Screening marker sensitivity: Optimizing eDNA- based rare species detection

Aim: Environmental DNA (eDNA)- based techniques are useful tools in disciplines such as conservation biogeography at local to global scales since they provide promising methods to locate organisms at low abundance. Here, we raise a largely overlooked issue that the marker (primer pairs and/or probes) sensitivity of eDNA- based detection should be optimized and reported to improve detection performance and result interpretation. Location: Global. Methods: We analysed 250 articles published between 2008 and 2019 that sought to detect animals from environmental water


| INTRODUC TI ON
Rare (low-abundance) species are critical for maintaining ecosystem functioning and services (Dee et al., 2019;Lyons et al., 2005).
Fast and accurate detection of rare species-such as newly introduced non-indigenous species or endemic endangered specieshas attracted the attention of ecologists and policymakers alike, as it underpins management (Bohmann et al., 2014;Dee et al., 2019;Jerde et al., 2011;Mehta et al., 2007;Sepulveda et al., 2020;Trebitz et al., 2017). For example, one fundamental question for conservation biogeographers is whether an introduced, non-indigenous species can establish and form self-sustaining populations in new areas, the answer to which is critical to triggering rapid responses. However, the abundance of the introduced propagules may be extremely low and be restricted to limited geographical ranges, challenging many traditional detection methods and impeding conservation attempts (Darling & Mahon, 2011;Dee et al., 2019;Harvey et al., 2009;Hoffman et al., 2011;Zhan & MacIsaac, 2015). Environmental DNA (eDNA)-based techniques have emerged as promising tools for targeting these species in a non-destructive manner while promising enhanced sensitivity. These techniques have the dual advantages of high environmental prevalence and ease of extraction from bulk environmental samples (Bohmann et al., 2014;Taberlet et al., 2012).
By using properly selected protocols, it is possible to discriminate a single species or profile an entire community from environmental samples (Bohmann et al., 2014). The former, which is the focus of the present study, is achieved by using species-specific primer pairs (and/or probes) of chosen genetic markers (hereafter markers) to amplify single species through PCR. Like many emerging techniques, eDNA-based species detection is imperfect and is affected by an array of technical issues that need to be disentangled (Ficetola et al., 2016;Roussel et al., 2015).
For example, a typical application of eDNA for detecting species from aquatic environments comprises multiple steps, and problems at any of these stages may impede detection and lead to false interpretations (Bohmann et al., 2014;Goldberg et al., 2016;Song et al., 2020). False negatives (type II errors), for instance, lead to incorrect assessments for "undetected" endangered species or nonnative species, leading to delayed conservation or prevention strategies, respectively (Darling & Mahon, 2011;Furlan et al., 2016;Xiong et al., 2016). Such uncertainties may impede full embracement of eDNA methods by policymakers and managers to trigger conservation applications (e.g. invasive species management) (Jerde, 2019;Sepulveda et al., 2020). Many case studies, particularly in aquatic systems, have focused on technical issues including primer design (Farrington et al., 2015;Wilcox et al., 2013), sample handling (Mächler et al., 2016;Moyer et al., 2014;Takahara et al., 2015), DNA extraction (Deiner et al., 2015), inhibition abatement (McKee et al., 2015) and selection of PCR method (Wood et al., 2019;Xia, Johansson, et al., 2018), thereby offering useful solutions to advance eDNA-based species detection. A recent review of critical considerations in detecting aquatic species via eDNA methods established important guidelines for such steps (Goldberg et al., 2016).
For marker selection, existing studies focus mostly on specificity for single species-based applications or generality for communitybased (or multispecies) applications (Deagle et al., 2014;Goldberg et al., 2016;Wilcox et al., 2013;Zhan et al., 2014). However, surprisingly little attention has been paid to screening sensitivity of markers, for which high sensitivity is usually expressed as a low limit of detection (LoD). This may unwittingly undermine the detection sensitivity of a given eDNA-based species detection protocol, which can be assessed as the lowest amount of target DNA that can be detected with 95% probability (Schultz & Lance, 2015). A developed protocol is likely-possibly inevitably-to be applied in subsequent studies. This has broad implications in conservation biogeography studies because developed and published markers can then be used by researchers in any locations where the species may occur (e.g. a transoceanic introduced non-indigenous species). A well-screened marker with robust sensitivity thus creates legacy value beyond the time and location of its origin.
To call attention to marker selection from a sensitivity perspective, we conducted a meta-analysis and addressed a deficit that insufficient attention has been paid to screening and reporting of the sensitivity of species-specific markers, which may harm the efficacy of eDNA as a tool in detecting low-abundance species. We first focused on marker sensitivity screening and the PCR method selection, which have been demonstrated to influence detection performance.
We then identified limitations that result from a dearth of information regarding LoD-which we define as the lowest amount of target DNA that can be detected-of markers used for eDNA-based species detection. These issues may result in use of suboptimal assays, thereby inflating false negatives and delaying management strategies. We demonstrated this by examining eDNA detection studies for species on the "100 of the world's worst invasive alien species" (Lowe et al., 2000).

