Species‐level biodiversity assessment using marine environmental DNA metabarcoding requires protocol optimization and standardization

Abstract DNA extraction from environmental samples (environmental DNA; eDNA) for metabarcoding‐based biodiversity studies is gaining popularity as a noninvasive, time‐efficient, and cost‐effective monitoring tool. The potential benefits are promising for marine conservation, as the marine biome is frequently under‐surveyed due to its inaccessibility and the consequent high costs involved. With increasing numbers of eDNA‐related publications have come a wide array of capture and extraction methods. Without visual species confirmation, inconsistent use of laboratory protocols hinders comparability between studies because the efficiency of target DNA isolation may vary. We determined an optimal protocol (capture and extraction) for marine eDNA research based on total DNA yield measurements by comparing commonly employed methods of seawater filtering and DNA isolation. We compared metabarcoding results of both targeted (small taxonomic group with species‐level assignment) and universal (broad taxonomic group with genus/family‐level assignment) approaches obtained from replicates treated with the optimal and a low‐performance capture and extraction protocol to determine the impact of protocol choice and DNA yield on biodiversity detection. Filtration through cellulose‐nitrate membranes and extraction with Qiagen's DNeasy Blood & Tissue Kit outperformed other combinations of capture and extraction methods, showing a ninefold improvement in DNA yield over the poorest performing methods. Use of optimized protocols resulted in a significant increase in OTU and species richness for targeted metabarcoding assays. However, changing protocols made little difference to the OTU and taxon richness obtained using universal metabarcoding assays. Our results demonstrate an increased risk of false‐negative species detection for targeted eDNA approaches when protocols with poor DNA isolation efficacy are employed. Appropriate optimization is therefore essential for eDNA monitoring to remain a powerful, efficient, and relatively cheap method for biodiversity assessments. For seawater, we advocate filtration through cellulose‐nitrate membranes and extraction with Qiagen's DNeasy Blood & Tissue Kit or phenol‐chloroform‐isoamyl for successful implementation of eDNA multi‐marker metabarcoding surveys.

eDNA-related publications have come a wide array of capture and extraction methods. Without visual species confirmation, inconsistent use of laboratory protocols hinders comparability between studies because the efficiency of target DNA isolation may vary. We determined an optimal protocol (capture and extraction) for marine eDNA research based on total DNA yield measurements by comparing commonly employed methods of seawater filtering and DNA isolation. We compared metabarcoding results of both targeted (small taxonomic group with species-level assignment) and universal (broad taxonomic group with genus/family-level assignment) approaches obtained from replicates treated with the optimal and a low-performance capture and extraction protocol to determine the impact of protocol choice and DNA yield on biodiversity detection. Filtration through cellulose-nitrate membranes and extraction with Qiagen's DNeasy Blood & Tissue Kit outperformed other combinations of capture and extraction methods, showing a ninefold improvement in DNA yield over the poorest performing methods. Use of optimized protocols resulted in a significant increase in OTU and species richness for targeted metabarcoding assays.
However, changing protocols made little difference to the OTU and taxon richness obtained using universal metabarcoding assays. Our results demonstrate an increased risk of false-negative species detection for targeted eDNA approaches when protocols with poor DNA isolation efficacy are employed. Appropriate optimization is therefore essential for eDNA monitoring to remain a powerful, efficient, and relatively cheap method for biodiversity assessments. For seawater, we advocate filtration through cellulose-nitrate membranes and extraction with Qiagen's DNeasy

