Long distance (>20 km) downstream detection of endangered stream frogs suggests an important role for eDNA in surveying for remnant amphibian populations

Study system and aims

The Armoured Mistfrog (Litoria lorica) is a Wet Tropics species that was considered to be extinct after severe chytridiomycosis-related declines in the early 1990s (Cunningham, 2002; Puschendorf et al., 2011). However, it was rediscovered as a single, small population in 2008 during research on populations of a co-occurring Endangered species, the Waterfall Frog (Litoria nannotis) (Puschendorf et al., 2011). These two frogs differ in size (L. lorica approximately 37 mm; L. nannotis approximately 55 mm long) but have near-identical ecologies, foraging side-by-side at night in the splash zone of waterfalls and cascades, and hiding by day in rock cracks in the flowing water (Puschendorf et al., 2011). The eggs and tadpoles of both species are also restricted to the same fast flowing sections, with the tadpoles of both species having suctorial mouth discs (Anstis, 2013; Puschendorf et al., 2011; C. Hoskin, 2021, unpublished data). Therefore, both species are semi-aquatic inhabitants of a very specific habitat—waterfalls and cascades.

The rediscovery of L. lorica triggered on-ground surveys of most potential habitat in the region, including surveying streams throughout the large catchment where L. lorica was rediscovered (Hoskin & Puschendorf, 2014; Hoskin & Puschndorf, 2021, in preparation) (Fig. 1). These surveys found populations of L. nannotis in all sections of suitable habitat but did not find any additional populations of L. lorica. Permission was obtained for a trial reintroduction of L. lorica, which involved translocating adults during three consecutive years (2013–2015) to establish a population in a discrete area of suitable habitat approximately 4 km upstream of the rediscovered population (Hoskin & Puschendorf, 2014; Hoskin & Puschendorf, 2021, in preparation) (Fig. 1). Based on regular monitoring of these two populations over the last five years, the rediscovered population is estimated to consist of approximately 1,000 frogs (plus an unknown number of aquatic tadpoles at any one time) along a 4 km stretch of stream, and the reintroduced population is estimated at about 100 frogs (and a small number of tadpoles at any one time) along an approximately 1 km section of stream (Hoskin & Puschendorf, 2014; Hoskin & Puschendorf, 2021, in preparation) (Fig. 1).

Here we aimed to test far-downstream distance eDNA detection of L. lorica and L. nannotis in order to determine the utility of eDNA as a method for surveying endangered stream frogs. We used three eDNA sampling methods (15 and 100 mL water volumes for precipitation, and large volume filtering of >1,000 L) to determine downstream detection at a series of sites kilometers downstream of the resolved lower limits of L. lorica and L. nannotis. This system is ideal for testing downstream detection distances for the following reasons. First, adults of the two species have daily contact with water, so eDNA can enter the stream at all times regardless of the presence of tadpoles. Second, the catchment containing L. lorica has been thoroughly surveyed (i.e., on foot, using head torches at night), including all areas of potentially suitable habitat for these two species (Hoskin & Puschendorf, 2014; Hoskin & Puschendorf, 2021, in preparation). These surveys have resolved downstream limits for the two species on all sections of streams (Fig. 1; details below). Third, L. nannotis is present and common in discrete upstream sections of all streams, whereas L. lorica is restricted to two sections—the small main population (approximately 1,000 frogs) and the very small reintroduced population (approximately 100 adults). We were therefore able to test eDNA detection at accurately calculated downstream distances from both species, and from both populations of L. lorica. We use these results to outline the feasibility of eDNA for detecting small upstream populations of stream frogs.

