Methods

Site Selection

Lake Way is a large ephemeral salt lake in central Western Australia that receives enough rainfall to fill approximately once every decade (Emerge 2020). Lake Way was selected because of its well documented macroinvertebrate fauna due to ongoing environmental monitoring targeted at assessing the impact of mining operations on the lake (Bennelongia Environmental Consultants 2020).

Sediments for Experiment 1 were collected from six sites in November 2023, while sediment for Experiment 2 was collected from four sites in November 2024 (Table S1; Figure 1). Sediment collection involved scraping the top 10 cm of sediment in a 0.5 m2 area, which was stored in a calico bag and transported to the laboratory where it was homogenised via sieving (mesh size = 500 μm) and kept at room temperature for 6 months until analysis (as per Campagna 2007). Collection sites were selected based on the evidence of a previous lake highwater mark with the intention of maximising the chances of collecting dormant resting eggs (see Rahman, Chaplin, Lawrie, and Pinder 2025).

FIGURE 1: Lake Way, a large ephemeral salt lake in central Western Australia. Sites where sediment samples were collected are coloured by their inclusion in either or both of the two experiments. Purple and red dots = Experiment 1 and red and green dots = Experiment 2 sites.

Experimental Design

All sample processing and extraction was undertaken in the Trace and Environmental DNA (TrEnD) laboratories, Curtin University, Western Australia, which are dedicated to working with samples of low DNA concentration.

Experiment 1

At each of the six sites (Figure 1), a homogenised sediment sample was subdivided into six biological replicates and subjected to two different pre‐DNA extraction procedures. Three biological replicates per site were extracted following the standard method (n = 18) and the remaining three were subjected to the sugar‐flotation method (n = 18).

The standard DNA extraction method comprised processing 250 mg of sediment (the maximum input specified by the manufacturer) using the DNeasy PowerSoil Pro Kit (Qiagen) as per the manufacturer's instructions. Sugar‐flotation permits a substantially larger sediment mass to be represented in one extraction than a column‐based kit can accept directly, and therefore each biological replicate comprised 30 g of sediment in 50 mL Falcon tubes that were then topped up with 20 mL of a sucrose solution (1:1 sugar and ddH2O as per Onbé 1978). Falcon tubes were then hand shaken vigorously for 20 s and centrifuged at 3667 rcf for 3 min. The sucrose solution supernatant was then poured into a sterilised collection cup of a peristaltic Pall Sentino Microbiology pump (Pall Corporation), washed with ddH2O and filtered through a Pall 0.45 μm GN‐6 Metricel mixed cellulose ester membrane. The presence of resting eggs from ostracods, cladocerans and anostracans on the filter membrane was confirmed via microscopy in preliminary trials. The filter membrane was then processed using the DNeasy PowerSoil Pro Kit as per the manufacturer's instructions. All materials were decontaminated between samples using a 10% bleach solution, and two blank extraction controls (reagents only) were processed through to sequencing to assess potential cross contamination between samples. Extractions were undertaken in a dedicated PCR‐free laboratory. The metabarcoding and bioinformatics methods were the same as presented in Experiment 2 (see below).

Experiment 2

Additional sediment samples were collected in November 2024 from three sites sampled for Experiment 1 (Site 2, Site 3 and Site 5) and an additional site (Site 7). Site selection for Experiment 2 was constrained by a partial flooding event in 2024 with the selected sites representing unflooded areas. Sediment from each site was added into pre‐decontaminated plastic containers (10% bleach) in four weight classes (50, 250, 500 and 1000 g) with three replicates per weight class. A 1:1.5 ratio of sediment to sucrose solution (1:1 sugar to ddH2O as above) was added to each sample container and mixed thoroughly with a decontaminated stirring stick for approximately 1 min per sample. Large sample volumes prevented centrifugation as per Experiment 1 but to minimise filter paper clogging with fine suspended solids, samples were left to settle for approximately 4 h prior to filtering. After settling, the sucrose solution was filtered through a 63 μm sieve and washed with ddH2O into a sterile collection vessel and filtered through a Pall 0.45 μm GN‐6 Metricel mixed cellulose ester membrane (47 mm) using a peristaltic Pall Sentino Microbiology pump (Pall Corporation). The filter paper was then added to a DNeasy PowerSoil Pro Kit zirconium bead tube and processed as per the manufacturer's instructions. Extractions were undertaken in a dedicated PCR‐free laboratory, and with two DNA extraction controls that were processed alongside the samples through to sequencing.

