R E SOU R C E A R T I C L E
Carrion fly-derived DNA metabarcoding is an effective tool
for mammal surveys: Evidence from a known tropical mammal
community
Torrey W. Rodgers1,2 | Charles C. Y. Xu3,4,5 | Jacalyn Giacalone6 | Karen
M. Kapheim2,7 | Kristin Saltonstall2 | Marta Vargas2 | Douglas W. Yu3,8 |
Panu Somervuo9 | W. Owen McMillan2* | Patrick A. Jansen2,10*
1Department of Wildland Resources, Utah State University, Logan, UT, USA
2Smithsonian Tropical Research Institute, Balboa, Panama
3State Key Laboratory of Genetic Resources and Evolution, Kunming Institute of Zoology, Chinese Academy of Sciences, Kunming, Yunnan, China
4Groningen Institute for Evolutionary Life Sciences, University of Groningen, Groningen, The Netherlands
5Redpath Museum and Department of Biology, McGill University, Montreal, QC, Canada
6College of Science and Mathematics, Montclair State University, Montclair, NJ, USA
7Department of Biology, Utah State University, Logan, UT, USA
8School of Biological Sciences, University of East Anglia, Norwich Research Park, Norwich, Norfolk, UK
9Department of Biosciences, University of Helsinki, Helsinki, Finland
10Department of Environmental Sciences, Wageningen University, Wageningen, The Netherlands
Correspondence
Torrey W. Rodgers, Department of Wildland
Resources, Utah State University, Logan, UT,
USA.
Email: torrey.w.rodgers@gmail.com
Funding information
Smithsonian Tropical Research Institute;
STRI Terrestrial Environmental Studies
Project; U.S. Department of Education
Math-Science Partnerships Project; private
funds from Gregory E. Willis; National
Natural Science Foundation of China, Grant/
Award Number: 31400470, 41661144002,
31670536, 31500305, GYHZ1754; Ministry
of Science and Technology of China, Grant/
Award Number: 2012FY110800; University
of East Anglia; State Key Laboratory of
Genetic Resources and Evolution at the
Kunming Institute of Zoology, Grant/Award
Number: GREKF13-13, GREKF14-13,
GREKF16-09
Abstract
Metabarcoding of vertebrate DNA derived from carrion flies has been proposed as
a promising tool for biodiversity monitoring. To evaluate its efficacy, we conducted
metabarcoding surveys of carrion flies on Barro Colorado Island (BCI), Panama,
which has a well-known mammal community, and compared our results against diur-
nal transect counts and camera trapping. We collected 1,084 flies in 29 sampling
days, conducted metabarcoding with mammal-specific (16S) and vertebrate-specific
(12S) primers, and sequenced amplicons on Illumina MiSeq. For taxonomic assign-
ment, we compared BLAST with the new program PROTAX, and we found that PROTAX
improved species identifications. We detected 20 mammal, four bird, and one lizard
species from carrion fly metabarcoding, all but one of which are known from BCI.
Fly metabarcoding detected more mammal species than concurrent transect counts
(29 sampling days, 13 species) and concurrent camera trapping (84 sampling days,
17 species), and detected 67% of the number of mammal species documented by
8 years of transect counts and camera trapping combined, although fly metabarcod-
ing missed several abundant species. This study demonstrates that carrion fly
metabarcoding is a powerful tool for mammal biodiversity surveys and has the
potential to detect a broader range of species than more commonly used methods.
K E YWORD S
Barro Colorado Island, biodiversity, camera trapping, environmental DNA, transect counts
*These authors jointly advised this work.
Received: 28 April 2017 | Revised: 11 July 2017 | Accepted: 25 July 2017
DOI: 10.1111/1755-0998.12701
Mol Ecol Resour. 2017;1–13. wileyonlinelibrary.com/journal/men © 2017 John Wiley & Sons Ltd | 1
1 | INTRODUCTION
Due to the rapid decline in biodiversity worldwide, there is an urgent
need for more efficient techniques to survey and monitor biodiver-
sity. Traditional surveys based on direct observation are costly and
time-consuming, detection rates can vary across observers, and rare
or cryptic species are often overlooked. Camera trapping has
emerged as a more efficient survey method, especially for large-bod-
ied vertebrates (Beaudrot et al., 2016), but it has limited ability to
detect small-bodied, arboreal and volant species. Moreover, camera
traps are expensive for use at large scales and are subject to damage
and theft. An efficient technique capable of detecting a wide range
of vertebrate species with limited bias would greatly increase our
ability to monitor vertebrate diversity, and to gauge the effectiveness
of conservation measures such as nature reserves.
One potential method has emerged from the field of environ-
mental DNA (eDNA) sampling, which uses trace amounts of DNA
from the environment for species detection (Bohmann et al., 2014).
This approach, particularly when used in conjunction with next-gen-
eration metabarcode sequencing, shows great potential for efficient
monitoring of biodiversity (Ji et al., 2013; Taberlet, Coissac, Hajiba-
baei, & Rieseberg, 2012; Yu et al., 2012). To date, the majority of
vertebrate eDNA research has focused on aquatic species because
eDNA is easy to collect from water (Thomsen et al., 2012). Collect-
ing eDNA from terrestrial vertebrates is more difficult. Researchers
have used soil (Andersen et al., 2012), browsed twigs (Nichols,
Konigsson, Danell, & Spong, 2012), prey carcasses (Wheat, Allen,
Miller, Wilmers, & Levi, 2016) and drinking water (Rodgers & Mock,
2015) for eDNA detection of terrestrial species, but these
approaches have limited scope. So far, the most promising approach
for use across a broad range of species is mass trapping and
metabarcoding of invertebrates that feed on vertebrates (Bohmann,
Schnell, & Gilbert, 2013; Calvignac-Spencer, Leendertz, Gilbert, &
Schubert, 2013). Such invertebrate “samplers” tested to date include
leeches (Schnell et al., 2015), mosquitos (Logue et al., 2016), dung
beetles (Gillett, Johnson, Barr, & Hulcr, 2016) and carrion flies (Calvi-
gnac-Spencer, Merkel et al., 2013; Lee, Gan, Clements, & Wilson,
2016; Lee, Sing, & Wilson, 2015; Schubert et al., 2014). We focus
on the latter.
Carrion flies are species from the families Calliphoridae and Sar-
cophagidae that feed and oviposit upon dead animals, open wounds
and faeces. When carrion flies feed, they ingest vertebrate DNA. Car-
rion flies are ideal candidates for eDNA surveys because they are
easy to trap, are ubiquitous worldwide and are believed to feed
opportunistically on vertebrates of all sizes, including terrestrial spe-
cies, volant species and species occupying the forest canopy. In a
remarkable study, Calvignac-Spencer, Merkel et al. (2013) used San-
ger sequencing to detect 20 different mammal species from just 115
flies collected in Co^te d’Ivoire and Madagascar. They also used Roche
GS FLX (“454”) pyrosequencing to metabarcode a subset of samples
and were able to detect the majority of species that had been
detected with Sanger sequencing, plus several others.
Although carrion fly metabarcoding is clearly a promising method,
there is still a need to quantify its effectiveness relative to conven-
tional methods. There is also a need to test new methods for assign-
ing taxonomies to sequence data. This step may be particularly error
prone in eDNA studies, because target amplicons for eDNA are
short due to the need to amplify degraded DNA and thus have
lower taxonomic information content. Also, because reference
sequence databases are incomplete, taxonomic assignment software
has a bias towards overconfidence, meaning that an operational tax-
onomic unit (OTU) sequence may be assigned to a species that is in
the database and fairly similar, when the correct species is not in the
database (Somervuo et al., 2017). Correct species-level identification
is especially important when applied to vertebrates, as incorrect
assignment of an OTU to an endangered species, as opposed to a
less endangered congener (or vice versa), can have a large impact on
conservation decision making.
