Detection of a new non-native freshwater species by DNA metabarcoding of environmental samples – first record of Gammarus fossarum in the UK

We report the discovery of a non-native gammarid, Gammarus fossarum (Koch, 1836) (Crustacea, Amphipoda), in UK rivers. Gammarus fossarum is a common freshwater gammarid in many parts of mainland Europe, but was previously considered absent from the UK. Gammarus fossarum was detected in a number of UK rivers following DNA metabarcoding of a mini-barcode region of the COI gene in macroinvertebrate kick samples, and environmental DNA (eDNA) from water and sediment samples. Subsequent morphological analysis and standard DNA barcoding showed that the species could be reliably identified and separated from Gammarus pulex (Linnaeus, 1758), the most dominant and widespread native freshwater gammarid in the UK. Our data demonstrate extensive geographical coverage of G. fossarum in the UK, spanning distant river catchments. At present there is no data to confirm the likely origin of G. fossarum’s introduction. Subsequent re-examination of historic archive material shows the species to have been present in the UK since at least 1964. This study is among the first to demonstrate the potential of eDNA metabarcoding for detection of new non-native species.


Introduction
Amphipods are successful invaders in freshwater ecosystems, with many invasive non-native species (INNS) having been observed to adversely impact indigenous species within Europe over the last century (Bij de Vaate et al. 2002;Grabowski et al. 2007).The introduction of non-native amphipods may not only lead to displacement of native congeners (e.g.Dick and Platvoet 2000;MacNeil and Platvoet 2005;Kinzler et al. 2009), but may also impact on ecosystem structure and functioning (MacNeil et al. 2011;Piscart et al. 2011;Constable and Birkby 2016) and introduce novel pathogens to newly colonised areas (Bacela-Spychalska et al. 2012).
Once non-native species are widely established, efforts to reduce their impacts are often problematic, hence management strategies are strongly focused on preventing introductions or spread (e.g. the "check, clean, dry" campaign in the UK).Early detection is key to such strategies, either to improve the success of eradication programs or to prevent further establishment and dispersal (Roy et al. 2014;Dejean et al. 2012).For freshwater macroinvertebrates, INNS detection methods typically rely on sampling programmes and morphological identification.However, the standard UK monitoring method for macroinvertebrates, a three minute kick sample, will typically recover 62% of families and 50% of species at a site (Furse et al. 1981).This can present considerable challenges when dealing with rare or elusive species.Morphological identification can also prove difficult when identifying taxonomically similar or cryptic species, or juvenile life stages, and is highly dependent on the taxonomical expertise of the investigator.Emerging molecular detection methods may provide significant benefit for detecting non-native species in aquatic environments (Darling and Mahon 2011;Lawson Handley 2015).
One new and rapidly developing method is the use of environmental DNA (eDNA) (Taberlet et al. 2012a, b;Rees et al. 2014;Lawson Handley 2015), which refers to cellular or extracellular DNA that can be extracted directly from environmental samples without prior separation of taxa (Taberlet et al. 2012a).Environmental DNA has been successfully used in numerous studies to detect specific taxa using a targeted approach based on standard or quantitative PCR (Dejean et al. 2012;Dougherty et al. 2016).In an alternative approach, called "metabarcoding", entire species assemblages are analysed by PCR with broadly conserved primers, followed by Next Generation Sequencing (NGS: see Lawson Handley 2015; Hänfling et al. 2016;Port et al. 2016;Valentini et al. 2016 for further detail).Environmental DNA metabarcoding has been successfully used in a small number of studies, for example, to describe entire communities of vertebrates (e.g.Lawson Handley 2015;Hänfling et al. 2016;Port et al. 2016;Valentini et al. 2016) and invertebrates (Deiner et al. 2016) from marine, lake and river samples.Metabarcoding has excellent potential as an early warning tool for detection of non-native species from samples collected from invasion pathways or natural/semi-natural habitats (Mahon and Jerde 2016;Lawson Handley 2015).For example, the technique was recently used as an early detection method for screening ship ballast, and detected non-indigenous zooplankton in Canadian ports (Brown et al. 2016).Environmental DNA metabarcoding has also identified non-native fish species present in samples from the live bait trade (white perch, Morone americana (Gmelin, 1789) Mahon et al. 2014) and in river samples (northern snakehead, Channa argus (Cantor, 1842) Simmons et al. 2015).However the number of applications of metabarcoding for detection of nonnative species has so far been limited.
In this paper we describe the detection of Gammarus fossarum (Koch, 1836), a newly recognised freshwater amphipod to the UK, using macroinverte-brate community and eDNA metabarcoding.The species was found in several UK rivers following a preliminary non-targeted sampling programme for macroinvertebrate communities based on metabarcoding of a 313 bp mini-barcode region of the cytochrome c oxidase subunit I (COI) gene, and was subsequently confirmed using a combination of morphological analysis and standard full-length COI DNA barcoding (via Sanger sequencing).This study demonstrates the power of eDNA metabarcoding for detection of non-native species in natural habitats.

