Monitoring invasive alien macroinvertebrate species with environmental DNA

Regular monitoring of ecosystems can be used for the early detection of invasive alien species (IAS), and provide information for management and preventing them from becoming established or advancing into new areas. Current methods of monitoring freshwater systems for IAS can be both financially costly and time‐consuming, with routine monitoring often carried out at low intensity and at only a small number of sites. In this study, we evaluate how environmental DNA (eDNA) metabarcoding for monitoring freshwater macroinvertebrate IAS compares to traditional kick‐net sampling as part of a national (Switzerland) and a catchment monitoring programme. Kick‐net sampling was more fruitful for the detection of several well‐known target macroinvertebrate IAS. However, eDNA samples proved complementary for the detection of IAS that belong to species often being unnoticed by traditional sampling due to methodological or taxonomic reasons. Specifically, the invasive jellyfish Craspedacusta sowerbii, hardly detectable using classic kick‐net sampling, was found to be widespread in both the national and the catchment‐scale monitoring with the eDNA method only. Our study shows that IAS detection using eDNA is easily implemented in both national‐ and catchment‐scale monitoring campaigns. However, successful detection of target IAS is still highly dependent on primer choice, species' biology, and availability of adequate markers. Specifically, multiple markers should be considered for detecting IAS from several different taxonomic groups, such as those under the ‘freshwater macroinvertebrate’ umbrella term. While eDNA is still developing in terms of these fundamental methodological requirements, surveillance for both target and non‐target IAS using eDNA is likely to increase efficiency in early detection of IAS in freshwater systems.


