2.1 Study Area

Our two study locations were in the Hill Country of central Texas. This region is notable for its karst topography and hills of limestone or granite. Many caves can be found throughout Hill Country and are used by different bat species at different times of the year. We targeted areas outside two caves in Texas Hill Country. Because these caves are located on private land, we obscure their names and precise locality information. Cave 1 in Comal County had a population of several hundred Myotis velifer (cave bats) between the spring and fall, with smaller populations of M. velifer and P. subflavus in the winter. Cave 2 in Williamson County had an overwintering population of P. subflavus, with generally fewer than 30 individuals. Our field efforts were carried out from October 2022 to January 2023.

2.2 Experimental Setup

Following Frick et al. (2023), at each cave, we set up a treatment site with an artificial prey patch with a 320–400 nm UV light (model 2851 L; BioQuip) and a paired unlit control site. The sites were 55 m away from the entrance of Cave 1 and 71 m away from the entrance of Cave 2. At each site, we monitored bat foraging and overall activity with passive acoustic monitoring using a bat detector (Song Meter Mini Bat Ultrasonic Recorders; Wildlife Acoustics, Maynard, MD, USA) mounted on a 3 m pole, 3 m from the UV light and eDNA sampler (described below). Detectors recorded from 30 min before sunset to 30 min after sunrise using a 256 kHz sample rate, minimum trigger frequency of 16 kHz, trigger level of 12 dB, and a trigger window of 3 s. While acoustic monitoring was simultaneous at both caves and all sites, differences in battery life resulted in varying survey efforts. Therefore, we only retained nightly data whereby both treatment and control groups were recorded simultaneously at a given cave location. This included recordings at Cave 1 for 47 nights and Cave 2 for 55 nights.

2.3 Acoustic Processing

Following Frick et al. (2023), we processed echolocation recordings using SonoBat Software v.4.4.5 (SonoBat, Arcata, CA, USA) set to the north central Texas bat species classifier for automated signal classification using Sonobatch. Acceptable call quality was set to 0.80, with a sequence decision threshold set to 0.9 and max number of calls per file set to 16. Feeding buzzes were identified using SonoBat 30. We manually confirmed the auto-identification of all P. subflavus recordings and confirmed species presence at each site. We tabulated feeding buzzes and overall activity based on each identified file being counted as a single pass.

2.4 Statistical Analysis

We assumed that if the detectors recorded feeding buzzes, the bats were in close enough proximity to the eDNA samplers to collect their DNA. Additionally, we assumed that increased activity at a site would increase our chances of detecting bats through passive air sampling. To confirm that bats used the artificial prey patch sites equal to or greater than the control sites, following Frick et al. (2023), we fit a generalized linear model with a negative binomial error distribution using the lme4 R package (Bates et al. 2015), with overall bat activity (number of files) as the response variable, the presence (treatment) or absence (control) of UV lights as a fixed effect, and the cave location as a mixed effect. We used the DHARMa R package (Hartig 2024) to verify conformance of the error distribution and non-effects of zero-inflation and deviance. As a null model, we also fit and verified models in a similar manner without UV presence as a fixed effect, accepting the model with the lowest Akaike Information Criterion (AIC) and a ΔAIC of 2 to the next highest value. We back-transformed the β estimate to derive the effect size (i.e., rate ratio) of light lure presence, calculating 95% confidence intervals for that estimate.