Sampling
Field surveys were carried out in March 2015 within 8 UK river catchments (Figure 1, Maps A-H, excluding E).At each site (n = 65) environmental variables including water depth, width, substrate type and surrounding habitat were recorded.Three sample types were collected at each site: a three minute macroinvertebrate kick sample (Murray-Bligh 1999) for identification by microscopy analysis and high molecular weight DNA extraction from pools of individuals; and water and sediment samples were collected for eDNA extraction.Two litres of water was sampled from the surface by collecting 4 × 500 ml from points across the river width using a sterile bottle.Sediment samples were collected from points across the river width using a trowel, and the material was placed in a 42 fluid oz.sterile Whirl-pak ® bag (Cole-Palmer, Hanwell, London).All sampling equipment was sterilized in 10% commercial bleach solution for 10 minutes then rinsed with 10% MicroSol detergent (Anachem, UK) and purified water between samples.Sample bottles filled with ddH 2 O were taken into the field and later filtered as sample blanks.

Macroinvertebrate community sample processing
All macroinvertebrates from each kick sample were sorted and identified to the lowest taxonomic level possible, before being stored in sterile 50 ml falcon tubes filled with 100% ethanol.For DNA extraction, samples were dried to remove the ethanol and the entire macroinvertebrate community was lysed in a Qiagen Tissue Lyser ® with Digisol (50mM Tris, 20M EDTA, 120 mM NaCl and 1% SDS) (3 × 30 sec).Samples were then incubated overnight at 55 °C with SDS and Proteinase K. DNA from a 200 μl subsample of the lysed tissue was extracted using the DNeasy Blood & Tissue Kit ® (Qiagen, Hilden, Germany) according to the manufacturer's protocol.S1 for further site information (Pebesma et al. 2005;Wickham. 2009;Bivand et al. 2013;Bivand et al. 2016;Gallic. 2016) Contains OS data © Crown copyright and database right (2016).
Environmental DNA sample processing Water samples were filtered within 24 hours through sterile 47 mm diameter 0.45 μm cellulose nitrate membrane filters and pads (Whatman, GE Healthcare, UK), using Nalgene filtration units attached to a vacuum pump.Sediment samples were stored at −20 °C within 12 hours of sampling.The sample was defrosted, mixed and 200 ml of sediment placed in a sterile measuring cylinder with 500 ml of molecular grade water, then inverted 10 times and left to stand for 30 s, the supernatant was then poured off into a sterile container.This procedure was repeated twice.Two hundred and fifty millilitres of the supernatant was then prefiltered through sterile 20 μm filter paper (Whatman, GE Healthcare, UK), and the filtrate subsequently filtered through 0.45 μm cellulose nitrate filters, as for the water samples.Filter papers were stored in sterile petri dishes at −20 °C until extraction.Filtration blanks (2 L purified water) were run before the samples for each filtration run to test for contamination at the filtration stage (n = 5).Filtration equipment was sterilized in 10% commercial bleach solution for 10 minutes then rinsed with 10% MicroSol detergent and purified water after each filtration.
Environmental DNA from both water and sediment samples was extracted using PowerWater ® DNA Isolation Kit (MoBio Laboratories, Inc. Carlsbad, USA) following the manufacturer's instructions.