Summary of traditional field surveys

The downstream limits for L. lorica and L. nannotis have been determined in detail in this catchment. Field surveys have been conducted in all areas of suitable habitat, specifically targeting L. lorica, L. nannotis and other ‘missing’ and endangered stream frogs (Hoskin & Puschendorf, 2014; Puschendorf et al., 2011; C. Hoskin, 2021, unpublished data). In particular, the objective of the surveys was to locate populations of stream frogs in the lower sections of each stream, based on the hypothesis that these areas are refuges from chytrid disease impacts due to warmer conditions associated with lower elevation and open canopy woodland (Hoskin & Puschendorf, 2014; Puschendorf et al., 2011). Some of these areas have been surveyed repeatedly over the last decade to confirm presence/absence at lower and upper range limits, and monitor numbers along key stream sections (Hoskin & Puschendorf, 2021, in preparation). Litoria lorica and L. nannotis require steep, rocky, permanently flowing streams. Adults of both species are readily located in surveys due to their restriction to waterfalls and cascades, and due to their reliable detection when present at a site (Puschendorf et al., 2011). All surveys were done by two highly experienced frog biologists (one of which was always C. Hoskin).

The basic procedure for field surveys was to locate the downstream limits for L. lorica and L. nannotis and then survey upstream to the rainforest. Suitable stream sections were identified from Google Earth imagery, accessed by helicopter or car, and then surveyed on foot. The streams in this region have a predictable morphology. Each is sourced above 1,100 m in the rainforest uplands. The stream aspect is initially low in the uplands, and then grade increases substantially as the stream flows out of the rainforest around 900 m elevation and steeply descends through fairly continuous (and often substantial) waterfalls and cascades on exposed granite bedrock through open canopy forest, down to about 500 m elevation, where each stream then abruptly flattens out. This abrupt transition from steep to relatively flat marks the end of suitable habitat for these species. Stream gradient is then consistently very low, and the stream consists of long pools separated by short, sandy riffle areas. Surveys extended to the lowest area of suitable habitat on each stream (i.e., the lowest waterfall/cascade), which is between 450 and 500 m elevation, and beyond for some distance along the flat stream. Litoria lorica and L. nannotis were never found beyond the lowest waterfall or cascade (Hoskin & Puschendorf, 2014; Hoskin & Puschendorf, 2021, in preparation). Some stream sections kilometers downstream from the downstream limit of suitable habitat for L. lorica and L. nannotis were also surveyed to confirm absence of these species. The results of surveys in this catchment are summarized in Fig. 1, showing the distribution of L. lorica (red) and L. nannotis (blue) on each stream section.

Stream water sampling

The present study was conducted in a permanently flowing stream catchment in the Wet Tropics of Queensland, north-east Australia. Tropical Australia exhibits high rainfall seasonality (Feng, Porporato & Rodriguez-Iturbe, 2013; Lough, 1993), with a wet season (generally December–April) and a dry season (generally May–November). This seasonality is evident in the monthly rainfall for the study region in 2018–2019 (Fig. S1). This study included two eDNA sampling trips. The most comprehensive was conducted during the wet season (8–10 April 2019) and included all sites shown in Fig. 1. These are: a site immediately downstream from the reintroduced L. lorica population (site 2) and then two sites further downstream from this population (sites 4, 5); a site about 10% into the main L. lorica population (site 6); three sites (sites 7, 9, 10) far-downstream from the main L. lorica population (10.0, 17.1 and 22.8 km, respectively); and three sites in tributaries where only L. nannotis has been found upstream in field surveys (sites 1, 3, 8) (Fig. 1). Site 2 (immediately downstream from the reintroduced population) and site 6 (at the upper end of the main L. lorica population) are comparable because there is an estimated 100 adult L. lorica immediately upstream from both sites (C. Hoskin, 2019, unpublished data). All sites and all three eDNA sampling methods (outlined below) were done on this trip.

A dry season sampling trip was also conducted to assess whether far-downstream detection was still possible during reduced stream flow. Sampling during the dry season was conducted on 24–25 October 2019, and only included a subset of the sites (sites 6–10) deemed most informative for comparing long distance eDNA transport. These sites were the four far-downstream sites (7–10), and site 6 as a ‘positive control’ (i.e., sampling water where both species occur). Environmental DNA sampling during this trip was only conducted using the 100 mL sampling method (see below), with five replicates taken at each site (Table S2). Water volume in the stream during this dry season sampling trip was observed to be markedly less than during the wet season trip, as seen in comparative photos at site 7 (Fig. S1).