| INTRODUCTION
There has been a steep increase in the number of invasive alien species (hereafter IAS) spreading to new areas throughout all ecosystems in recent years due to the increase in trade, tourism, and travel (Hulme, 2006(Hulme, , 2009IPBES, 2019;Sutherland et al., 2013). This particularly concerns freshwater ecosystems, which, as highly diverse habitats, are simultaneously facing increasing threats from other anthropogenic pressures (Dudgeon, 2019;Dudgeon et al., 2006) and are prone to be invaded by species from many taxonomic groups, including fish and invertebrates (Baltazar-Soares et al., 2019;De Ventura, Kopp, Seppälä, & Jokela, 2017). To prevent IAS from successfully invading a new habitat and for subsequent management, early detection is paramount, also to reduce the financial costs associated with any possible control attempts (Hulme, 2006). However, to do this, regular and accurate monitoring must be carried out (Dudgeon, 2019).
Current freshwater monitoring methods rely on sight or capture of specimens (e.g., electro-fishing, kick-net, or Surber sampling) with subsequent morphological identification either in the field (i.e., fish) or via microscope (i.e., macroinvertebrates). However, these methods may not be the most suitable tools as species in low abundance (e.g., IAS at early phase of establishment) are often missed. Freshwater macroinvertebrates are typically collected by kick-net sample, which aims to sample a representation of the community from subhabitats within a designated sample site (Barbour, Gerritsen, Snyder, & Stribling, 1999). This method is highly standardised but was developed to allow ecological assessment of rivers by using prior knowledge of specific macroinvertebrate community preferences and pollution tolerances, rather than early detection of macroinvertebrate IAS. Furthermore, cryptic, or closely related taxa, juvenile or damaged specimens may not be identified correctly, or only identified to a coarse taxonomic level, which may lead to incorrect or unsuccessful detection of an IAS via morphological methods (Haase et al., 2006;Mandelik et al., 2010;Blackman et al., 2017). Finally, the surveyed 'macroinvertebrates' are purely defined by their size and lifestyle (generally meaning benthic invertebrates that can be seen by the naked eye), and thus are biased and likely overseeing small species.
Developments of molecular tools for the identification of taxa via DNA, either from tissue or environmental samples (bulk and environmental DNA, respectively), offer potential solutions to the limitations of current approaches. Using environmental DNA (hereafter eDNA) to identify hidden biodiversity, including IAS from simple water sample collection, is a growing trend and 'game-changer' regarding biomonitoring (Taberlet, Coissac, Hajibabaei, & Reiseberg, 2012;Lawson-Handley, 2015;Deiner et al., 2017). In 2008, Ficetola and colleagues used DNA extracted from pond water to detect the American bull frog, a prominent IAS in Europe. Since this study, research application of eDNA has ballooned to include different taxonomic groups by either a species-specific method (conventional PCR, qPCR and ddPCR), or whole communities such as fish (Hänfling et al., 2016;Pont et al., 2018), macroinvertebrates (Fernández, Rodríguez-Martínez, Martínez, Garcia-Vazquez, & Ardura, 2019;Brantschen et al., 2021), and zooplankton (Brown, Chain, Zhan, MacIsaac, & Cristescu, 2016;Djurhuus et al., 2018) using a metabarcoding approach. A key advantage of eDNA sampling, which makes it particularly well suited for IAS monitoring, is its scalability. Water eDNA sample collection can be quick and simple to collect (dependant on water body type, water volume, and equipment). These samples can be processed in large numbers, therefore have an associated reduction in cost per sample compared to traditional sampling methods (Lacoursière-Roussel et al., 2018). Furthermore, monitoring programmes can cover large areas from catchment campaigns to national monitoring schemes, a scalability generally lacking for traditional methods .
Recently, several private companies have begun to offer eDNA services, such as protected species monitoring (e.g., great crested newt), and eDNA-based macroinvertebrate IAS monitoring has a large potential to transition from a still mostly academic to a more applied use. This is for example highlighted by the continued development of eDNA approaches by North American agencies and researchers for the detection of invasive Dreissenid mussels, primarily by a speciesspecific approach (Gingera, Bajno, Docker, & Reist, 2017;Sepulveda et al., 2020). The focus has remained on the species-specific approach, as it is thought to be more sensitive than a general metabarcoding approach Harper et al., 2018;Simmons, Tucker, Chadderton, Jerde, & Mahon, 2015). However, the metabarcoding approach can lead to the detection of unknown biodiversity previously not recorded, including 'unexpected' invasive or non-native species (Blackman et al., 2017;Simmons et al., 2015), and while a species-specific approach is an effective tool for monitoring known IAS, it does not take full advantage of the potential to detect these 'unexpected' or non-target IAS. Although several studies have compared eDNA metabarcoding and traditional kick-net sampling for freshwater macroinvertebrate (e.g., Fernández et al., 2019;Laini et al., 2020;Mächler et al., 2019), none have focused specifically on IAS within this group. Therefore, to assess the complementarity and potential advantages of eDNA to traditional kick-net methods for macroinvertebrate IAS detection, there first needs to be standardised testing of these tools at a large scale .

Most
national freshwater macroinvertebrate sampling programmes include a list of established IAS, which incorporates potential IAS that have been determined using a 'horizon scanning' approach (Altermatt, 2012;Roy et al., 2014). In Switzerland, 50 freshwater macroinvertebrate IAS are currently known to be present, and this list is supplemented with a further 32 taxa listed as 'likely to occur' due to documented invasions in neighbouring countries (See Table S1 and Wittenberg, 2005;Altermatt, 2012, Altermatt, Alther, Fišer, & Švara, 2019. While this is a suitable strategy for predicting IAS, it does not consider those species that may arrive 'unexpectedly'. Within the list of 82 taxa (present or likely to become present), several taxa are also not recorded as part of the standard kick-net survey methodology (Table S1) KgaA, Darmstadt, Germany). Filters were sealed with Luer fitting and placed in a cool box for transport to the lab where samples were stored at À20 C until further processing. Negative controls consisting of 500 ml of ddH 2 O were filtered in the field and treated in the same way as samples (n = 8).

