Do the evolutionary interactions between moths and bats promote niche partitioning between bats and birds?

Abstract Ecological theory suggests that the coexistence of species is promoted by the partitioning of available resources, as in dietary niche partitioning where predators partition prey. Yet, the mechanisms underlying dietary niche partitioning are not always clear. We used fecal DNA metabarcoding to investigate the diets of seven nocturnal insectivorous bird and bat species. Low diet overlap (2%–22%) supported resource partitioning among all species. Differences in diet corresponded with species identity, prey detection method, and foraging behavior of predators. Insects with ultrasonic hearing capabilities were consumed significantly more often by birds than bats, consistent with an evolved avoidance of echolocating strategies. In turn, bats consumed a greater proportion of noneared insects such as spruce budworms. Overall, our results suggest that evolutionary interactions among bats and moths translate to dietary niche partitioning and coexistence among bats and nocturnal birds.


| INTRODUC TI ON
Aerial insectivores like birds and bats are decreasing at alarming rates across North America (Spiller & Dettmers, 2019), in part due to simultaneous declines of aerial insects (Sánchez-Bayo & Wyckhuys, 2019). Niche theory predicts that in resource-limited environments, species that occupy the same guild will partition dietary resources to avoid competitive exclusion (MacArthur & Levins, 1964). Such partitioning is often underpinned by variations in morphology or behavior that allow species to exploit different resources (Schoener, 1974). Dietary niche partitioning related to prey size (Vesterinen et al., 2018), predator morphology, and echolocation behavior (Emrich et al., 2014) is evident among many sympatric bat species.
If and how dietary partitioning occurs among co-occurring nocturnal insectivorous birds and bats is less clear, but by identifying the processes that promote the coexistence of aerial insectivores, we can better predict future community dynamics.
Interactions between bats and moths provide a model system for studying the evolution of predator-prey relationships (Hofstede & Ratcliffe, 2016;Waters, 2003). Prey capture by bats is often dependent on echolocation behavior and how insects respond (Fenton & Fullard, 1979). Moths with ultrasound-sensitive ears can hear echolocation calls at distances up to 100 m (e.g., noctuids; Miller & Surlykke, 2001) and avoid predation through evasive maneuvers or sounds (Dunning et al., 1996). This adaptation arose independently in moths at least six times (Hofstede & Ratcliffe, 2016). In turn, some bats echolocate at low intensities or high enough frequencies to go undetected by moths (Faure et al., 1990;Hofstede & Ratcliffe, 2016 Yet, how evolutionary interactions between moths and bats may extend to dietary resource partitioning between bats and nocturnal insectivorous birds is unknown (Yack et al., 2020). Nocturnal birds often use visual cues and possess adaptions for silent flight that enable them to evade detection by insects (Clark et al., 2020). These adaptations may allow them to exploit resources that bats cannot.
For example, eared moths can only detect the cyclic wingbeats of approaching birds within 2.5 m (Fournier et al., 2013), perhaps making moths more vulnerable to predation by visually-oriented insectivores. The distributions of bats and nocturnal insectivorous birds suggest that they may interact. However, little research exists on if, or to what extent they may partition prey resources, or the underlying mechanisms (Fenton & Fleming, 1976).
We used fecal DNA metabarcoding to analyze the diets of seven co-occurring nocturnal aerial insectivores (hereafter NAIs). We compared the diet composition and richness of three nocturnal birds: Myotis evotis (Western Long-eared Myotis). Despite differences in prey detection methods used by these insectivores (Table 1), previous studies using microscopy of fecal samples have reported broad similarities in the insects they consume, primarily moths and beetles (Agosta, 2002;Csada et al., 1992;Ober & Hayes, 2008;Reynolds & Linkhart, 1987;Todd et al., 1998;Whitaker, 1995). However, NAI diets and available prey can vary across regions and over time, hampering cross-study comparisons. Additionally, traditional methods of prey analysis in feces primarily result in prey identification to only the order or family level, which masks resource partitioning at finer taxonomic resolutions.
As with differences in prey detection methods, NAIs in this study also display different foraging behaviors. For example, Flammulated Owls (Goggans, 1985) and Common Poorwills (Brigham & Barclay, 1992) are sit-and-wait predators (Table 1). Both use their legs to launch after prey from the ground or perches, a foraging behavior not found in insectivorous bats. Modifications of the pelvis that allow bats to hang from perches and fly prevent bats from jumping into flight (Schutt et al., 1997). Instead, the bats in this study hunt by foraging insects while in flight, termed "aerial hawking" (Saunders & Barclay, 1992), or, as in Long-eared Myotis, sometimes also by gleaning insects from the ground and foliage (Faure & Barclay, 1994).
Like bats, Common Nighthawks are also aerial hawkers and prey on insects at a wide range of heights above ground and over great distances in a single foraging bout (Clark et al., 2020).
Despite clear differences in prey detection and foraging behavior of insectivores, it is not always clear if or to what extent these differences translate to differences in diet. Insectivores with different foraging behaviors may still target the same prey (Brigham & Fenton, 1991;Kent & Sherry, 2020). Prey movement may also overlap with the foraging range of more than one predator species (Remmel et al., 2011). Still, foraging behaviors and prey detection methods that do correspond to dietary differences may decrease interspecific competition among NAIs.
To our knowledge, this is the first study to use fecal DNA metabarcoding to investigate the diets of multiple, distantly related, cooccurring NAIs. Our objectives were two-fold. First, we developed a reference barcode database from 56,191 locally collected arthropod specimens to provide more accurate taxonomic assignments of potential prey items than possible in previous studies. We then used DNA metabarcoding of fecal samples to determine the degree to which NAI diets differ in richness and composition. We expected that differences in diet would depend on NAI species identity and correspond with (1) prey detection methods (i.e., echolocation or visual hunting) and (2) differences in foraging behavior (i.e., aerial hawking or sit-and-wait predators).