2.5 eDNA Sample Collection

At prey patch sites where light lures were deployed, we collected eDNA with the Tisch Model TE-200 Passive Air Sampler (Figure 1) loaded with two types of eDNA accumulation media: polyurethane foam (PUF, TE-1014, 14 cm diameter, 1.27 cm thickness) and a Purafilter 2000 Vent Filter (VF, PFVENT-12, 10 × 30 cm) for comparison. PUF and VF media were placed in the same sampler, stacked on top of one another (Figure 1). Before loading media, we sprayed the sampler with a 5% bleach solution, allowing it to sit for 5 min before wiping with clean paper towels and then wiping with a sterile alcohol pad. Guidance for field blanks is limited for airborne eDNA; however, in this study, we collected field blanks by loading the media into the sampler, closing it, then immediately retrieving the media with freshly gloved hands, and storing it in a plastic re-sealable bag. As these may not be considered true field blanks, we categorized these as t0 (time = 0) samples. For field-tested samples, we loaded and deployed samplers approximately 3 m downwind of the UV light. We collected and replaced the eDNA media at intervals ranging from 1 to 5 weeks from October 2022 through January 2023. Variability in sampling intervals was unavoidable due to logistical constraints (e.g., driving distance, difficult terrain, and scheduling conflicts). Each field sample medium was placed into a separate re-sealable bag. Samples were immediately stored in a freezer (−20°C) prior to laboratory processing. In total, media were derived from 24 eDNA samples for testing with 5 PUF (+1 t0) and 5 VF (+2 t0) media from Cave 1 and 6 PUF (+1 t0) and 4 VF (no t0 due to loss between field collection and shipment) from Cave 2.

2.6 DNA Extraction

We compared three methods of DNA extraction. We performed DNA extraction directly from airborne media and used two indirect methods, whereby media were soaked in aqueous suspension to resuspend eDNA and then concentrated using filtration or pelleted via centrifugation (Johnson et al. 2019; Lynggaard et al. 2022; Allen et al. 2023; Frere et al. 2023). We performed all work in a Type II biological safety cabinet. All supplies used to manipulate or make direct contact with airborne media were first UV decontaminated, soaked in 50% bleach (Goldberg et al. 2016), and then in 70% molecular grade EtOH. All other surfaces and supplies were decontaminated with 10% bleach or DNA AWAY (ThermoFisher Scientific, Waltham, MA, USA). We used filtered pipette tips for DNA extraction and all downstream experiments. We included extraction blanks, which we prepared alongside batches of samples with the same reagents.

For direct extraction, due to the large sizes of the air sampling media, we excised ten 1 × 1 cm sections from each of six randomly sampled media (n = 60 excisions). That is, we retained 3 PUF and 3 VF media, including one field blank (Cave 1 = 5 and Cave 2 = 1). We extracted DNA from the direct excisions using DNeasy Blood and Tissue protocols (Qiagen, Valencia, CA, USA), modifying with an overnight lysis and a 100 μL DNA elution volume.

We extracted from all 24 eDNA media using the filtration and pelleting methods. For the filtration method, we excised narrow strips (1 × 8 cm) and transferred them to a 50 mL vial containing a custom mixture of differently sized zirconium beads (Wagner et al. 2022): 1 g of 1.0 mm beads (BioSpec Products 11079110Z) and 2 g of 0.1 mm beads (BioSpec Products 11079101Z). Each vial also contained 30 mL PCR grade water. To dislodge eDNA-bearing particulates, vials were subject to vigorous bead beating at 1000 rpm for 5 min. We decanted the water in each vial directly into the barrel of a sterile, single-use 60 mL syringe (Fisher Scientific: 14-955-461). Some residual suspension would be absorbed by the air sampling media, so we used sterile, single-use cotton tip applicators to squeeze out the remaining volume. Syringes were attached via Luer lock to reusable filter casings, which housed a 25 mm cellulose acetate membrane with a pore size of 0.45 μm (Whatman WHA70000002). Following syringe filtration, filters were transferred to a 2 mL vial containing 1.0 mm (0.5 g) and 0.1 mm (1 g) zirconium beads (Wagner et al. 2022). Filters were rolled flatly against the contour of the inside of a vial such that the filtered particulates faced inward. We added 720 μL buffer ATL and 80 μL proteinase K to each bead vial containing a filter. The filters were then subjected to bead beating at 1400 rpm for 15 min. Samples were lysed overnight at 56°C and then vortexed in the same manner prior to DNA purification. We continued the Blood and Tissue protocol as recommended by the manufacturer. The entire volume of lysate was run through spin-columns, and DNA was eluted 2 × 50 μL using the same eluate (total volume = 50 μL). To facilitate elution, buffer AE was prewarmed to 56°C (Wagner et al. 2022) and then incubated at 37°C for 15 min prior to each elution spin (Lynggaard et al. 2022).