With these goals in mind, we conducted a field test on Barro
Colorado Island (BCI) in Panama, where the vertebrate fauna is well
documented, particularly for mammals. We collected carrion flies
for metabarcoding and compared results with data sets from annual,
diurnal transect counts (distance sampling) and semi-continuous
camera trapping of the mammal community for 8 years leading up
to, and concurrent with, fly collection. In addition, we compared
taxonomic assignments between the most commonly used method,
BLAST (Altschul, Gish, Miller, Myers, & Lipman, 1990) and the new
method PROTAX (Somervuo et al., 2017), which uniquely takes into
account the possibility that an OTU sequence belongs to a species
that is not in the reference database, thus avoiding overconfident
assignments. We find that metabarcoding of carrion flies is an
effective, but imperfect, method for surveying mammal communi-
ties, and PROTAX substantially outperformed BLAST for taxonomic
assignment.
2 | MATERIALS AND METHODS
2.1 | Study site
Field work was conducted on Barro Colorado Island, a 1,560-ha
island in the Panama Canal waterway (Figure 1). BCI (9°100N,
79°510W) sits within Gatun Lake, an artificial body of water created
in 1912 by the damming of the Chagres River to create the Panama
Canal, and is part of the protected 54-km2 Barro Colorado Nature
Monument. BCI has 108 known mammal species including 74 bats,
and 34 nonvolant species, (Glanz, 1982 and current expert informa-
tion). However, some of these species, such as jaguar (Panthera
onca), puma (Puma concolor) and jaguarundi (Herpailurus yagouar-
oundi), are only infrequent visitors to the island.
2.2 | Fly collection
Flies were collected between 10 February 2015 and 5 May 2015
in three trapping sessions totalling 620 fly-trap days. All flies col-
2 | RODGERS ET AL.
lected were used in the metabarcoding analysis. First, flies were
collected from 10 February to 14 February using a variety of trap
types within 1,000 m of the laboratories on BCI, to determine the
best methods for sampling. Sampling methods tested included net-
ting flies above covered bowls of pork, and the trap types
described in Calvignac-Spencer, Merkel et al. (2013). Based on
results from this testing, we used a trap type modified from www.b
lowflies.net/collecting.htm for all subsequent trapping. Traps were
baited with raw pork hung in a cup surrounded by fine cloth net-
ting to keep flies from landing directly on the bait, and bait was
replaced every 2–3 days. From 15 February to 23 February, flies
were then collected from traps placed along trails at 16 trap loca-
tions, with one trap per location, spaced approximately 200 m apart
in a nonuniform grid (Figure 1). Second, from 28 February to 4
March, traps were placed along trails at the same 16 trap locations
as session 1, but with two traps per location. Third, from 12 March
2015 to 5 May 2015, flies were collected from traps placed in a
transect crossing the island with 16 trap locations, each containing
two traps, placed every 250 m along a trial in a roughly straight
line (Figure 1), for 10 sampling days. In all three collection efforts,
flies were removed from traps once or twice daily and placed in a
40°C freezer within 2 hr of collection, because DNA degradation
causes detection success to decline 24 hr after fly feeding (Lee
et al., 2015).
2.3 | Library preparation and sequencing
DNA was extracted from flies using the GeneMATRIX Stool DNA
Purification Kit (Roboklon, Berlin, Germany). To reduce extraction
costs, up to 16 flies were pooled for each extraction. Flies were first
cut into several pieces with sterile scissors and placed into 2.5 ml
Polypropylene 96 Deep Well Plates along with stainless steel, 5/32”
grinding balls (OPS Diagnostics, Lebanon, NJ, USA). Up to four flies
and 100 ll of lysis buffer per fly were added to each well. Plates
were shaken on a TissueLyser II (Qiagen, Germantown, MD, USA)
until fly tissue was homogenized in the lysis buffer. An equal volume
of the resulting homogenate per fly was then pooled in a total vol-
ume of 160 ll and added to the extraction kit bead tube. Extractions
then proceeded following manufacturer’s recommendations. For flies
collected in the first trapping session, each extraction included 16
79°50'0"W
79°50'0"W
79°52'0"W
79°52'0"W
9°
10
'0
"N
9°
10
'0
"N
9°
8'
0"
N
9°
8'
0"
N
0 1 2 3 40.5
Kilometers
Republic of Panama
Legend
Fly traps - Feb 15 to Mar 4 
Fly traps - Mar 12 to May 5 
Trails
F IGURE 1 Location of carrion fly trapping on Barro Colorado Island, Panama
RODGERS ET AL. | 3
flies. For flies collected in the second and third trapping sessions,
each extraction contained all flies collected at the same trap location
on the same day, up to a maximum of 16 flies. To examine whether
pooling had an effect on species detection, 24 flies were also
homogenized and extracted individually, and 20 ll of homogenate
from each of these same 24 flies was also pooled and extracted in
two samples containing 12 flies each. Single-fly and pooled samples
were all sequenced at the same depth. All extractions included blank
controls that were included in the PCR step to test for contamina-
tion.
PCR was performed on all samples using a mammal-specific pri-
mer set (16Smam1, 16Smam2; Boessenkool et al., 2012) targeting
130–138 bp (including primers) of the mitochondrial 16S rRNA
locus, and a pan-vertebrate primer set (12SV5F, 12SV5F; Riaz et al.,
2011) targeting 140–143 bp (including primers) of the mitochondrial
12S locus. Reactions also included human blocking primers (16S-
mam_blkhum3; Boessenkool et al., 2012; 12S_V5_blkhum; Calvig-
nac-Spencer, Merkel et al., 2013) and Sus blocking primers
(16Smam_blkpig, 12S_V5_blkpig; Calvignac-Spencer, Merkel et al.,
2013) to decrease competition from contaminating DNA from pork
bait. Primers also included a 5’ addition of a 33 bp Illumina-specific
sequence for addition of adapters in a second round of PCR. A
minimum of two PCR replicates per sample were performed in
10 ll volumes. For 16S, reactions included 0.2 uM of each primer,
1 lM of human blocking primer (5x) 4 lM of Sus blocking primer
(20X), 200 lM dNTP, 4 mM MgCl2, 1X PCR buffer, 1.25 U Plat-
inum
 Taq polymerase (Invitrogen) and 3 ll of template DNA. For
12S, volumes were the same, except MgCl2 was reduced to
2.5 mM. Cycling conditions were 10 min at 95°C, followed by 42
cycles (16S) or 47 cycles (12S) of 30 s at 95°C, 30 s at 64°C, and
1 min at 72°C, with a final extension of 10 min at 72°C. Following
the initial PCR, sample replicates were pooled, and a second PCR
was conducted to add Illumina flow cell binding sequences and
unique 8-bp sample-specific indexes to each end of each amplicon.
Reactions included 1 ll each of the forward and reverse Illumina
index primers, 2.5 ll of 10X QTaq buffer, 1.5 mM MgCl2 and 1.25
units of Qiagen Qtaq. Cycling conditions were 3 min at 94°C, fol-
lowed by six cycles of 45 s at 94°C, 60 s at 50°C and 1 min at
72°C, with a final extension of 10 min at 72°C. All PCR reactions
were prepared in a UV-sterilized laminar flow hood and included
no-template negative controls to test for contamination. None of
the extraction negative controls and no-template PCR negative con-
trols showed any sign of amplification when run on an agarose gel,
so they were not included in sequencing libraries. Samples were
purified and normalized using SequalPrepTM Normalization Plates
(Thermo-Fisher Scientific, Waltham, MA, USA). All samples were
pooled in equimolar concentration, and pools were concentrated
using standard Ampure XP beads (Beckman Coulter, Indianapolis,
IN, USA). The concentrated pool was quantified using a Qubit fluo-
rometer and quality checked using a Bioanalyzer High Sensitivity
DNA Kit (Agilent, Santa Clara, CA, USA). The library was sequenced
on an Illumina MiSeq as 30% of a 500 cycle 2 9 250 Paired-end
sequencing run.
