Casting a broader net: Using microfluidic metagenomics to capture aquatic biodiversity data from diverse taxonomic targets

Environmental DNA (eDNA) assays for single‐ and multi‐species detection show promise for providing standardized assessment methods for diverse taxa, but tech‐ niques for evaluating multiple taxonomically divergent assemblages are in their in‐ fancy. We evaluated whether microfluidic multiplex metabarcoding on the Fluidigm Access Array™ platform and high‐throughput sequencing could identify diverse stream and riparian assemblages from 48 taxon‐general and taxon‐specific meta‐ barcode primers. eDNA screening was paired with electrofishing along a stream continuum to evaluate congruence between methods. A fish hatchery located mid‐ way along the stream continuum provided a dispersal barrier, and a point source for non‐native White Sturgeon ( Acipencer transmontanus ). Microfluidic metabarcoding had 87% accuracy with respect to


| INTRODUC TI ON
Land managers survey and monitor many biota to meet diverse management objectives. These activities typically involve multiple agencies and objectives that address questions focusing on shared geography. Riparian areas can be migration corridors for amphibians or water-dependent mammals, while catchments may harbor pathogens that affect wildlife (e.g., bat white-nose syndrome [Flory et al., 2012]; amphibian chytridiomycosis [Olson et al., 2013]), forest health (sudden oak death from Phytophthora [Hansen et al., 2012]), or human community health (Tiedemann, 2000). These overlapping management concerns on a common catchment require multiple teams with technical expertise to address basic questions of species detection. Typically, biotic surveys are conducted with limited cross-taxon integration due to the difficulty of coordinating across disciplines, managers, and ownership (private, state, federal).
Environmental DNA (eDNA) analysis has emerged as a powerful method for detecting stream obligate and riparian species, with the capacity to bridge diverse disciplines and inform multiple management objectives. Originally developed for characterizing microbial (Venter et al., 2004) and fungal (Anderson and Cairney, 2004) communities, eDNA analysis has expanded to include diverse eukaryotes, including plants (Willerslev et al., 2003), invertebrates (Hajibabaei et al., 2011;Thomsen, Kielgast, Iversen, Wiuf, et al., 2012), and vertebrates (Andersen et al., 2012;Thomsen, Kielgast, Iversen, Møller, et al., 2012). Methods for eDNA analysis have evolved from assays targeting one to a few well-characterized taxa (e.g., qPCR, digital PCR; Nathan et al., 2014), to "metabarcoding" assays that identify scores of taxonomic targets per sample (Deiner et al., 2016;Thomsen, Kielgast, Iversen, Møller, et al., 2012;Valentini et al., 2016;Wilcox et al., 2018). Metabarcoding approaches have been shown to provide detection accuracies equivalent or better than traditional sampling methods (Deiner et al., 2016;Thomsen, Kielgast, Iversen, Møller, et al., 2012;Valentini et al., 2016; but see Cilleros et al., 2019), and a much larger taxonomic spectrum per assay. Metabarcoding represents a technological leap, but it is not without limitations, such as the challenge of independently validating unexpected (false) positives and negatives, the difficulty detecting rare species, or the difficulty of identifying taxa to the species level using universal DNA metabarcoding genes (Deiner et al., 2016). From the perspective of land management agencies, metabarcoding assays based on single markers share a limitation in that they address a fraction of the broad spectrum management questions asked by land managers.
Incorporating multiple barcoding and metabarcoding gene targets into a single assay offers independent observations for taxon presence/absence, and more accurate biodiversity estimates across highly diverse taxa (Drummond et al., 2015;Elbrecht et al., 2017;Evans et al., 2017;Gibson et al., 2014;Stat et al., 2017). Here, we evaluate the performance of microfluidic PCR in combination with eDNA metabarcoding to simultaneously evaluate up to 48 loci in 2,304 nanoliter-scale PCRs per array. PCR products ranging from 149 to 406 bp in length were amplified using the microfluidic Fluidigm Access Array™ (Fluidigm Corporation, 2016), and products were sequenced by Illumina massively parallel sequencing. Highlevel multiplexing allows simultaneous screening with taxon-general and taxon-specific genes in one assay, providing multiple markers for taxonomic inference (e.g., phylum to species), and a mechanism to independently validate presence and absence observations (Brown et al., 2016).
We tested eDNA samples collected from a stream continuum ~11 km in length ( Figure 1). This continuum includes an impassable barrier to hatchery fish, which creates a break in the distribution of hatchery and wild stocks for select species and allows measurement of eDNA transport distance from hatchery residents. Multiple primer sets targeting taxon-general and taxon-specific mitochondrial genes were designed for the detection of fish (salmonids, sculpins, lamprey, sturgeon), amphibians (frogs, salamanders), invertebrates (crayfish, mayflies, stoneflies), oomycete (Phytophthora, Saprolegnia), and fungal pathogens (Batrachochytrium, Pseudogymnoascus). Presence/absence detection by eDNA is directly compared to electrofishing to evaluate the accuracy and specificity of eDNA microfluidic metabarcoding as a qualitative and semi-quantitative proxy for species-level field identification.