For the pellet-based centrifugation method, the entirety of the remaining eDNA media was placed in an individual resealable plastic bag and filled with 50 mL of PCR-grade water. We then alternated between vigorously shaking the bag and massaging the filter for 1 min to suspend the eDNA in solution (Allen et al. 2023). We transferred the PCR water into a new 50 mL vial using a sterile serological pipette and centrifuged at 3000× g for 20 min to create an eDNA pellet. Following the removal of the supernatant, we resuspended the pellet in 180 μL buffer ATL and transferred the solution to a microcentrifuge tube containing 20 μL proteinase K before subjecting the samples to overnight lysis at 56°C. We continued the Blood and Tissue protocol as recommended by the manufacturer. The entire volume of lysate was run through spin-columns, and DNA was eluted 2 × 50 μL to a total volume of 100 μL.

2.7 DNA Metabarcoding

PCR master mixes were prepared in a DNA-free room, with genomic DNA template added in a decontaminated PCR hood (UV, DNA AWAY) in a pre-PCR room and amplicon libraries prepared in a post-PCR room. For the pelleting extraction batch, we prepared libraries in a separate high-sensitivity room (intended for RNA library preparation and targeted enrichment) where bat-derived samples have never been worked with. In the high-sensitivity room, we worked in a decontaminated laminar flow PCR hood (Lynggaard et al. 2024), using only dedicated, freshly opened reagents. Our primary objective was bat identification, so we selected a bat-specific primer set (Walker et al. 2016) for DNA metabarcoding. The primers, SFF145f and SFF351r, amplify a 202 bp insert of cytochrome oxidase subunit I. The primers were modified with 5' universal tails for two-step library preparation and flexible, high-fidelity indexing (Colman et al. 2015). PCR reactions were carried out in 15 μL reactions with 8.46 μL PCR-grade water, 1.5 μL 10X Mg-free PCR buffer (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA), 1.5 mM MgCl₂, 0.2 mM each dNTP, 0.03 U/μL Platinum Taq polymerase, 0.16 μg/μL bovine serum albumin (Ambion, Life Technologies, Carlsbad, CA, USA), 0.2 μM each 5'-modified primer, and 3 μL DNA template (neat). Cycling conditions for this primer set consisted of 94°C for 5 min; 5 cycles of 94°C for 1 min, 45°C for 1.5 min, and 72°C for 1 min; 35 cycles of 94°C for 1 min, 60°C for 1.5 min, and 72°C for 1 min; and a final extension step at 72°C for 10 min. Samples derived from the direct extraction were run without PCR replicates but were re-prepared if any bat detection did occur. For this sequencing batch, we included a PCR mock community containing the genomic DNA of nine bat species (Leptonycteris nivalis, Eptesicus fuscus, Eumops perotis, Lasionycteris noctivagans, Lasiurus cinereus, Myotis occultus, Nyctinomops macrotis, Tadarida brasiliensis, and Euderma maculatum). Each member of the mock community was standardized to 1 ng/μL and pooled at equal volume. However, later recognizing that the mock community used in this sequencing batch contained members that could occur in the study area and confound results if contamination occurred, we changed the positive control to one containing gBlocks synthetic COI (657 bp) templates of up to seven bat species not occurring in the United States or Canada. These sequences were downloaded from the Barcode of Life Database (Ratnasingham and Hebert 2007) and included Otomops martiensseni (ABBWP311-07), Scotophilus dinganii (SKBET055-07), Desmodus rotundus (ABTVC131-05), Phyllostomus latifolius (ABGYG450-06), Rhinolophus blasii (ABBWP069-06), Rhinolophus hipposideros (SKBPA561-08), and Vespertilio murinus (SKBPA194-07). Combinations of these members were run alongside the filtration (seven mock members) and pellet samples (five mock members—excluding P. latifolius and D. rotundus for the requirements of a separate, co-occurring eDNA project). All samples deriving from filtration or pelleting were prepared in PCR duplicate. All samples, regardless of extraction method, were run alongside PCR negative controls, whereby PCR-grade water was substituted for DNA template. We used a second PCR step to extend unique indices and sequencing adapters to each library in a non-combinatorial manner. Reactions were carried out in 25 μL reaction volumes with 8.5 μL PCR-grade water, 12.5 μL 2X Kapa HiFi HotStart ReadyMix (Roche Sequencing, Wilmington, MA, USA), 1 μL 10 μM primers containing a unique 8 bp index and i5 or i7 Illumina adapters (Colman et al. 2015), and 2 μL amplicon from the previous PCR step. Cycling conditions consisted of 98°C for 2 min; 15 cycles of 98°C for 30 s, 60°C for 20 s, and 72°C for 5 min; and a final extension of 72°C for 5 min. Following verification of amplification with gel electrophoresis, libraries were normalized using a SequalPrep Normalization Plate Kit (Thermo Fisher Scientific). Pooled libraries were sequenced on an Illumina MiSeq v2 500 cycle kit (Illumina, San Diego, CA, USA) among three sequencing runs.

