Environmental (e)DNA has rapidly become a powerful biomonitoring tool, particularly in aquatic ecosystems. This approach has not been as widely adopted in terrestrial communities where the methods of vertebrate eDNA collection have varied from the use of secondary collectors such as blood-feeding parasites and spider webs to washing surfaces of leaves and soil sampling. Recent studies have demonstrated the potential of direct collection of eDNA from air sampling, but none have tested how effective airborne eDNA sampling might be in a biodiverse environment. We used three prototype samplers to actively sample a mixed neotropical bat community in a partially controlled environment. We assess whether airborne eDNA can accurately characterize a high-diversity community with skewed abundances and to determine if filter design impacts DNA collection and taxonomic recovery. Our study provides evidence for the accuracy of airborne eDNA as a detection tool and highlights its potential for monitoring high-density, diverse assemblages such as bat roosts. Analysis of air samples recovered >91% of the species present and some limited relationship between species abundance and read count. Our data suggest this method can accurately depict a diverse mixed mammal community, particularly when the location is contained (e.g., a roost, den or burrow) but also highlights the potential for secondary transfer of eDNA material on clothing and equipment. Our results also demonstrate that simple, inexpensive, battery-operated homemade air samplers can collect an abundance of eDNA from the air, opening the opportunity for sampling in remote environments.

Air Sample Collection and Filtration

We collected air samples from 25 April to 7 May 2022 (13 sample days), in Orange Walk District, Belize. We used three prototype models of air samplers for eDNA collection using computer fans and 3D-printed attachments bolted to the fans. Bats were captured in the riverine preserve forest of Lamanai Archeological Reserve, the forests near the Lamanai Outpost Lodge, and the forest fragment of the Ka’kabish Archeological Project using mist nets and harp traps (Permit # FD/WL/1/21(18), York ACC 2021-10).  We deployed samplers in the classroom field lab where captured bats were brought from the field and processed each night. eDNA was sampled by simultaneously filtering air using the 3 prototypes described. Samplers were turned on when the first bats were brought into the classroom. They ran for 6–8 hours depending on the arrival time of the first bats in the empty classroom each night, and we removed the filters the next morning. We did not sample during the day as the bats captured were released before sunrise and the room was used for other activities during the day. Filters were folded with the exposed side in and frozen in clean sample bags for later analysis. Before each sampling night, we cleaned the 3D-printed filter frames with a 50% bleach solution and then water. We collected 37 individual samples (12vLarge n = 13, 12vSmall n = 12, 5v n = 12, one extra 12vLarge filter was collected on a day with minimal bat sampling) over the course of 14 days. KN95 masks and gloves were worn by all researchers when handling filters or when bats were present in the classroom. Over the sample period, 880 individual bats of 35 species were present in the room.

DNA Extraction from Filter Material

We performed all laboratory procedures inside a sterilized AirClean 600 PCR workstation while wearing N95 masks, gloves, and coats. We cleaned all instruments using a 10% bleach solution, followed by a 10% ethanol solution, and then rinsed these with sterilized water before each use and between each sample. We cleaned the workstation with a 25% bleach solution and 30 minutes of UV light before handling any eDNA products. We extracted all filter samples using a Qiagen Blood and Tissue extraction kit as follows. We cut out a 2cm² piece from the centre of the filter and incubated it overnight at 56℃ in 180 µL ATL buffer and 20 µL protein kinase K. After incubation, we removed the filters using clean forceps and spun the filters in QiaShredder spin columns for 3 minutes at 13,000 RPMs. For the rest of the extraction, we followed the manufacturer’s guidelines. We processed extraction blanks (an extraction without any filter) along with the samples to act as a negative control during extraction. We quantified the DNA concentration of each sample using a Nanodrop. We froze extracted DNA until use.

PCR and Sequencing

We amplified a region of the mitochondrial 16S gene for each sample using the mam1 (‘5-CGGTTGGGGTGACCTCGGA-3’) and mam2 ('5-GCTGTTACCCTAGGTAACT-3’) primers (~90 bp insert)adapted for the Illumina MiSeq sequencing platform. Each mix included 7.5µL of mix QIAGEN Multiplex PCR Master Mix, 1.25µL of ddH2O, 1.75µL of human blocking primer (5’ – GCGACCTCGGAGCAGAACCC-spacerC3 – 3’), 0.75µL of each forward and reverse primer tagged with CS1 and CS2 adaptors and 3µL of extracted DNA. Because two facilities were used to avoid batch effects (below) we use different adaptors with slight variation in PCR cycling to achieve the same PCR success. We conducted the first PCR using the following cycle conditions: 95℃ for 10min, 40 cycles of 95℃ for 12s, 59℃ for 30s, 70℃ for 25s, and a final 72℃ for 10min. We performed two additional rounds of PCR under the same conditions but with the number of cycles increased to 45 and the overhang adaptors changed to those specified by Illumina in their 16S metagenomic library preparation guidelines. We included negative (no template) and positive (salmon) controls in each PCR.