| Study site
We surveyed five sites in Fall Creek, a fifth-order stream draining 78 km 2 , and a tributary to the Alsea River in the Coast Range of

| Water sampling and DNA extraction
Water samples were collected immediately prior to electrofishing surveys at the downstream blocknet used for the electrofishing survey. Preliminary trials (data not shown) indicated that 3 L water samples filtered through 0.45-μm cellulose nitrate filters (Sterlitech, Kent, WA, USA) gave significantly better results than smaller samples (e.g., 1 L with 0.45 μm filters), and comparable results to larger volumes through 1.0-μm high-volume filtration capsules (e.g., 60 L through a Pall Envirochek ® Sampling Capsule). We filtered 3 L of stream water (2 L thalweg + 1 L adjacent slack water) per replicate using a peristaltic pump (Proactive Pegasus Alexis, Fairborn, OH, USA). At site 3, all 3 L were collected in a pool at the outflow from OHRC. Each site was represented by eight independent filter replicates to capture taxa with low detection probabilities (Ficetola et al., 2015). To prevent cross-site contamination, we always entered sites downstream from where samples were taken, and equipment (bottles, tweezers, waders) was decontaminated with a 50% bleach solution followed by a triple rinse of deionized water. Filters were stored in 5-ml vials on wet ice during collection and transport, frozen at −20°C within 6 hr of collection, and stored at -20°C until DNA extraction. DNA was extracted using the MoBio (Qiagen, Hilden, Germany) Power Water DNA extraction kit per manufacturer's instructions. This kit has a step specifically designed to remove polymerase inhibitors, such as humic acids, which are commonly found in freshwater ecosystems (Matheson et al., 2010;Wetzel, 1993

| Primer design for microfluidic PCR amplification
We used the Fluidigm 48.48 Access Array™ (Fluidigm, San Francisco, CA, USA;Fluidigm Corporation, 2016) to amplify taxongeneral and taxon-specific target genes. The Access Array uses integrated fluidic circuits and a 4-primer amplicon tagging scheme in which target-specific primer pairs amplify 48 different targets in combination with sample-specific barcoded primer pairs in 48 different samples. This allows for the simultaneous amplification of barcoded targets in each of the 2,304 individual reaction chambers. For more information, see the online user guide (Fluidigm Corporation, 2016). We designed amplification primers with Fluidigm compatible annealing temperatures of 58-60°C, and target amplicon lengths ranging from 149 to 406 bp (lower bound to meet postamplification removal of primer-adapter dimers; upper bound to limit amplicon length for paired-end sequencing).
Forward and reverse amplification primers were modified by the addition of 5' common sequence tags (CS1, CS2; www.fluid igm. com) which serve as the binding site for the addition of the P5 and P7 Illumina sequences and dual-index multiplex barcodes.
We designed taxon-general universal metabarcoding primers that amplified diverse classes of organisms (e.g., ray-finned fishes [Teleostei] and amphibians [Batrachia], Chondrostei fish, mussels [Bivalvia], and insects), by targeting the 12S rDNA region used by Valentini et al. (2016), as well as 16S, 18S, and internal transcribed spacer (ITS) rDNA. To design taxon-specific primers, we focused on taxonomically informative genes typically used for species barcoding such as cytochrome C oxidase 1 (COI), cytochrome B (CytB), NADH dehydrogenase 2 (ND2), D-loop, and beta-tubulin. Primers for these genes were designed to amplify multiple related species (e.g., salmonids) in gene regions that included diagnostic polymorphisms for species-specific identification ("barcode gaps"; Hebert et al., 2004). The "barcode gaps" were identified using the sliding window analysis in SPIDER (version 1.5.0; Brown et al., 2012). Windows showing maximum intertaxon divergence were targeted for primer development.
All primers were computationally screened for primer compatibility, annealing temperature, and off-target activity. Those primer pairs that could be validated with positive control DNA were screened following manufacturer's instructions (Fluidigm Corporation, 2016) with minor modifications (see Appendix S1 for additional details on alignments, primer development, primer screening, and primer validation). Primer sequences are shown in File S1.