Environmental DNA field sampling methods

We used three eDNA sampling methods during the wet season sampling event: (1) direct water collection and preservation of 15 mL samples; (2) direct water collection and preservation of 375 mL samples (from which 100 mL was subsequently sub-sampled); and (3) on-site filtration of large volumes of water. The different methods were used to ultimately assess far-downstream detection success against feasibility of sampling remote populations (i.e., amount and weight of gear/samples carried on foot).

For the 15 mL water sampling method, a new 50 mL Falcon tube was used to decant a 15 mL sample of stream water into another 50 mL Falcon tube containing ten mL of Longmire’s preservative solution (Longmire, Maltbie & Baker, 1997) at the Longmire’s: water ratio used by Edmunds & Burrows (2020) (Fig. 2). A total of five replicates of the 15 mL samples were taken in close proximity at each site (Table 1). Additionally, a field blank (control) was taken at each site, by conducting the same procedure but decanting 15 mL of MilliQ water rather than stream water into the tube containing the Longmire’s solution.

The 100 mL sampling method was performed by using a new bottle of 500 mL capacity to take a 375 mL sample from the stream and decant it into a 500 mL bottle containing 125 mL of Longmire’s preservative solution (Fig. 2). A field blank was included for each site, by decanting 375 mL of MilliQ water into a 500 mL bottle containing 125 mL of Longmire’s solution. Between two and four replicate 375 mL stream water samples were collected at each site, depending on how far the walk was to sites (i.e., the limitation being carrying water sample weight over long distances) (Table 1). A total of two replicate 375 mL samples were collected at sites 1–4, three replicates at site 6, and four replicates at site 5 and sites 7–10 (Table S2; Table 1). In the laboratory, each 375 mL sample was decanted into five aliquots of 20 mL each for eDNA extraction (Fig. 2). Therefore, a total of 100 mL of each 375 mL replicate was screened for eDNA and hence this method is hereafter termed the ‘100 mL sampling method’. Field work was conducted under Queensland Government permit WITK18662017 (to C.J.H).

For the on-site filtration method, water was filtered through a deep filter layer of 1 µm nominal pore size, using a sampler system described by Stevens (2020) and used by Lewis et al. (2018). A flow meter was attached to record the volume of water processed. Filtering was conducted at five sites with relatively close road access because this filtering device was large and heavy (approximately 5 kg). A single replicate was collected at each site, consisting of water filtration for a period of 1 h. The mean total volume of filtered water across the five sites was 1,480 L (± 111.11 SE) (see Table 1 and Table S2 for filtered volumes at each site). After filtration was completed, the filter was preserved in 700 mL of Longmire’s solution diluted to 25% in MilliQ water. An equipment blank was included for each site, which involved preserving a new, clean filter in 700 ml of Longmire’s buffer diluted with water to 25%, with the procedure conducted on-site. In order to allow the eDNA in the filters to resuspend, filters in Longmire’s solution were stored indoors at room temperature for 1 week after filtration. After this period, each water sample was mixed by inversion ten times in order to ensure that all eDNA was resuspended before taking an aliquot for extraction. A 20 mL aliquot from the diluted Longmire’s solution in which the filter had been preserved was then taken and eDNA was extracted (Fig. 2).

