps) that were opened at dusk and closed at midnight. All captured bats were placed individually in clean calico bags and transferred to a holding area until the following dusk. Bat identification was undertaken using Churchill (2009) and corroborated by echolocation recording on release of each individual. Fecal pellets from each bat (2-8 pellets) were collected from the bag with sterilized tweezers. Each fecal sample was stored in a labeled sterile 1.5-ml microcentrifuge tube and frozen (−4°C) within 2 hr. Samples were then transferred to a −20°C freezer until DNA extraction. Fecal samples left in traps by escaped bats were noted as "unknown" species.
Invertebrate light traps were set in the center of cotton crops to estimate prey abundance in parallel with bat surveys at Site 2 ( Figure 1).
The custom-made light traps were specially designed to be positioned above a growing crop (Peter Gregg, personal communication, 2016). A large fiberglass cone (750 mm diameter × 500 mm high, smooth on the inside and sourced from Fiberglass Moulding Pty Ltd) was placed inside a 60-L garbage bin, with a strip of 12-V ultraviolet LED lights taped to an aluminum ring resting inside the cone to attract flying invertebrates. The LEDs were connected to a 12-V battery and 10-W solar panel (SGM-10W, Solar Australia). Traps were emptied each morning.
Invertebrate collections were immediately transferred into 70% ethanol and dried in the laboratory at 40°C prior to identification to arthropod order (>1 mm) and weighing. Invertebrates <1 mm were removed from further analysis. The percentage of each invertebrate order was then pooled with other light trap samples from the same night and farm. Given that most insectivorous bat species feed on night-flying insects, light trapping is commonly used to measure insect abundance and understand bat-prey dynamics (Froidevaux, Fialas, & Jones, 2018;Gonsalves, Law, et al., 2013b;Krauel et al., 2018;McCracken et al., 2012). To avoid the attractiveness (or avoidance) of bats to light traps, they were placed at least 250 m from where bats were captured (Froidevaux et al., 2018).
The taxonomy of OTUs was assigned using the NCBI database nucleotide Basic Local Alignment Search Tool (Altschul, Gish, Miller, Myers, & Lipman, 1990). OTU sequences were assigned to reference sequences at species level with a minimum identity threshold of 97%, reflecting natural intraspecific divergence (Alberdi et al., 2018;Elbrecht & Leese, 2017), and an e-value < 1 e−20 . There is no general rule as to dealing with identity thresholds, read errors, or discarding sequences (Pompanon et al., 2012). In this study, an OTU was classified as "unknown" when a taxonomic assignment fell below the similarity threshold or when a match was not found in the NCBI database (see Appendix S1 for further details). Those unable to be identified to species level were identified to order. Based on the taxonomic identification of OTUs, the relative abundance of sequences assigned to order and species level was used as a proxy to semiquantify the relative abundance and richness of prey species in each sample (Deagle et al., 2018). Each taxonomically assigned prey species was then allocated to one of six categories ("pest," "beneficial," "pest likely," "beneficial likely," "unknown," or "neutral") based on literature detailing their impact in natural and agricultural systems, and in cotton specifically (Appendix S2).

| Statistical analysis
All statistical analyses, diversity indices, and species accumulation curves were performed using PRIMER v7 software (Clarke & Gorley, 2015) with the PERMANOVA+ add-on (v.1.0.8; PRIMER-E, Plymouth Marine Laboratory, UK) for the analysis of nonparametric multivariate or univariate ecological datasets (Anderson, 2001). Two relative abundance datasets were used to analyze dietary composition: (a) the OTU dataset (Appendix S3) and (b) the taxonomically assigned prey species dataset (728 unique species). Dietary richness was initially explored by examining the differences in the frequency of OTUs and taxonomically assigned prey species between bat fecal samples.
OTU relative abundance was calculated using the proportion of total reads in a sample (determined by sequencing). The richness and relative abundance of each prey taxon was calculated using the OTUs taxonomically assigned to species level (728 unique species) and expressed as a percentage of identified prey species per bat sample.
To ensure that estimates of total species and subsequent species diversity calculations were reliable, species accumulation curves were constructed for the number of species and the number of OTUs in bat fecal samples, randomizing the samples 9,999 times. Four nonparametric richness estimators were used: Chao2, Bootstrap, Jackknife 1, and Jackknife 2. These emphasize the incidence of rare species, an accepted approach for estimating species richness in ecological studies (Gotelli & Colwell, 2011;Cardoso, Rigal, Borges, & Carvalho, 2014). A K-dominance curve (cumulative relative abundance against the log species rank) was constructed measuring abundance trends and inventory diversity using both datasets (results are shown in Appendix S4).
Permutational analysis of variance (PERMANOVA) was used to test for differences in richness and relative abundance of OTUs, arthropod orders, and prey species on Bray-Curtis matri-

