Cryptic biodiversity: A portfolio‐approach to coral reef fish surveys

Biodiversity conservation and management requires surveillance that captures the full spectrum of taxa. Here, we showcase the potential for a portfolio of visual, extractive, and molecular methods for detecting previously hidden components of tropical fish biodiversity in an economically and culturally valuable marine site that spans a tropical‐temperate ecotone—the Ningaloo Coast World Heritage Area. With scale and practicality in mind, we demonstrate how environmental DNA (eDNA) methods deployed in a stratified sampling design can yield a more comprehensive monitoring program for species presence than current alternatives (e.g., extractive sampling via anesthetic). eDNA from filtered water samples detected up to six times as many cryptobenthic fish species per site than samples collected with anesthetic, indicating it is a potentially powerful tool for assessing biodiversity of tropical fishes. However, there were also species that were only found when using anesthetic and the contribution of cryptobenthic species to overall diversity of the fish assemblage was unexpectedly low, suggesting not all cryptobenthic fish species have been detected with eDNA. There were also distinct differences in cryptobenthic assemblages both among sites and sample depths (2–3 m) when using eDNA from filtered water, suggesting this technique may be able to identify fine scale spatial differences in cryptobenthic fish assemblage. eDNA collected from water detects the most cryptobenthic species and is therefore an efficient tool for rapidly assessing biodiversity, but extractive techniques may still be required for biological and monitoring studies, and when combined with eDNA sampling provides the most comprehensive assessment of cryptobenthic fishes.