For data processing, we removed priming regions and read-through using cutadapt v4.4 (Martin 2011). Using QIIME2 v2023.5 (Bolyen et al. 2019), we ran DADA2 (Callahan et al. 2016) by truncating R1 and R2 reads to 202 bp (based on initial inspection of quality plots), filtering by quality, joining paired-end reads, and removing chimeric sequences. All other DADA2 parameters were run with default settings. We then retained amplicon sequence variants (ASVs) that were exactly 202 bp in length, matching the expected insert size for this primer set (Walker et al. 2016, 2022). To avoid PCR errors, we post-clustered ASVs into operational taxonomic units (OTUs) using LULU (Frøslev et al. 2017; Walker et al. 2022). As part of our LULU parameters, we used the default minimum match threshold of 84% to detect co-occurrence between putative parent ASVs and putatively erroneous ASVs within the same sample. Note that settings between 84% and 90% for this detection threshold are typically adequate for COI (Liu et al. 2020; Brandt et al. 2021). If co-occurrence was detected, an erroneous ASV would be classified as such if its read abundance was at all lower than its putative parent (ratio type = “min”; minimum ratio = 1) and would only be aggregated with its parent ASV if this abundance relationship was frequent across the dataset (minimum relative co-occurrence rate = 0.99). We also verified that LULU-curated OTUs in our dataset produced the same list of taxa as the original ASVs to ensure we alleviated redundancy while preserving taxonomically distinct variants. OTU taxonomy was first classified using Naïve Bayes classification (Bokulich et al. 2018) against the bat-specific, Species from Feces reference library (Walker et al. 2016) with a 90% confidence threshold, elevated from its recommended threshold (Walker et al. 2016) to prioritize precision. Any OTU not classified to the species level using Naïve Bayes classification was screened with a BLAST homology search (Altschul et al. 1990) against the National Center for Biotechnology Information's (NCBI) Genbank (Benson et al. 2009) non-redundant nucleotide collection (nr/nt) database. We derived consensus classifications from those results using the least common ancestor (LCA) method in MEGAN6 v2.11.2020 (Huson et al. 2007) with previously described parameters (Walker et al. 2022). We set a conservative read abundance threshold (Garrett, Watkins, Francis, et al. 2023), excluding any classifications falling below the total number of reads observed among PCR negative controls (185 reads), regardless of which sequencing run it was derived from. We excluded any expected mock community OTUs that occurred in field-collected samples sequenced alongside them (metadata available in Data S3). We also excluded OTUs aligning to known pseudogenes or other sources of nuclear DNA (all were human detections). We removed detections of fish species (class = Actinopteri), like Phoxinus phoxinus (Common minnow) and Cottus spp. (Sculpin), which amplified in DNA extraction blanks or were not expected in the study area. Our laboratory very rarely works with eDNA from fish, and we have previously noted unexpected detections of Actinopteri in samples and controls of other internal projects. We identified an unexpected bat species (Myotis lucifugus) occurring in five libraries and at a high read count (median = 66,631 reads) in both direct and filtration extraction batches but not in PCR negative controls and DNA extraction blanks. This is a commonly processed bat species in our laboratory, so we omitted those OTUs as potential contaminants. We speculated that this could be a contaminant from our standard amplicon room. When we subsequently moved the pelleted extraction panel to the high-sensitivity room and laminar flow PCR hoods (described above), we no longer identified this species in our sequencing data.