| INTRODUC TI ON
Environmental DNA (eDNA) metabarcoding is defined as the simultaneous identification of a multitude of species through next-generation sequencing from environmental samples (soil, sediment, water). eDNA surveys are gaining attention as a novel, noninvasive, time-efficient, and cost-effective monitoring method (Thomsen & Willerslev, 2015).
Circumventing visual species observation by applying a genetics approach might be especially rewarding to survey the marine biome, where traditional monitoring methods tend to survey only a subset of the community due to time constraints, the inaccessibility of the environment, lack of taxonomic expertise, and funding limitations (Yamamoto et al., 2017). To date, aquatic eDNA has shown great promise as an alternative monitoring method in aquatic environments (Hunter et al., 2015) and in assessing their eukaryotic biodiversity (Stat et al., 2017;Thomsen et al., 2012).
The recognized potential of eDNA research has led to an exponential increase in eDNA-related publications ( Figure 1). However, a meta-analysis of the current literature also shows a broadening spectrum of protocols being used to capture and extract eDNA from aquatic samples (Figure 1a; Supplement 1). Prior to qPCR amplification for end-point analysis, eDNA research follows three general steps. First, a predetermined volume of water is sampled from the habitat of interest. Second, the volume of water is concentrated in a capture step using either precipitation (Dejean et al., 2012), centrifugation (Klymus, Richter, Chapman, & Paukert, 2015), or filtration (Port et al., 2016). Third, DNA is extracted by one of a wide range of commercial kits and modified extraction protocols.
The choice of capture and extraction method is known to significantly influence DNA yields obtained from freshwater samples (Deiner, Walser, Mächlera, & Altermattac, 2015;Eichmiller, Miller, & Sorensen, 2016;Hinlo, Gleeson, Lintermans, & Furlan, 2017;Spens et al., 2016). However, the influence of variation in protocols for the emerging field of marine eDNA is less well established. Recent work suggests that only the capture step in eDNA metabarcoding workflows significantly influences DNA yields from marine samples (Djurhuus et al., 2017). DNA yield in turn affects metabarcoding results in freshwater samples (Deiner et al., 2015), but not in marine samples (Djurhuus et al., 2017). Although findings appear inconsistent between ecosystems, standardization and optimization of capture and extraction methods in eDNA metabarcoding research are desirable to improve comparability between studies (Goldberg, Strickler, & Pilliod, 2015) and the sensitivity of metabarcoding assays.
To date, most eDNA metabarcoding projects, including the recent work on marine eDNA protocol optimization, have used broadspectrum single-marker OTU analyses for biodiversity assessment.
Blood & Tissue Kit or phenol-chloroform-isoamyl for successful implementation of eDNA multi-marker metabarcoding surveys.

K E Y W O R D S
biodiversity assessment, eDNA, extraction, filtration, metabarcoding F I G U R E 1 Meta-analysis on current aquatic eDNA literature displaying (a) the total number of described methods and (b) the proportion of aquatic eDNA research performed in freshwater and marine ecosystems However, OTUs are known to overinflate biodiversity estimates in certain cases (Diaz et al., 2012). Additionally, conservation and management decisions still rely on data at the species, rather than the OTU level (Baker et al., 2016;Keith et al., 2015), making biodiversity assessments conducted using OTUs less directly applicable to conservation management. The field of eDNA research is therefore moving toward massive scale species detection through targeted multi-marker metabarcoding. Although the impact of eDNA capture and extraction protocols on a broad-scale metabarcoding approach has been determined (Djurhuus et al., 2017), no research has yet been performed to determine the effect on a targeted multi-marker metabarcoding approach allowing species-level identification in the marine environment. This information is much needed to maximize the utility of eDNA in marine applications ( Figure 1b).
Here, we investigated the performance of the most prevalent capture and extraction protocols for marine water samples using total DNA yield measurements. As filtration is the most commonly used capture protocol for seawater, we evaluated the filtration capture performance for a range of membrane types and pore sizes.
We also compared the DNA extraction performance for several extraction protocols. We determined the effect of different capture and extraction protocols on the resulting DNA yield by comparing metabarcoding results from both an optimal and low-performance capture and extraction protocol. For taxonomic breath, we employed four metabarcoding assays. We assessed OTU and species richness per replicate and the overall detected biodiversity within both treatments for each metabarcoding assay. Our results clearly demonstrate the benefits of optimizing and standardizing eDNA metabarcoding methods prior to widespread implementation in the marine environment.