and distribution patterns.As coral reefs are increasingly threatened by stressors, such as a changing climate (Hughes et al. 2003;Carpenter et al. 2008;Muir et al. 2022), there is a pressing need to document and improve our understanding of these significant ecosystems through marine survey techniques that capture a broad cross-section of biodiversity.
Fish are often used as sentinels for understanding and monitoring marine biodiversity because of their socio-economic importance (e.g., human consumption, tourism) and fundamental role in maintaining ecosystem functions (Holmlund and Hammer 1999).With more than 4000 coral reef fish species described globally, fishes are a diverse group which are more taxonomically resolved than other marine taxon, such as invertebrates (Fisher et al. 2015).Many fishes are prominent and easily observed, particularly the diurnal, demersal species that are typically targeted in biodiversity assessments (Sale 1991).However, fish diversity surveys rarely incorporate adult fishes less than 50 mm in length that are visually and/or behaviorally cryptic and typically referred to as cryptobenthic (Ackerman and Bellwood 2000;Brandl et al. 2018).This is despite cryptobenthic fishes comprising nearly half of all fish species on coral reefs (Depczynski and Bellwood 2004;Brandl et al. 2018).Estimates of cryptobenthic fish biomass and productivity also suggest they are important conduits for trophic flows (Depczynski and Bellwood 2003;Brandl et al. 2019), which is a core function for maintaining coral reef ecosystem health (Morais et al. 2020).The short life span of many cryptobenthic species (months to only a few years) suggests they may be sensitive to environmental change and stressors (Wilson 2004;Bellwood et al. 2006;Depczynski and Bellwood 2006), making them potentially suitable indicator species for monitoring as their population assemblages can undergo phase-shifts in response to disruptions (Bellwood et al. 2012).However, many cryptobenthic fishes remain undescribed and their substantial contribution to current measures of fish diversity and ecological significance are underestimated (Allen 2015;Brandl et al. 2018).
Standard visual fish survey methods are unsuitable for accurately monitoring cryptobenthic fishes because their behavior and visually camouflaged nature makes them hard to detect, attributes further compounded by the complex and diverse micro-environments where they often live (Depczynski and Bellwood 2004).For example, nearly 30% of the known fish biodiversity (> 1500 spp.) of Western Australia's Kimberley region, including cryptobenthic and pelagic species, have only been recorded by extractive methods (Moore et al. 2020).Accordingly, surveys of cryptobenthic fish assemblages typically consist of using ichthyocides (e.g., rotenone) or terminal anesthetics (e.g., clove oil, Ackerman and Bellwood 2002, Finquel MS-222;Robertson and Smith-Vaniz 2008).While the extractive nature of these methods is critical for confirming taxonomic identities and generating DNA reference libraries (Moore et al. 2020), they are often unsuitable for repeated monitoring and may not always align with conservation objectives to limit threats to population persistence or recovery.Furthermore, identification of collected fish requires taxonomic expertise at a time when these skills are declining (Noss 1996;Hutchings 2019).Indeed, the taxonomy of cryptic fishes is often poorly resolved, and many species remain undescribed with approximately 30 new discoveries expected each year over the coming decades (Brandl et al. 2018).Likewise, underwater video methods are generally only suitable for larger, conspicuous species and require high water clarity and extensive laboratory analysis of images (Murphy and Jenkins 2010).
An effective fish monitoring survey will enable detectability of target species, have sufficient power to detect change, and maintain consistency over the long-term (Lindenmayer et al. 2020).Molecular genetic techniques are emerging worldwide as a new tool to monitor fish assemblages because they have potential to offer a non-invasive and accurate way to identify species that can be deployed at large scales (Taberlet et al. 2018;Jeunen et al. 2019;Bessey et al. 2020).DNA is shed into the water by fishes through excreted cells, tissue, feces, or decaying matter.This environmental DNA (eDNA) can be collected and sequenced, which enables species to be identified by comparison to a reference database of DNA sequences (Thomsen and Willerslev 2015).Collection of aquatic eDNA is primarily through water filtration (Tsuji et al. 2019), although alternative approaches extract eDNA from sediment samples (Koziol et al. 2019) or use devices suspended in the water column to passively collect eDNA (Bessey et al. 2020).Selecting the best eDNA collection method is target-taxa specific and an important consideration for experimental design.Cryptic and nocturnal fish species detection is possible using eDNA techniques (Bessey et al. 2020;Stat et al. 2019), yet its utility against more established methods remains unknown.Understanding the variance of these methods and how they compare to conventional surveys will help determine how these new methods can assist cryptobenthic fish biodiversity assessments to provide more complete descriptions of total fish biodiversity.
Our objective was to develop a scalable and reliable method for measuring and monitoring cryptic fauna in a coral reef setting using fish as a model.The Ningaloo Coast World Heritage Area was used as a case study because of its unique location as a tropical-temperate ecotone.We used a combination of conventional and emerging fish survey techniques to evaluate the benefits and limitations of each method.Five survey methods, including anesthetic collection (conventional method), eDNA collection through water filtration, sediment samples and passive collection (emerging method), as well as commonly used, standard underwater visual census (UVC) were evaluated for fish species diversity, cryptobenthic specific detection, and variance in method.

Study site and design
Sampling was conducted in shallow lagoon waters (< 3 m), adjacent to the backreef and within the Ningaloo Marine Park, Western Australia (Fig. 1a).Four sites were chosen based on habitat which consisted primarily of sand, rubble, dead coral, and a small proportion of macroalgae or live coral.This type of benthos is common at the study site and at other locations has been found to support a high diversity and abundance of cryptobenthic reef fishes (Depczynski and Bellwood 2004).Samples were obtained daily between 8 am and 4 pm from March 2 to 6, 2020, using five survey methods (Treatment): underwater visual surveys (UVC; Fig. 1b), fish collection using the anesthetic clove oil (Fig. 1c), collection of sediment samples for eDNA extraction (Fig. 1d), active pumping of water across a membrane for eDNA collection from surface and atdepth water samples (Fig. 1e), and deployment of membranes at the surface and at-depth (just above the reef) to passively collect eDNA (Fig. 1f).Each sampling method was repeated five (n = 5) times at each site.

UVC survey
UVC was used as the most widely deployed method in general for surveying fishes.Scuba divers used a 5-m radius point count (n = 5) at each site to record the presence of all fish species encountered (Fig. 1b).Larger bodied fish were recorded first before the diver swum around the survey area recording smaller bodied and sedentary species.Any fish that entered the area after surveys commenced were not recorded.

