Results and Discussion
In this study, we have demonstrated that the presence of our targeted
threatened species, the koala, and its co-occurring terrestrial
mammalian community can be detected from the collection of airborne eDNA
under natural conditions. All 11 sampled sites detected the presence of
terrestrial mammals (Figure 2) and using a custom taxonomic assignment
strategy, we identified the presence of nine taxonomic families, most of
which, apart from the Macropodidae and Phalangeridae
(kangaroos/wallabies and possums, respectively), were assigned to the
species level, including our primary target, the koala. Moreover, we
show that airborne eDNA enables the detection of both native and
introduced species simultaneously, highlighting the utility of this
untargeted sampling approach for the identification of potential
biodiversity threats. Dogs, for instance, are a known threat to koalas
(Beyer et al. 2018) and were detected from a high abundance of reads
across our 11 sampled sites (Figure 2). We also identified common, but
problematic, invasive species including the black rat (Rattus
rattus ) and the hare (Lepus europaeus ) (Barney et al. 2021,
Finlayson et al. 2022). We discuss below how continued methodological
optimisation will enable the resolution of some teething problems.
First, and most problematic, is the presence of high level of
non-informative co-sampled DNA (e.g. human and domesticated animals).
Even with taking the utmost care and using published human DNA blocking
primers (Vestheim and Jarman 2008b), we found that many reads were
identified as human (Supplementary Table 3). We found that 53% of
reads, on average, were lost per sample as they were identified as human
DNA and a total of 5 samples were lost as only human DNA was amplified
(99% human reads; Supplementary Table 1). While we know that human
co-sampling and contamination is a common re-occurring problem when
using eDNA (Harper et al. 2019, Leempoel et al. 2020), we also found
that, under natural conditions, precious sequencing reads are further
lost to non-informative domesticated species, such as cows and horses
(Figure 2). This is despite these animals not being identified in faunal
transects or prior human led field surveys at the sampling location. The
presence of such high abundance of non-informative co-sampled airborne
DNA is a challenge we ought to tackle as it likely outcompetes the lower
abundance of airborne DNA particles shed by and therefore collected from
lower-density species of interest including threatened, endangered, or
cryptic elusive species. Our high level of non-informative co-sampled
DNA might be one of the reasons why our only sample which detected the
presence of koala DNA (Site 11. Figure 2) was the filter located
directly beneath a tree occupied by a koala while two additional
sampling sites (Site 5 and 2) positioned 50 - 30 meters away from koalas
failed to detect their DNA. Whilst this high level of sensitivity in
detection range allows us to confirm koala presence with a high degree
of confidence, the use of alternative human DNA blocking primers
(Boessenkool et al. 2012) coupled with targeted primers to enrich the
DNA abundance of low density or low biomass target species prior to
Illumina sequencing may help increase our rate and range of detection of
low-density, low-biomass species like the koala. A targeted qPCR assay
approach will require additional investment in development but when used
in concert with an untargeted approach will allow for a more inclusive
representation of the occupancy of less abundant species.
Second, we found that our ability to assign species level taxonomic rank
was limited by the small target sequence length. While small sequence
length is often best when dealing with degraded DNA, a characteristic of
eDNA (Beng and Corlett 2020), we identified it imposed some limitations
to the taxonomic assignment of our ASVs. For instance, the lack of
genetic variation between the mountain and common brushtail possum (1 bp
difference) made it difficult for us to assign our ASV to either species
with high certainty even though our ASV was 100% similar to the
mountain brushtail possum. This is because 1 bp difference between
sequences could easily fall within the margin of sequencing error
(Stoler and Nekrutenko 2021). Similarly, we were not able to disentangle
ASV 19 from a red-neck wallaby or a grey kangaroo because these
reference sequences differed by 1 bp. In contrast, we found that some of
our ASVs (ASV 14 and ASV 23) contained levels of genetic variation high
enough to make taxonomic assignments challenging. The Macropididae ASV
#14, for instance, differed from both the Red-Necked and the Swamp
Wallaby reference sequence by 5 bp. It, however, phylogenetically
clustered with the swamp wallaby (Figure 1) because of a shared
conserved region separating them both from the red-necked wallaby. A
similar trend was identified for our detected ring-tailed possum ASV
(ASV #23) which, while it clustered with our reference ring-tailed
possum sequence, nonetheless differed from it by 6 bp. Ecological
surveys of our sampling site detected the presence of those three
species indicating that, while imperfect at this stage, our taxonomic
assignments are likely correct. To continue to improve the robustness of
taxonomic assignment, we recommend considering the geographical
provenance of reference sequences because of expected geographic
patterns of genetic diversity. The ring-tailed possum 16S reference
sequence publicly available, for instance, came from an animal located
in Western Australia. We therefore propose that prior to deploying
airborne eDNA for detection of a suite of target terrestrial species, it
is critical to develop a relevant mitochondrial genomic database to
ensure accurate taxonomic resolution can be reached. This will include
assessing if the targeted sequence region(s) contain sufficient
variation to disentangle closely related species and are of geographic
relevance to your sampling location. Challenges to obtain material and
data from sensitive species groups can be overcome by engagement with
stakeholders, researchers, museums and the community. This is
particularly important to the deployment of airborne eDNA technology in
natural settings where, unlike zoological facilities (Clare et al.
2022), many closely related species overlap with each other and are only
differentiable by a few base pairs resulting in less refined taxonomic
resolution or potential misattributed taxonomic assignment.
Last and similar to other eDNA studies (Lusk 2014, Xing et al. 2022), we
demonstrate the importance of filtering stringency in the management of
DNA contamination (see Figure S1). DNA extracted from a skin biopsy of
the Indo-Pacific bottlenose dolphin (Tursiops aduncus ) was used
to estimate and control for DNA contamination in our downstream
bioinformatic pipeline, as this species is not found at or near the
sampling location. Like many other studies, we did identify a
significant amount of DNA cross-contamination which was present in our
dolphin sample when we only used forward reads in our bioinformatic
pipeline. We, however, found that the merging of ASVs got rid of all DNA
cross-contamination which we underline as a necessary step in any future
eDNA studies (Supplementary Figure 1).