| MATERIAL S AND ME THODS
We performed two comparison experiments to assess how multiple capture and extraction protocols affect DNA yield ( Figure 2

5.
Capture protocol key Designing an optimal (high DNA yield) and low-performance (low DNA yield) protocol based on filtration and extraction performance experiments first experiment, we evaluated the capture performance of several commonly employed filter membranes and pore sizes. In the second experiment, we evaluated the performance of the most commonly used extraction protocols according to our meta-analysis (Supplement 1). From both comparison experiments, we developed an optimal (high DNA yield) and a low-performance (low DNA yield) protocol. We analyzed metabarcoding data obtained from technical replicates treated with the optimal or low-performance protocol to determine the effect of protocol choice and DNA yield on multimarker eDNA metabarcoding research in the third experiment.

| Sample collection and contamination prevention
Prior to field and laboratory work, we sterilized all equipment using a 10-min exposure to 10% bleach solution (Prince & Andrus, 1992).
We decontaminated all sampling bottles (2 L, HDPE Natural, EPI Plastics) by rinsing twice with ultrapure water (UltraPure™ DNAse/ RNAse-Free Distilled Water, Invitrogen™), submerging in 10% bleach for ten minutes, and rinsing twice again with ultrapure water. We checked for contamination at each step in the process. We filled two 2 L bottles with ultrapure water and placed them among the sampling bottles to test for contamination during sampling. We added negative filtration controls by filtering 500 ml ultrapure water and included two extraction blanks consisting of 500 µl ultrapure water.
All laboratory work prior to amplification was performed in a PCRfree designated clean room.
Water samples were collected from Otago Harbor, Dunedin, New Zealand (latitude 45°52′49.33″S, longitude 170°30′22.58″E). We collected surface samples from a pontoon to reduce introducing bias from sediment with higher DNA concentrations mixing with the water column (Turner, Uy, & Everhart, 2015). Samples were collected at the same time for each experiment and transported back to the laboratory within one hour of collection and mixed by inversion to ensure an even distribution of DNA before aliquoting into technical replicates.

| Capture performance
We compared capture efficiency of the three most frequently used The volume of water able to be processed depends on pore size and filter material. We therefore repeated this experiment by filtering water through each of the five filters until clogging occurred.
The volume filtered before clogging was 500 ml for the 0.2-µm PC filters, 2,500 ml for the 1.2-µm PC filters, 1,000 ml for the 0.2-µm CN filters, 3,000 ml for the 1.2-µm CN filters, and 5,000 ml for the 1.6-µm GF filters. All samples were extracted with the PCI protocol, and DNA extracts were stored at −20°C until further analysis.

| Extraction performance
We compared DNA extraction performance across seven proto- PCI alcohol DNA extraction protocol (Renshaw et al., 2015), (f) Silica extraction protocol (Alawi, Schneider, & Kallmeyer, 2014;Ogram, Sayler, & Barkay, 1987), and (g) Magnetic Beads extraction protocol (Oberacker et al., 2018). A detailed description of each procedure can be found in Supplement 2. Prior to comparison, each treatment was optimized for DNA extraction from marine water samples.
Based on the results from the capture experiment, we sampled seventy 1,000 ml technical replicates, ten for each extraction method.
Each sample was vacuum filtered through a 1.2-µm CN filter. The final elution volume for each DNA extraction protocol was 200 µl to ensure accurate comparison across methods. DNA extracts were stored at −20°C until further analysis.

| DNA yield measurement and statistical analysis
Capture and extraction performance were based on total DNA concentration measured in triplicate for each sample on a Qubit ® 2.0 Fluorometer (Qubit ® dsDNA HS Assay Kit, Invitrogen™). We used a one-way analysis of variance (ANOVA) to compare capture and extraction performance across the different treatments tested. Post hoc comparisons for ANOVA were performed using Tukey-Kramer's test.
All statistical tests were performed in R (http://www.R-project.org).