2.8 qPCR Detection of P. subflavus

We designed candidate qPCR assays targeting P. subflavus using cytochrome oxidase I (COI) sequences. We used all existing 657 bp variants for P. subflavus from BOLD and designed hydrolysis probe-based assays (Data S1) using both PrimerQuest (Integrated DNA Technologies) and PrimerProspector v1.0.1 (Walters et al. 2011). Selecting 3 candidate assays for testing (Table 1), we optimized qPCR conditions and evaluated specificity and sensitivity according to recommended procedures (Klymus, Ramos, et al. 2020; Klymus, Merkes, et al. 2020). We optimized assay conditions using a gBlocks synthetic sequence of P. subflavus (BOLD process-id: GBMA4410-13), which we diluted to 10,577 copies/μL. Each assay included a FAM-labeled, double-quenched hydrolysis probe (31ABkF1 with ZEN quencher, Integrated DNA Technologies). We ran all qPCR experiments on a QuantStudio 7 Pro Real-Time PCR System (ThermoFisher Scientific, Waltham, MA, USA). We first checked optimal annealing temperatures in 15 μL qPCR reaction volumes with 7.5 μL 2X Environmental Mastermix 2.0 (Life Technologies, Carlsbad, CA, US), 900 nM each primer, 250 nM hydrolysis probe, and 3 μL DNA template. PCR-grade water was added to the remainder of the reaction volume. Cycling conditions included a hot start cycle of 95°C for 10 min; 45 cycles of 95°C for 15 s, a gradient of annealing temperatures (46°C, 49°C, 51°C, 54°C, 57°C, and 60°C) for 30 s on the same qPCR run, followed by an extension step of 72°C for 30 s. After selecting cycling parameters, we identified optimal primer and probe combinations in 16 combinations. Forward and reverse primer concentrations were varied at 100, 300, 600, and 900 nM, with probe concentrations held constant at 250 nM (Wilcox et al. 2015). Using optimized qPCR parameters, we screened each assay against non-target panels that were selected based on in silico PCR, mismatch analysis, and potential co-occurrence with the target species (Table 2) of gBlocks templates (dilutions ranging from 185,670–257,404 copies/μL) and in-house gDNA (1 ng/μL) in triplicate. Any amplification from a non-target sequence was sequenced for verification with Illumina sequencing according to previously described methods (Walker et al. 2022). We estimated the limit of detection (95% detection in a single qPCR replicate) using probit modeling (Klymus, Merkes, et al. 2020). This method allowed us to predict an effective limit of detection (LOD) of up to eight technical replicates per sample to inform experimental design. We used a 5-level, 10-fold standard curve of the P. subflavus synthetic sequence, which included 381,908, 38,191, 382, 39, and 4 copies/reaction, and ran with 20 replicates per level. We also used these standard curves to estimate the limit of quantification (LOQ), regression coefficients, and PCR efficiency.

Using the best performing assay (Table 1), DNA samples derived from airborne sampling media, field blanks, and extraction blanks were screened in triplicate in 15 μL reactions with an internal positive control to identify PCR inhibition. Reactions included 7.5 μL of 2X Environmental Mastermix 2.0 (Life Technologies), 250 nM hydrolysis probe, assay-specific primer concentrations (Table 1), 0.3 μL internal positive control (IPC; Life Technologies), 1.5 μL IPC assay pre-mix, 3 μL DNA template (neat), and PCR-grade water making up the remaining reaction volume. Each qPCR run included the same positive control and three PCR negative controls, whereby PCR grade water was substituted for DNA template. We ran a six-level standard curve with the same copy numbers as noted above (range: 3,819,080–4 copies/reaction). Reactions were run at 95°C for 10 min, followed by 45 cycles of assay-specific cycling conditions (Table 1). We would determine if inhibition influenced the reaction if the IPC in an eDNA sample was delayed relative to the PCR negative controls by one quantification cycle (Cq). A reaction resulting in a plausible detection would then be sequenced for verification as described above.