| Sample collection and processing
We collected fecal samples from NAIs May through September during 2017 and 2018. We captured bats monthly after evening emergence in mist nets set over dry land, streams, and ponds. We placed bats in individual paper bags to collect their fecal pellets. Six additional bat species occur in the study area but were excluded from this study due to low sample sizes. We collected fresh fecal samples from Flammulated Owls, Common Poorwills, and Common Nighthawks captured in mist nets on or near breeding territories. Brown Bats (Maxell, 2015). Based on observational and telemetry data, home and foraging ranges overlapped for all species.
The Canadian Centre for DNA Barcoding (CCDB) performed all DNA extractions, amplification, and sequencing. DNA extraction and PCR amplification followed CCDB protocols as described in Moran et al. (2019). Samples were incubated overnight in a lysis buffer, concentrated by centrifugation, dried, and finally eluted using a Tris-HCl elution buffer. The CCDB also processed negative extraction and PCR controls in parallel with samples. All negative controls ensured that contamination did not occur. The cytochrome C oxidase 1 (CO1) region was amplified from each sample using the arthropod-specific primers, ZBJ-ArtF1c_t1 and ZBJ-ArtR2_t1 , as described previously (Moran et al., 2019;Prosser & Hebert, 2017). Following amplification, samples were pooled and purified. The CCDB performed sequencing on an Ion Torrent PGM following standard protocols (Prosser & Hebert, 2017).