2.4 | Sequence filtering
After demultiplexing, paired ends were merged for each sample using
PEAR v. 0.9.6 (Zhang, Kobert, Flouri, & Stamatakis, 2014) with default
parameters, except minimum overlap (-v) = 100, minimum length (-
n) = 100 and quality score threshold (-q) = 15. We then separated
12S and 16S sequences and removed primers and remaining adapters
with CUTADAPT v. 1.3 (Martin, 2011), keeping sequences with a mini-
mum length (-m) of 10. Sequences had been previously trimmed to
minimum 100. Mean  1 SE length after trimming was
109.78  0.13 for 12S and 93.70  0.08 for 16s. Sequences were
subsequently pooled and filtered for chimeras with the iden-
tify_chimeric_seqs.py script and the usearch61 method within the QIIME
environment (Caporaso et al., 2010; Edgar, 2010). We used the pick_-
de_novo_otus.py script with the uclust method to cluster sequences
into OTUs (Operational Taxonomic Units) at 97% similarity, create a
table of OTU frequency in each sample, and pick representative
sequences (Caporaso et al., 2010; Edgar, 2010). We removed OTUs
with fewer than 20 sequences using the filter_otus_from_otu_table.py
script (Caporaso et al., 2010). A threshold of 20 sequence reads was
chosen because all OTUs with fewer than 20 reads had no similarity
to any vertebrate species and were thus likely artefactual. To remove
OTUs from contaminants such as the pork bait, human DNA and bac-
teria, we conducted an initial BLAST search and removed all OTUs
with a top hit for Sus, Homo or bacteria. For 12S, we used this initial
BLAST search to separate OTUs into mammal, bird and reptile groups
for downstream taxonomic assignment.
2.5 | Taxonomic assignment
For mammalian taxonomic assignment, we compared two methods:
the most commonly used method BLAST (Altschul et al., 1990) and the
new probabilistic taxonomic placement method PROTAX (Somervuo
et al., 2017). For nonmammals, only BLAST was used. BLAST searches
were conducted against the National Center for Biotechnology Infor-
mation (NCBI) nonredundant database with default settings except
for e-value ≤1e-6, and per cent identity ≥95%. We processed the BLAST
output in two ways for each OTU: selecting the top hit, and using the
lowest common ancestor (LCA) algorithm in MEGAN 5.11.3 (Huson,
Auch, Qi, & Schuster, 2007; default settings except top per cent = 5),
a software program designed to nonsubjectively “choose” the appro-
priate taxonomic assignment among multiple BLAST hits.
For PROTAX, we generated probabilities of taxonomic placement
for each OTU at four taxonomic ranks (order, family, genus and spe-
cies) (Somervuo, Koskela, Pennanen, Henrik Nilsson, & Ovaskainen,
2016). PROTAX is a statistical wrapper that processes the output of
one or more other taxonomic assignment methods and takes into
account the uncertainty of taxonomic assignment contributed by
species that do not have reference sequences. Briefly, this is
achieved by training a model against a reference data set that
includes all available reference sequences for the taxon of interest,
plus a full Linnaean taxonomy for that taxon which includes all
named species, including those that do not have reference
4 | RODGERS ET AL.
sequences. By taking into account the taxonomic information con-
tent of both the gene sequence being assigned and the reference
sequence database, PROTAX removes the inherent “over-assignment”
bias of other taxonomic assignment software. This makes PROTAX
especially suited for markers that have highly incomplete reference
databases, such as 16S and 12S.
For the reference sequence databases, we downloaded all avail-
able mammalian mitochondrial 16S and 12S sequences from Gen-
Bank, randomly truncated each species to a maximum of 10
sequences via a Fisher–Yates shuffle, removed ambiguous bases and
used ecoPCR within OBITOOLS 1.2.6 (Boyer et al., 2016; Ficetola et al.,
2010) to extract the amplicon regions used in this study (Clark,
Karsch-Mizrachi, Lipman, Ostell, & Sayers, 2016). This resulted in
5,243 16S sequences representing 1,733 species and 5,395 12S
sequences representing 1,935 species. For the taxonomic database,
we downloaded full ranks for the 6,711 species in the NCBI Mam-
malia taxonomy. We are aware that the NCBI taxonomy is not com-
plete for the Mammalia, but the important benefit is that the names
are consistent between the sequence and taxonomic databases, and
even this incomplete Mammalia taxonomy has ca. 3.8X and 3.4X the
number of species as represented in the 16S and 12S reference
sequence databases, respectively, which allows us to contrast PROTAX
with BLAST.
To start, all OTU representative sequences were pairwise com-
pared to the reference sequences using LAST version 744 (Kielbasa,
Wan, Sato, Horton, & Frith, 2011), and the PROTAX model used the
maximum and second-best similarities. We parameterized three PRO-
TAX models, one unweighted model in which all mammal species
were given an equal prior weight of being in the sampling location,
and two weighted models in which species independently known to
be present in Panama (Panama-weighted) or BCI (BCI-weighted)
were given a 90% prior probability of being in the sampling location,
with all other species being given a prior probability of 10%. The
accuracy and bias of the trained PROTAX model at each taxonomic
rank were estimated by plotting the cumulative predicted probabili-
ties against the cumulative number of cases in which the outcome
with the highest probability was correct when training the model.
For all methods (BLAST top hit, BLAST plus MEGAN, and PROTAX
unweighted and weighted), we estimated accuracy by calculating the
percentage of OTUs assigned to a genus or species known from BCI
(rate of correct assignment) and the percentage of OTUs assigned to
a genus or species known to not be present on BCI (rate of clear
false positives). We also counted the number of mammal species
known from BCI identified with each method. Given that we only
used wild-caught flies, we cannot directly test the probability that an
OTU was assigned incorrectly to a species known from BCI. How-
ever, the PROTAX output does give us an estimated probability of cor-
rect assignment for each OTU at each rank.
2.6 | Transect counts and camera trapping
We compared results from fly metabarcoding with results of two tra-
ditional survey methods. First, diurnal mammal transect counts were
carried out yearly from 2008 to 2015 during January and/or Febru-
ary of each year. Transects covered all trails on BCI (Figure 1) each
year, with a mean distance of 120 km walked per year. An average
walking rate of 1 km/hr was maintained. Record of each mammal
sighting included date, time, location, species and number of individ-
uals. All censuses were conducted between 06:40 and 12:00 and
were thus unlikely to detect nocturnal animals. In 2015, the year fly
sampling was conducted, 151.1 km was surveyed from 24 January
to 22 February. Transect counts started before, but overlapped with
fly and camera trap sampling; however, the presence of mammals on
BCI does not vary throughout the year, so this slight difference in
timing should not affect the methods comparison.
Second, camera traps were operated continuously across BCI
from 2008 to 2015. A mean of 25 (range 19–34) trail cameras
(PC900 and RC55 Reconyx Inc., Holmen, Wisconsin) were mounted
at knee height, with a spacing of 500–1,000 m. Thus, they were unli-
kely to capture arboreal or volant species. Additional cameras were
deployed for shorter periods. Cameras were checked at least once
every 6–7 months and replaced or repaired if no longer functioning.
This effort resulted in a total of 39,151 total camera days. For the
period concurrent with fly collection, February 10 2015 to May 5
2015, 26 cameras were active for the entire period, resulting in a
total effort of 1,967 camera days.
2.7 | Methods comparison
We compared the total number of mammal species detected by fly
metabarcoding with the total number of species detected by concur-
rent (2015) and long-term (2008–2015) transect counts and camera
trapping. In addition, we fit species accumulation curves for each of
the 2015 surveys using the specaccum function in the R package VE-
GAN 2.4-1 (Oksanen et al., 2016) with method = random and 10,000
permutations.