PCR, library prep and sequencing
We chose to use COI for metabarcoding because this region has the broadest taxonomic coverage for macroinvertebrates in public sequence databases and is the most widely used DNA barcode for taxonomic discrimination in this group.A 313 bp fragment ("mini-barcode") was targeted using the primers described in Leray et al. (2013).For library preparation we used a nested tagging protocol, modified from the Illumina 16S two-step metabarcoding protocol (Illumina 2011) as outlined in Kitson et al. (2015).
In the first step, PCRs were performed with modified versions of the primers jgHCO2198 TAIA CYTCIGGRTGICCRAARAAYCA and mICOIintF GGWACWGGWTGAACWGTWTAYCCYCC (Leray et al. 2013).In addition to the standard primer sequence, primers included one of eight unique forward or 12 unique reverse 8-nucleotide Molecular Identification Tags (MID), plus a bridge site, which acts as a binding site for PCR 2 (see Kitson et al. 2015 for full details).PCRs were carried out in 25 μl volumes with MyFi High-Fidelity Taq (Bioline, UK) containing: 10 μM of each primer, and 2 μl of undiluted DNA template.PCRs were performed on an Applied Bio-systems Veriti Thermal Cycler with the following profile: initial denaturation at 95 °C for 1 min, followed by 45 cycles of denaturation at 98 °C for 15 s, annealing at 51 °C for 15 s and extension at 72 °C for 30 s, with a final extension time of 10 min at 72 °C.This included PCR and filtering blanks (n = 3 and n = 5, respectively) and single species positives: Triops cancriformis (Bosc, 1801) (n = 2) and Harmonia  axyridis (Pallas, 1773) (n = 2).PCR products were confirmed by gel electrophoresis on a 2% agarose gel stained with ethidium bromide.PCRs were carried out three times and then pooled.Pooled PCR products were then purified using the E.Z.N.A Cycle Pure Kit ® (VWR International, Leicestershire).
In the second PCR step, Illumina adapters and additional forward and reverse MID tags were added in a second PCR with 10 μM of each tagging primer and 2 μl of purified PCR product.PCR settings were: initial denaturation at 95 °C for 3 min, followed by 12 cycles of denaturation at 98 °C for 20 s, annealing at 72 °C for 1 min and extension at 72 °C for 5 mins, with a final extension time of 10 mins at 4 °C (Kitson et al. 2015).
Samples were then classified into five categories based on the strength of band produced on ethidium bromide-stained agarose gels.Negative controls (including filtration blanks) produced no bands on the agarose gel so were categorised with samples with the lowest band strengths when being added to the library.All positive control (i.e.extracted tissue) samples were categorised as high band strength.Volumes of the samples were then pooled according 5 band strength categories: 10 μl for the lowest band strength, then decreasing volumes of 8 μl, 6 μl, 4 μl, and 2 μl for increasing band strength.The library was then pooled and cleaned using AMPure XP beads following the recommended manufacturer's protocol (Agencourt AMPure XP, Beckman Coulter Inc. US).The library was run at a 12 pM concentration on an Illumina MiSeq, at the in-house facility at the University of Hull, using the 2 × 300 bp V3 chemistry.