| Constructing a DNA barcode library from local Arthropoda
In 2017 Figure A1).
DADA2 is sensitive to single base-pair differences among sequences and produces unique "amplicon sequence variants" (ASVs). The median base pair quality score for all sequences was maintained above 25. Denoised sequences shorter than 100 bp were removed from analyses. This resulted in a total of 1,450,971 quality-filtered sequences. We then clustered sequences into operational taxonomic units (OTUs) based on a 97% sequence similarity threshold (Vamos et al., 2017), using the VSEARCH plugin (Rognes et al., 2016). We removed sequences only occurring in a single sample or that were represented by fewer than 0.001% of sequences to limit artifactual sequences.
We determined taxonomic assignments using our local DNA barcode library and the BLAST plugin within QIIME2, with a coverage value of 0.7 and sequential percent matching identities of 100%, 99%, 98%, and 97%. If taxonomy could not be assigned to our local database using these parameters, we used a global COI database compiled from BOLD and GenBank and a pretrained RDP classifier (Porter & Hajibabaei, 2018;Wang et al., 2007). The BOLD accession ID associated with each taxonomic identification is indicated where available. We verified all taxonomic identifications based on the plausibility that they may occur within or nearby the study area. All sequences not matching to Arthropoda using either the global COI database or the local database were removed from further analyses, resulting in a total of 1,147,127 sequences with assigned taxonomy.
We rarefied samples at 500 sequences per sample, which was sufficient to adequately characterize most species within each sample ( Figure A2). In total, 77% of OTUs recovered from NAI fecal samples matched 97% or greater with locally collected specimens, while 23% were assigned taxonomy using the RDP classifier.
All statistical analyses were conducted in RStudio Version  (Benjamini & Hochberg, 1995). Insect taxa <200 reads were removed prior to analysis.
To determine resource partitioning among NAI diet composition, we analyzed both relative read abundance and presence/absence data. Presence/absence data are considered a more conservative option in insectivore fecal analyses (Jusino et al., 2019). However, presence/absence data can also overestimate the importance of prey consumed in small quantities, and it is generally thought that relative read abundances provide more accurate population-level data (Deagle et al., 2019). Even so, we chose to analyze both relative read abundance and presence/absence data and found similar results. We performed all compositional comparisons on either Bray-Curtis distances of Hellinger transformed relative read abundances, or Raup-Crick transformed presence/absence data using the vegan package (Oksanen et al., 2019).
Because differences in diet composition can stem from differences among group centroids or group dispersions, we tested for both at the OTU level. We assessed differences in diet dispersion (distance from mean) among species, prey detection methods, and foraging behaviors using the betadisper() function in the vegan package (Oksanen et al., 2019). We observed no differences in data dispersion among species or groups of species (p > .1).
A perMANOVA analysis was performed to test the effects of prey detection method, foraging behavior, species identity, sampling month, plant community, the presence of a water body at the sampling site, and all interactions, on abundance and presence/absence data using the adonis2 function, with permutations constrained within collection year. We applied forward selection to successively add predictor variables that significantly (p < .05) improved model fit. To additionally test for differences in diet for each species pair, we ran pairwise analyses using the "pairwiseAdonis2" function in the pairwiseAdonis package (Arbizu, 2021), and adjusted for multiple comparisons (Benjamini & Hochberg, 1995). We performed a principal coordinate analysis (PCoA) using the "cmdscale" function to visualize diet variation among species. For each NAI, we also calculated diet turnover among samples using the "turnover" function in the vegetarian package (Charney & Record, 2012). Diet turnover within each NAI species was calculated based on Shannon beta diversity where zero equals no difference between samples and one represents completely different samples. Standard error was estimated for diet turnover through bootstrapping, with 500 iterations.
Overlap in diet was calculated based on the proportion of OTUs common to each species pair.
To accommodate non-normal error distributions associated with richness and diversity metrics, we used generalized linear regression using the "glm" function with Gaussian or Poisson distributions to assess variation based on species, plant community, and collection month. The Akaike information criterion was used to select the best models. We used a two-way Anova (car package, Fox & Weisberg, 2019) with a type II sum of squares for unbalanced data to test significance of predictor variables (Table A8 in Appendix 1). Where significant, the "emmeans" function in the emmeans package was used for pairwise analyses of diversity metrics between each species (Lenth et al., 2021).

| Dietary partitioning corresponds with predation of eared insects
Moth families that contain species with ears (Miller & Surlykke, 2001) were maximally associated with the diets of NAIs that hunt visually (p < .01; Figure 1a; Table 2). The most abundant eared family, F I G U R E 1 Variation in composition and richness of insectivore diets. (a) The percent relative sequence abundance of arthropod families found in the diets of seven nocturnal aerial insectivores. The size of points indicates the percent relative sequence abundance within each species and red outline indicates arthropod families significantly associated with the diet of an individual insectivore. Asterisks indicate the prey families maximally associated with each predator ( † p ≤ .07; *p ≤ .05) based on indicator species analyses ( Other eared moth families, including Geometridae, Sphingidae, and Erebidae, also occurred significantly more often in visual hunters' diets, but rarely in bat diets (0%-15%).
Conversely, the noneared moth family Tortricidae (mostly spruce budworm), was the most abundant family consumed more often by echolocators than by visual hunters (p < .001). We found Tortricidae

| Dietary breadth and turnover
From the fecal samples of all seven NAI species, we identified 73 arthropod families, 165 genera, and 382 OTUs. Silver-haired bats had the widest diet breadth at the order (10) and family (36) levels ( Table   A7 in Appendix 1), whereas Common Poorwills consumed the greatest number of insect genera (75) and putative species or OTUs (154).
We detected the fewest total OTUs in Common Nighthawk samples (50). Flammulated Owls had the highest variation or turnover among

samples, whereas Common Poorwills, Common Nighthawks, and
Silver-haired bats had the lowest (Table 1). Long-legged Myotis had the most OTU-rich diet on average (Figure 1c), consuming more prey