ces. PERMANOVA is a distance-based nonparametric test with
Pseudo-F and p-values obtained using permutation techniques (Anderson, 2001, ). PERMANOVA is recommended for examining complex community composition datasets with small sample sizes (Anderson, 2001;McArdle & Anderson, 2001). Unrestricted permutations of raw data were completed using 9,999 permutations.
PERMANOVAs were conducted to assess the temporal variability in dietary composition among months of the cotton-growing season at Site 2 (n = 5: November-March), and the spatial variability between farms at Site 1 (December 2014) and Site 2 (December 2015; see results in Appendix S5). A two-way PERMANOVA was conducted with bat sex (two groups: male and female) and bat species with ≥ 6 samples (three groups: Vespadelus vulturnus, Nytophilus geoffroyi, and Chalinolobus gouldi) as factors. PERMANOVAs were also conducted on the prey species dataset with pest status of natural and modified systems (six groups: "pest," "beneficial," "pest likely," "beneficial likely," "unknown," or "neutral") and cotton pest status (six groups: as before) as factors. Following PERMANOVAs, PERMDISP was used to examine the homogeneity of dispersion (Anderson, 2006), based on mean distance to group centroid for all groups within each factor (9,999 permutations). Univariate analysis was undertaken for important single-factor comparisons using a Euclidean distance matrix (Anderson, ). Variability in composition between months (β diversity) was measured using PERMDISP on a Bray-Curtis similarity matrix (presence/absence data), which is equivalent to the Sørensen index (Anderson, 2006;Anderson, Gorley, & Clarke, 2008).
To examine how representative consumed prey species were in relation to the prey available in the cotton landscape, average taxonomic distinctness (AvTD) and variation in taxonomic distinctness (VarTD) within each bat fecal sample were examined. AvTD and VarTD measure the average taxonomic breadth and evenness of a sample between every pair of species and the mean value recorded in the species dataset under random sampling (Clarke & Warwick, 1999, 2001Ellingsen, Clarke, Somerfield, & Warwick, 2005). AvTD (Delta+) and VarTD (Lambda+) were calculated from the prey species dataset (728 species) and compared to 95% probability limits under random sampling using the TAXDTEST function, based on the Linnaean relatedness (see Appendix S1 for further information).

| Bat species richness
In total, 58 individual insectivorous bats of seven species (all Vespertilionidae) were captured at the two sites (Table 2). Of these, 12 were male, 12 female, and the remainder of unknown sex (due to not being sexed or escaping traps prior to sexing). Two species, Chalinolobus picatus and Vespadelus baverstocki, are endangered in NSW, yet are relatively common in inland and arid areas. The diversity of insectivorous bats known to forage over cotton crops is considerably higher than those species captured in this study (approximately 20 species) and highlights the difficulty of trapping echolocating bats in open areas where they may detect and avoid traps. of the relative abundance in the OTU dataset (Table 3), with the remaining 28% not meeting the species identity threshold.

| OTU richness
The species accumulation curves showed a similar pattern across all richness estimators, with the observed number of prey items detected close to saturation (see Appendix S4). All estimators reached 100% observed prey richness (asymptote) at 42 samples. Our study thus provided a reliable basis for assessing total prey species richness.
V. vulturnus notably consumed larger volumes of smaller-sized prey items (tachinid flies, mosquitoes, gnats) than other bat species, while both V. vulturnus and N. geoffroyi consumed the most lygaeid bugs.
Chalinolobus morio (chocolate wattled bat, 8.9 g) and V. baverstocki (inland forest bat, 4.6 g) consumed the most Noctuid moths in volume and richness, while C. gouldii (Gould's wattled bat, 13.8 g) consumed TA B L E 4 Relative species abundance (%) and species richness (%) of pest and beneficial arthropod orders detected in bat fecal samples the greatest volume and richness of larger prey, such as crickets (Gryllidae). Chalinolobus gouldii also consumed a smaller range of prey items, with a greater relative abundance of Coleoptera than V. vulturnus or N. geoffroyi (Appendix S7). Significant differences in richness were detected between C. gouldii and N. geoffroyi using the species incidence matrix (Pseudo-F = 1.339, p (perm) = .023), indicating that rare prey items were responsible for the differences in diet.
Taxonomically, the analysis of AvTD showed that 43% of bat fecal samples had significantly lower AvTD than expected in relation to those species available in the habitat (p ≤ .05; Figure 3a). These  (Table S6-2).