| Multiplex PCR amplification and massively parallel sequencing
We analyzed eight replicate samples per five sites, submitting normalized samples that each contained 15 ng/μl of input eDNA. After dilutions and loading across the array, the actual amount of eDNA in each reaction chamber was 0.09405 ng. We also included two positive controls and one negative control. Positive controls consisted of a multi-target standard that included either 2 × 10 6 or 2 × 10 4 molecules of each amplicon per Fluidigm reaction. Positive and negative controls were supplemented with "inert" (nontarget) genomic DNA from two gymnosperms not native to Fall Creek (Pinus lambertiana Dougl. and Ginkgo biloba L.) to bring the final DNA concentration to 15 ng/μl. This was done to make PCR conditions in the controls similar to eDNA samples. See Appendix S1 for additional details.  (Table S1). Bovine serum albumin (BSA) was added at 0.2 μg/μl to alleviate PCR inhibition from contaminants in environmental DNA (Romanowski et al., 1993;Widmer et al., 1996) and to mitigate inhibitor-driven bias (Valentini et al., 2016).

| Bioinformatic analysis
Sequences were demultiplexed using sample-specific dual barcodes and target-specific amplification primers using dbcAmplicons (version 0.8.6; Settles and Gerritsen, 2014). See Appendix S1 for additional details.
Overlapped sequences were individually taxonomically assigned using Centrifuge (version 1.0.4-beta, accessed 2018-10-07; Kim et al., 2016), which in the absence of a full-length perfect match gives higher scores to matches with longer stretches of identical sequence.
We used a custom database developed from sequences downloaded from National Center for Biotechnology Information (NCBI) nucleotide (nt) database in January 2018 using queries listed in Table S2.
The highest scoring match between query and target was reported; if multiple sequences scored equally, the lowest taxonomic rank containing all highest scoring hits was reported.
Preliminary Centrifuge analyses classified some reads as belonging to species related to those from western Oregon, but known to not occur in the Pacific Northwestern United States. From these analyses, a list of 337 species, genera, and hybrids was developed to exclude from future classification (File S2). Notable excluded taxa include Oncorhynchus × Salmo; marine Cottus species; genera closely related to Cottus; and species and subspecies of Oncorhynchus, Cottus, Rana, Acipenser, and Phytophthora found outside of the Pacific Northwestern United States. We chose this strategy recognizing that worldwide databases such as NCBI contain error and bias (Langdon, 2014;Nilsson et al., 2006), classification algorithms represent a tradeoff between efficiency and phylogenetic accuracy, and that even the taxon-specific loci used here may not accommodate perfect species classification in all cases. For other regions or contexts, a different set of excluded taxa may be chosen, based on attributes such as geographic range of the unexpected taxa relative to the study, phylogenetic relationships between unexpected taxa and expected taxa, and the invasion potential of the unexpected taxa.
See the Discussion for more consequences of database representation and accuracy.
For each sample × primer combination, the number of reads classified to each taxon was counted. Reads appearing in the negative control, other than those originating from the Ginkgo primer, were used to calculate a minimum read threshold. These reads may be the result of index hopping, sequencing error in the indices, contamination, or amplification of organisms present in the DNA extraction of the gymnosperm samples added to the negative control. The threshold was set at the 95th percentile of read counts for taxon × primer combinations appearing in the negative control. Counts below this threshold were dropped from all other sample × primer combinations. Counts above the threshold were considered "positive hits" for that taxon in that sample × primer combination. To assess primer efficiency for each primer pair, we compared the total number of reads generated in the 2 × 10 6 and 2 × 10 4 copy positive controls across all primer pairs and ranked them. We examined Pearson correlations between read counts as a function of the length of the amplicon and the GC% of the primers ( Figure 2, File S1).