Environmental DNA extractions

Upon arrival to the laboratory, eDNA was extracted via a glycogen-aided isopropanol based precipitation protocol (Villacorta-Rath et al., 2020) in a dedicated eDNA laboratory at James Cook University (JCU), Australia. For all extractions, 20 mL sample aliquots were mixed with 5 µL glycogen (200 mg/mL), 20 mL isopropanol and 5 mL NaCL (5M). Samples were then incubated overnight at 4 °C and subsequently centrifuged at 6,750 g for 10 min to form a pellet. The supernatant was then discarded and pellets were dissolved in 600 µL of lysis buffer (guanidinium hydrochloride and TritonX; pH 10), transferred into a 2 mL DNA LoBind^® Tube (Eppendorf), and frozen for up to four months at −20 °C to enable processing all samples as soon as possible. Environmental DNA present in the samples was lysed at 50 °C for 5 h and a subsequent precipitation step was carried out by adding 1 µL glycogen and 1,800 µL polyethylene glycol (PEG) buffer to the samples. Samples were centrifuged at 20,000 g for 30 min to form a pellet that was then washed twice using 70% ethanol. After the ethanol washes, the pellet was dried and eDNA was resuspended in 100 µL MilliQ water. Finally, eDNA was purified using the Qiagen DNeasy^® PowerClean^® Pro Cleanup kit and eluted in 100 µL elution buffer. Given that we sampled very large volumes of water with our filtration system, samples were highly pigmented (Fig. S2). After eDNA extraction and purification, samples still contained a high level of coloration and therefore we applied a 1:20 dilution to result in a sample that did not exhibit any coloration. We applied inhibition tests (Jane et al., 2015) on the final 1:20 dilution.

Screening of the 100 mL samples from the wet season sampling was complicated by a problem in the laboratory. A mechanical failure of the real-time PCR machine during screening of L. nannotis from sites 1–5 and 9–10 resulted in complete failure of those samples. Re-extraction from those water samples was conducted 5 months later but this resulted in minimal detections. Given the high detection frequencies of sites 6, 7 and 8 (27–46%; Table 2), which were successfully screened during the first round of qPCR, it was concluded that the eDNA had degraded in the intervening months. This was despite the fact that the samples were mixed with Longmire’s solution and stored at room temperature in the dark, which had been demonstrated to protect eDNA from degradation for at least 8 weeks (Edmunds & Burrows, 2020).

Real-time PCR (qPCR)

We used two different species-specific primer pairs targeting the cytochrome c oxidase subunit I (COI) mitochondrial gene of L. lorica and L. nannotis (Table S3; Edmunds et al., 2019). The limit of detection (LOD) was estimated using a ten-fold serial dilution of double-stranded synthetic DNA fragments (qBlocks^TM Integrated DNA Technologies Pty Ltd., New South Wales, Australia) synthesized to match the target fragments of each of the study species, ranging from 2.82 × 10⁷ to 0.70 copies/µL for L. lorica and 2.17 × 10⁷ to 0.55 copies/µL for L. nannotis (Edmunds et al., 2019). Additionally, the LOD was determined using a seven-fold serial dilution of genomic DNA (gDNA) derived from toe pad tissue from each of the target species. For L. lorica, gDNA dilutions ranged between 1.24 and 3.1 × 10⁻⁵ ng/μL, and between 6.45 × 10⁻⁴ and 1.6125 × 10⁻⁵ ng/μL for L. nannotis (Edmunds et al., 2019). Between four and six technical replicates per dilution were used, and the LOD was set at the lowest standard with 95% or greater detection (Klymus et al., 2019). Based on the serial dilutions, the LOD was determined to be 4.9 × 10⁻⁵ ng/μL or 2 copies/reaction for L. lorica, and 4.8 × 10^-5 ng/μL or 2 copies/reaction for L. nannotis.