Cryptobenthic fish survey using anesthetic
Following the visual assessment for conspicuous fishes, small cryptobenthic fishes were collected from a small plot ($ 0.5 m 2 ) within the UVC survey area ensuring each plot was discrete and within similar habitat.We used a fine mesh tent net and anesthetized (clove oil) all fish within the tent following the methods of Depczynski and Bellwood 2004 (Fig. 1c).All fishes collected were euthanized and preserved (frozen at À 20 C) for identification and to enable a reference DNA sequence for each species.We classify fish as cryptobenthic fish if they are less than 50 mm in length, are visually and/or behaviorally cryptic and commonly associated with the benthos, or are in the following families: Apogonidae, Blenniidae, Gobiidae, Pseudochromidae, Tripterygidae.We also note fish that are often neglected in common fish survey monitoring, such as underwater visual and underwater video surveys (Table S1).

eDNA collection
Five replicate water samples were collected by divers from the surface and just above the benthos (2-5 cm) in sterile 1 L bottles at each site to assess fish presence using active eDNA filtration.Water samples were immediately filtered on-site, aboard a boat, using a vacuum pump (Smith-Root eDNA Sampler Backpack; Fig. 1e), where the water was poured through a sterile tube over a cellulose ester membrane (0.45 μm, 47 mm Pall GN-6 Metricel ® ).Sterile tubing and filter holders, as well as new membranes, were used for each individual sample.Each day, a field control, consisting of 500 mL of deionized water was filtered over a membrane in the same manner as all other water samples.Membranes were placed into individual zip-lock bags upon retrieval using sterile tweezers and immediately placed on ice.
At each site, passive deployment devices (Fig. 1f), each containing a cellulose ester membrane (0.45 μm, 47 mm Pall GN-6 Metricel ® ), were attached to a weighted line enabling eDNA collection at both the bottom and surface of the water column.Five weighted lines with collection devices at the bottom and surface were deployed from the boat, at least 50 m apart, at each site and retrieved after approximately 24 h.Upon retrieval, membranes were immediately placed into individual zip-lock bags and stored on ice.
Five replicate sediment samples were collected by divers at each site using a sterile Falcon ® tube (50 mL), which was opened underwater and used to scoop the top 5 mm of sediment ($ 30 g mainly sand, Fig. 1d) from the sea floor.The lid was used to secure the sediment inside the tube and prevent any further water and particles from entering.
All collection and deployment apparatus were sterilized by soaking in 10% bleach solution for at least 15 min and rinsed in deionized water.

eDNA extraction from membranes and sediment
All cellulose ester membranes were flash frozen (À 80 C) and crushed into small pieces that were placed in a 2 mL Eppendorf tube in preparation for extraction.Total nucleic acid was extracted from the membranes using a DNeasy Blood and Tissue Kit (Qiagen; Venlo, Netherlands), with an additional 40 μL of Proteinase K used during a three-hour digestion period at 56 C on rotation (300 rpm).DNA was eluted into 200 μL AE buffer.
All sediment samples were extracted using a DNeasy PowerSoil Kit (Qiagen; Venlo, Netherlands), where samples were thawed, and an $ 0.4 g subsample was weighted out and vortexed (bead beating with Solution C1) with a mill grinder (Retsch MM400, Dusseldorf, Germany) for 10 min.Our methods followed the manufacturer's standard protocol for this kit.
All extractions took place in a dedicated DNA extraction laboratory using a QIAcube (Qiagen; Venlo, Netherlands), where benches and equipment were routinely bleached and cleaned.