We visualized all PCR products, including all controls (positive, negative, and extraction blank) using a 1.5% agarose EtBr gel and run at 110V for 1 hour. Given the unknown nature and efficiency of air sampling for such an ecologically complex community, we employed extra sequencing controls. To independently confirm sequencing, avoid batch effects, and detect potential sequencing contamination, the replicates were sent to two separate sequencing facilities for library building. The first CS1/CS2 tagged PCR replicate was sent to Barts and the London Genome Centre, UK, where the products were indexed. The samples were quantified on a TapeStation D100 (Agilent) and normalised and pooled for sequencing using an Illumina MiSeq V3 Micro 2x300 cycle run. Two other replicates were sequenced and indexed separately as follows. A sequencing library was prepared from the purified amplicons, indices were added following Illumina’s 16S Metagenomic Sequencing Library Preparation protocol but using 1X DreamTaq PCR Master Mix (Thermo Scientific). Indexed PCR products were again purified using Mag-Bind® TotalPure NGS (Omega Bio-tek) magnetic beads. The purified index products were quantified using a Qubit dsDNA BR Assay Kit, normalized and pooled. The pooled PCRs were sized using a TapeStation D1000 ScreenTape System (Agilent). The libraries were sequenced on an Illumina MiSeq with a V3 MiSeq Reagent kit, 300 cycles. The final library was loaded at 10 pM with a 20% PhiX control spike and sequenced at the NatureMetrics laboratory in London, UK. Reads were demultiplexed in preparation for bioinformatic analysis and exported as FASTQ files.

Bioinformatics Methods and Statistical Analysis

We processed the demultiplexed sequences using the DADA2 pipeline in RStudio.The forward and reverse reads were filtered, trimmed to 90bp, and errors in the sequence data were removed based on the learned error rates generated by the DADA2 learnErrors function. We removed primer sequences using cutadapt 3.7 in paired-end mode. We merged paired reads and generated amplicon sequence variants (ASVs). Any chimaeras detected at this stage were removed and the ASVs were exported as a FASTA file. We used BLAST to compare individual ASVs to the full nucleotide collection in NCBI to evaluate likely taxon of origin and use updated taxonomic designations following BatNames (https://batnames.org), Mammal Species of the World (http://www.departments.bucknell.edu/biology/resources/msw3/browse.asp), and Catalogue of Life (https://www.catalogueoflife.org/). We removed ASVs matched to human DNA, which was ubiquitous in the sampled environment. Next, we removed ASVs matched Black-tailed Jackrabbit (Lepus californicus) as these are likely from a previous sample processed in our lab. A single ASV matched to Cervus canadensis nannodes was also removed as it does not occur in our study area and was present in samples that had previously been processed in the lab. All matches to fish species and one match to a Harbour porpoise (Phocoena phocoena) were considered contaminates during the sequencing process as the sequencing facility primarily processes aquatic samples. We then disregarded the positive control, Salmo salar. The ASVs that had a match greater than 95% percent identity (100% overlap) to either bats or other taxa known to be in the area were kept for further scrutiny. Four fell below this but were retained based on known room occupancy and matches to sister taxa where the local species was not represented in the database. We retained all ASVs matched to a bat species known to be in the classroom regardless of read count. For matches to other taxa known to be in the area, ASVs with read counts below 20 were discarded which is the threshold that captures all true positives (bats we know were in the classroom and non-target species that have been seen in the area).

We divided then the ASVs into high-quality detections, low-quality detections, and very low-quality detections. High-quality detections are those that would be retained using a highly conservative filtering approach of excluding any ASV detection with a read count below the highest read count found for any taxa detected in the negative controls. Low-quality detections were those which fell below the read count of the highest negative control contaminant but were matched with bat taxa known to be present in the classroom. Similar to low-quality detections, very low-quality detections would normally be excluded based on very low read count, but they matched species known to be in the area and had no other potential source in our lab. Any identifications that did not fall into the above categories, such as non-Neotropical bat species were fully explored to try and determine the source and are discussed further below.

As the data were not normally distributed, we performed a Kruskal-Wallis ANOVA to compare both the total read counts (library size) and DNA concentration per prototype sampler design to test whether read count or DNA concentration differed by sampler type (5v, 12vSmall 12vLarge). We used Pearson’s correlation to assess the relationship between cumulative species abundance over the two-week sampling period and recovered cumulative read count per species across all three replicates.

To further independently verify the collection of bat eDNA from air, a subset of five samples was sent to a collaborating lab at Northern Arizona University. This analysis used COI and 12S markers instead of the 16S region used on the full sample set.

DADA2

Cutadapt

BLAST