| Macroinvertebrate collection and determination
At each site, kick-net samples were collected by sampling eight microhabitats following Stucki (2010). Coarse organic particles, debris, amphibians, and fish were removed from the sample, and the remaining material was pooled and stored in 85% molecular grade ethanol. Identification was carried out by experienced taxonomists in the laboratory following the IBCH Labor-Protokoll (Stucki, 2010) except for the genera of Ephemeroptera, Plecoptera, and Trichoptera where individuals were further identified to species-or complex-level where taxonomically necessary.  Table S2 and Mächler et al., 2019;Carraro, Mächler, Wüthrich, & Altermatt, 2020;Blackman, Ho, Walser, & Altermatt, 2021). Environmental DNA samples consisting of 2 Â 500 ml of water was collected directly from the river with single-use sterile syringes and filtered through 0.22 μm Sterivex filters and sealed with Luer fittings and stored in a cool box for transporting to the lab where samples were stored at À20 C until further processing (n = 40). All samples were collected using sterile gloves from the bank of the river without entering the watercourse (to avoid contamination). Negative controls consisting of 500 ml of ddH 2 O were filtered on each day of sampling (n = 7) and were treated in the same way as samples.

| Macroinvertebrate sampling
Kick-net samples were collected from 20 sites in the Thur catchment in Summer 2016 (Carraro, Stauffer, & Altermatt, 2021;Mächler et al., 2019). The method of collection differed from the national monitoring method in terms of sampling effort and seasonality: the sampling protocol was simplified to a total of three samples covering the prevailing substrates and fieldwork was done in July instead of March/April as scheduled by Stucki (2010). Identification was carried out by experienced taxonomists in the laboratory following the protocol introduced above.

| eDNA extraction and library preparation
DNA extractions from filters were performed in a clean room environment at Eawag, Switzerland (Deiner, Walser, Mächler, & Altermatt, 2015). The DNA was extracted using the QiAgen PowerWater Sterivex Extraction Kit (Qiagen, Germany). Filters from different sites were extracted in random batches including field and filter control that were treated equally to the samples. Extractions were performed as described by the manufacturer protocol. DNA was eluted into 100 μl of elution buffer and stored until further processing at À20 C. Both sets of samples (national and catchment) from this study used the same library preparation. A two-step library preparation method was used targeting a 313 bp fragment of the COI barcode region with the degenerative primer pair: mICOIintF and jgHCO2198 (Table S3, Geller et al. 2013;Leray et al. 2013). These primers were modified to include the Nextera transposase sequences. A synthetic DNA strand, which amplified with the primer sequences, was used as PCR positive control (Table S4).
Samples and controls were randomized over all 96-well PCR plates (four plates for the national sampling and three plates for the catchment sampling).
Each PCR reaction consisted of SigmaFree water, the provided 1x Buffer I (Thermo Fisher Scientific, MD), BSA (0.1 mg/μl), dNTP (0.2 mM), MgCl 2 (1 mM), mICOIintF, and jgHCO2198 primers (0.5 μM each) and the polymerase AmpliTaq Gold 360 (1.25 U/μl) in a total volume of 25 μl. Exactly 2 μl of DNA template was used in each reaction. A touchdown protocol was used as follows: 95 C for 10 min, denaturation of DNA at 95 C for 15 s, annealing at 62 C for 30 s, followed by extension at 72 C for 30 s. For the first 16 cycles the annealing temperature was reduced by one degree per cycle, for the next 25 cycles, the annealing temperature remained at 46 C, followed by a final extension at 72 C for 5 min before the plates were cooled down to 10 C. All PCR were carried out in triplicate and PCR products were checked for amplification with the AM320 method on the QiAxcel Screening Cartridge (Qiagen, Germany); we did not experience any issues of inhibition. First round PCR products were cleaned