Specimen confirmation -microscopy and standard DNA barcode sequencing:
Verification of the results from DNA metabarcoding was carried out using a combination of morphological identification and standard DNA barcoding (by Sanger sequencing).
Gammarus fossarum is a well-studied diverse species complex, which has three well established cryptic species (types A, B and C) with a further 36-53 different cryptic lineages being identified through phylogenetic studies (Weiss et al. 2014;Copilaş-Ciocianu and Petrusek 2015).Species within this complex are known to differ in their ecology both in terms of their environmental requirements and geographic distribu-tions (Copilaş-Ciocianu and Petrusek 2015; Eisenring et al. 2016).The G. fossarum complex belongs to the G. pulexgroup, which means it has small oval or kidney shaped eyes (less than twice as long as wide) and the pereopods 5-7 are armed with spines and few setae (Pinkster 1972).Within the UK, these features alone would help to separate it from G. duebeni, G. tigrinus and G. zaddachi.It can be distinguished from all five known UK freshwater Gammarus residents by examining uropod III.In G. fossarum the ratio length of the endopod versus the exopod is about 0.5, whilst in the other five it is >0.5, typically 0.75 (see Figure 2B and 2D respectively).Another feature of G. fossarum is that only the inside margin of the exopod has plumose setae, whilst the other five have plumose setae on both inner and outer margins (see Figure 2C and 2E respectively).The latter feature should however be used with caution, as plumose setae on the outer margin of the exopod can show up in very old males of G. fossarum (Meijering 1972).
A post hoc morphological examination of UK Gammarus specimens was carried out to confirm the presence of G. fossarum.Since the entire macroinvertebrate samples from the original sampling program had been lysed for metabarcoding, new specimens were collected by hand net from two catchments where G. fossarum was detected by metabarcoding in close proximity to previously sampled sites; River Taff, Wales (n = 38) on 7/6/2016 and River Frome, England (n = 39) on 27/6/2016.Additional, archived specimens obtained from the Nailbourne (Little Stour catchment), England (n = 2) on 20/4/2013, were also analysed; (see Table 1 and Figure 1, Maps: E, F and G).Collected individuals were then subject to morphological examination and identified using Karaman and Pinkster (1977), Eggers and Martens (2001) and Piscart and Bollache (2012).
Microscopic identification was carried out on all specimens collected for morphological confirmation.Both G. fossarum (n = 37) and G. pulex (n = 1) were identified from individuals collected from the River Taff and only G. fossarum (n = 39) was found in a sample from the River Frome.Standard DNA barcoding was performed on some of the individuals identified morphologically as G. fossarum (n = 3) and G. pulex (n = 1) from the River Taff, and G. fossarum from the Nailbourne (Little Stour catchment) (n = 2).DNA was extracted using the DNeasy Blood & Tissue Kit ® (Qiagen, Hilden, Germany) according to the manufacturer's protocol.The full length COI DNA barcoding fragment was amplified (Folmer et al. 1994) using the following protocol: PCRs were performed in 25 μl volumes with MyTaq (Bioline, UK), 10 μM of each primer and 2 μl of DNA template.The PCR profile consisted of: initial denaturation at 95 °C for 1 min, followed by 35 cycles of denaturation at 95 °C for 15 s, annealing at 50 °C for 15 s and extension at 72 °C for 10 s, with a final extension time of 10 min at 72 °C.PCR products were checked on agarose gels and commercially sequenced using HCO2198 (Macrogen Europe, Amsterdam, Netherlands).