qPCR assays were performed on a QuantStudio™ 3 or QuantStudio™ 5 Real-Time PCR System (Thermo Fisher Scientific, Scoresby, VIC, Australia Pty Ltd) in white 96 or 384-well plates, respectively, and sealed with optical films (Thermo Fisher Scientific, Scoresby, VIC, Australia Pty Ltd). Presence of L. lorica and L. nannotis was screened through twelve technical qPCR replicates of each field replicate, representing 36% of the total available DNA elution volume for each species. This is a much higher number of technical replicates than most eDNA studies use (reviewed by Rees et al., 2014). High field replication is needed in order to avoid false negative detections (Furlan et al., 2016), so by thoroughly screening each field replicate we maximised the chances of detecting the target species eDNA in the available replicate samples. The downside of increasing the number of technical replicates is that the false positive rate also increases; however, screening field and extraction controls, and qPCR controls, can account for this problem (Bustin et al., 2009). Additionally, a triplicate positive control consisting of gDNA of the target species and three no-template controls (NTC) was used (Bustin et al., 2009). The NTC samples did not contain the target species DNA and their lack of amplification indicated that no contamination was introduced during plate handling. Each qPCR assay consisted of 3 µL of template DNA and 7 µL of master mix (5 µL PowerUp SYBR Green Master Mix; 0.5 µL forward primer at 10 µM; 0.5 µL reverse primer at 5 µM; 1 µL MilliQ^® water). Thermal cycling conditions were as follows: initial denaturation and activation at 95 °C for 2 min, then 55 cycles of 95 °C for 15 s, and 60 °C for 1 min. A subsequent melt curve analysis was performed to generate dissociation curves by transitioning from 60 to 95 °C, at 0.15 °C s⁻¹.

All plates were analysed with a common fluorescence threshold of 0.2 using QuantStudio™ Design and Analysis Software (version 1.4.2; Thermo Fisher Scientific, Scoresby, VIC, Australia Pty Ltd.) before export and subsequent analyses in Microsoft Excel. Samples were considered putative positive detections if: (1) the amplification curve crossed the common fluorescence threshold within 50 cycles; (2) the amount of eDNA was above the LOD; and (3) the melt curve analysis showed a dissociation temperature peak at 78.52 °C (±0.62–99% confidence interval) for L. lorica and 79.66 °C (±0.75–99% confidence interval) for L. nannotis. Amplicons from putative positive detections were sequenced via dual direction Sanger sequencing at the Australian Genome Research Facility (AGRF) to confirm that they were true detections. Amplicon sequences from the samples considered putative positive detections were considered as true detections if there was ≥99% pairwise identity to in-house L. lorica sequences (no sequences on GenBank), with the targeted COI section being: CGACACTTATTATGTTGTAGCCCATTTCCATTATGTATTGTCTATAGGAGCTGTATTCGCCATTATAGC (Edmunds et al., 2019). This section of L. lorica sequence is invariant among L. lorica sequences and has at least three mismatches to other frog species. For L. nannotis, positive detections were considered true detections if they exhibited ≥97% pairwise identity with the COI gene of the species (GenBank accession number JN130908 matching between positions 219–286 bp; accession number JN30913 matching between positions 167–286 bp; in-house sequences).

Inhibition test

We tested inhibition in water samples by spiking 80 copies of artificial DNA into triplicated samples from the sites without presence of L. lorica (sites 1, 3 and 8) and 8 copies of artificial DNA into all field controls. Additionally, we spiked the same number of DNA copies into three technical replicates containing only MilliQ water. A sample was considered inhibited if it exhibited a shift in Ct values of three or more cycles when compared to the spiked MilliQ water (Cao et al., 2012).

Calculating detection frequencies and downstream detection distances

Detection frequency for each species was calculated as the proportion of qPCR replicates of each sampling method at each site that yielded a positive detection, in relation to the total number of qPCR replicates at that site. Detection frequencies for each eDNA capture method were calculated separately for the two sampling trips. Standard errors of these proportions were calculated by dividing the standard deviation of the number of positive qPCR detections by the square root of the total number of qPCR replicates (Calter & Calter, 2011). Downstream detection distances were calculated by using the path distance function in Google Earth to measure the distance from the downstream limit of the species on each stream (from field surveys) to the eDNA sampling site. Of particular relevance were the downstream distances from the two populations of L. lorica (Fig. 1).