DNA metabarcode amplification for fish detection
Our amplification procedures have been previously detailed (Bessey et al. 2021).Briefly, one-step quantitative polymerase chain reactions (qPCR) were performed in duplicate for each sample using 2 μL of extracted DNA and a mitochondrial DNA 16S rDNA universal primer set targeting fish taxa (16SF/D 5 0 GACCCTATGGAGCTTTAGAC 3 0 and 16S2R-degenerate 5 0 CGCTGTTATCCCTADRGTAACT 3 0 ; Berry et al. 2017;Deagle et al. 2007), with the addition of fusion tag primers unique to each sample that included Illumina P5 and P7 adaptors.A single round of qPCR was performed in a dedicated PCR laboratory.Quantitative PCR reagents were combined in a dedicated clean room and included 5 μL AllTaq PCR Buffer (QIAGEN; Venlo, Netherlands), 0.5 μL AllTaq DNA Polymerase, 0.5 μL dNTPs (10 mM), 1.0 μL Ultra BSA (500 μg μL À1 ), SYBR Green I (10 units μL À1 ), 0.5 μL forward primer (20 μM) and 5.0 μL reverse primer (20 μM), 2 μL of DNA and Ultrapure™Distilled Water (Life Technologies) made up to 25 μL total volume.Mastermix was dispensed manually, and qPCR was performed on a CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad, California, USA) using the following conditions: initial denaturation at 95 C for 5 min, followed by 40 cycles of 30 s at 95 C, 30 s at the primer annealing temperature 54 C, and 45 s at 72 C, with a final extension for 10 min at 72 C. All duplicate qPCR products from the same subsample were combined prior to library pooling.A sequencing library was made by pooling amplicons into equimolar ratios based on qPCR C q values and sequenced on an Illumina Miseq platform (Illumina; San Diego, USA).The final libraries were size selected using a Pippin Prep (Sage Science, Beverly, USA) and purified using the Qiaquick PCR Purification Kit (Qiagen; Venlo, Netherlands).The volume of purified library added to the sequencing run was determined by quantifying the concentration (Murray et al. 2015) using a Qubit 4 fluorometer (ThermoFisher Scientific).The library was unidirectionally sequenced using a 300 cycle MiSeq ® V2 Reagent Kit and standard flow cell.

DNA metabarcoding sequence data processing
Our DNA sequence data processing has been previously detailed (Bessey et al. 2021).It is available online (https://data.csiro.au/collection/csiro:46025),directly follows the procedure described at https://pythonhosted.org/OBITools/wolves.html, and is briefly outlined here.Data generated by Illumina sequencing were processed using OBITools command "ngsfilter" to assign each sequence record to the corresponding sample based on tag and primer.Then "obiuniq" was used to dereplicate reads into unique sequences.Reads less than 190 bp and with fewer than 10 counts were discarded.Denoising was performed using "obiclean" to retain only sequences with no variants containing a count greater than 5% of their own and is a final step in cleaning the sequences for PCR and sequencing errors.Sequences were assigned to taxa using "ecotag" and a result table was generated using "obiannotate."Our reference database was built on 8 March 2021 using ecoPCR to extract all sequences from the EMBL (European Molecular Biology Laboratory) that could be amplified in silico by our universal fish primer assay (detailed in the "Build a reference database" section of https://pythonhosted.org/OBITools/wolves.html).Only fish species with identities ≥ 90% and whose sequence variants could be assigned to at least family (and lower) were included.Taxonomic assignments were designated to species for identities > 97%, to genus for identities < 97% but > 95%, and to family for identities < 95%.
We provide our assigned taxon name, identities, GenBank accession number and number of reads for each unique sequence, ensuring complete transparency in taxonomic assignment (Appendix I-Assigned Taxa).
Quality control measures-PCR controls, reference sequences, and field controls PCR controls PCR plates included blank laboratory extraction controls when reagents bottles were half full (extraction reagents used with no DNA template), as well as PCR negative (2 μL of DI water used rather than DNA template) and positive controls (West Australian Dhufish; Glaucosoma hebraicum) for each plate.All extraction controls had < 5 reads of any fish taxa, except one that had 411 reads of our positive dhufish control (Appendix II-PCR Controls).Similarly, all PCR negative controls had < 5 reads of any fish taxa, except one that had 16 reads (12 of a specific variant) of our positive dhufish control.As a conservative approach, we exclude all variants with ≤ 20 reads, and ultimately remove the positive dhufish control from all samples because it was only used to confirm our PCR amplification was successfully detecting fish.All positive PCR controls matched dhufish with 100% identify and all other fish species with ≥ 10 reads detected in a positive control were excluded from our analyses (e.g., Choerodon rubescens, Choerodon, and Sparidae).

