At Palmyra Atoll, the environmental DNA (eDNA) signal on tidal sand flats was associated with fish biomass density and captured 98%–100% of the expected species diversity there. Although eDNA spilled over across habitats, species associated with reef habitat contributed more eDNA to reef sites than to sand-flat sites, and species associated with sand-flat habitat contributed more eDNA to sand-flat sites than to reef sites. Tides did not disrupt the sand-flat habitat signal. At least 25 samples give a coverage >97.5% at this diverse, tropical, marine system.

In August 2017, water samples were taken from 22 sites (Figure 1; Supporting Information Table SS1) by filling sterile Nalgene jars with surface water while wearing nitrile-free gloves. Depth and tide direction were noted for the intertidal sites (though, regrettably, temperature was not recorded). Five sites on submerged reef (2 on the fore reef, and 3 on the inner reef, 1 sample at each reef site corresponding to 2 pooled 250-ml samples taken within 10 m), 1 site in the deep-water lagoon (which is ignored here) and 16 intertidal sand-flat sites (1 sample of 250-ml per site, because higher turbidity limited the volume that could be passed through a filter) were sampled. Unfortunately, the difference in sample volumes taken from the two habitats means that when comparing reef sites to sand-flat sites, the relative proportion (rather than total counts) of fish species per sample associated with reef habitat or sand-flat habitat had to be analysed. All 16 sand-flat sites on outgoing tides and 7 of these sites again on incoming tides were sampled. The 29 samples put on ice were transported to the Palmyra Atoll lab, where they were filtered within 2 h. Because of the remote location and lack of vacuum pumps, the samples were pushed through a 0.22-um Sterivex™ filter unit, using a sterile 50-ml syringe. All filters were preserved using Longmire's solution (Longmire et al., 1997; Renshaw et al., 2015) and kept refrigerated until extraction in February 2019 (the authors did not collect fish, perform surgical procedures, stress fish in experiments, cause lasting harm to sentient fishes or involve sentient un-anaesthetised animals).

DNA extractions and PCR preparations were performed in a pre-PCR room, separate from all areas where post-PCR products are used. For each filter, DNA was extracted separately from both the individual filter capsules and the Longmire's preservation solution within the capsules, resulting in 58 samples (e.g., two extractions for each of the 29 samples). After lysis buffer (720 μl buffer ATL for capsule and 180 μl for the solution extraction; Qiagen Inc., Hilden Germany) and proteinase K (80 μl and 20 μl, respectively) was added, the samples were incubated at 56°C for 24 h on a rotating incubator. A modified DNeasy™ Blood & Tissue protocol (following Spens et al., 2017) was used to extract the eDNA before amplification. Each set of 29 extractions had its corresponding extraction blank: a new Sterivex™ filter capsule with 720 μl buffer ATL and 80 μl proteinase K was used for the filter set, and a sterile microcentrifuge tube with 180 μl of ATL and 20 μl of Proteinase K was used for the Longmire's solution set.

DNA was PCR amplified using the following primer sets with an Illumina Nextera™ adaptor modification: MiFish 12S Universal and MiFish 12S Elasmobranch (Miya et al., 2015), 18S (V8-V9; Bradley et al., 2016) and CO1 (mlCOIintF and jgHCO2198; Leray et al., 2013). PCR amplification was carried out in triplicate using a 15- μl reaction volume containing 3 μl of extract, 7.5 μl of QIAGEN Multiplex Taq PCR 2x Master Mix, 4.20 μl of dH2O and 0.15 μl of each primer (10 μmol l^–1). Each set of 29 extractions had its corresponding PCR blank. For these, dH₂O was used instead of DNA extract. Although CO1, 18S, MiFish 12S Universal and MiFish 12S Elasmobranch were sequenced, only the 12S data were analysed here, because these barcodes had the best coverage for fishes (CO1 detected 48 species, and 18S detected 3); nevertheless, the other barcodes give insight into different taxa at Palmyra Atoll.

Thermocycler settings for 12S reactions included an initial denaturation at 95°C for 15 min, followed by 13 touchdown cycles of (a) 94°C for 30 s; (b) annealing beginning 69.5°C for 45 s, decreasing by 1.5°C until 50°C was reached; and (c) extension at 72°C for 45 s. Subsequently, 35 additional cycles were run with a 50°C annealing temperature, followed by a 10-min final extension at 72°C. PCR replicates were pooled after verifying successful amplification. Then the PCR products were cleaned using Serapure magnetic beads (Faircloth & Glenn, 2014) to remove <100 bp fragments. Cleaned samples were quantified using the Qubit dsDNA BR Assay (Thermofisher Scientific, Waltham, MA, USA). Even numbers of molecules were pooled from each primer set for indexing. Dual i7 and i5 Illumina Nextera indices (Illumina, San Diego, CA, USA) were incorporated by amplification using Kapa HiFi HotStart ReadyMix (Kapa Biosystems, Wilmington, MA, USA) in 25-μl reactions of 12.5-μl mix, 0.625 μl of each index and sample volumes amounting to 10 ng of DNA for each sample. The thermocycler setting began with a 95°C incubation for 5 min, followed by 12 cycles of 98°C for 20 s, 56°C for 30 s and 72°C for 3 min, and ended with a 72°C incubation for 5 min. Indexed PCR products were again bead-cleaned and quantified. DNA libraries were pooled evenly and submitted for sequencing to the Technology Center for Genomics & Bioinformatics (TCGB) at the University of California, Los Angeles, where they were run on an Illumina MiSeq™ with Reagent Kit V3 (2 × 300 bp) at a goal depth of 40,000 reads in each direction per marker per sample.