OTUs than Common Nighthawks, Big Brown Bats, and Long-eared
Myotis (p < .001; Table A8 in Appendix 1). NAI species, collection month, year, and plant community were significant predictors of dietary richness. Species identity had the greatest influence.

| Eared moths are eaten more often by nocturnal birds than bats
In this study, we observed previously unreported dietary partitioning among co-occurring nocturnal aerial insectivorous birds and bats. Flammulated Owls also fly quietly and possess relatively long wings that allow them to move quickly (though perhaps without much agility) throughout the forest canopy (Johnson, 1997). Rather than aerial hawking, Flammulated Owls, like Common Poorwills, primarily use a sit-and-wait hunting strategy. This consists of flying from a perch inside the tree crown to capture insects resting in other areas of the same crown or adjacent trees (Reynolds & Linkhart, 1987). Together, these results indicate that birds that can ambush prey, rather than alert them with echolocation calls, can initiate successful attacks on eared insects at closer ranges.
The lower occurrence of eared moths in bat diets demonstrates the effectiveness of moth adaptations to bat predation (Hofstede & Ratcliffe, 2016). Still, Long-legged and Long-eared Myotis tended to consume eared moths at higher rates than the other bats in this study. Long-legged Myotis makes echolocation calls at higher frequencies and detects prey at greater distances than Big Brown Bats and other myotis species, which may give it an advantage (Fenton & Bell, 1979;Saunders & Barclay, 1992). Alternatively, Long-eared Myotis uses passive hearing and low-amplitude calls while gleaning, which are undetectable by some eared moths (Faure et al., 1990).
Gleaning by Myotis species evolved subsequent to echolocation strategies (Morales et al., 2019) and may be a counteradaptation to reduce detection by eared prey (Razak, 2018). However, gleaning may also have evolved as a general adaptation to hunting in cluttered areas (Brinkløv et al., 2010). An obvious counterstrategy to eared prey would be for bats to use a sit-and-wait hunting strategy.
However, the physiology of most bats precludes them from leaping into flight (Schutt et al., 1997).
In addition to moths, ultrasonic hearing via tympanal organs has evolved independently within at least eight other insect orders, including Orthoptera, Mantodea, Blattodea, Hemiptera, Hymenoptera, Coleoptera, Neuroptera, and Diptera (Göpfert & Hennig, 2016;Hoy & Robert, 1996). Besides serving to detect and avoid predators, insect hearing has also evolved as a means of communication (Hoy & Robert, 1996). In Neuroptera, green lacewings can detect ultrasonic frequencies and avoid predation by bats (Miller, 1975), and a recent study indicates a similar ability in Myrmeleontidae of the Neuroptera (antlions) (Holderied et al., 2018). However, no insect family with known tympanal hearing abilities was significantly associated with bat diets in this study. Other insect families have evolved different mechanisms of hearing (e.g., Culicidae; Hoy & Robert, 1996), however these insects did not appear to avoid detection by bats more than birds.