Reference sequences
All fishes collected during the survey were identified.Each species was identified by unique morphological characteristics in accordance with published taxonomic literature and by comparison with vouchers accessioned at the Western Australian Museum.During morphological analyses, we identified species that currently do not have a 16S reference DNA sequence available on GenBank, or whose voucher specimen are from other countries or localities (Appendix III).We cut a tissue sample from specimens of these species for Sanger sequencing of a c. 570 bp fragment of the 16S gene, enabling us to obtain a reference sample for our metabarcoding analyses (Appendix III-Ref Consensus).The DNA of a small tissue sample from each specimen was extracted in the same fashion as our membrane samples (detailed above in eDNA extraction from membranes and sediment), of which a 20 μL sample ($ 10 ng μL À1 concentration) was sent to the Australian Genome Research Facility (AGRF) in Brisbane for sanger sequencing (16SarL; Forward Primer 5 0 -3 0 , CGCCTGTTTATCAAAAACAT and 16SbrH; Reverse Primer 5 0 -3 0 , CCGGTCTGAACTCAGATCACGT; annealing temperature 55 C; Palumbi 1996).Consensus sequences (the most common nucleotides at each position using both forward and reverse sequences) for each sample were constructed in Geneious Prime 2021.2.2, where the consensus threshold was set to "Highest Quality (60%)," assign quality was set to "Highest," and call sanger heterozygotes > was set to "50%." We then matched the DNA metabarcoding results to that of our reference consensus sequences to ensure these species were included in our data set.In all cases, the DNA metabarcoding results showed a 100% identity match to the variant with the most abundant read count (Appendix IV-Reference Samples; indicated by a 1 in Consensus Match).The identically matching segment of the consensus sequence is highlighted in red lower-case text in Appendix III.