| Bioinformatics
After completion of each Illumina MiSeq PE300 (600 cycles) run, the data were demultiplexed (MiSeq Reporter V2.4) and reads were quality checked using usearch v11.0.667 (Edgar, 2010), FastQC (Andrews, 2015), and MultiQC (Ewels, Magnusson, Lundin, & Kaller, 2016). Raw reads were first 3 0 -end-trimmed, merged, and full-length primer sites were removed using usearch v11.0.667 (Edgar, 2010 Table S1). We used a Chi-square test on IAS that were detected by both methods to determine if there was a statistical difference in eDNA and kick-net sampling methods. We also tested whether the number of IAS detected by the two methods increased with upstream drainage area (km 2 ), using a generalised linear model (GLM) with Poisson regression. All analysis was carried out in R version 3.6.3 (R Core Team, 2021). Map projections were made in SwissRiverPlot (Alther & Altermatt, 2021) for the national sampling campaign and tmap (Tennekes, 2018) for the catchment sampling campaign.
To confirm the detection of Craspedacusta sowerbii, we constructed a phylogenetic tree based on the 13 hydrozoan sequences from our data set (including 5 C. sowerbii sequences) and all NCBI records of tissue extracted samples from published papers. All analysis was conducted in MEGA version X (Kumar, Stecher, Li, Knyaz, & Tamura, 2018;Stecher, Tamura, & Kumar, 2020). All available published tissue derived sequences registered in NCBI (https://www.ncbi. nlm.nih.gov/) using the COI barcode region were downloaded (n = 13). We trimmed and aligned the 26 sequences using MUSCLE (Edgar, 2004). Sequences were mapped using a neighbour-joining (Saitou & Nei, 1987) and maximum composite likelihood method (Tamura, Nei, & Kumar, 2004) with 1,000 bootstrap replicates (See Figure S10 and Table S5 for NCBI accession numbers and source).

| RESULTS
The MiSeq runs generated 26.4 and 17.56 million raw reads from the national-and catchment-scale campaigns, respectively (the full description of both library outputs can be found in Brantschen et al. 2021 andBlackman et al. (2021) for the national-and catchment-scale campaigns, respectively). Of the 50 known macroinvertebrate IAS taxa listed as being in or at risk of arriving in Switzerland, 13 IAS taxa were detected over the national-and catchment-scale monitoring campaigns using eDNA and kick-net sampling (Table 1 and Table S6).
Of the 13 taxa at the national scale, one macroinvertebrate IAS was only detected with eDNA, nine taxa were detected with kick-net only, and three taxa were detected with both methods (Figure 1, Table 1).
At the catchment level, three IAS taxa were detected in total, one taxon was detected with eDNA only, one taxon was detected with kick-net only, and one taxon was detected with both methods ( Figure 1, Table 2).

| IAS detection with both methods
Only three taxa were detected by both sampling methods:   were detected at the most sites and with the highest read numbers and densities using eDNA and kick-net sampling, respectively (See Table S7 for further information).
Of those taxa not detected with both methods, kick-net sampling was more successful for the detection of macroinvertebrate IAS (n = 9). At the national-scale sampling campaign, this included a range of taxa that were identified morphologically to different taxonomic levels, that is, four family-and five species-level identifications (See  Table S6). The remaining taxa were found at only one site each (Janiridae, Cambaridae, Echinogammarus ischnus, Physella heterostropha, and Pacifastacus leniusculus). In the catchment-scale sampling campaign, only one IAS was found with kick-net and not with eDNA, namely a flatworm of the Dugesiidae family, which was found morphologically at two sites but not identified to species level.
Although eDNA sampling only detected one species that was not found with kick-net sampling, this finding is particularly interesting.
Craspedacusta sowerbii, also known as the Peach blossom jellyfish, is widespread in Europe (Jankowski, Collins, & Campbell, 2008) and has been recorded in Switzerland since 1962 (Balvay, 1990). However, it is hardly ever (if at all) detected by kick-net sampling due to its size and form (polyp and medusa). However, in both national-(48 or 52% of sites) and catchment-scale (12 or 60% of sites) campaigns, we detected this IAS using eDNA at a high rate (Figure 4), indicating it being relatively widespread. Detection using eDNA often requires further verification to confirm the taxonomic assignment. This confirmation for Craspedacusta sowerbii can be found in Figure S10.