3 | RESULTS
3.1 | Transect counts and camera trapping
In the 2015 transect counts, we detected 13 mammal species,
whereas in 2008–2015 transect counts, we detected 17 mammal
species. With concurrent 2015 camera trapping, we detected 17
mammal species, whereas in 2008–2015 camera trapping, we
detected 26 mammal species. Species detected by the two methods
overlapped only partially, so that the 2015 data sets combined
detected a total of 22 mammal species, and the 2008–2015 data sets
combined detected a total of 30 mammal species (Table 2).
3.2 | Fly metabarcoding
3.2.1 | Fly collection, extraction and sequencing
A total of 1084 flies were collected and pooled into 102 pooled
samples and 24 single-fly samples. We obtained 2,780,574 initial
RODGERS ET AL. | 5
reads, which were reduced to 2,288,009 after quality filtering
(1,504,440 16S; 783,569 12S). OTU clustering, removal of OTUs
with fewer than 20 reads, and removal of OTUs assigned to human,
Sus (pork bait), and bacteria resulted in a final set of 54 OTUs for
16S and 63 OTUs for 12S (49 mammal, 12 bird and two reptile
OTUs; OTU sequences in Appendix S1). After quality filtering and
removal of human, Sus and bacterial reads, mean number of reads
per OTU was 8,401 (Range = 20–336,679; SD = 46,578) for 16S
and 4,585 (Range = 20–78,841; SD = 13,029) for 12S, and mean
number of reads per sample was 2,525 (Range = 1–14,716;
SD = 3,098) for 16S and 2,218 (Range = 1–25,575; SD = 4,582) for
12S for pooled samples. For single-fly samples, mean number of
reads per sample was 3,744 (Range = 1–11,173; SD = 3,195) for
16S and 87 (Range = 0–2,069, SD = 422) for 12S.
3.2.2 | Mammal species detection
The number of mammal species detected differed between assign-
ment methods (Figure 2a). BCI-weighted PROTAX resulted in the
detection of 20 total mammal species (Table 1), the most of any
method (16 species from 53 OTUs with 16S, 13 species from 37
OTUs with 12S and nine species with both markers). Assignment of
multiple OTUs to the same species is expected in metabarcoding as
the sequence-clustering step typically applies a single similarity
F IGURE 2 Comparison of taxonomic
placement methods for assigning mammal
OTUs to species from metabarcoding of
carrion flies collected on Barro Colorado
Island (BCI). (a) Number of mammal species
detected from BCI using each taxonomic
method. (b) Percentages of OTUs assigned
to genera or species known to occur on
BCI. (c) Numbers of OTUs assigned to
genera or species known not to exist on
BCI (clear false positives)
6 | RODGERS ET AL.
threshold across all OTUs, which can result in species splitting. Fif-
teen of these 20 species had PROTAX probabilities of >0.9, but several
had somewhat low PROTAX probabilities despite their known presence
on BCI (Table 1). Panama-weighted and unweighted PROTAX resulted
in detection of 19 total mammal species (13 with 16S and 12 with
12S).
BLAST top hit and BLAST plus MEGAN both resulted in detection
of fewer mammal species (15 and 13 respectively). The percentage
of OTUs assigned to a species known from BCI was higher for
PROTAX than for BLAST (Figure 2b). Rates of mammal OTUs assigned
to species not present on BCI (clear false positives) were generally
higher for BLAST than for PROTAX. Some clear false positives were
still present with all assignment methods, although no clear false
positives at the genus level occurred with BCI-weighted PROTAX
(Figure 2c).
3.2.3 | Effect of fly pooling
From the 24 flies that were extracted both singly and in pools, four
total mammal species were detected from the pooled samples,
whereas 10 total mammal species were detected from the same 24
flies when extracted singly. From the single-fly extractions, a range
of 1–4 species were detected from each fly.
3.2.4 | Nonmammal species detection
The 12S primers also detected several birds and one lizard. We were
only able to assign one OTU to the species level, wattled jacana
(Jacana jacana). Two OTUs were assigned to birds at the genus level,
one to an antshrike species (genus Thamnophilus, likely atrinucha or
doliatus; both known from BCI), and one to a trogon species (genus
Trogon, of which five species are known from BCI). Several OTUs
were assigned the family Anatidae; however, these were a 100%
match to many species within that family, and six species from
Anatidae are known from BCI. We also assigned one reptile OTU to
the whiptail lizards (family Teiidae) which is likely either Ameiva fes-
tiva or Ameiva leptophrys, the two Teiidae species known from BCI.
3.3 | Methods comparison
In 2015, we detected a greater number of mammal species in 29 days
of carrion fly metabarcoding than were detected by either 29 days of
transect counts or 84 calendar days of camera trapping carried out
concurrently (metabarcoding = 20 species; camera trapping = 17 spe-
cies, transect counts = 13 species, Figure 3). We detected more mam-
mal species with fly metabarcoding in 2015 than with 8 years of
transect counts from 2008 to 2015, but fewer than with 8 years of
TABLE 1 Twenty mammal species detected from metabarcoding of carrion flies on Barro Colorado Island Panama in 2015, along with
PROTAX (BCI-weighted) estimated probabilities of correct assignment at genus and species rank, and the percentage of fly pool samples that
each species was detected from
Species name Common name Genus prob. Species prob. % of samples
Alouatta palliata Mantled howler monkey .996 .996 98
Ateles geoffroyi Geoffroy’s spider monkey .983 .983 85
Tamandua Mexicana Northern tamandua 1.000 .997 29
Cebus capucinus White-faced capuchin .997 .979 26
Caluromys (derbianus)a Derby’s woolly opossum .993 NA 22
Saguinus (geoffroyi)a Geoffroy’s tamarin .400 NA 21
Sciurus granatensis Red-tailed squirrel .994 .994 21
Proechimys (semispinosus)a Tome’s spiny-rat .457 NA 20
Marmosa robinsoni Robinson’s mouse opossum .982 .98 19
Procyon (cancrivorus)a Crab-eating raccoon .934 NA 16
Potos flavus Kinkajou .979 .978 11
Didelphis marsupialis Common opossum .984 .957 11
Pecari tajacu Collared peccary .979 .978 6
Leopardus (pardalis)a Ocelot .817 NA 3
Pteronotus (gymnonotus or parnellii)a Big naked-backed bat or Parnell’s mustached bat .230 NA 3
Canis (latrans or familiaris)a Domestic dog or coyote .942 NA 1
Bradypus variegatus Brown-throated sloth 1.000 .998 1
Choloepus hoffmanni Hoffmann’s two-toed sloth 1.000 .999 1
Dasyprocta punctata Agouti .958 .924 1
Tonatia (saurophila)a Stripe-headed round-eared bat .894 NA 1
a
PROTAX assignment only at genus level, thus species probabilities were not estimated. Species names in parentheses are the species of the identified
genus that occurs on BCI.
RODGERS ET AL. | 7
camera trapping (transect counts 2008–2015 = 17; camera trapping
2008–2015 = 26). Using all three methods combined, we detected a
total of 27 mammal species in 2015, and 34 mammal species from
2008 to 2015. Visual inspection of the species accumulation curves
from 2015 (Figure 4) suggests that additional carrion fly sampling
effort, or greater sequencing depth, would likely have resulted in a
greater number of species detections, whereas the transect count and
camera-trap data sets were nearing asymptotes.
Of the 20 species detected with fly metabarcoding (Table 1),
four were not detected by camera traps or transect counts from
2008 to 2015. These included Derby’s woolly opossum (Caluromys
derbianus), a species from the genus Canis that was a 100% match to
reference sequences of domestic dog (C. domesticus) and coyote
(C. latrans), and two bat species. Conversely, metabarcoding did not
detect three species commonly detected with transect counts or
camera trapping: paca (Agouti paca), white-nosed coati (Nasua narica)
and red brocket deer (Mazama americana; Table 2). Moreover, the
Central American agouti (Dasyprocta punctata), a large rodent that is
by far the most common mammal on BCI, was detected in just one
metabarcoding sample, and only by the 12S marker.