Field controls
Minimal field contamination was detected in filtered deionized water samples (day 1 = 4 reads, day 2 = 9 reads, day 3 = 608 reads, day 5 = 4 reads) except on day 4 (Appendix V-Field Controls), when 4280 reads of fish were detected; predominately G. cauerensis (4259 reads).As this fish was the main species collected every day during surveys and detected in the booty water of all divers (Appendix VI-Dive Booty Water), we could not reasonably exclude it from the data.Therefore, as a conservative approach, we include only samples that contained > 4500 reads in total because that exceeds the reads of the highest negative field sample.A total of 4,958,971 sequence reads passed our quality control measures which consisted of 2645 variants of fish (Appendix I-Assigned Taxa), with an average read count per sample of 39,377 (min = 4857 and max = 155,593).
We used a permutation MANOVA (adonis) to examine spatial differences in eDNA collected from surface and bottoms (depth) waters between survey methods (Treatment; excluding underwater visual survey and sediment because they did not detect adequate cryptic species; see Results).Treatment and depth were fixed factors, while site was random (distance = "jaccard" and permutations =10,000 which ignores joint absences and focusses on proportion of shared species).We also conducted individual permutational multivariate analyses of variance on each survey method by site to determine if all treatments were able to distinguish spatial differences in fish assemblages among the four sites.A pairwise comparison was used to explore significant effects within the general model.We also determined the mean number of cryptobenthic species detected per treatment and site, where the standard deviation was calculated using sample replicates and the coefficient of variation was calculated as the standard deviation divided by the mean and multiplied by 100 and expressed as a percentage to describe the dispersion around the mean.We used metaMDS to produce a visual representation of the similarities of fish communities detected between eDNA survey methods, where we removed any rare species detected in fewer than five samples from the analyses.
Cryptobenthic specific collection methods using anesthetic identified 13 species of which three were not considered cryptic (28% of all cryptic species), and four were unique to this method.The most frequently collected fishes by anesthetic were G. cauerensis (18/20 replicates) and E. prasina (4/20), with all other species being collected during two or less replicates (Table S1).The most frequently detected cryptobenthic species by active eDNA methods were G. cauerensis (16/20) and Gobiodon sp. or unidentified species from the family Blennidae (13/20) in bottom waters, compared to Gobiodon axillaris (18/20) and Cirripectes sp.(18/20) in surface waters.Species most detected by passive eDNA methods were E. bimaculata/distigma Fig. 2. Venn diagram showing how many fish taxa were detected by each survey method for all fish species, as well as for cryptic fish species only.
(4/20) and G. axillaris (3/20) in bottom waters, compared to G. axillaris (9/20) and Cirripectes sp.(6/20) in surface waters.As hypothesized, UVC sampling was not effective to survey cryptobenthic fishes, with only one species (G.cauerensis) being observed and this method was therefore excluded from statistical comparisons.Sediment survey methods were also excluded from statistical analyses as they likewise only detected one cryptobenthic species (G.cauerensis; Fig. 2; right).
Both anesthetic collection of species and passive eDNA collection identified approximately two species of cryptobenthic fishes per site (Fig. 3; raw data provided in Appendix VI).However, active eDNA detection ranged from 6 to 12 species, depending on the site.The level of dispersion around the mean number of cryptic species observed was generally less for active eDNA filtration (< 40%) as compared to both passive eDNA and anesthetic methods (> 40%).All survey methods  were able to distinguish site differences (Table 1), except for passive eDNA collection from bottom water.
Cryptobenthic fish assemblage comparisons Cryptobenthic fish assemblages detected by eDNA methods differed significantly by survey method (treatment: active vs. passive), depth (surface vs. bottom) and location (site; Table 1).Active eDNA assemblages differed between surface and bottom water collections (Fig. 4a, Table 2), as well as from passively collected samples.Passive samples were characterized by a large amount of multivariate dispersion compared to actively collected samples.The spatial separation in species assemblages was mainly driven by Cirripectes sp., G. axillaris, Gobiodon citrinus, and Salarias fasciatus being more associated with surface water collections, while G. cauerensis was more associated with water collected just above the benthos (Fig. 4b).

Discussion
Our study revealed 70 fish species out of the 210 detected that could be classified as either cryptobenthic (32) or neglected by commonly deployed survey methods (38).Surveys along the Kimberley coast of Western Australia likewise found a similar number of fish species would be considered cryptobenthic or often overlooked by conventional methods (Moore et al. 2020), suggesting that cryptobenthic species make up approximately a third of the fish species detected in North Western Australia.Our estimate is less than the 50% predicted given the number of described cryptobenthic fishes relative to other species (Brandl et al. 2018), and the proportion of cryptobenthic fishes collected from similar habitats on the Great Barrier Reef (Depczynski and Bellwood 2004), yet they remain a significant portion of the fish diversity overall.The relatively low proportion of cryptobenthic fishes in our study may relate to surveys being limited to backreef/lagoon habitats.Large mobile fish will traverse multiple reef microhabitats, while many cryptobenthic fishes are likely to associate with specific microhabitats (Depczynski and Bellwood 2004;Ahmadia et al. 2012a,b) or reef zones (Wilson 2001;Goatley et al. 2016).Thus, increased sampling across  different habitats may increase the representation of cryptobenthic species within the Ningaloo fish assemblage.