Bioinformatics
Processing of Illumina read data and taxonomic assignment were performed using a custom bioinformatics pipeline (metaBEAT, v.0.97.7-global; see Github reference 1) as described previously (Hänfling et al. 2016), with minor modifications.For each sample, raw Illumina sequences were filtered to retain only read pairs containing the expected forward/reverse in-line barcode combination (perfect matches only) using the program process_shortreads from the Stacks v1.20 program suite (Catchen et al. 2013) and subsequently quality trimmed using the program Trimmomatic v0.32 (Bolger et al. 2014).Specifically, read quality was assessed across 5 bp sliding windows starting from the 3'-end, and reads were clipped until the per window average read quality reached a minimum of phred 30.Any reads shorter than 100 bp after the quality clipping were discarded.To remove PCR primers and spacer sequences the first 30 bp of the reads was clipped off.Remaining sequence pairs were merged into single high quality reads using the program FLASH v1.2.11 (Magoč and Salzberg 2011).For any read pairs not merged successfully, only the forward read was retained for downstream analyses.Sequences were clustered at 97% identity using vsearch v1.1 (see Github reference 2).Any clusters represented by less than three sequences were excluded from further analyses, as these likely represent sequencing error.Each of the remaining distinct sequence clusters was collapsed to a single representative sequence (aka centroid).Only centroid sequences of the expected length as determined by the primers (313 bp ± 5%) were retained for downstream analyses.To obtain a final set of non-redundant (nr) queries for taxonomic assignment, centroid sequences across all samples were clustered globally at 97% identity using vsearch v1.1.The global set of nr queries was subjected to a BLAST (Zhang et al. 2000) search (blastn) against a custom reference database consisting of gammarid sequences from Weiss et al. (2014) and two CO1 sequences from T. cancriformis (GenBank accession numbers EF189678.1 and JX110644.1)and H. axyridis (accession numbers KU188381.1 and KU188380.1),respectively.Taxonomic assignment was performed using a lowest common ancestor (LCA) approach.In brief, after the BLAST search the algorithm identifies the most significant matches to the reference database (top 10% bit-scores) for each of the query sequences.If only a single taxon is present in this list of matches then the query is assigned directly to this taxon.If more than one taxon is present, the query is assigned to the lowest taxonomic level that is shared by all taxa in the list.
Queries yielding best BLAST matches below a bitscore of 80 or with less than 85% identity were binned as "unassigned".To assure full reproducibility of our analyses we have deposited the entire workflow in an additional dedicated Github repository (see Github reference 3).To reduce the possibility of false positives based on our single species positive samples and in order to obtain a conservative estimate of the distribution of G. fossarum in the UK, we only report G. fossarum as present at a given site if it was supported by at least 1% of the total quality trimmed reads per sample.

Phylogeny
Phylogenetic analysis was performed to further confirm the identity of the putative Gammarus sp.sequences obtained as part of the current study.We downloaded a previously published CO1 dataset (Weiss et al. 2014;Copilaş-Ciocianu and Petrusek 2015) from Genbank, comprising 89 sequences of G. fossarum, six G. pulex (Linnaeus, 1758) sequences and a single sequence each from four further outgroup species (G.balcanicus (Schaferna, 1922), G. glabratus (Hou and Li, 2003), G. roeselii (Gervais, 1835) and G. tigrinus (Sexton, 1939) (Radulovici et al. 2009;Hou et al. 2011;Feckler et al. 2012;Weiss et al. 2014).This set of previously published sequences was extended by the sequences obtained via standard full-length DNA barcoding and mini-barcode metabarcoding.Prior to phylogenetic analysis we extracted the most abundant sequence, i.e. haplotype, from each sample from the initially obtained 97% sequence clusters assigned to G. fossarum and G. pulex, respectively.Nucleotide sequences of G. fossarum and G. pulex used in the phylogenetic analysis were deposited in Genbank (GenBank accession KY464959-KY464977).Phylogenetic analysis was performed in the Reprophylo environment (Szitenberg et al. 2015).In brief, sequences were aligned using the program MAFFT v7.123b (Katoh and Standley 2013) and the alignment was trimmed using the program trimAl v1.2rev59 (Capella-Gutiérrez et al. 2009).Maximumlikelihood tree inference was performed using RAxML v8.0.12 (Stamatakis 2014).The full, detailed analysis is provided as Jupyter notebook in the dedicated Github repository (Github reference 3), which also contains the alignment underlying the phylogenetic tree and further supplementary information.