The primary aim of this study was to determine downstream detection distances for each species, with a particular focus on far-downstream distances (i.e., >10 km, >20 km). Beyond this overall aim, we could also assess determinants of detection success where sufficient data was obtained. The three determinants assessed were: (1) whether precipitation from larger water volumes (i.e., 100 mL vs. 15 mL) yielded higher detection frequencies in L. lorica wet season samples; (2) whether detection frequencies were higher for the considerably more abundant species, L. nannotis; (3) whether there was higher probability of detecting L. lorica during the wet season than during the dry season. The methods for assessing these are described in the sections below. Other determinants of detection success could not be assessed due to insufficient numbers of comparisons. This was primarily due to the mechanical failure of the real-time PCR machine, which impacted L. nannotis wet season screening for the 100 mL sampling method. This precluded testing for higher detection frequencies for the 100 mL vs. 15 mL method for L. nannotis, or testing for higher detection frequencies for L. nannotis vs. L. lorica using the 100 mL sampling method. The water filtering technique was used at too few sites to test for differences in detection success compared to the 100 mL and 15 mL precipitation methods.

Was the more abundant species more readily detected?

We had sufficient data from the 15 mL wet season sampling to test the prediction that L. nannotis would be more readily detected than L. lorica. This prediction was based on the fact that there is estimated to be at least an order of magnitude more L. nannotis upstream of all these sites (Table S5). The difference in detection frequencies for the two species was tested using a Wilcoxon signed-rank test, comparing the detection frequencies for each species at sites where both species could both have been detected (i.e., sites 2, 4–7, 9, 10; Fig. 1). Occupancy models were also run to estimate the probability of detection of each species during the 15 mL wet season sampling. The modelling was conducted as described above, but with ‘species’ added as a covariate (Supplemental Informations 2 and 3).

Was detection probability higher in the wet season than the dry season?

The wet and dry season sampling was not designed to test for significant differences in detection frequencies with stream flow (cf. Jane et al., 2015; Shogren et al., 2018) but rather the two sampling trips were conducted to determine if far-downstream detection was still possible under low stream flow conditions of the dry season. Lower detection chances were assumed based on longer eDNA transport times to downstream sites under dry season (low flow) conditions. There were not enough comparisons to statistically test for a difference in detections between the wet and dry season sampling. However, occupancy models were run to estimate the probability of detecting L. lorica in the 100 mL sampling during the wet and dry seasons. The modelling was conducted as described above, but with ‘season’ added as a covariate (Supplemental Informations 2 and 3).

Inhibition tests

Mean Ct values for all spiked field samples and field controls did not show a shift in Ct values of more than one cycle (Tables S6 and S7), indicating that no inhibition was present in the qPCR assays.

Detection frequencies and downstream detection distances for the three methods

All positive detections satisfied the following conditions: (1) amplification occurred within 50 cycles; (2) amount of eDNA was above the LOD; (3) the melt curve peak corresponded to that of each species; and (4) amplicon sequences from positive detections had >97% and >99% match with targeted COI regions for L. lorica and L. nannotis, respectively. The qPCR of all field control samples and extraction control samples were devoid of L. lorica and L. nannotis DNA (Table S5).

Precipitation of the 15 and 100 mL samples, and water filtration gave positive detections of L. lorica and L. nannotis at most of the sampling sites where each of the species is known to occur at the site or upstream (Table 2; Fig. 1; Table S4). Of particular note are the far-downstream detections (10, 17.1 and 22.8 km) for both species using all three methods during the wet season sampling (Table 2; Figs. 1 and 3), and for the 100 mL sampling method during the dry season sampling (Fig. 3; Table S4). Importantly, we did not detect L. lorica eDNA at the three sites where this species is not known to occur upstream (i.e., sites 1, 3, 8; Table 2; Fig. 1; Table S4).

Comparing detection success from the three eDNA sampling methods during wet season sampling

Did precipitation from larger water volumes yield higher detection frequencies?