3.1 Acoustics

Bats were consistently active at both sites but were not significantly more or less active at artificial prey patches compared to unlit control sites (Figure 2, Data S2). Despite an increasing trend for artificial prey patches, models considering the presence of light lures failed to outcompete null models that were absent of that covariate for both feeding buzzes (ΔAIC = 0.37) and search-phase calls (ΔAIC = 0.78).

We recorded eight bat species at both caves over the duration of this study (Figure 3, Data S2). T. brasiliensis had the highest activity, followed by L. noctivagans. P. subflavus and M. velifer were consistently within the top four most active bats throughout the sampling duration. M. velifer foraging activity was markedly higher at artificial prey patches compared to control sites (Figure 3).

3.2 DNA Metabarcoding

We detected three bat species (Figure 4, Data S3) in several libraries (Table 3), all of which were detected from the VF media and spanning deployments no longer than one week. OTU read counts were high for bat detections, ranging from 3869 to 243,789 reads. Two detections were derived from a single VF sample among two of its 1 × 1 cm direct excisions. These were also the most credible bat detections based on replicate amplification. L. noctivagans was the only detection occurring in 2/2 PCR replicates, whereas E. fuscus (big brown bat) was only detected in 1/2 PCR replicates of a separate excision. We detected T. brasiliensis using the pelleted extraction method but only from a t0 sample that was loaded into the sampler and retrieved but not deployed.

Despite more bat detections with the direct extraction batch, the pelleted extraction method was the most effective at recovering vertebrate eDNA from the collection media, yielding 19 species among both sites and having recovered vertebrate data from 81.25% (39/48) of the libraries and 83.33% (20/24) of the samples (Table 3). This centrifugation-based method, which utilizes an entire portion of an eDNA medium, was the only method that yielded non-bat wildlife taxa such as black vulture (Coragyps atratus), cedar waxwing (Bombycilla cedrorum), ringtail (Bassariscus astutus), white-crowned sparrow (Zonotrichia leucophrys), hispid cotton rat (Sigmodon hispidus), and deer (Odocoileus spp.). The other extraction methods only yielded frogs and a variety of domesticated mammals, the origin of which can also be attributed to common contaminants in eDNA datasets (Klepke et al. 2022; Garrett, Watkins, Simmons, et al. 2023). Although direct extraction recovered bat species, it was less effective than the pelleted extraction method, only recovering taxonomic data from 9 of 68 libraries in two of the six randomly sampled media for excision. Despite visible target bands from gel electrophoresis, the filtration method only yielded taxonomic data from a single library, which failed to yield any wildlife detections.

3.3 qPCR

Following testing and validation, we ultimately screened all samples for the presence of P. subflavus using the PESU1 assay (Table 1). While neither candidate assay co-amplified from the non-target panel (Table 2), PESU1 was the most sensitive (LOD = 6 copies, assuming three qPCR replicates [Table 4]) and did not co-amplify from the non-target panel (Table 4). Using PESU1, we detected P. subflavus from two samples (Table 3), collected after 1–2 weeks in the sampler. Similarly to the metabarcoding dataset, we only detected P. subflavus from VF media excised for direct extraction. The target species was detected in two of the PCR replicates, and each detection was sequence-verified to P. subflavus (Data S4). The other detection came from the same sample where L. noctivagans was detected via metabarcoding, albeit from a different excision. The other two extraction methods failed to yield detections of the target species. The filtration batch was deemed inconclusive due to sequence-verified detections of the target species in both DNA extraction blanks. These amplifications re-occurred following a rerun, suggesting that P. subflavus DNA contaminated the reagents or supplies for that DNA extraction batch. No other DNA extraction blanks nor any PCR negative controls amplified. For all qPCR experiments, we observed some Cq delay in the IPCs of six reactions (direct excisions), which gave non-detections. However, the rest of the reactions were absent of PCR inhibition.