| Agreement in taxon presence/absence between electrofishing and eDNA
Agreement between detection by electrofishing and eDNA was assessed for all sites excluding the hatchery outflow. For this analysis, only taxa detected by electrofishing are considered, and detection of a taxon's DNA in one replicate was considered a detection by eDNA for that taxon at that site. Though electrofishing itself does not have 100% accuracy, we considered the electrofishing detections as "true" when calculating the sensitivity (or "true-positive" rate), specificity ("true-negative" rate), and accuracy ([true-positive + true-negative]/all observations) of eDNA detections, using R (version 3.5.2; R Core Team, 2018) and the caret package (version 6.0-80; Kuhn, 2008). Additionally, for genera with multiple species detected by electrofishing (e.g.,

Oncorhynchus, Cottus, Rhinichthys), detections by electrofishing
and eDNA were grouped and evaluated at the genus level, and agreement among methods assessed similarly.
F I G U R E 2 Primer efficiency across primer sets. (a) Read counts from 36 target loci derived from positive control reactions containing 2 × 10 4 or 2 × 10 6 template molecules per target. Vertical black line represents the 2 × 10 6 molecule control median count of 14,970 reads. Each locus is represented by one primer pair except "Salmonid Species ND2" and "Salmonid Species COI," which are averages of three and two primer sets, respectively. Positive control DNA was not available for all primers; see File S1 for a complete list of primers. (b) Biplot of sequence yield from the two positive controls, with linear trendline, trendline equation, and correlation test statistics. (c) Biplot of sequence yield (2 × 10 6 molecule control) as a function of amplicon length, in bases. (d) Biplot of sequence yield as a function of forward + reverse primer %GC-content were developed for ND2 (3 sets) and COI (2 sets) so that all species likely to occur in Fall Creek could be sampled. Agreement between electrofishing and eDNA was assessed separately for each of the taxon-specific targets; agreement was also assessed for "consensus detection," requiring any two loci for a positive detection, or requiring all three loci for a positive detection.

| Metagenomic count variation over a stream gradient
To examine variability of read abundance across the stream gra-  Ondov et al., 2011). For this presentation, taxon counts above the minimum read threshold were summed across replicates within a site and for all primers within a replicate.

| eDNA yield and metabarcoding results
The   Although we screened for off-target activity during primer development, some primer pairs showed substantial off-target amplification (File S1

| Stream complexity and defining the "hidden world"
Based on NCBI taxonomy, 3.2 million DNA sequences from field samples were classified into 878 predicted taxa (including sequences classified both at the species level and at higher taxonomic levels), including 20 phyla, 68 classes, 157 orders, 302 families, 448 genera, and 647 species (Figure 3, Files S3 and S4).
Overall representation of taxonomic groups was roughly proportional to the number of primers used in targeted PCR amplifica- Amphibians stand out as a surprisingly underrepresented class.
Amphibians accounted for seven taxa and 25,848 sequences (0.80% of total), and this low representation contrasts the large number of primer pairs (12) used to screen for amphibian taxa, and their abundance in the electrofishing survey (especially site 5).
A final observation is that the sole mammal surveyed in this analysis-North American Beaver (Castor canadensis Kuhl), a known resident of Fall Creek-accounted for 4,318 sequences (0.13%) and was detected at sites 1 and 2.

| Electrofishing results: Species presence and abundance
Thirteen species were identified by electrofishing in Fall Creek, including fishes (9), amphibians (3), and crayfish (1; Figure 4). Taxon distributions followed two trends, with most taxa either distributed across all four sites, or found in the three downstream sites (1, 2, 4).

| Comparison between electrofishing and eDNA
All 13 species of fishes, amphibians, and crayfish that were identified with electrofishing were identified by eDNA (Figure 5b- Figure 5), but requiring agreement among two loci reduced accuracy to 0.25. No sculpin species was supported by detection from all three loci. but they could also represent the undescribed Phytophthora taxon "Oaksoil" (Brasier et al., 2003;Hansen and Delatour, 1999;Sims et al., 2015). Phytophthora sequence counts showed a strong linear relationship (r = 0.841; F 1,38 = 92.11, p < 0.001) with total sequences from sample libraries ( Figure 6). Normalizing sequence counts by total library counts (e.g., Phytophthora reads per million total reads   (Figure 7f). This species is only present in OHRC raceways and eDNA was not detected in the downstream sites, indicating that transport of sturgeon eDNA is spatially limited, or that our assay has a high detection limit for this taxon.