4 | DISCUSSION
In total, we detected a greater number of mammal species with car-
rion fly metabarcoding than with transect counts or camera trapping
carried out during the same general time frame with similar levels of
effort. Thus, carrion fly metabarcoding appears to be an effective,
albeit imperfect, method for surveys of mammalian diversity. Of the
20 species detected with fly metabarcoding (Table 1), four were not
detected by camera traps or transect counts. Two were bat species
that are unlikely to be detected by camera traps due to their size
and are unlikely to be detected by diurnal transect counts. The third,
Derby’s woolly opossum (Caluromys derbianus), is arboreal and
nocturnal, but it has been commonly observed on BCI at night feed-
ing on nectar of canopy flowers. The fourth and most unexpected
detection was a species from the genus Canis, which was a 100%
match to reference sequences of both domestic dog (C. domesticus)
and coyote (C. latrans). Domestic dogs are not permitted on BCI, but
they are present on the mainland <1 km away at the nearest point.
Coyotes have never been detected on BCI but have been expanding
in Panama (Bermudez, Gonzalez, & Garcia, 2013) and have recently
been photographed by camera traps in the adjacent Soberania
National Park (www.teamnetwork.org). It is possible that a fly fed on
a canid on the mainland and then flew to BCI. Little is known about
flight and foraging distances for carrion flies; however, Lee (2016)
detected a blowfly movement distance of 3 km in tropical Malaysia.
It is also possible that ex situ contaminating DNA from a domestic
dog was introduced in the laboratory or was present in PCR
reagents, although this seems unlikely.
Although fly metabarcoding detected more mammal species than
the traditional techniques, a clear shortcoming was failure to detect
three abundant species that were commonly detected by transect
counts and camera trapping. This included a common rodent (Agouti
paca), the most common carnivore (Nasua narica), and a common
ungulate (Mazama americana). Metabarcoding did, however, detect
F IGURE 3 Number of mammal species detected by alternative
sampling methods on Barro Colorado Island, Panama
F IGURE 4 Species accumulation curves for three different
survey methods used to sample mammal diversity on Barro Colorado
Island, Panama in 2015
8 | RODGERS ET AL.
TABLE 2 Mammal species detected by carrion fly metabarcoding, camera trapping and diurnal transect counts on Barro Colorado Island,
Panama. For metabarcoding, values represent number of samples in which a species was detected. For camera trapping and transect counts,
values are number of individuals detected
Species Common name
Metabar-
coding Camera trapping Transect counts
Habita
16s 12s 2015 2008–2015 2015 2008–2015
Order Primates
Alouatta palliata Mantled howler monkey 99 67 6 315 1,996 D A
Ateles geoffroyi Geoffroy’s spider monkey 87 17 4 32 136 D A
Cebus capucinus White-faced capuchin 27 2 4 147 34 429 D A
Saguinus geoffroyi Geoffroy’s tamarin 21 17 94 D A
Order Carnivora
Potos flavus Kinkajou 11 1 N A
Procyon cancrivorus Crab-eating raccoon 2 14 8 52 N T
Leopardus pardalis Ocelot 3 329 5,801 6 C T
Canis sp. Domestic dog or coyote 1 1
Nasua narica White-nosed coati 1,064 7,358 18 178 D T
Puma concolor Puma 6 N T
Panthera onca Jaguar 9 C T
Eira barbara Tayra 1 54 4 D T
Herpailurus yagouaroundi Jaguarondi 3 5 C T
Order Rodentia
Dasyprocta punctata Agouti 1 2,975 72,719 302 1,821 D T
Sciurus granatensis Red-tailed squirrel 10 15 20 1,769 68 735 D ST
Proechimys semispinosus Tome’s spiny-rat 5 18 114 1,074 1 2 N T
Agouti paca Paca 108 2,412 N T
Heteromys desmarestianus Desmarest’s spiny pocket
mouse
44 N T
Coendou rothschildi Rothschild’s porcupine 3 N A
Diplomys labilis Rufous Tree Rat 4 N A
Hydrochoerus hydrochaeris Capybara 18 N T
Order Didelphimorphia
Didelphis marsupialis Common opossum 1 11 57 1,065 N ST
Marmosa robinsoni Robinson’s mouse opossum 19 1 7 N ST
Caluromys derbianus Derby’s woolly opossum 2 21 N ST
Order Pilosa
Choloepus hoffmanni Hoffmann’s two-toed sloth 1 1 2 C A
Bradypus variegatus Brown-throated sloth 1 1 2 C A
Tamandua mexicana Northern tamandua 30 20 359 2 27 D ST
Order Artiodactyla
Pecari tajacu Collared peccary 6 1,040 12,731 25 383 C T
Mazama temama Central American red brocket 304 4,794 9 48 D T
Order Cingulata
Dasypus novemcinctus Nine-banded armadillo 5 169 N T
Cabassous centralis Northern naked-tailed armadillo 2 N T
Order Chiroptera
Tonatia saurophila Stripe-headed round-eared bat 1 N V
Pteronotus gymnonotus
or parnellii
Big naked-backed bat or Parnell’s
mustached bat
3 N V
(Continues)
RODGERS ET AL. | 9
two other rodent species, red-tailed squirrel (Sciurus granatensis) and
Tome’s spiny-rat (Proechimys semispinosus), and two confamilial carni-
vores, crab-eating raccoon (Procyon cancrivorus) and kinkajou (Potos
flavus; Table 1).
There are three potential explanations for why we were unsuc-
cessful in detecting these three common species with metabarcod-
ing. The first possible explanation is that reference sequences for
those species are missing from the reference database. Only 59%
and 62% of mammal species and 82% and 85% of genera known
from BCI are represented in our 16S and 12S reference databases,
respectively. However, all three undetected species were present in
our reference databases to varying extents: paca was represented in
our database by 12S, coati was represented by both markers, albeit
only at the genus level (South American coati; Nasua nasua), and red
brocket deer was represented by both markers at the species level.
Given that PROTAX was able to place 98% of 16S OTUs, and 74% of
12S OTUs to at least the genus level (Figure 2), missing reference
sequences are not a strong explanation for our failure to detect
these species. However, in less-studied areas, incomplete reference
databases are more likely to hamper species assignment, and thus,
investment in building reference databases should continue.
The second potential explanation for why some common species
were not detected with metabarcoding is that there were mis-
matches in our 16S and 12S primers with binding sites for these
species. This could result in failed PCR amplification, even if DNA
from these species was present in fly samples. For paca, all nine ref-
erence sequences on GenBank have one mismatch in the forward
primer, 6 bp from the 3’ end. One mismatch is unlikely to completely
prevent amplification but could reduce PCR efficiency, especially in
mixed samples with other targets. For red brocket deer, no primer
mismatches are observed for 12S. For 16S, the forward primer has
one mismatch with all GenBank reference sequences, but at the far
5’ end. We could not evaluate primer mismatches for white-nosed
coati at the species level, but both 16S and 12S primer sets have
one 5’ mismatch to the congener Nasua nasua. Further, currently
available GenBank reference sequences may not account for local
sequence variants with primer mismatches.
Lastly, it is likely that fly feeding preferences may contribute to
sampling bias. Although carrion flies feed on carcasses of dead ani-
mals, we suspect that the main source of eDNA is faeces. Primates
were the most commonly detected species, with mantled howler
monkey (Alouatta palliata) and Geoffroy’s spider monkey (Ateles geof-
froyi) identified in 98% and 85% of samples, respectively. Monkeys,
and particularly howler monkeys which feed on leaves, produce an
abundant quantity of soft scat that can be easily consumed by flies,
while rodents such as agoutis and pacas, and ungulates such as red
brocket deer, have smaller and harder scats that may not be as
attractive to flies. It is notable that some rodent species were
detected nonetheless, but this could have been from carrion, and
not from faeces. Likewise, Calvignac-Spencer, Merkel et al. (2013)
were able to detect forest ungulates with carrion flies. Further
research into the source of carrion fly-derived eDNA and how this
affects detection is needed. Fly metabarcoding does seem to work
especially well for primates, as all four primate species present on
BCI were detected, and Calvignac-Spencer, Merkel et al. (2013) also
detected many primates with carrion fly metabarcoding in Cote
d’Ivoire and Madagascar.