Comparison of data from eDNA/DNA and microscopy analysis
A correlation was performed to compare the Gammaridae abundance data generated from the kick sample microscopy analysis and the DNA/eDNA metabarcoding.Specifically, the relationship between DNA/eDNA data (read count) and data from microscopy analysis (biomass calculated from average Gammaridae specimen weight) was investigated by calculating Pearson's Correlation Coefficient in R v3.1.3 (R Core team 2013).Note that G. fossarum and G. pulex sequencing data have been combined here as the species were not distinguished during the initial morphological determination.

Metabarcoding survey
The total sequence read count passing quality control, before removal of chimeric sequences, was 4,290,271.We quantified the level of possible contamination using sequence information from single species positive samples, which enabled us to choose a suitable threshold level (1% of total sample reads) for filtering and removal of low level contamination.This conservative threshold is comparable to recent, similar studies (e.g.Hänfling et al. 2016;Port et al. 2016).After applying this threshold, over the 195 samples the total read count was 933,457.
Gammarus fossarum was detected in 28 sites in total: 25 via metabarcoding, 1 site by morphological identification, 1 site by standard DNA barcoding and 1 site by morphological identification and DNA barcoding (See Table 1 and Supplementary material Table S1).Of the 25 metabarcoding samples, G. fossarum was found in: 25 DNA macroinvertebrate samples, 8 water eDNA samples and 9 sediment eDNA samples.G. pulex was detected in 27 of the sites in the metabarcoding DNA macroinvertebrate samples only and a single site using Sanger sequencing.
A full breakdown of gammarid sequences per sample and proportion of gammarid biomass per sample are included in Supplementary material Table S1.A further 36 freshwater macroinvertebrate families were detected by metabarcoding: data from these non-gammarid species form part of a wider macroinvertebrate data set which is being analysed separately and will be published elsewhere.
The average read count of the samples with gammarid species present was 3512.At those sites the proportion of G. fossarum reads per sample ranged from 1.68 -100% in the macroinvertebrate DNA, 1.67 -55.35% in the water eDNA and 1.59 -18.05% in sediment eDNA samples (Table S1).Similarly, G. pulex reads ranged from 1.65 -97.41% in the DNA macroinvertebrate samples.There was a significant positive correlation between the percentage of Gammarus biomass in the sample, and the percentage of Gammarus sequence reads (Pearson's r = 0.747, df = 46, P = 1.098 × 10 -9 , Supplementary material Figure S1).Importantly, Gammarus sequences were detected when gammarids constituted as little as 2.6% of the total biomass (Table S1).

Verification of Gammarus fossarum by microscopy
Gammarus fossarum was not identified morphologically in any samples surveyed in March 2015 prior to metabarcoding.Of the 38 gammarid specimens recovered from the River Taff on 7/6/2016, 37 G. fossarum morphological identifications were made.Adult males ranged between 8-12 mm (n = 21) and adult females 7-10 mm (n = 15).Four females were ovigerous.The other gammarid specimen encountered was a male G. pulex (13 mm).Of the 39 gammarid specimens collected from the River Frome on 27/6/2016, all were identified as G. fossarum morphologically.Adult males of this population ranged from 8-11.5 mm (n = 24) and adult females 7-9 mm (n = 15).Again, four ovigerous females were recorded.The relative abundance of size distribution in the two sampled populations can be seen in the Supplementary information (Figure S2).The two individuals collected from the Nailbourne on 20/4/2013 were not verified using microscopy as the specimens were too heavily damaged for morphological identification.

Verification of Gammarus fossarum by DNA barcoding
Morphological identifications were confirmed by DNA sequencing for specimens collected from the River Taff (n = 4): 3 specimens of G. fossarum and a single G. pulex.The individuals collected from the Nailbourne (n = 2) were also both identified as G. fossarum using subsequent DNA barcoding (see Table 1).