| Literature surveys
To generate a quality data set to address the issues above, we focused on studies using species-specific markers to detect animal species from water samples because these studies represent the most extensive application of eDNA-based methods for detecting low-abundance species. Specifically, we searched the ISI K E Y W O R D S aquatic ecosystems, environmental DNA, false negatives, limit of detection, marker screening, optimization, PCR method, sensitivity Web of Science (WoS) using "eDNA" or "environmental DNA" and "species detect*" as keywords from 2008 through 2019, which encompasses the latest studies since macro-organism detection via eDNA was first used (Ficetola et al., 2008). We narrowed our survey to 11 WoS categories that comprise the majority of articles in aquatic ecosystems (Table S1, supporting information).
Through paper-by-paper identification, we removed articles that were beyond the scope of this study, such as those exclusively focusing on species barcoding (i.e. not real environmental samples) or detecting algae and bacteria, etc., or that purely focusing on concept validations with aquarium-based samples (i.e. validations in field samples were retained), resulting in 250 retained articles (Tables S1-2).

| Criterion extraction
For each article, criteria pertaining to source of marker used (i.e. designed in study, cited from existing study or combined when a mixed source is identified), status of sensitivity screening of mark-  [ddPCR] or combined when different methods were used in a single article) were extracted. We examined the methods and results of each article to extract the above information, and we also examined their supplementary materials if relevant information was not available in the main article but was directed to them. Articles that studied multiple species were broken down to single species, resulting in a case-based data set of 388 species, which we used to calculate the prevalence of taxonomic groups (Tables S1-2). For cases that used cited assays, corresponding information was incorporated from the original studies. To investigate the evolution of these efforts, we used time series linear models to relate the portion (%) of each criterion to the year of publication, with the analysis performed in R (version 3.6.2) (R Core Team, 2019) and the time series analysis using "dplyr" package (version 0.8.3). Since no records were identified in 2009-10, and only one was available in 2008 (see results), we used articles in 2011-19 for time series analysis to avoid missing values.

| Examples of the worst invasive species
To highlight potential problems of unscreened (i.e. as to sensitivity) markers being used in later studies, we reviewed articles detecting aquatic animal species listed on the "100 of the world's worst invasive alien species" (Lowe et al., 2000). We considered these species as examples because they represent concerns globally, which was critical to avoid biases in species selection and to simplify literature survey. We believe that studies on the selected species represent a snapshot of how sensitivity screening of markers is treated at a broad scale, though we might have missed some species not on the list but which have received extensive studies such as the Asian carp (Jerde, 2019). We identified articles in which the aquatic animal species on the list were detected via eDNA approaches using originally designed markers by using the same keywords for constructing the above article data set, together with species name (either Latin or common name). We retained several markers not originally designed for eDNA samples sensu stricto (e.g. plankton samples or gut content) because we found subsequent studies used them for eDNA samples (see results). Based on these records, we calculated the number of times each original marker was used in other studies. To achieve this, we reviewed all articles citing each of these original articles and only retained studies using the marker.
Among the 165 studies that used newly designed markers, only