| Temporal variation in bat diet
Dietary composition based on relative abundance of prey species dif-   Figure 3b). Dietary β diversity did not vary significantly among months (F = 2.842, p (perm) = .256).

| D ISCUSS I ON
Our study demonstrates the complementary role that qualitative and semiquantitative interpretation of metabarcoding sequence reads can provide in uncovering the nuances of prey items consumed by insectivorous bats. Our results support a growing body of global evidence illustrating the significant role that insectivorous bats play in arthropod pest control. Our results also emphasize the service, rather than disservice, bats provide to agriculture, consuming a diet comprised of around 1% relative abundance and richness of beneficial insects (predators, parasitoids, pollinators).
Thus, based on pest versus beneficial insect consumption alone, the benefits of bat-mediated insect suppression in crops outweigh any disservice. Importantly, 19 cotton-specific pest species, in three arthropod orders, were detected. The most abundant arthropods in the diet of insectivorous bats were pests of summer crops, suggesting that bats were sourcing arthropods from cotton and other summer-grown crops. Furthermore, insectivorous bats consumed few unique species and many species only once, supporting evidence that bats exploit preferred locally abundant taxa (such as large pest moth population influxes) in agriculture while simultaneously consuming a wide selection of available prey (Krauel et al., 2018).
that bats relying on specific invertebrate influxes (i.e., Helicoverpa sp. and other Lepidoptera) to meet energetic requirements are not significantly affected by Bt-cotton in Australian agroecosystems due to their ability to adjust foraging behavior to hunt moths elsewhere in the landscape.
Bats also frequently consumed soft-bodied flies, but these do not contribute greatly to overall diet in terms of volume, as found in similar studies (Gonsalves, Bicknell, Law, Webb, & Monamy, 2013a;Rydell, McNeill, & Eklöf, 2002;Wetzler & Boyles, 2017). Small prey items such as mosquitoes and gnats were most common in the diet of V. vulturnus (the smallest species of bat), while larger prey items (such as crickets) were most common in the diet of C. gouldii (a larger aerial-hawking bat) and N. geoffroyi (a gleaning bat). The ability of these bats to detect insect prey is constrained by echolocation call structure and explains the variation in prey choice (Waters, Rydell, & Jones, 1995). Smaller bats tend to use high-frequency-modulated "soft" echolocation and are restricted to detecting smaller prey items, typically dipterans and moths 5-10 mm in length (Gonsalves, Bicknell, et al., 2013a;Møhl, 1988;Robert & Brigham, 1991).
However, several studies on larger aerial-hawking bats that echolocate using low-frequency and long-duration calls show that they consume a range of prey sizes (Waters et al., 1995). In addition, bats use echolocation counterstrategies (such as reducing the amplitude or shifting frequencies) to exploit difficult-to-catch prey that are typically tympanate "hearing" insects (Goerlitz, Hofstede, Zeale, Jones, & Holderied, 2010;Miller & Surlykke, 2001  TA B L E 6 (Continued) Rolfe, Kurta, & Clemans, 2014;Wickramasinghe, Harris, Jones, & Vaughan Jennings, 2004). This further demonstrates how echolocation avoidance by tympanate insects can be overcome by bats in order to exploit large food resources (Goerlitz et al., 2010;Miller & Surlykke, 2001;Surlykke & Kalko, 2008). For example, B. barbastellus (an aerial-hawking bat) consumes mainly tympanate moths using low-intensity echolocation, thus detecting moths at a closer distance allowing them less time to respond or escape capture (Goerlitz et al., 2010). Other strategies to counter moth hearing include shifting echolocation frequency out of the audible range of prey (Fenton & Fullard, 1981), broadening the echolocation beam when in the final hunting phase (Jakobsen, Olsen, & Surlykke, 2015) (Fullard, Ratcliffe, & Guignion, 2005). Furthermore, crickets are unable to exhibit echolocation evasion responses while on the ground, that is, not in flight (Fullard et al., 2005), and are thus susceptible to gleaning bats such as N. geoffroyi.
The diversity of echolocation in bats and thus detection of insect prey provide strong support for the maintenance of bat functional diversity in agricultural areas. Increased predator functional diversity improves natural pest control (Barbaro et al., 2017;Greenop, Woodcock, Wilby, Cook, & Pywell, 2018) and is important for the suppression of a range of pest insect species in crops. This is because bats exert different pressures on different insects, mediated by echolocation constraints and thus the detection and capture of prey (Waters et al., 1995). The magnitude of bat insect pest suppression changes during the night, with bats' timing roost emergence to coincide with access to preferred prey (Swift, Racey, & Avery, 1985), but is dependent on predation risk, light intensity, and life stage (Duvergé, Jones, Rydell, & Ransome, 2000;Rydell, Entwistle, & Racey, 1996). Farmers wishing to benefit from the insect pest control service provided by bats in the cotton-growing landscape can incorporate bat-mediated insect suppression into existing IPM strategies by managing a diversity of noncrop habitat and roosting sites to support different bat species foraging over crops.
The general pattern of decline in dietary α diversity over the cotton-growing season reflected the change in arthropod community composition (increased Coleoptera and reduced Lepidoptera) as the crop matured ( Figure 2). The changing structure of the arthropod community over the cotton-growing season is likely to drive temporal and spatial α and β diversity in bat diets, as bats have less oppor-