| Noneared prey partitioning among bats and birds
Though these results show a clear link between the ultrasonic hearing of moths and their higher occurrence in bird diets compared with bats, the partitioning of noneared insects is less clear. Moths in the family Tortricidae lack hearing organs (Fullard & Napoleone, 2001). This may explain why bats consumed Tortricidae in such high amounts and more often than nocturnal birds. The most commonly consumed Tortricidae moths, spruce budworms, tend to fly near treetops (Soutar & Fullard, 2004). Bat species in this study are known to forage in or near the forest canopy (Faure & Barclay, 1994;Menzel et al., 2005). Common Nighthawks that hunt high above the ground and Flammulated Owls that hawk from tree perches would also still encounter spruce budworm. Indeed, 11% of nighthawks and 31% of Flammulated Owls consumed Tortricidae in this study. However, for Common Poorwills that generally hunt only up to three meters above ground (Brigham & Barclay, 1992), spruce budworm may often be out of range. This would explain why Common Poorwills preyed on Tortricidae moths less often than all the other NAIs. proportions of Limoniidae (Table 4; Table A4 in Appendix 1), and Culicidae (mosquito family; Figure 1a) which may be more available to aerial hawkers that can forage over water bodies, than to sit-andwait predators. Indeed, Culicidae was not found in the diet of any Common Poorwill or Flammulated Owl in our study. Previous investigations found that Common Poorwills only consumed prey >5 mm in length, despite a higher abundance of smaller insects in the environment, potentially due to visual constraints (Bayne & Brigham, 1995). We did not find any evidence contradicting this. However, since we used DNA instead of morphology to identify prey, we were unable to definitively determine prey size in many cases.
Previous studies suggest that variation in echolocation calls leads sympatric bat species to detect different prey resources, enabling coexistence (Razgour et al., 2011). However, such diet partitioning has not been shown empirically among the assemblage of bats in our study. Although overall diet composition did not differ or only marginally differed among Big Brown Bats, Long-legged Myotis and Long-eared Myotis (perMANOVA), we observed low overlap in the insect taxa consumed (18%-22%), suggesting some specialization. This pattern indicates that although these bats consume high abundances of the same species (i.e., spruce budworm), coexistence may be promoted due to differences in species consumed at lower frequencies. This hypothesis was also supported by stronger differences among species when analyzing presence/absence data compared with relative abundances, which is less sensitive to rare

| Conservation implications
North American avifauna have decreased in abundance by approximately 29% since 1970 (Rosenberg et al., 2019). Aerial insectivores are even more threatened (Nebel et al., 2010;Spiller & Dettmers, 2019). Bats face conservation threats globally and regionally (Frick et al., 2020). Though many factors contribute to declining population trends, decreases or changes in food availability play a role, making identification of key food sources important (Rosenberg et al., 2019;Spiller & Dettmers, 2019 especially, were often not resolved beyond the family level here, yet were the most common order found in the diet of four of the seven NAIs. Crane flies constitute the majority of prey for various wildlife, including snails, salamanders, and other Arthropoda (Lunghi et al., 2020), in addition to the species observed here (Table 4). A recent study found that crane fly abundance was a key predictor of the persistence of multiple sympatric bird species, and explained 39% of observed bird abundance (Carroll et al., 2015). This suggests that any decline in crane fly populations may be paired with future declines in avian populations. Monitoring crane fly populations may help identify high conservation priority areas as these insects are susceptible to plant community degradation and loss (Yadamsuren et al., 2015) and changes in water quality (Morse et al., 1994). Crane fly larvae in particular, are susceptible to desiccation (Pritchard, 1983), and prolonged drought or extreme heat caused by ongoing climate change may harm crane fly populations (Carroll et al., 2011). The importance of crane flies in NAI diets highlights the need for expanded analyses on crane fly ecology and conservation, especially as many species have yet to be described (Marshall, 2012).
Knowledge of NAI diets can also identify regulators of unwanted pests such as western spruce budworm, cutworm moths, and Douglas fir tussock moths that cause crop and forest damage.
Western spruce budworm in particular, is a common conifer defoliator that reduces tree growth in the Pacific Northwest (Fierravanti et al., 2019). Because NAIs consume pests like spruce budworm in high and variable proportions, future research into the possible cascading effects on forest biomass and soil carbon retention may have global implications (Schmitz et al., 2017). Overall, our findings indicate that the evolutionary interactions between bats and moths may promote the coexistence of multi-phyla predator communities. Future management practices that promote both eared and noneared prey insects may add stability to already threatened insectivore populations.

ACK N OWLED G M ENTS
The authors thank Mike McTee, Beau Larkin, and Ylva Lekberg, who provided valuable comments on earlier drafts of this manuscript.
They are grateful to Dr. Jon K. Gelhaus for identification of crane fly specimens. They also thank MPG Ranch for funding this research.

CO N FLI C T O F I NTE R E S T
None declared. Note: Insect families significantly associated with each foraging behavior using the "multipatt" function and 9999 permutations. p-values were corrected for multiple comparisons. Insect families containing eared moths are indicated with an asterisk.

TA B L E A 7 Dietary niche breadth
Orders Families Genera OTUs