Effectiveness of survey methods for cryptobenthic fish
We found active eDNA filtration of water samples detected the most fish species (193), of which 59 (31%) were cryptobenthic (22) or not typically surveyed (33).Conversely eDNA from sediment samples detected very few fish species, possibly because of the small volume of sediment used when following the QIAGEN DNeasy PowerSoil standard protocol for sampling soils.Active eDNA filtration generally had the least dispersion around the mean number of cryptobenthic species detected, which could enable better power to detect change in assemblage composition over time.Similarly, passive collection of eDNA from surface waters and anesthetic collections also detected distinct fish assemblages among the four survey sites, although there were differences in species composition among survey methods.Anesthetic methods observed 13 cryptobenthic species, of which four were unique to this method and would otherwise not have been detected.Our anesthetic method was also microhabitat specific to rubble habitats and although time consuming, surveying multiple microhabitats could reveal different cryptic species.Indeed, a full assessment of cryptobenthic fish in Ningaloo reef requires surveys of microhabitats other than rubble, including coral and algae dominated areas.That there were some species only found in anesthetic surveys indicates that not all cryptobenthic fish present at a site were detected by eDNA methods.We also note that anesthetic methods can vary with respect to the sample area and the type of anesthetic or ichthyocide applied (Ackerman andBellwood, 2000, 2002), which may increase the number and diversity of fish collected using these methods.Furthermore, fish-specific autonomous reef monitoring structures (FARMS) can attract cryptobenthic fish species present on local coral reefs (Brandl et al. 2023), facilitating standardized, efficient collections.
Both collection and UVC methods provide a quantitative community density assessment and size estimates of individual fish which is currently unachievable with eDNA methods.However, eDNA methods provide a broader understanding of both cryptic and non-cryptic fish presence within an area.For example, active eDNA methods detected 21 cryptobenthic fishes that were not detected during anesthetic surveys, including species of Gobiodon which are obligate coral dwellers (Munday 2000) and unlikely to have been collected from anesthetic sampling of rubble habitats.Our passive eDNA collection method detected a similar proportion (31%) of cryptic or typically unsurveyed fishes (15 and 18, respectively), but only approximately half (56%) of the fish species detected overall by active filtration (108 compared to 193).It is possible that the long submersion time (24 h) of passive collectors led to DNA degradation in our study, as shorter sampling periods can detect higher richness in tropical waters (Bessey et al. 2021).Eliminating active filtration undoubtedly makes the eDNA methods more efficient, increasing deployment capacity and scalability, however further studies are required to optimize deployment times, which may vary according to different environmental sea conditions.eDNA as a viable survey method for cryptobenthic fish Active filtration of eDNA enabled detection of cryptobenthic species and provided the most repeatable characterizations of species assemblages.eDNA samples can also be stored for extensive periods of time allowing for future reanalysis.These three factors contribute to it being an effective biodiversity monitoring survey tool (Lindenmayer et al. 2020).However, for eDNA methods to be a viable survey option for cryptobenthic fishes, an adequate voucher collection of these species with expertly verified identifications is required to build a globally accessible DNA reference database (Berry et al. 2021).Furthermore, ongoing taxonomic work in these cryptobenthic fishes is essential to establish species boundaries and provide a framework for a robust DNA library.Our study revealed 12 cryptic fish species that would not have been detected without collection for DNA reference material.Even with DNA from the 12 species collected during our study, we estimate that of the six cryptobenthic families we recorded during surveys, only 12/62 (19%) genera and 32/143 (22%) species expected in North Western Australia (Allen and Swainston 1988) were detected using eDNA (Table S2).To address the deficiency in reference DNA material, a combination of both collection and molecular genetic methods will be required to see the fruition of eDNA methods as a robust survey tool for assessing diversity of these fishes.The combination of well-preserved reference samples for morphological identification, along with DNA sequencing to create reference sequences, for all cryptobenthic fishes, including rare species, will ultimately lead to more informative non-lethal collection data in the future.Indeed, there is currently a nationwide need and effort in Australia and elsewhere to barcode all species (Weigand et al. 2019 and https://research.csiro.au/dnalibrary/).There are also initiatives to quantify the existing species gaps in publicly available genetic reference databases for multiple metabarcoding primers (Marques et al. 2021).Nevertheless, these initiatives are likewise limited by the need for species checklists compiled at appropriate spatial scales to assess how effective eDNA has been in detecting species.
An unexpected outcome of our active and passive eDNA methods was the ability to detect species differences between surface and bottom waters which were only three meters apart.Consistent with information on habitat association for cryptobenthic fishes, the surface waters were characterized by species that are typically found on reef tops or among the branches of coral colonies (Munday 2000;Wilson 2001;Depczynski and Bellwood 2004).The combination of surface water contamination and emerging evidence that passive eDNA collection happens rapidly (in the first 5 min; Bessey et al. 2021) may explain why no spatial differences in assemblage structure were detected from samples collected from deeper water when using the passive technique.However, other eDNA studies have shown that communities far from one another tend to be less similar than those nearby (O'Donnell et al. 2017;Jo et al. 2019), with a recent study indicating that vertical zonation patterns in species composition may be present on smaller spatial scales (e.g., 4 m apart) than horizontal community structure (Jeunen et al. 2020).These findings imply that sampling strategies will need to incorporate more than just surface water collections for eDNA methods if a full understanding of the community is to be obtained, especially in areas where permanent water column stratification exists.