For the precipitation method, L. lorica detection frequencies were usually higher using 100 mL samples than 15 mL samples, and generally by an order of magnitude (Table 2; Fig. 4A). For example, at the furthers three sites that L. lorica could have been detected (sites 7, 9 and 10), percentage of positive detections using the 100 mL samples were 37.5%, 16.7% and 12.5%, respectively, vs. 3.3%, 3.3% and 0% for the 15 mL samples. This difference was expected based on the fact that the 100 mL sampling method screened about five times as much water at each site. A Wilcoxon signed-rank test, using the paired average probabilities for each sampling method at the seven sites where L. lorica could have been detected (i.e., sites 2, 4–7, 9, 10) (Table 2; Fig. 1), revealed the 100 mL sampling method to be significantly better for detecting L. lorica than the 15 mL method (mean difference 1.38, z = −1.99, W = 1, P < 0.05). The site occupancy model revealed higher probability of L. lorica detection at all sampling sites during the wet season using the 100 mL sampling compared to the 15 mL sampling (Fig. 4A).

In real terms, when one is comparing positive detection or not, the 100 mL sampling method gave L. lorica detections at five of the seven possible sites, whereas the 15 mL method gave detections at three of the seven sites (Table 2). Additionally, the 100 mL sampling method gave L. lorica detections at both sites where only about 100 adults are present directly upstream (i.e., sites 2 and 6), whereas no L. lorica detections were made at these sites using the 15 mL method (Table 2). Finally, the 100 mL sampling method gave 12.5% L. lorica detections across replicates at the furthest site (site 10; 22.8 km downstream); whereas the 15 mL method gave no L. lorica detections at this site.

Comparison of L. nannotis detection from precipitation of 100 mL samples vs. 15 mL samples was not possible due to a laboratory issue during screening some of the 100 mL water samples, detailed above. These sites are shown as no data in Table 2 (dashes) and Fig. 1B (0 symbols). The only comparison that can be made is the observation that the detection frequency for L. nannotis at sites 6–8 was markedly higher for the 100 mL sampling method than the 15 mL method (Table 2). Using the 15 mL sampling method, L. nannotis was detected at all but one site (site 9; 17.1 km downstream) but detection frequencies were low (<10% at six out of the nine sites with detections; Table 2).

Was the more abundant species more readily detected?

We assumed that L. nannotis would be more readily detected than L. lorica at sites they could both be detected because there is estimated to be at least an order of magnitude more L. nannotis than L. lorica upstream of all these sites (C. Hoskin, 2019, unpublished data; Table S5). This was reflected in the percentage of positive eDNA detections across sites using the 15 mL methods, with L. nannotis detections being noticeably higher (Table 2). Further, when using the 15 mL samples, sites 2, 5, 6 and 10 did not show positive L. lorica eDNA detections, but all of these sites had positive detections of L. nannotis eDNA (43.3%, 3.3%, 25%, 5%, respectively). A Wilcoxon signed-rank test, using the paired average probabilities for each species at the seven sites where both species are present upstream (i.e., sites 2, 4–7, 9, 10) (Table 2; Fig. 1), revealed that L. nannotis was indeed more readily detected (mean difference 16.05, z = −1.89, W = 1.5, P < 0.05). Site occupancy modelling also revealed a higher probability of detecting L. nannotis than L. lorica eDNA at sites up to 4 km downstream from the source population (Fig. 4B). At sites located 10 km downstream of the source populations and beyond, the eDNA detection probability for both species was very similar (Fig. 4B).

In real terms, when one is comparing positive detection or not, L. nannotis was detected using the 15 mL sampling method at six of the seven possible sites, whereas L. lorica was only detected at three of the seven sites (Table 2). A similar comparison could not be tested for the 100 mL sampling method due to the mechanical failure of the qPCR machine during the L. nannotis screening for most sites (detailed above). For the 100 mL samples, only two sites could be compared (sites 6 and 7), and the percentage of positive eDNA detections was higher for L. nannotis at both of these (Table 2).

Was detection probability higher in the wet season than the dry season?