| D ISCUSS I ON
Using eDNA metabarcoding via microfluidic multiplex PCR and high-throughput sequencing, we successfully detected all species that were identified by electrofishing along a stream continuum, though not always at the same site. Detection of salmonid species has the highest detection accuracies, relative to electrofishing, because they were targeted with taxon-specific and taxon-general primers. Other fishes (sculpin, dace, lamprey) show equal detection accuracies at the level of genus or family, while being more challenging to identify to the species level. This is either due to the use of taxon-general universal primers, limited genetic divergence in these closely related species, or because databases lacked representatives for local populations and species (e.g., sculpin, discussed below). We also detected ecologically important PCR as a method for assessing a large number of diverse taxa relevant to stream and riparian communities. This highly multiplexed approach significantly extends the reach of traditional "metabarcoding" (Cilleros et al., 2019;e.g., Deiner et al., 2016;Thomsen, Kielgast, Iversen, Møller, et al., 2012;Valentini et al., 2016), and it offers a high degree of specificity that exceeds early results from hybridization-based enrichment (Wilcox et al., 2018).

| Microfluidic metabarcoding: Successes and challenges, by taxonomic group
While environmental DNA detected most targeted taxa, some detections occurred with high accuracy while others showed low correspondence with electrofishing results. Disagreement between electrofishing and eDNA observations may be due to several factors. Species detected by eDNA, but not electrofishing, may be present nearby but not captured, such as when animals are located upstream and their DNA is sampled from downstream transport.
Electrofishing also shows estimates of mark-recapture efficiencies of only 4%-25% in streams (Bayley and Peterson, 2001;Rosenberger and Dunham, 2005), and it is most efficient in shallower water with average stream habitat conditions, and for larger fish (Price and Peterson, 2010). Electrofishing can also miss fishes with low capture probabilities, such as those possessing coarse scales (cyprinids) or lacking swim bladders (sculpins). Species observed by electrofishing, but not eDNA, may be a result of bias against PCR amplification by primers used in the study, sampling heterogeneity, or bias against specific taxa by the methods used in water collection, filtering, or DNA isolation. Some primer sets also amplified off-target taxa, and a high relative abundance of nontargeted:targeted species may lead F I G U R E 5 Summary of presence/absence detection for samples and replicates at Fall Creek sites 1, 2, 4 and 5. Summaries for select genera are shown in the upper panel, and summaries for specific taxa shown in the lower panel. (a) Presence (black) and absence (white) for fish and amphibians based on electrofishing. "Unidentified Oncorhynchus" refers to YOY Rainbow or Coastal Cutthroat Trout, "Unidentified Cottus" refers to Riffle or Reticulate Sculpin, "Unidentified Rhinichthys" refers to Longnose or Speckled Dace. (b) Presence (black) and absence (white) for fish and amphibians based on all eDNA markers combined, with individual detections shown for each of eight replicates. Gray boxes indicate that the taxon was not observed in that replicate, but was detected in other replicate(s) at that site. "Unidentified Oncorhynchus," "Unidentified Cottus," "Unidentified Cyprinidae," and "Unidentified Rhinichthys" may refer to any species within those taxa, respectively. (c) Presence/absence for fish based on 12S rDNA metabarcoding primers targeting teleosts. (d) Presence/absence for salmonids and cottids based on COI primers targeting those taxa. (e) Presence/absence for salmonids based on taxon-specific ND2 primers. (f) Presence/absence for cottids based on taxon-specific CytB primers to an under detection of target species for some primers (File S1).
Finally, the detection limit of the Access Array with eDNA is not known, so under detection may be due to insufficient target DNA.