| Metabarcoding
Based on the performance of the different treatments included in the capture and extraction experiments, we designed an optimal (fil- Library preparation followed the protocol described in Ref. Stat et al., 2017). Briefly, we chose five random technical replicates per treatment to be analyzed using four metabarcoding assays targeting two fragments of the 16S rRNA gene region, one fragment of the cytochrome c oxidase subunit I (COI) gene region, and one fragment of the 18S rRNA gene region ( Forward and reverse reads from the eukaryotes (18S) and eukaryotes (COI) metabarcoding assays were merged using default settings in PEAR v 0.9.10. (Zhang et al., 2014). Reads were separated by barcode and assigned to samples using Geneious v 9.1.6. (Kearse et al., 2012) with no mismatches allowed. Amplicons matching the primer, and sequencing adapter for the single-end assays, at a 100% level were retained for further analysis. The remaining amplicons were filtered using USEARCH (Edgar, 2010) based on a maximum error of 0.1, minimum length of 100 for single-end reads and 300 for paired-end reads, and removal of singleton sequences and sequences containing ambiguous bases. Reads passing quality filtering were clustered at 97% to generate an OTU table, according to TA B L E 1 Metabarcoding assays and the respective primer sets used for biodiversity detection standard settings in USEARCH. OTUs with single-observation occurrence were discarded from the dataset. OTUs passing all quality filtering steps were queried using BLASTn against the full NCBI database.
Taxonomic data for each metabarcoding assay were obtained as follows: (a) all BLAST hits with similarity >97% were retained per OTU, (b) lowest taxonomic level across all remaining BLAST hits for each OTU was used, and (c) species identification required an existing record of the species occurring in New Zealand (Ayling, 1987;Cook & Archer 2010). These criteria led to species identification for both targeted metabarcoding assays, and genus and family identification for the eukaryotes (COI) and eukaryotes (18S) metabarcoding assays, respectively. BLAST hits resulting in unicellular picoplankton and bacteria were discarded, as the focus of this study was on multicellular eukaryotes.
We compared the overall OTU and taxonomic biodiversity retrieved by both optimal and low-performance treatments, as well as the OTU richness and species richness retrieved per technical replicate for each metabarcoding assay. Student's t test was used to compare the difference in OTU and species richness between the optimal and low-performance eDNA protocol.

| Extraction performance
The choice of extraction method significantly affected the total amount of DNA recovered from 1,000 ml of seawater (F (6,63) = 81.83, p < 0.0001) (Figure 4)
Species and OTU detection from targeted fish and crustacean assays were strongly affected by protocol choice and the differing DNA yield. The increased total and target DNA obtained using the optimal protocol resulted in a ~5-fold increase in average OTU and species richness per replicate for both assays ( Figure 5).
Besides increased richness, increased DNA yields also resulted in more consistent OTU and species detection across replicates (Supplement 6). For both assays, many species were left undetected when the low-performance protocol was used, with six versus two fish; and fourteen versus five crustacean species detected for the optimal and low-performance protocols, respectively ( Figure 6; Supplement 6). In contrast to the targeted assays,