For species assignment, PROTAX, especially weighted PROTAX, out-
performed other assignment methods. With weighted PROTAX, we
were able to assign nearly all OTUs to genus, and most to species,
especially with 16S. For those OTUs that we could only place at the
genus level, we could assign all but two to species based on prior
knowledge of the vertebrate community on BCI, as only one mem-
ber of each genus is present on the island. Weighted PROTAX resulted
in fewer clearly false positives (species not known from BCI) than
either BLAST or unweighted PROTAX (Figure 2c). However, even
weighted PROTAX produced a few false-positive assignments at the
species level. In cases where it is important to be conservative with
species assignment, and eliminate or minimize false assignments,
selecting a high probability threshold (e.g., 0.9 or 0.95) may be desir-
able. All but one of our false-positive assignments from weighted
PROTAX at the species level, and all at the genus level, had PROTAX
probabilities of <0.9 (Appendix S2). Assigning such a cut-off value,
however, is a trade-off between minimizing false positives and false
negatives. We detected several species known from BCI at assign-
ment probabilities of <0.9 (Table 1), and so with a higher probability
threshold we would have considered those species as undetected
even though they are present. The results from our weighted data
sets suggest that an effective way to simultaneously reduce false-
positive (overly confident) and false-negative (overly conservative)
assignments in PROTAX is to use expert knowledge to assign high prior
probabilities to species known to exist in the region (Figure 2).
As evidenced by the 24 flies that we analysed individually and in
pools, pooling reduced the number of species detected. When flies
were extracted individually, we detected 10 species, including ocelot
(Leopardus pardalis), Hoffmann’s two-toed sloth (Choloepus hoff-
manni), collared peccary (Pecari tajacu), kinkajou, crab-eating raccoon,
red-tailed squirrel and all four primates. In the two pooled samples
from these same 24 flies, we only detected collared peccary, kinka-
jou and the two most commonly detected primate species.
TABLE 2 (Continued)
Species Common name
Metabar-
coding Camera trapping Transect counts
Habita
16s 12s 2015 2008–2015 2015 2008–2015
Order Perissodactyla
Tapirus bairdii Baird’s tapir 2 54 1 N T
aHabit: D = diurnal, C = cathemeral, N = nocturnal, A = arboreal, ST = semi-terrestrial, T = terrestrial, V = volant.
10 | RODGERS ET AL.
Hoffmann’s two-toed sloth was only detected in one of the single-
fly extractions and was not detected in any of the pooled samples.
Thus, if we had not extracted some flies singly, this species would
have been missed entirely. Pooling is more likely to affect detection
of rare species, as their DNA may be outcompeted during PCR by
more abundant DNA of common species. It is also likely that species
with more primer mismatches will be outcompeted by species with
greater PCR efficiency if DNA of two such species exist in the same
pool. Pooling of flies, however, allows for far fewer DNA extractions
and individual PCRs, which substantially reduces labour and cost (up
to 16 fold in this example), which may allow many more flies to be
processed, ultimately increasing the number of species detected. If a
pooling strategy is employed, and there are particular target species
of concern for which detection is essential, it may be desirable to
include species-specific primers in addition to general primers (Schu-
bert et al., 2014). It is also possible that the high number of PCR
cycles we employed led to “PCR runaway,” in which common ampli-
cons became exponentially abundant in pooled samples at the
expense of less-common amplicons. Thus, we advise future studies
to optimize the number of PCR cycles with quantitative PCR prior to
metabarcoding (Murray, Coghlan, & Bunce, 2015). Finally, the dis-
crepancy in species detection between pooled and single-fly samples
could have been the result of insufficient sequencing depth. Because
single-fly and pooled samples were sequenced at the same depth, it
is possible that for pooled samples, sequencing depth was not suffi-
cient to detect all species in the sample.
The use of two different markers (12S and 16S) resulted in the
detection of more species than either marker alone. Thus, we recom-
mend using multiple markers to improve species detection. The
inclusion of the 12S marker allowed us to detect more mammal spe-
cies and also allowed detection of birds and reptiles. This marker,
however, appears to have relatively poor information content for
discriminating birds and reptiles at the species level. Thus, if the goal
is to detect nonmammalian vertebrates, it may be preferable to
employ additional group-specific markers. Multiple markers opti-
mized for different groups or families could be run simultaneously,
which should increase detection in each group. For mammals, 12S
species assignment were generally lower confidence than with 16S
(Appendix S2), and a greater proportion of 12S OTUs could not be
assigned at the genus or species level (Figure 2b). The standard cyto-
chrome oxidase I (COI) barcode region is by far the most repre-
sented in reference databases, but lack of conserved sites in the
coding region of COI make primer design for amplification of short
fragments from degraded DNA difficult for a wide range of taxa
(Deagle, Jarman, Coissac, Pompanon, & Taberlet, 2014). Thus, mito-
chondrial gene regions such as 16S and 12S are more appropriate
for metabarcoding studies targeting degraded DNA.
Choice of bait, trap configuration and trap distribution may have
an impact on carrion fly metabarcoding results. We chose to use
pork for bait because we wanted to target flies that feed on mam-
mals, and because a Sus blocking primer has been previously
designed for both markers we used (Calvignac-Spencer, Merkel et al.,
2013). We made an effort to keep bait separate from flies during
sampling, but even with the blocking primers, Sus DNA was still
commonly amplified. Sus OTUs accounted for 68% of total 16S
reads and 43% of total 12S reads in our data set. In a small explora-
tory study (unpublished data), we found that increasing Sus blocking
primer concentrations to 209 as opposed to the 59 used by Calvig-
nac-Spencer, Merkel et al. (2013) led to increased detection of non-
Sus DNA. If pork is used as bait, we recommend taking as much care
as possible to reduce contact of flies with the bait to reduce con-
tamination, and ensure enough sequencing depth so that non-Sus
amplicons are still sequenced even if they are in the minority. Use of
other baits such as chicken or fish, or nonbiological commercial fly
baits may mitigate these concerns, but the 12S marker will still
amplify chicken or fish DNA, and the 16S marker will also amplify
fish DNA (Cannon et al., 2016). Finally, most of the fly samples we
collected were from a relatively clustered area (Figure 1) whereas
transect counts and camera trapping were more widely distributed
throughout the island. It is possible that more widely distributed fly
sampling may have further improved species detection.
For maximal species detection, metabarcoding, camera trapping
and transect counts could be used simultaneously, as they are com-
plementary. All three techniques combined detected more species
than any one of the techniques alone. Conveniently, conducting fly
collection and transect counts simultaneously would add little addi-
tional field time or cost. During this study, we visually observed two
common species missed by metabarcoding, white-nosed coati and
red brocket deer, as well as numerous others, while collecting flies
(unpublished data). Also, if the goal was to survey biodiversity of
both vertebrates and invertebrates, fly traps could be paired with
other types of insect traps, for example pitfall traps, and fly DNA
and bulk DNA from traps could be amplified with invertebrate pri-
mers and included in sequencing runs (Ji et al., 2013; Yu et al.,
2012).
5 | CONCLUSIONS
This field test confirms that carrion fly-derived DNA metabarcod-
ing is a powerful tool for mammal biodiversity surveys, already on
par with other commonly used methods. A relatively small effort
(29 days of fly sampling conducted by one individual) detected
the majority of the nonvolant mammal species resident on BCI,
and a greater number of mammal species than were detected by
camera trapping or diurnal transect counts with similar sampling
effort. Nonetheless, carrion fly metabarcoding failed to detect
three abundant species that were easily detected with other
methods, and metabarcoding might not provide reliable estimates
of species relative abundance (Schnell et al., 2015). Thus, we do
not advocate that carrion fly metabarcoding replace current meth-
ods, but rather add to the tools available for biodiversity surveys.