| Multiple taxon-specific and taxon-general markers improve detection accuracy and reduce detection uncertainty: Examples from salmonids
Salmonids showed near-perfect agreement between electrofishing and eDNA, except a single discrepancy at site 5. We attribute this discrepancy to false-negative detection with electrofishing, based on the strength of eDNA evidence for the presence of Rainbow Trout (detection in 7 of 8 replicates; detection by three genes [12S rDNA, COI, ND2]). Uncertain morphological classification of juveniles probably contributes to this discrepancy, but only four individuals were classified as YOY at this site so the impact of their DNA would be modest. The more likely explanation is that Rainbow Trout are present site 5, a finding that expands the presence of this species further upstream than is generally recognized (Baumsteiger et al., 2005).
Salmonids highlight a trend that we observed study-wide: taxon-specific barcoding markers (COI, ND2, CytB) provide more precise classifications than taxon-general universal markers (12S, 16S, and 18S rDNA). With salmonids, the difference in classification precision is due to the minimal sequence divergence between Coastal Cutthroat and Rainbow Trout at 12S rDNA (e.g., one segregating polymorphism). Under these conditions, Centrifuge classifies many Coastal Cutthroat Trout sequences as either "Rainbow Trout" or "Oncorhynchus," leading to fewer replicates indicating Coastal Cutthroat Trout presence based on taxon-general markers relative to taxon-specific markers. A similar lack of precision was observed for sculpins (identified as "Unidentified Cottus") and daces ("Unidentified Cyprinidae"). This finding mirrors those from studies that also show greater resolution with taxon-specific versus taxongeneral barcode markers (Drummond et al., 2015;Evans et al., 2017).
The importance of basing eDNA presence/absence estimates on observations from multiple independent markers (e.g., Evans et al., 2017) can also be illustrated using salmonids. In microfluidic PCR, DNA is amplified in a separate chamber for each primer pair, so each reaction provides an independent estimate of taxon presence. These independent observations can be used to devise detection criteria that range from lenient (e.g., detected with genes A, B or C) to restrictive (e.g., detected with genes A, B, and C). Multiple loci increase the likelihood that low abundance or difficult to detect taxa can be observed, and they can be used to independently corroborate unexpected observations, such as novel observations for occupancy or range (e.g., Rainbow Trout presence at site 5). Although the use of multiple loci did not improve the detection accuracy for salmonids, they have the potential to improve accuracy, as we show with sculpins (Table 1).

| Database representation and accuracy influence classification of novel diversity: An example from sculpins
Sculpins also showed near-perfect agreement between electrofishing and eDNA, with Riffle and Reticulate sculpin detected at sites 1, 2, and 4 by both methods. eDNA indicated the presence of sculpin CytB sequences at site 5, where it was not detected by electrofishing. As observed with Rainbow Trout, this likely indicates that sculpins extend further upstream than previously documented.
eDNA also indicated the presence of abundant COI and CytB haplotypes that classify to Prickly Sculpin (C. asper Richardson), a morphologically distinctive species that does not occur in Fall Creek.
We classified these sequences as "Prickly Sculpin-like," but they likely represent haplotype lineages that have yet to be attributed to Riffle or Reticulate Sculpin. While this may be a "false-positive" with respect to Prickly Sculpin, these sequences still represent a "true-positive" detection of a sculpin species that has yet to be documented in the database. Further investigation into the genetic and morphological variation in sculpins is warranted to clarify taxo- or incorrect classifications. Longnose and Speckled Dace were observed by electrofishing at three sites, but eDNA only observed Speckled Dace at two sites, and Longnose Dace at one site. 12S rDNA sequences classified as "Unidentified Cyprinidae" were observed at two of these sites, and they are likely attributable to one or both of these species. In contrast to other taxa, we surveyed dace using only one taxon-general metabarcoding marker (12S rDNA), and this locus may lack the precision to discriminate closely related Cyprinidae.
Pacific Lamprey was detected by electrofishing and eDNA using taxon-general (16S rDNA) and taxon-specific (CytB) primers, but it was also imprecisely classified by eDNA to family (Petromyzontidae), even though it is the sole lamprey species in Fall Creek. In this instance, recognition of the genus Entosphenus distinct from Lampetra in NCBI has resulted in nearly identical sequences representing different genera; the most parsimonious resolution is to move all sequences up the taxonomic hierarchy to family Petromyzontidae. Two actions-the application of taxon-specific markers and modifying the reference database to reflect finer taxonomic subdivisions (e.g., subfamily Lampetrinae) or different taxonomic concepts (e.g., a broader definition of Lampetra)will improve the detection accuracy in future assays.