Phylogeny
The phylogeny (Figure 3) is congruent with the findings of the morphological identification.The G. cf.fossarum and G. cf.pulex sequences cluster with their respective lineages (identified in Weiss et al. 2014;Copilaş-Ciocianu and Petrusek 2015).Gammarus fossarum sequences obtained by both metabarcoding and standard DNA barcoding show little divergence and cluster together in the phylogeny, indicating closely related sequences.The G. fossarum sequences obtained in the current study group with high statistical support within Clade 11, as defined using the distance based Automatic Barcode Gap Discovery (ABGD) approach in Weiss et al. (2014).Sequences further group in a subclade with samples from southwestern Germany, Southern Black Forest and Eastern Sauerland in Germany, i.e. clade 14, as delineated using the tree-based GMYC in Weiss et al. (2014).Aligning the UK G. fossarum specimens within Clade 11 confirms previous studies which show this clade to be the most widely distributed across Europe within the species complex (Copilaş-Ciocianu and Petrusek 2015; Weiss and Leese 2016).

Discussion
Non-targeted detection by direct and environmental DNA metabarcoding has the potential to revolutionise early warning systems for non-native species, but this utility of the new technology has so far been demonstrated only a limited number of times (Mahon et al. 2014;Brown et al. 2016).In this study,  S2, for accession numbers and origin of individual sequences).GMYC -General Mixed Yule Coalescent, ABGD -Automatic Barcode Gap Discovery (Puillandre et al. 2011) indicate the approaches used by Weiss et al (2014) to detect the different clades in their study.
G. fossarum, a newly recognised non-native species for the UK, was detected during the course of a wider metabarcoding survey of macroinvertebrate communities.The identification of G. fossarum was subsequently confirmed by microscopy and standard DNA barcoding.The sequences generated from this study indicate that the UK populations of G. fossarum sampled here fall within the previously identified Clade 11, sensu Weiss et al. (2014), of this highly diverse species complex (Figure 3).Importantly this is the most widely distributed clade within the G. fossarum complex (Weiss et al. 2014;Copilaş- Ciocianu and Petrusek 2015;Weiss and Leese 2016).
Gammarus fossarum was found in seven distant river catchments within the UK, indicating a widespread distribution (Figure 1).Initial detection of G. fossarum was made using non-targeted metabarcoding of macroinvertebrate DNA, water eDNA and sediment eDNA samples.Of the sites where G. fossarum was detected using this method (n = 25), G. fossarum was detected in all 25 DNA macroinvertebrate samples (100%), in 8 of water (32%) and 9 sediment (36%) samples.The lower detection of G. fossarum in eDNA samples compared to macroinvertebrate samples is not surprising due to the dilution of eDNA and effects of flow on DNA availability in lotic systems.
At 23 of the 28 sites (including post hoc samples) where G. fossarum was present it was the only Gammaridae species detected.This suggests it is not only widespread in the UK but could also be the dominant gammarid in some locations, possibly even having displaced the native G. pulex locally.With the new species discovery, recent re-examination of historical archived gammarid samples was undertaken from available Environment Agency and Natural History Museum (NHM), London, collections.Material from the Environment Agency had overlooked records of G. fossarum dating back to 2005 from the River Len, Maidstone, Kent (51.2619ºN; 0.56451ºE) whilst re-examination of material from the NHM revealed the earliest record to date, 1964 from the River Darent, Kent.This shows that G. fossarum has remained undetected and overlooked by conventional means for a substantial length of time.