Metabarcoding

The DNA extracts from both experiments were quantified using a Qubit 4.0 fluorometer (Life Technologies) and then qPCR amplified using the fwhF2/fwhR2n primer set (F: 5′‐GGDACWGGWTGAACWGTWTAYCCHCC‐3′; R: 5′‐GTRATWGCHCCDGCTARWACWGG‐3′), hereafter referred to as fwh, designed for the amplification of approximately ~205 bp of the cytochrome oxidase subunit 1 (COI) (Vamos et al. 2017) and Crust16S primer set (F: 5′ GGGACGATAAGACCCTATA 3′; R: 5′ ATTACGCTGTTATCCCTAAAG 3′), designed for the amplification of approximately ~170 bp of 16S (Berry et al. 2017). These assays were selected because they have been designed for the amplification of aquatic macroinvertebrates (fwh) including crustaceans (Crust16S). Prior to metabarcoding, the optimal DNA dilution level (neat, 1:10 or 1:100 dilution) for each sample was determined via qPCR. The qPCRs consisted of a 25 μL reaction comprised of: 1× PCR Gold Buffer (Applied Biosystems), 0.25 mM dNTP mix (Astral Scientific, Australia), 2 mM MgCl2 (Applied Biosystems), 1 U AmpliTaq Gold DNA polymerase (Applied Biosystems), 0.4 mg/mL bovine serum albumin (Fisher Biotec), 0.4 μM forward and reverse primers, 0.6 μL of a 1:10,000 solution of SYBR Green dye (Life Technologies), 2 μL template DNA and PCR H2O up to 25 μL. qPCRs were performed on StepOne Plus instruments (Applied Biosystems) with the following cycling conditions for fwh: 95°C for 5 min, followed by 50 cycles of: 95°C for 30 s, 50°C for 30 s, 72°C for 45 s, then a melt‐curve analysis of: 95°C for 15 s, 60°C for 1 min, 95°C for 15 s, finishing with a final extension stage at 72°C for 10 min. The cycling conditions for Crust16S were the same except that the annealing temperature was 51°C.

The DNA metabarcoding was performed using a one‐step qPCR with fusion primers and the same reagents and cycling conditions as described above but all fusion‐tagged qPCR reactions were prepared in dedicated clean room facilities using an automated QIAgility robotics platform (Qiagen). Fusion tagged qPCRs included at least one non‐template control (reagents only; NTC) and a positive control (Libellulidae, Tramea limbata) included in every run with controls processed alongside samples through to sequencing for both fwh (extraction controls = 4, NTC = 3, positive controls = 3) and Crust16S (extraction controls = 4, NTC = 2, positive controls = 1). Each biological replicate was repeated in two PCR technical replicates to account for detection sensitivity. Each technical replicate contained a unique combination of two 8 bp multiplex identifier‐tags (MID‐tag), one on each of the forward and reverse primers as described in Koziol et al. (2019) and van der Heyde et al. (2020).

Sequencing libraries were prepared separately for the fwh and Crust16S assays, by combining samples into mini‐pools based on the qPCR amplification results from each sample. Samples were assigned to mini‐pools according to comparable ΔRn fluorescence endpoints, with ΔRn values compared only among samples amplified within the same qPCR run. The mini‐pools were then quantified for DNA concentration (ng/μL) using a QIAxcel fragment analyser (Qiagen) and combined at approximate equimolar concentrations to form the sequencing libraries. Amplicons in each library were then size selected to 150–400 bp (Pippin Prep: Sage Sciences) using a 2% Agarose Gel Cassette (Sage Science), cleaned using a QIAquick PCR purification kit (Qiagen), quantified by Qubit 4.0 Fluorometer (Thermo Fisher), and diluted to 2 nM. All libraries were sequenced on an Illumina NextSeq 2000 sequencing platform following the manufacturers protocols, utilising a NextSeq1000/2000 P1 Reagents 300 cycle kits, with a final library molarity of 650 pM containing 24% PhiX.

Bioinformatics

Sequence data was processed using the automated workflow ‘eDNAFlow’ (Mousavi‐Derazmahalleh et al. 2021) on the Setonix cluster at the Pawsey Supercomputing Centre (Pawsey Supercomputing Research Centre 2023), with parameters ‐‐maxTarSeq '10', ‐‐minLen '100', ‐‐minsize '8', ‐‐qcov '100', ‐‐perc_identity '85' for COI and '90' for 16S, and ‐‐minMatch_lulu '97'. The ZOTUs (zero‐radius operational taxonomic units) were compared using BLASTN against the NCBI (GenBank) and an in‐house custom reference database generated from known taxa collected from previous rehydration trials and active sampling of the seven studied sites (see Supplementary Table 2 for GenBank accession numbers). A minimum pairwise identity threshold of 85% was applied for COI and 90% for 16S. When two or more taxa matched a ZOTU and their percentage identities differed by ≤ 0.5%, the assignment was reduced to the lowest taxonomic rank shared by those taxa. All taxonomic assignments were assessed manually, and species‐level attributions were limited to matches > 92% for COI and > 97% for 16S.