| Detection of amphibians: Importance of detection probability, primer specificity, and assay sensitivity
The accuracy of amphibians detection across sites was moderately lower than overall eDNA accuracy (75% vs. 87%, respectively), but amphibians were also detected in fewer replicates per site, and at sequence counts far lower than fish, invertebrates, or pathogens. We . This likely indicates a low detection probability in electrofishing (e.g., low numbers) or eDNA sampling (e.g., poor assay specificity or sensitivity). Assaying a larger number of replicates or modifying sample collection methods could be one solution to improving accuracy when detection probabilities and falsepositive rates are low (Ficetola et al., 2015;Strobel et al., 2017).
While the ability of amphibians to traverse the aquatic/terrestrial interface can make continuous detection in water samples difficult, poor assay detection for amphibians is also at least partly responsible for lowered detection. For example, 1.4% of the reads from 12S rDNA "Universal Salamander" primers derived from the order Caudata, with the remaining sequences dominated by fish. Our assay included other species-specific salamander primers that ranked highly in amplification efficiency (Coastal Giant Salamander ND2, rank 3 of 40; Roughskinned Newt ND2, rank 13 of 40; File S1), suggesting that primer specificity and efficiency do not fully account for our unreliable detection of amphibians. The influence of a third factor, assay sensitivity, is not well understood. The Fluidigm Access Array is typically used for sequence characterization (e.g., Brown et al., 2016), not quantitative or semi-quantitative analysis. Future studies including quantitative methods like qPCR or ddPCR will help us understand these limits.

| Microfluidic metabarcoding extends opportunities to examine diverse, unrelated communities: Examples from invertebrates and aquatic oomycetes
Over one-third of DNA sequences obtained in this analysis derived from phyla Arthropoda (aquatic insects and crayfish) and Oomycota (water molds; Figure 3, Files S3 and S4). The only taxon from these phyla evaluated by electrofishing, Signal Crayfish, was detected at 100% per site accuracy using eDNA. Our field identification efforts did not extend to aquatic insects, but two universal primer sets targeting 16S rDNA produced 395,584 sequences that were classified into Ephemeroptera (46.4%), Plecoptera (25.2%), and Diptera (19.6%), and 17 of 19 identified genera are confirmed residents of western Oregon (Table 3).
Microfluidic metabarcoding also revealed diverse oomycetes and forest pathogens. Phytophthora was our specific focus, as these waterborne oomycetes include significant pathogens responsible for root/crown rot and stem cankers on oaks and oak relatives (P. ramorum Werres), Port-Orford Cedar (P. lateralis (Mont.) de Bary), and other trees and shrubs (Hansen et al., 2012). Saprolegnia, or similarly cosmopolitan oomycetes may also provide "internal positive" controls that can be used in lieu of spike-in controls (Tourlousse et al., 2017) for sample validation, or as a check for errors in collection (e.g., filtration) or DNA extraction methods.
Our assay included primers for additional animal pathogens, specifically amphibian chytrid fungus (Batrachochytrium dendrobatidis Longcore, Pessier & Nichols; Olson et al., 2013) and the fungus that causes white-nose syndrome in bats (Pseudogymnoascus destructans (Blehert & Gargas) Minnis & Lindner; Blehert et al., 2009). We did not detect these pathogens at Fall Creek, where they are not known to occur, but this assay allows primers for these and other pathogens to be included with routine screening of other eDNA targets.

| CON CLUS ION
We demonstrate that eDNA metabarcoding, using multiple primer pairs via microfluidic multiplex PCR and high-throughput sequencing, can successfully detect the presence of multiple species across Eukarya, validating the ability to define multiple aspects of aquatic biodiversity from a single sample. Although eDNA detected all taxa that were detected by electrofishing with an overall accuracy of 87%, eDNA additionally detected insects, a mammal, and oomycete pathogens. The ability to target multiple genetic loci across multiple taxa allows for independent observations of taxa across loci, allows for fine taxonomic resolution as well as broad detection at higher taxonomic levels, and allows for detection of common microorganisms that can act as proxies for per-sample positive controls. This work broadens the scope of eDNA research by informing conservation decisions for a wide range of taxonomic groups, including common, endangered, rare, and cryptic species, enabling data-driven prioritization and evaluation of management actions across the aquatic community.

ACK N OWLED G M ENTS
We thank electrofishing crews, colleagues who contributed tis-