Our results suggest some methodological modifications that will
likely increase the detection power of carrion fly metabarcoding,
including more complete reference sequence databases, optimiza-
tion of PCR conditions, multiple custom markers, greater sampling
RODGERS ET AL. | 11
effort and greater sequencing depth. Individual fly metabarcoding
could also increase detection power, but at a larger cost. Never-
theless, we conclude that carrion fly metabarcoding is an impor-
tant tool, with the potential to greatly improve researchers’ ability
to efficiently monitor biodiversity. As sequencing costs drop fur-
ther and as reference databases of complete mitochondrial gen-
omes becomes readily available (Tang et al., 2014), this method
promises to allow for the rapid characterization and monitoring of
vertebrate communities.
ACKNOWLEDGEMENTS
This work was funded by a Smithsonian Tropical Research Institute
fellowship (TR). Camera trapping and transect counts were funded
by the STRI Terrestrial Environmental Studies Project, the U.S.
Department of Education Math-Science Partnerships Project (JG),
and private funds from Gregory E. Willis. We also thank Gregory E.
Willis for years of mammals census field work. DY was supported by
the National Natural Science Foundation of China (31400470,
41661144002, 31670536, 31500305, GYHZ1754), the Ministry of
Science and Technology of China (2012FY110800), the University of
East Anglia, and the State Key Laboratory of Genetic Resources and
Evolution at the Kunming Institute of Zoology (GREKF13-13,
GREKF14-13, GREKF16-09).
DATA ACCESSIBILITY
Raw sequence data are available on the NCBI Sequence Read
Archive under Accession number PRJNA382243. OTU sequences
are available in the supplementary information. Scripts for the PRO-
TAX analysis are available on Dryad under https://doi.org/10.5061/
dryad.bj5k0.
AUTHOR CONTRIBUTIONS
T.W.R. designed the study, carried out all fly collection field work
and laboratory work, coordinated the data analysis, and wrote the
first draft of the manuscript. C.C.Y.X. performed the PROTAX analysis;
J.G. collected the camera trap and transect count data; K.M.K. con-
ducted the sequence filtering and OTU generation bioinformatics
work; K.S. helped with design and implementation of laboratory
work; M.V. helped with laboratory work management and conducted
MiSeq sequencing; D.W.Y. supervised the PROTAX analysis; P.S. pro-
vided scripts and helped with the PROTAX analysis; and P.A.J. helped
with field study design. W.O.M. and P.A.J. jointly advised this work.
All authors contributed to manuscript writing and editing.
REFERENCES
Altschul, S. F., Gish, W., Miller, W., Myers, E. W., & Lipman, D. J. (1990).
Basic local alignment search tool. [Article]. Journal of Molecular Biol-
ogy, 215(3), 403–410. https://doi.org/10.1006/jmbi.1990.9999
Andersen, K., Bird, K. L., Rasmussen, M., Haile, J., Breuning-Madsen, H.,
Kjaer, K. H., . . . Willerslev, E. (2012). Meta-barcoding of ‘dirt’ DNA
from soil reflects vertebrate biodiversity. Molecular Ecology, 21(8),
1966–1979. https://doi.org/10.1111/j.1365-294x.2011.05261.x
Beaudrot, L., Ahumada, J. A., O’Brien, T., Alvarez-Loayza, P., Boekee, K.,
Campos-Arceiz, A., . . . Andelman, S. J. (2016). Standardized assess-
ment of biodiversity trends in tropical forest protected areas: The
end is not in sight. PLoS Biology, 14(1), e1002357. https://doi.org/10.
1371/journal.pbio.1002357
Bermudez, C. S. E., Gonzalez, D. D., & Garcia, S. G. (2013). Ticks (Acari:
Ixodidae, Argasidae) of Coyotes in Panama [Article]. Systematic and
Applied Acarology, 18(2), 112–115.
Boessenkool, S., Epp, L. S., Haile, J., Bellemain, E., Edwards, M., Coissac,
E., . . . Brochmann, C. (2012). Blocking human contaminant DNA dur-
ing PCR allows amplification of rare mammal species from sedimen-
tary ancient DNA. Molecular Ecology, 21(8), 1806–1815. https://doi.
org/10.1111/j.1365-294x.2011.05306.x
Bohmann, K., Evans, A., Gilbert, M. T. P., Carvalho, G. R., Creer, S.,
Knapp, M., . . . de Bruyn, M. (2014). Environmental DNA for wildlife
biology and biodiversity monitoring. Trends in Ecology & Evolution, 29
(6), 358–367. https://doi.org/10.1016/j.tree.2014.04.003
Bohmann, K., Schnell, I. B., & Gilbert, M. T. P. (2013). When bugs reveal
biodiversity. Molecular Ecology, 22(4), 909–911. https://doi.org/10.
1111/mec.12221
Boyer, F., Mercier, C., Bonin, A., Le Bras, Y., Taberlet, P., & Coissac, E.
(2016). obitools: A unix-inspired software package for DNA metabar-
coding. Molecular Ecology Resources, 16(1), 176–182. https://doi.org/
10.1111/1755-0998.12428
Calvignac-Spencer, S., Leendertz, F. H., Gilbert, M. T. P., & Schubert, G.
(2013). An invertebrate stomach’s view on vertebrate ecology Certain
invertebrates could be used as “vertebrate samplers” and deliver
DNA-based information on many aspects of vertebrate ecology.
BioEssays, 35(11), 1004–1013. https://doi.org/10.1002/bies.
201300060
Calvignac-Spencer, S., Merkel, K., Kutzner, N., Kuhl, H., Boesch, C., Kap-
peler, P. M., . . . Leendertz, F. H. (2013). Carrion fly-derived DNA as a
tool for comprehensive and cost-effective assessment of mammalian
biodiversity. Molecular Ecology, 22(4), 915–924. https://doi.org/10.
1111/mec.12183
Cannon, M. V., Hester, J., Shalkhauser, A., Chan, E. R., Logue, K., Small, S.
T., & Serre, D. (2016). In silico assessment of primers for eDNA stud-
ies using PrimerTree and application to characterize the biodiversity
surrounding the Cuyahoga River. Scientific Reports, 6, 1–11. https://d
oi.org/10.1038/srep22908
Caporaso, J. G., Kuczynski, J., Stombaugh, J., Bittinger, K., Bushman, F.
D., Costello, E. K., . . . Knight, R. (2010). QIIME allows analysis of
high-throughput community sequencing data. Nature Methods, 7(5),
335–336. https://doi.org/10.1038/nmeth.f.303
Clark, K., Karsch-Mizrachi, I., Lipman, D. J., Ostell, J., & Sayers, E. W.
(2016). GenBank. Nucleic Acids Research, 44(D1), D67–D72. https://d
oi.org/10.1093/nar/gkv1276
Deagle, B. E., Jarman, S. N., Coissac, E., Pompanon, F., & Taberlet, P.
(2014). DNA metabarcoding and the cytochrome c oxidase subunit I
marker: Not a perfect match. Biology Letters, 10(9), 1–4. https://doi.
org/10.1098/rsbl.2014.0562
Edgar, R. C. (2010). Search and clustering orders of magnitude faster
than BLAST. Bioinformatics, 26(19), 2460–2461.