Conclusions
With scale and practicality in mind, we demonstrate how environmental DNA methods deployed in a stratified sampling design can yield a comprehensive assessment of the diversity of cryptobenthic and non-cryptic fishes on coral reefs.Collection of samples for eDNA analysis is also rapid, compared to current alternatives (e.g., extractive sampling via anesthetic), especially when using the passive method, but appears to be quite sensitive to vertical zonation in the water column.Nevertheless, collection by anesthetic provides a quantitative assessment which may be informative for studies that investigate shifts in community composition, length frequency, biomass, density, or metabolic rates (Depczynski and Bellwood 2003;Ackerman et al. 2004).Quantitative assessments of species across an entire assemblage are currently unachievable with eDNA, though possible for some species (Tillotson et al. 2018) and may be developed in the future.Moreover, fish collection facilitates additional biological studies that can provide information on key ecological processes (Depczynski et al. 2007;Brandl et al. 2019;Morais et al. 2020) and are essential for establishing a framework for a robust DNA library.Ultimately, the collection of voucher samples and DNA sequencing for reference of all species will be necessary to achieve non-destructive survey methods for the future.Environmental DNA provides an effective way of surveying species presence with minimal effort in the field, especially when using passive collectors, and are well suited to biodiversity studies.However, until the development of more quantitative eDNA methodologies that can reliably relate to fish biomass and/or densities are established, cryptobenthic fish surveys should reflect specific study objectives, which may go beyond the requirements of estimating species diversity.

Data Availability Statement
Raw sequences of all assigned taxa, consensus sequences, and statistical R script are provided in the supplementary materials and at https://data.csiro.au/collections/collection/CIcsiro:46025v1 for the bioinformatic script and reference database.

Fig. 1 .
Fig. 1.(a) Study site for fish surveys conducted using (b) underwater visual census, (c) conventional anesthetic collection of whole cryptobenthic species caught with a circular net, (d) sediment samples, (e) active eDNA filtration, and (f) passive eDNA collection.

Fig. 3 .
Fig. 3.The mean number of cryptobenthic fish species detected per site by each survey method (n = 5).Error bars represent standard deviation, and the coefficient of variation is indicated as a percentage above each treatment.

Fig. 4 .
Fig. 4. (a) Nonmetric multidimensional scaling plot of the autotransformed read count data (distance = jaccard) of cryptic fish species by survey method (active = blue circle, passive = pink square) and depth (surface = solid shape, bottom = open shape) and (b) showing species dissimilarities by water depth.

Table 1 .
Permutational multivariate ANOVA results (permutations = 10,000).The main test is presented at the top (All), with strong site effects explored below for each of the different methods.

Table 2 .
Pairwise comparison results of the permutational multivariate ANOVA to determine which methods differ significantly for cryptobenthic fishes.
Note: The * indicates the p-adjusted value is statistically significant at alpha = 0.05.