The curated ZOTU table from lulu (Frøslev et al. 2017) was filtered using a multi‐step decontamination process using metabar (Zinger et al. 2021) in R Studio (R Core Team 2024). ZOTUs that were most abundant in at least one control were considered contaminants and removed using the contaslayer function, while ZOTUs occurring at a relative abundance below 0.5% within an individual PCR technical replicate were considered potential tag‐jumps and removed with the tagjumpslayer. PCR technical replicates were aggregated by biological replicate with taxa in at least one biological replicate at a site recorded as a positive detection.

Bioinformatic outputs were manually verified and automated taxonomic assignments refined for each primer dataset as follows: Non‐arthropod sequences were removed (e.g., rotifer, fungal, or diatom ZOTUs).ZOTUs assigned to terrestrial arthropod species were removed (e.g., Arachnida or Blattodea).Filtered ZOTUs sequences of uncommon, broadly assigned or not previously recorded taxa at a site were manually verified and reassigned accordingly.

Rehydration Trials

The same sediment samples from the four sites in Experiment 2 (Site 2, Site 3, Site 5 and Site 7) were used in the rehydration trials following the methods of Campagna (2007). Specifically, 500 g of sediment was placed into a 20 L glass aquarium with approximately 18 L of reverse‐osmosis water, which was aerated using an airstone and maintained under a 12‐h light/dark cycle in a temperature‐controlled room set to 25°C. Temperature, electrical conductivity, pH and percentage saturation of dissolved oxygen were recorded three times per week using a multiparameter water quality meter (Table S7). Aquaria were not topped up with water after the initial flooding event with water allowed to gradually evaporate over the course of the experiment. Aquaria were monitored for 28 days; then any remaining water was removed, and the sediments were dried using heat lamps. Dry sediments were subjected to a second filling cycle. This process is necessary to assess for the presence of ‘bet hedging’ taxa within the egg bank which may not respond to the initial filling event (Pinceel et al. 2021). All emergent animals were harvested at the end of each filling cycle by pouring the water from each aquarium through a 53 μm sieve. All harvested taxa were identified to the lowest possible taxonomic level.

Data Analysis

Data for each assay were pooled into separate presence/absence datasets per experiment with species detected by both assays counted only once. To examine how sediment weight influenced the detection of taxa in Experiment 2, species richness was analysed using linear mixed‐effects models in R Studio (R Core Team 2024) using the lme4 package (Bates et al. 2015). Sediment weight was first modelled as a continuous variable to test for an overall linear relationship between sediment weight and species richness, while including site as a random intercept to account for non‐independence among samples from the same location. Because sediment weights represented discrete treatment levels (50, 250, 500 and 1000 g), a second model was fitted with sediment weight as a categorical factor to enable post hoc pairwise comparisons among weight levels. Tukey‐adjusted comparisons of estimated marginal means were performed using the emmeans package. Model assumptions were assessed by inspection of residual against fitted plots and homogeneity of residual variance among weight classes was tested using Levene's test. No violations of model assumptions were detected. The effect of sediment weight, site, and their interaction on community composition was analysed using permutational multivariate analysis of variance (PERMANOVA) based on Bray‐Curtis dissimilarities (999 permutations) using the vegan package (Dixon 2003).

To test whether increasing sediment weight reduced among‐replicate variability in community composition within each weight class, we quantified multivariate dispersion using PERMDISP (betadisper function in the vegan package). Bray‐Curtis dissimilarities were calculated from presence‐absence community matrices. Because sites differed strongly in community composition (see below), PERMDISP was run separately for each site, with sediment weight (50, 250, 500 1000 g) as the grouping factor. This approach was taken to ensure that differences reflected variability between weight classes rather than inherent community differences among sites.

For the comparison between metabarcoding and rehydration trials, we limited the metabarcoding data to the 1000 g treatment of Experiment 2 with replicates pooled to the site level. This treatment was selected because it was the most diverse of the four sediment treatments analysed (see Section 3). To account for differences in taxonomic resolution between the metabarcoding and the morphological analyses, all species‐level identifications were brought up to genus prior to comparison. A Kruskal–Wallis test was used to determine whether median taxa richness differed significantly between the two sampling methods.