| Capture performance
Processing aquatic samples in eDNA studies starts with reducing the volume and capturing DNA by filtration, for which membrane type and pore size of the filter vary between studies (Supplement 1). Membranes with small pore sizes (0.2 µm) produced an increased DNA yield, irrespective of the membrane type used for the same filtered volume of water. However, total DNA yield does not distinguish between eukaryotic target DNA and bacterial DNA. Bacteria are capable of passing through the 1.2 and 1.6 µm pore sizes (Turner, Miller, Coyne, & Corush, 2014); thus, the increased DNA yield seen when 0.2-µm filters were used could be attributed to more microorganisms being captured by that pore size, but not others (Stat et al., 2017). If true, this would render comparisons between different pore sizes invalid. Taking this into account, although we did not a detect a difference in DNA yield between the two membranes with 0.2 µm pores, cellulose-nitrate filters with 1.2 µm pores significantly increased DNA yield compared to other filters with larger pore sizes tested. Sample volume usually exceeds the limit of filters with smaller pores (Uchii, Doi, & Minamoto, 2016); thus, we also filtered the maximum volume possible to determine volume limit and maximum DNA yield for each filter. In this case, cellulose-nitrate filters outperformed polycarbonate filters of the same pore size both in maximum volume filtered and in DNA yield. Although the glass-fiber filter tested in this study had the largest pore size and allowed the largest sampling volume to pass through, DNA yields for this filter were significantly lower than cellulose-nitrate membranes, as has been found in other studies (Djurhuus et al., 2017). Filtration was the sole method tested in this study, as it is currently the most prevalent capture method in aquatic eDNA research.
Previous research performed in freshwater ecosystems (Deiner et al., 2015;Eichmiller et al., 2016;Hinlo et al., 2017) shows filtration outperforms both centrifugation and precipitation. We recommend using cellulose-nitrate filtration to capture DNA in future eDNA studies conducted in temperate marine environments, a finding in agreement with that for the freshwater ecosystem (Hinlo et al., 2017). These findings are suggesting some commonalities across aquatic eDNA sampling environments. Choice of filter pore size for a given study will be dependent on the number of suspended particles and the turbidity of the water being sampled (Table 2). Recently, the Sterivex™ filtration system has gained popularity and was found to be the optimal capture strategy in freshwater eDNA research (Spens et al., 2016). In Sterivex™ systems, the plastic housing surrounding the filter circumvents the need for a vacuum filtration manifold, allowing for in-field filtration strategies.
The faster processing time of samples might aid in halting eDNA degradation until extraction. However, the significant increase in cost per filter might make Sterivex™ filters less desirable in some applications.
Comparative studies in the marine environment between the Sterivex™ filtration system and vacuum filtration are needed to determine the advantages and drawbacks of both capture protocols.  (Table 2).

| Metabarcoding approach
Optimizing the capture and extraction step in our marine eDNA protocol resulted in a ninefold difference in total DNA yield compared to the poorest performing methods. Although total DNA yield is not necessarily indicative of target DNA yield (Stat et al., 2017), lower Ct-values were recorded with increased total DNA yield for all four metabarcoding assays tested in this study (Supplement 4). Protocol choice, which resulted in differing total DNA yield and starting template, significantly affected downstream metabarcoding analyses.
Unfortunately, due to the inherent connection, we were unable to determine whether the protocol used, or the obtained DNA yield  (Murray et al., 2015;Saulnier, Decker, & Haffner, 2009), through intensified stochasticity in early PCR amplification cycles, and the protocol used (Eichmiller et al., 2016;Hinlo et al., 2017), through the introduction and removal of specific contaminants and inhibitors, are known to affect downstream metabarcoding analyses.
Neither of the broad-scale metabarcoding assays tested (COI and 18S) displayed a difference in the average OTU richness per replicate between the optimal and low-performance protocol. This result is in agreement with the only other evaluation of marine eDNA processing protocols (Djurhuus et al., 2017). The average family richness for the eukaryotes (18S) assay per replicate, however, did show a significant difference in favor of the low-performance protocol. Despite a lack of impact in these broad-scale metabarcoding assays, a highly variable metabarcoding output was retrieved between replicates within each treatment and between treatments. The variability in metabarcoding results led to differences in the overall biodiversity retrieved between the optimal and low-performance protocol (Supplement 6).
Environmental samples contain a highly complex eDNA signal, due to the presence of DNA signatures from the entire community.
The enormous complexity is likely linked to the variability seen in our broad-scale metabarcoding assays and illustrates the difficulties of using universal primer sets to assess the true biodiversity in complex eDNA samples, such as those commonly sourced from marine systems. More intensive sampling and deeper sequencing would be required to accurately represent the biodiversity using universal primers, both of which would add time and financial costs to any study. Although we note that sequencing costs continue to decrease, as the technology advances. Many universal primer sets are also unable to generate species-level assignments in metabarcoding studies, due to incomplete reference datasets and current sequencing constraints (Soergel, Dey, Knight, & Brenner, 2012). Thus, currently, some universal metabarcoding assays result in low taxonomic resolution. OTU analyses, which make up the bulk of eDNA metabarcoding research to date, have been proposed to alleviate some of the problems associated with universal primer sets (Yoon et al., 2016).
However, OTU metrics are known to inflate biodiversity estimates (Diaz et al., 2012). Moreover, OTU analyses do not provide specieslevel information that may be required in certain instances of ecosystem conservation (Baker et al., 2016;Keith et al., 2015). The field of eDNA research is therefore moving toward species detection at massive scale through targeted multi-marker metabarcoding, making the performance of targeted assays much more important to the future of the discipline.
While targeted assays may be more relevant for biodiversity surveys, they are also more sensitive to protocol choice. The increased DNA yield obtained through optimization of the capture and extraction step led to higher OTU and species richness estimates for both targeted assays. Our results clearly show, for the first time, that targeted metabarcoding assays amplifying low-abundance taxonomic groups (i.e., fish and crustaceans) are far more sensitive to protocol choice and the resulting reduced DNA yields compared to broadscale metabarcoding assays. Stochasticity in the early cycles of qPCR TA B L E 2 Recommendation and critical considerations of laboratory protocol for future marine eDNA research We strongly suggest using the phenol-chloroform-isoamyl extraction procedure to reduce cost and contamination risk