| Detection of the "worst invasive alien species"
Twenty aquatic animal species are listed in the "100 of the world's worst invasive alien species" (Lowe et al., 2000). As of 31 December 2019, 14 of these species have been recorded in Web of Science as having been detected at least once each from eDNA samples (including plankton samples). Brown trout, bullfrog and common carp were each represented by more than one marker, resulting in a total of 19 species-specific markers ( Figure 3; Table S4). Three (17.6%) of these markers were screened to achieve low LoD and thereby optimize detection sensitivity, and LoD was reported for nine (47.4%) of them. Seven (36.8%) of the markers were subsequently used in other studies, even though only one (14.3%) had been screened previously for sensitivity and two (28.6%) had LoD reported (Figure 3).
In addition, a few markers (e.g. fishhook water flea, green crab) not originally designed for eDNA samples sensu stricto were also used in later studies for eDNA sample-based detection (Figure 3; Table S4).

| D ISCUSS I ON
Rapid and accurate locating of rare species is vital for conservation applications (e.g. rapid response to introduced non-indigenous species). eDNA-based techniques serve as promising tools for detecting to fully embrace such tools, notably due to uncertainties associated with eDNA results (Jerde, 2019;Sepulveda et al., 2020). Use of optimal markers following sensitivity screening of multiple candidates can reduce false negatives (e.g. Farrington et al., 2015;. Despite this, an overwhelming 84.2% of articles that used newly designed markers seemingly overlooked this crucial step, contrasting the increasing use of more sensitive detection platforms such as qPCR (Figure 2b, Table S3). Furthermore, the proportion using sensitivity-screened markers did not change over time ( Figure 2c, p =.186, Table S3), which seems more problematic than the low overall rate per se given the increasing number of eDNA applications. Similarly, 46.7% of studies using newly designed markers did not report their LoDs, a pattern that did not decline over time (Figure 2d, p =.288; Table S3). The absence of knowledge on LoD prevents researchers from identifying the boundary at which false-negative rate increases, which in turn compromises eDNAbased species detection programmes. For example, two recent studies detecting fishes using both eDNA methods and the American Fisheries Society's standard sampling assessment (gillnetting, boat electrofishing and snorkelling) reported overall lower detection rates with eDNA than with standard methods (Perez et al., 2017;Ulibarri et al., 2017). Use of eDNA-based technique as a detection tool was challenged in these programmes, partially owing to the lack of optimization of both sampling methods and assay design (Perez et al., 2017;Ulibarri et al., 2017). In contrast, a recent study reported an improvement in detection performance for Asian carp following a switch to more sensitive markers (Song et al., 2020). To some extent, the above examples imply false negatives associated with eDNA methods, which may delay the initiation of conservation strategies (e.g. rapid response to invasive species). Therefore, eDNA-based species detection requires careful protocol design and execution to improve confidence in the results obtained (Furlan et al., 2016;Jerde, 2019;Jerde et al., 2011;Loeza-Quintana et al., 2020;Sepulveda et al., 2020;Song et al., 2020).
High marker sensitivity is typically expressed by a low LoD, which has a theoretical minimum at one copy per PCR (Bustin et al., 2009). Though it may be challenging to achieve this theoretical limit, eDNA markers should be screened to achieve an LoD as low as possible to maximize detection probability. In a more applied perspective, a lower LoD for eDNA-based detection can result in a greater probability of a rare species being detected through eDNA earlier than traditional detection (Jerde, 2019).
However, prior knowledge about LoD is almost impossible until tested (Farrington et al., 2015;. Therefore, a sensitivity screening step is required to obtain a robust marker with a low LoD. To achieve this, designing multiple candidate markers for sensitivity screening may be required. Different genes with multiple markers based on each can increase the probability of finding highly sensitive ones. Different marker design platforms can also be considered to generate different high-score candidates, which may exhibit varied amplification efficiency as these design tools often stem from different algorithms Goldberg et al., 2016). Even though further studies are required,