Gammarus fossarum is indigenous and widespread in mainland Europe, and typically inhabits springs and upper reaches of mountainous streams, with G. pulex being more dominant in lower river sections (Nijssen 1963;Goedmakers 1972;Karaman and Pinkster 1977;Chen et al. 2012).This distribution pattern is linked to G. fossarum's comparative preference for shallower streams and higher current velocities, and its reduced tolerance of low dissolved oxygen conditions (Meijering 1971;Peeters and Gardeniers 1998).It may also be found in middle sections of rivers and is able to coexist with G. pulex (Janetzky 1994;Piscart and Bollache 2012;Copilaş-Ciocianu et al. 2014).In such areas of coexistence, G. fossarum will often occupy faster flowing areas where vegetation is sparse or absent, and G. pulex will be found near marginal shore zones, with reduced currents and rich vegetation growth (Karaman and Pinkster 1977).The distributions of G. fossarum in this study covered a range of habitats, mainly lowland rivers (altitude <90 m) with the exception of the Nailbourne spring, adjacent to Saint Ethelburga Well and Maiden Newton on the Upper Frome, with altitudes of 106 m and 109 m, respectively (see Supplementary information).The river depths at G. fossarum locations were shallow, seldom reaching more than 20 cm.It is important that further exploration of UK upland systems is undertaken as the sites surveyed for this study were mostly lowland, and at this stage are an indication of habitat suitability rather than preference for G. fossarum in the UK.Of our five study sites where G. fossarum and G. pulex coexisted, all had a mean depth >20 cm and featured both fast and slow currents as well as vegetative marginal areas, however there appears to be no other pattern in the distribution of sites where both species were found to co-exist.Four of the five sites were from the metabarcoding samples, the percentage read count for both species varied substantially, hence no species dominance can be inferred from this data (see Supplementary material Table S1).
Gammarus fossarum is the third non-native freshwater gammarid to be found in the UK within the last six years, following the discoveries of Dikerogammarus villosus in 2010 (MacNeil et al. 2010) and Dikerogammarus haemobaphes in 2012 (Aldridge 2013).The record is rather unforeseen and the species has not been included on the UK's nonnative species watch list with more focus being placed on Ponto-Caspian species that have invaded western Europe (Gallardo and Aldridge 2015).A detailed risk assessment of the threat that G. fossarum poses to native Gammaridae within the UK does not currently exist; further research into how G. pulex and G. fossarum co-exist within UK habitats should be carried out to decide if this action is warranted.However, the importance of this discovery as a new non-native species to the UK should not be overlooked as it has important implications for future ecological assessments.
In conclusion, we detected a newly recognised non-native species to UK fauna using non-targeted DNA metabarcoding, and confirmed its presence using microscopy and standard DNA barcoding.It is well known that the effectiveness of INNS control or management relies heavily upon early detection (Lodge et al. 2006;Vander Zanden et al. 2010).In future, for other species, non-targeted monitoring of high risk invasion pathways using eDNA may ensure that early eradication or containment are possible management options (Davis 2009;Hulme 2009;Jerde et al. 2011;Thomsen et al. 2012;Lawson Handley 2015).It is important that future research should now focus on establishing the true distribution, ecology and potential implications of G. fossarum within the UK, as well as exploring how the nontargeted eDNA metabarcoding approach can be used to detect non-native species.

Figure 2 .
Figure 2. Picture of Gammarus fossarum found in the River Taff, UK, 7/6/2016, A) male adult specimen, B) male uropod III and C) male plumose hairs on inside of exopod of uropod III (↗); and picture of male Gammarus pulex features for comparison D) uropod III and E) plumose hairs on inner and outer edge of exopod of uropod III (↗) (Photographs by D. Constable).

Figure 3 .
Figure 3. Maximum likelihood phylogenetic tree for the COI gene -based on sequences obtained from previously published and newly obtained Gammaridae sequences.The mini-barcode (metabarcoding) and standard COI barcode sequences from this study are represented in blue and red, respectively.(See supplementary material TableS2, for accession numbers and origin of individual sequences).GMYC -General Mixed Yule Coalescent, ABGD -Automatic Barcode Gap Discovery(Puillandre et al. 2011)  indicate the approaches used byWeiss et al (2014) to detect the different clades in their study.

Table 1 .
Specimen identification and identification method for morphologically identified and DNA barcoded specimens.(*Specimens collected from the River Frome were subject to morphological identification only.**Specimens collected from Nailbourne were DNA sequenced only due to damaged specimens).