Ficetola, G. F., Coissac, E., Zundel, S., Riaz, T., Shehzad, W., Bessiere, J.,
. . . Pompanon, F. (2010). An In silico approach for the evaluation of
DNA barcodes. Bmc Genomics, 11(1), 434. https://doi.org/10.1186/
1471-2164-11-434
Gillett, C. P. D. T., Johnson, A. J., Barr, I., & Hulcr, J. (2016). Metagenomic
sequencing of dung beetle intestinal contents directly detects and
identifies mammalian fauna. bioRxiv. https://doi.org/10.1101/074849
Glanz, W. E. (1982). The terrestrial mammal fauna of Barro Colorado
Island: Censuses and long-term changes. In E. G. Leigh, A. S. Rand, &
D. M. Windsor (Eds.), The ecology of a tropical forest: Seasonal rhythms
12 | RODGERS ET AL.
and long-term changes (pp. 455–468). Washington, DC: Smithsonian
Institution Press.
Huson, D. H., Auch, A. F., Qi, J., & Schuster, S. C. (2007). MEGAN analy-
sis of metagenomic data. Genome Research, 17(3), 377–386. https://d
oi.org/10.1101/gr.5969107
Ji, Y., Ashton, L., Pedley, S. M., Edwards, D. P., Tang, Y., Nakamura, A.,
. . . Yu, D. W. (2013). Reliable, verifiable and efficient monitoring of
biodiversity via metabarcoding. Ecology Letters, 16(10), 1245–1257.
https://doi.org/10.1111/ele.12162
Kielbasa, S. M., Wan, R., Sato, K., Horton, P., & Frith, M. C. (2011). Adap-
tive seeds tame genomic sequence comparison. Genome Research, 21
(3), 487–493. https://doi.org/10.1101/gr.113985.110
Lee, P. S. (2016). A blowfly-derived dna approach to assess diversity of trop-
ical mammals. PhD thesis, University of Malaysia. Retrieved from
http://studentsrepo.um.edu.my/6715/
Lee, P. S., Gan, H. M., Clements, G. R., & Wilson, J. J. (2016). Field cali-
bration of blowfly-derived DNA against traditional methods for
assessing mammal diversity in tropical forests. Genome, 59(11), 1008–
1022. https://doi.org/10.1139/gen-2015-0193
Lee, P. S., Sing, K. W., & Wilson, J. J. (2015). Reading mammal diversity
from flies: The persistence period of amplifiable mammal mtDNA in
blowfly guts (Chrysomya megacephala) and a new DNA mini-barcode
target. PLoS ONE, 10(4), 1–12. https://doi.org/10.1371/journal.pone.
0123871
Logue, K., Keven, J. B., Cannon, M. V., Reimer, L., Siba, P., Walker, E. D.,
. . . Serre, D. (2016). Unbiased characterization of anopheles mosquito
blood meals by targeted high-throughput sequencing. PLOS Neglected
Tropical Diseases, 10(3), e0004512. https://doi.org/10.1371/journal.
pntd.0004512
Martin, M. (2011). Cutadapt removes adapter sequences from high-
throughput sequencing reads. EMBnet Journal, 17(1), 10–12. https://d
oi.org/10.14806/ej.17.1.200
Murray, D. C., Coghlan, M. L., & Bunce, M. (2015). From benchtop to
desktop: Important considerations when designing amplicon sequenc-
ing workflows. PLoS ONE, 10(4), e0124671. https://doi.org/10.1371/
journal.pone.0124671
Nichols, R. V., Konigsson, H., Danell, K., & Spong, G. (2012). Browsed
twig environmental DNA: Diagnostic PCR to identify ungulate spe-
cies. Molecular Ecology Resources, 12(6), 983–989. https://doi.org/10.
1111/j.1755-0998.2012.03172.x
Oksanen, J., Blanchet, F. G., Friendly, M., Kindt, R., Legendre, P., McGlinn,
D., . . . Wagner, H. (2016). vegan: Community Ecology Package. R pack-
age version 2.4-1.
Riaz, T., Shehzad, W., Viari, A., Pompanon, F., Taberlet, P., & Coissac, E.
(2011). ecoPrimers: Inference of new DNA barcode markers from
whole genome sequence analysis. Nucleic Acids Research, 39(21),
e145. https://doi.org/10.1093/nar/gkr732
Rodgers, T. W., & Mock, K. E. (2015). Drinking water as a source of envi-
ronmental DNA for the detection of terrestrial wildlife species. Con-
servation Genetics Resources, 7(3), 693–696. https://doi.org/10.1007/
s12686-015-0478-7
Schnell, I. B., Sollmann, R., Calvignac-Spencer, S., Siddall, M. E., Yu, D.
W., Wilting, A., & Gilbert, M. T. P. (2015). iDNA from terrestrial hae-
matophagous leeches as a wildlife surveying and monitoring tool -
prospects, pitfalls and avenues to be developed. Frontiers in Zoology,
12, 1–14. https://doi.org/10.1186/s12983-015-0115-z
Schubert, G., Stockhausen, M., Hoffmann, C., Merkel, K., Vigilant, L.,
Leendertz, F. H., & Calvignac-Spencer, S. (2014). Targeted detection
of mammalian species using carrion fly-derived DNA. Molecular Ecol-
ogy Resources, 15(2), 285–294. https://doi.org/10.1111/1755-0998.
12306
Somervuo, P., Koskela, S., Pennanen, J., Henrik Nilsson, R., & Ovaskainen,
O. (2016). Unbiased probabilistic taxonomic classification for DNA
barcoding. Bioinformatics, 32(19), 2920–2927. https://doi.org/10.
1093/bioinformatics/btw346
Somervuo, P., Yu, D. W., Xu, C., Ji, Y., Hultman, J., Wirta, H., & Ovaskai-
nen, O. (2017). Quantifying uncertainty of taxonomic placement in
DNA barcoding and metabarcoding. Methods in Ecology and Evolution,
8(4), 398–407. https://doi.org/10.1111/2041-210x.12721
Taberlet, P., Coissac, E., Hajibabaei, M., & Rieseberg, L. H. (2012). Envi-
ronmental DNA. Molecular Ecology, 21(8), 1789–1793. https://doi.
org/10.1111/j.1365-294x.2012.05542.x
Tang, M., Tan, M., Meng, G., Yang, S., Su, X., & Liu, S. (2014). Multiplex
sequencing of pooled mitochondrial genomes – A crucial step toward
biodiversity analysis using mito-metagenomics. Nucleic Acids Research,
42, https://doi.org/10.1093/nar/gku917
Thomsen, P. F., Kielgast, J., Iversen, L. L., Wiuf, C., Rasmussen, M., Gil-
bert, M. T. P., . . . Willerslev, E. (2012). Monitoring endangered
freshwater biodiversity using environmental DNA. Molecular Ecology,
21(11), 2565–2573. https://doi.org/10.1111/j.1365-294x.2011.
05418.x
Wheat, R. E., Allen, J. M., Miller, S. D. L., Wilmers, C. C., & Levi, T.
(2016). Environmental DNA from residual saliva for efficient noninva-
sive genetic monitoring of brown bears (Ursus arctos). PLoS ONE, 11
(11), e0165259. https://doi.org/10.1371/journal.pone.0165259
Yu, D. W., Ji, Y., Emerson, B. C., Wang, X., Ye, C., Yang, C., & Ding, Z.
(2012). Biodiversity soup: Metabarcoding of arthropods for rapid bio-
diversity assessment and biomonitoring. Methods in Ecology and Evo-
lution, 3(4), 613–623. https://doi.org/10.1111/j.2041-210x.2012.
00198.x
Zhang, J., Kobert, K., Flouri, T., & Stamatakis, A. (2014). PEAR: A fast and
accurate Illumina Paired-End reAd mergeR. Bioinformatics, 30(5), 614–
620. https://doi.org/10.1093/bioinformatics/btt593
SUPPORTING INFORMATION
Additional Supporting Information may be found online in the sup-
porting information tab for this article.
How to cite this article: Rodgers TW, Xu CCY, Giacalone J,
et al. Carrion fly-derived DNA metabarcoding is an effective
tool for mammal surveys: Evidence from a known tropical
mammal community. Mol Ecol Resour. 2017;00:1–13.
https://doi.org/10.1111/1755-0998.12701
RODGERS ET AL. | 13