PCR methods cPCR qPCR
Current formats 0.5 ~ 0.01 ng, 7.25 × 10 -11 ng/μl, 1 zooid, 5 D-hinge-stage larvae, 50 copies 1 individual per 100 m 2 , 2 copies, 3 copies per reaction, 1 larva, 0.013-3.2 pg/μl, <10 copies 4 μl −1 Suggested formats # of copies # of copies TA B L E 1 Current report formats for detection limit of markers in literature and suggested report formats for single species detection from eDNA samples F I G U R E 3 Number of uses in other studies for 19 designed markers of 14 aquatic animal species on "the 100 worst invasive species in the world" (Lowe et al., 2000). Grey bars represent markers that were screened for sensitivity to optimize detection, while white bars were unscreened markers. An asterisk indicates that the limit of detection (LoD) of the markers was reported, otherwise it was unreported. The dashed line indicates that the genetic marker has accumulated no uses as of 31 December 2019 (Web of Science)

Rana catesbeiana or Lithobates catesbeianus (Bullfrog) b
Rana catesbeiana or Lithobates catesbeianus (Bullfrog) a vitro and in situ) to confirm specificity and sensitivity before formal use (Darling & Mahon, 2011;Goldberg et al., 2016;Klymus et al., 2019;Langlois et al., 2020;Roussel et al., 2015;Wilcox et al., 2013). It is important to keep in mind that highly sensitive markers are sensitive to trace levels of target eDNA regardless of its source, thus requiring strict quality control to avoid contamination (Darling & Mahon, 2011;Goldberg et al., 2016).

Number of uses of cited geneƟc marker
We observed that designed markers were used more frequently than cited ones in almost all years (Figure 2a), although over 36% of the original markers in the "worst aquatic invaders" case study were utilized in at least one subsequent study (Figure 3) than those used at the original design site (Wilcox et al., 2013). We expect that a similar situation applies to marker sensitivity (see below for more discussion), and both require associated information reported. Almost 53.3% of studies reported the LoD of their assays used, though formatting varied across studies (Table 1).
Diverse characterization of the LoD and reporting formats can also impede their use in future studies (Bustin et al., 2009;Darling & Mahon, 2011) because they render difficult comparisons of marker efficiency for a single species or among laboratories (Klymus et al., 2019). Therefore, it is vital to ensure that the sensitivity of newly designed markers be screened, tested and reported in publications (MacDonald & Sarre, 2017;Veldhoen et al., 2016).
In addition, their testing and reporting should be formatted in a standard manner such as the number of copies of the target fragment, which can be derived by using gBlocks gene fragments (i.e. synthesized DNA fragments, Langlois et al., 2020). For example, Klymus and colleagues (2019) reported comprehensive guidelines in calculating and reporting LoD for qPCR, which dominates current eDNA-based applications (Figure 2b).
In addition to enhanced efforts in sampling, inhibitor abatement and quality control (Wilcox et al., 2013;Wood et al., 2019), the use of robust markers and more sensitive detection platforms provide additional room to improve detection of low-abundance species from eDNA samples (Song et al., 2020;. However, it is important to interpret the changes of criteria investigated in the present study from an evolving perspective. For example, the increasing use of qPCR may reflect the collective pursuits of highly sensitive and quantitative tools Hunter et al., 2016;Xia, Johansson, et al., 2018). Our study may have underestimated the proportion of studies using sensitivity-screened and LoD-known markers if the authors did, in fact, test but not report these results.  (2016) established comprehensive guidelines and recommended minimum information that should be considered in eDNA-based aquatic species detection. Sepulveda et al., (2020) proposed a decision-support framework to assist decision-making in aquatic species management based on eDNA results. Additionally, we encourage authors to screen sensitivity and report LoD for any newly developed markers.
As rare species detection becomes increasingly based upon eDNAbased techniques, we believe that such routine efforts can add additional value to their work by benefitting future users and advance the use of eDNA in conservation studies and applications.

ACK N OWLED G EM ENTS
We are grateful for the constructive comments of two anonymous

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13262.

DATA AVA I L A B I L I T Y S TAT E M E N T
All supporting information can be found online for this article.