| Metabarcoding and taxonomic identity limitations
The limitations of molecular methods for dietary studies are well recognized (Krehenwinkel et al., 2017;Pompanon et al., 2012), and interpreting estimates of taxon abundance from sequence reads remains a challenge. The number of relative reads (OTUs) may not perfectly reflect dietary prey composition as a result of species differences in the amount of DNA per unit of mass or volume; digestibility and hence DNA degradation during digestion; amplification success; and recovery bias. However, it can be used to offer a semiquantitative view of diet composition and variation (Deagle et al., 2018).
Taxonomic resolution in metabarcoding-type studies of diets is limited by the choice of primer. Broad-spectrum primers may amplify nontarget species including amplicons from the predator, gut parasites, and symbionts (Deagle et al., 2007;Pompanon et al., 2012). However, primer mismatches may lead to the overrepresentation of some prey taxa or preferential amplification of other taxa and thus provide significant differences in read abundances across taxa (Deagle et al., 2007;Krehenwinkel et al., 2017 can provide a more reliable qualitative and quantitative recovery of DNA reads and thus reveal species diversity (Elbrecht & Leese, 2017;Krehenwinkel et al., 2017). Nevertheless, the Zeale primers show no significant arthropod amplification bias . Despite this, it is currently disputed whether abundance estimates can be derived from metabarcoding due to taxon-specific PCR amplification biases (Krehenwinkel et al., 2017).
One solution is to use taxon-specific correction factors, which allow species abundances to be predicted from sequencing data (Thomas et al., 2016). However, with a generalist predator consuming a diverse diet, characterizing the taxonomic composition of a large community of prey would not be feasible. Several recent studies have shown a strong correlation between input DNA and recovered read counts for most arthropod taxa (Giner et al., 2016;Krehenwinkel et al., 2017). Thus, no correction factor was applied to our data and we used the number of reads assigned to a species on the NCBI database as a proxy for the semiquantification of prey items in terms of relative abundance (Deagle et al., 2018).
While our estimates of relative abundance may differ from the true biomass proportions of taxa in the diets of the bats, we are confident in the major patterns of variation observed.

| CON CLUS ION
Australian insectivorous bats have a diverse diet in transgenic cotton landscapes (made up of a few key species and many unique species) dominated by Lepidoptera in a major cotton production zone in inland eastern Australia. This suggests that insectivorous bats are capable of selective foraging on preferred taxa (moths). Selective foraging behavior that adjusts to the available prey over the growing season benefits both bats and agriculture, as bats have access to a wider prey resource and can target pest moth population outbreaks.
Our results show that bat diets were dominated by pest arthropods of economic importance. This provides evidence for growers to integrate insectivorous bats into pest management programs, provided that bats and their habitat are conserved. Vegetation management strategies that incorporate insectivorous bat habitats and bat-friendly farm management practices are vital in maintaining and maximizing these important pest control services in intensive farming regions.

ACK N OWLED G M ENTS
Firstly, we would like to thank the anonymous reviewers who helped improve this manuscript. We would also like to thank the growers

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
HK, RS, RR, and NR designed the research and experiments. HK planned the fieldwork, collected and analyzed the data, and drafted the manuscript. All authors contributed substantially to the analytical approach and manuscript revisions and gave final approval for publication.