Extraction method
Targeted assays amplify only a subset of the biodiversity present in an eDNA sample but allow species identification and obtain consistent results. Universal assays can be used to give a general overview of the biodiversity amplification, intensified in low copy number situations (Saulnier et al., 2009), likely reduced the number of detected species and the consistency between replicates when the low-performance protocol was used. Most affected was the fish (16S) assay, which led to amplification issues and inconsistent detection results between replicates with lower DNA yields. Multiple species known to occur in the Otago Harbour (Ayling, 1987) were only detected when the optimal capture and extraction methods were used (e.g., New Zealand blueback sprat (Sprattus antipodum); yellow-eye mullet (Aldrichetta forsteri); chinook salmon (Oncorhynchus tshwaytscha); and common triplefin (Forsterygion lapillum)). A similar pattern was seen in the crustacean

| CON CLUS ION
Targeted metabarcoding assays amplifying low-abundance taxonomic groups appear more affected by protocol choice and the resulting DNA yield difference than broad-scale metabarcoding assays.
A shift toward assays allowing species identification in eDNA metabarcoding studies, therefore, requires optimization of eDNA capture and extraction protocols prior to commencing marine eDNA studies to reduce cost and to increase efficiency and applicability. Our results suggest that capture via cellulose-nitrate membranes and extraction via Qiagen's DNeasy Blood & Tissue Kit present an optimal protocol, but that phenol-chloroform extraction may be preferable to reduce cost and lower contamination risk depending on the circumstances.
Targeted metabarcoding, after optimization and standardization, has the potential to address the challenges of assessing and monitoring biodiversity in the vast and inaccessible marine biome. Using our optimal capture and extraction protocol, we obtained data on 20 species, 16 genera, and 49 families in a time-efficient and cost-effective manner, clearly illustrating the power and potential of aquatic eDNA monitoring in the marine environment when correctly optimized.

ACK N OWLED G M ENTS
We thank the editor and two anonymous reviewers for their constructive comments, which helped us to improve this manuscript.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N
Gert-Jan Jeunen: This person made substantial contributions to the conception, acquisition, analysis, and interpretation of the data. This person also made substantial contributions in drafting the manuscript. This

DATA ACCE SS I B I LIT Y
All sequence data and bioinformatics scripts will be archived on Sequence Read Archive (SRA) upon acceptance of the paper.