| Sample method detection over catchment size
To test the success of both methods to detect IAS at different scales, we plotted the upstream drainage area against the total number of IAS taxa found by either method (Figure 5). There is no change in the   Although Craspedacusta sowerbii has previously been recorded in Switzerland (Balvay, 1990), due to its size and morphology, it is unlikely to be found in routine kick-net sampling, whereas eDNA metabarcoding is better suited for its detection. The finding of eDNA signals of this species at such a large extent in both sampling campaigns within Swiss rivers was surprising and alarming, as it indicates the possibility of major knowledge gaps. We therefore carried out further stringent quality controls, to prevent false positives from the high-throughput sequencing data (Darling, Pochon, Abbott, Inglis, & Zaiko, 2020). By constructing a phylogenetic tree of the C. sowerbii sequences produced in this study and reference sequences from NCBI of DNA tissue extracts, we find further evidence our identification is correct. However, like similar studies seeking to confirm 'unexpected' IAS, we would recommend physical collection of this species as a next step (Blackman et al., 2017). We are not able to indicate which form of the freshwater jellyfish (e.g., polyp or medusa) was detected in this study (and picked up in the eDNA samples). However, it is highly likely that the signal we detect is from the polyp form, as they can persist on river substrates, whereas flow velocity is a limiting factor in suitable habitat conditions for the free-floating medusa stage (Gasith, Gafny, Hershkovitz, Goldstein, & Galil, 2011). Both forms of C. sowerbii are zooplankton consumers, and although some work suggests it has minimal effect on a freshwater ecosystem (Dodson & Cooper, 1983), it is likely to influence algae grazer populations and therefore have cascading effects on food webs due to algal accumulation, especially during jellyfish 'bloom' events where large numbers appear in a relatively short amount of time (Gasith et al., 2011;Jankowski, Strauss, & Ratte, 2005). This dataset and other eDNA metabarcoding sources are valuable resources, complementing also classical surveys, to map the extent of C. sowerbii's occurrence in Switzerland and Europe wide. However, these data should be used in conjunction with eDNA models (e.g., Carraro et al., 2020Carraro et al., , 2021 to reflect both the spatial extent to which the eDNA signal represents and sources of the signals both in the rivers where it was detected, and any lentic body outflows, which could also contribute to the signal. Encouragingly, the most common taxa found in both data sets, P. antipordarum, was found at similar scales with no significant differ-  we increase the area, which is screened for IAS to a river stretch, rather than a single site. Combining this added benefit with the overall decreasing sample cost with increasing sample number Lacoursière-Roussel et al., 2018), eDNA has the potential to be a highly beneficial monitoring method for IAS detection in particular.
In our dataset, we have several positive kick-net detections where IAS have not been detected with eDNA, which may be criticised as 'false negative'. We note this being a common pattern for eDNA metabarcoding studies in riverine systems for multiple reasons.
Firstly, understanding low DNA quantity available (either from shedding rates or low biomass; Barnes & Turner, 2015) or optimisation of the eDNA sampling protocol (Mächler, Deiner, Fabienne, & Altermatt, 2016;Muha, Robinson, Garcia de Leaniz, & Consuegra, 2019). Rather than sight or capture of specimen, for a positive detection via eDNA, sufficient DNA must be available in the water column. However, certain taxa due to their morphology (e.g., body armour or shell) may produce very low quantities of DNA (Martins et al., 2020). For example, Blackman et al. (2020) successfully detected Dreissenidae in a river in the United Kingdom using the same primers and sampling method as this study. However, here we do not detect Dreissenidae with eDNA metabarcoding, only with kick-net sampling. Blackman et al. (2020) noted a correlation between decrease in the metabarcoding read number and number of Dreissena rostriformis bugensis mussels found at sampling sites in their study. In this study, at sites where Dreissenidae were detected by kick-net between 1 and 11 specimens were collected. We therefore assume that this was not sufficient biomass or density, in relation to the size of the river, for the successful detection via eDNA metabarcoding with COI in our study, and further developments in our methodological understanding are still needed.  (Taberlet, Bonin, Zinger, & Coissac, 2018;Martins et al., 2020). Decisions when applying an eDNA metabarcoding approach to the detection of macroinvertebrates need to be weighed up between taxonomic resolution and potential taxon detection. However, by collecting eDNA samples in the first place, we uncover the potential to apply multiple primers (both universal and species specific) to the same samples (while sufficient sample remains) and repeatedly test for the presence of different taxa, which is another advantage of using this approach .
The kick-net sampling strategy showed a positive increase in the number of IAS taxa detected with increased drainage area. Finding a higher number of IAS in larger water bodies is consistent with the increased opportunity for introductions in major water bodies, which have been reengineered and connected to new areas due to trade and transport across Europe (Leuven et al., 2009). The river Rhine in particular is a major source of invasive macroinvertebrate species due to its connection to areas such as the Ponto Caspian regions (Leuven et al., 2009). Therefore, these larger rivers (the Rhine, Rhone, and Aare) are where IAS become established and then spread upstream into other tributaries either by human intervention or natural means. Contradictorily, we do not find a correlation between the number of IAS detected and drainage area with eDNA sample collection. This likely reflects not only the comparable lower volumes of water collected in larger rivers to those collected at sites with smaller drainage area at the top of the catchment but could also the heterogeneously dispersed DNA across a river width (Macher & Leese, 2017). Increased detection may be achieved by sampling multiple locations within a site to capture eDNA (Macher & Leese, 2017), in a similar method to the multihabitat approach of a kick-net sample.
Although the taxa detected in this study will likely be impossible to be eradicated once established, it is important to use appropriate and timely methods to ensure the ecological integrity and possible management of the systems in which they occur. For example: while the impacts of P. antipodarum are limited in low numbers, it has high fecundity and has been recorded to impact primary production and nutrient in large densities (Goldberg, Sepulveda, Ray, Baumgardt, & Waits, 2013;Hall Jr, Dybdahl, & VanderLoop, 2006;Hall Jr, Tank, & Dybdahl, 2003), similarly C. fluminea/fluvialis in large numbers out-competes native unionid species and other filter-filters for space and food (Schmidlin & Baur, 2007). Accurate monitoring is therefore important to detect potential changes in freshwater ecosystems and ensure better understanding of the advance of IAS in Switzerland. As mentioned previously, a potential benefit of using eDNA metabarcoding for IAS detection would be the ability to upscale our monitoring campaigns to monitor at finer resolutions. This would enable us to include a higher numbers of sites and key IAS pathways, which are either not currently monitored or unsuitable to check with established methods.

| CONCLUSION
Our findings support the complementary use of both eDNA and kick-net sampling for macroinvertebrate IAS detection in freshwater river systems. We demonstrated the added benefit of eDNA as a complimentary tool to kick-net sampling. While not all IAS within the macroinvertebrate group can be detected by using a single primer, traditional and molecular methods do overlap for several common macroinvertebrate IAS. We especially see two major benefits of eDNA metabarcoding. Firstly, the ability to detect unexpected or overlooked IAS where traditional kick-net sampling and morphological identification may not be suitable. Secondly, by using an eDNA metabarcoding approach, monitoring can be upscaled both in terms of number of samples collected and the area in which they represent. These benefits justify the integration of eDNA metabarcoding as a complementary tool also for routine biomonitoring programmes.