The primary objective of conducting eDNA sampling during both the wet and dry season was to determine whether far-downstream eDNA detection was possible under both high stream flow (wet season) and low stream flow (dry season). Dry season sampling just focussed on key subset of the sites. Importantly, both L. lorica and L. nannotis were also detected at most far-downstream sites during the dry season sampling (Table S4). In the dry season sampling, the percentage of positive L. lorica eDNA detections at the main population (site 6) was 6.7% and 1.7% at both of the two furthest downstream sites (17.1 and 22.8 km); but there were no detections at site 7 (10 km) (Table S4). Litoria nannotis percentage of positive eDNA detections in the dry season sampling was 3.3% at site 7 (10 km) and 1.7% at sites 9 and 10 (17.1 and 22.8 km downstream, respectively); but there were no detections at site 8 (Table S4).

Litoria lorica detection frequencies were higher during the wet season (high flow) sampling than the dry season (low flow) sampling at all four sites they could have been detected at (sites 6, 7, 9, 10; Fig. 3; Table S4). The site occupancy models showed slightly higher probability of L. lorica eDNA detection during the wet season than during the dry season (Fig. 4C). Statistical testing using a Wilcoxon signed-rank test was not possible due to too few site comparisons. Comparison of L. nannotis wet vs. dry season sampling was not possible due to the laboratory issue outlined above (Table S4).

Implications for finding populations of threatened and ‘missing’ amphibian species

Our study shows that rare aquatic species can be detected over 20 km downstream, including a Critically Endangered frog population of about 1,000 adults. This is a significant advance on previous studies of small population detection using eDNA, which have generally demonstrated detection distances of hundreds of meters rather than kilometres (Olson, Briggler & Williams, 2012; Stoeckle, Kuehn & Geist, 2016; Wacker et al., 2019). Further, our study shows that reliable far-downstream eDNA detection is not only possible using filtration of large volumes but also by precipitating water volumes of 100 mL. In contrast, smaller water volumes of 15 mL were not reliable for far-downstream detection of these frogs. The 100 mL sampling method was also successful for detecting a very small population of about 100 individuals of L. lorica immediately downstream. None of the three methods were successful for detecting the very small population kilometres downstream, suggesting there may be a population size threshold for long distance eDNA detection using the methods and replicate numbers we used. Further studies should be conducted on other taxa, either using wild populations or caged individuals, to determine population size thresholds for far-downstream eDNA detection.

Our results have applicability to the hundreds of Critically Endangered and ‘missing’ stream frogs, globally, many of which are/were found in remote mountainous areas (Scheele et al., 2019). We suggest the following methodology. First, a water filtration method is recommended if a stream site is accessible (e.g., where there is a road crossing), and we recommend filtering large volumes of stream water (>1,000 L) to detect rare species. Second, the method that best balances reliability and practicality is collecting moderate water volumes (e.g., individual 100 mL samples, or 500 mL and then precipitating eDNA from 100 mL subsamples). Collecting water samples does not require training or special equipment and many bottles can be carried in a backpack. Third, although we could not test the effect of stream flow statistically, the data suggests that water collection during higher flow periods (e.g., the wet season in our system) gives higher detection than during low flow periods. However, we predict lower detection during flooding conditions due to the dilution of eDNA in massive water volumes. Fourth, we suggest screening in readily accessible downstream areas and then working upstream from there if: (i) a positive detection is made, or (ii) there is strong reason to believe a very small population remains undetected upstream (e.g., the last known historic site or an unconfirmed recent sighting).

In our system, the next step will be screening downstream sites in all catchments where L. lorica was known to occur before disease-induced declines, or likely occurred but was undetected prior to declines. The only two known populations of L. lorica occur in the catchment we studied herein, on the western Carbine Tableland, but the species was previously known from a major catchment on the eastern Carbine Tableland, and several streams in the Thornton Peak uplands (Cunningham, 2002). The downstream detection distances found in the present study suggest that we can use filtration or the 100 mL sampling method to sample accessible downstream areas of all major catchments flowing off these two upland areas to fairly reliably search for ‘missing’ populations across the entire historic distribution.

More broadly, a key remaining question is what positive eDNA detections say about abundance and distance from the population. This could be further tested in either of these species, by incorporating more sites, more repeat sampling across seasons, and additional data on environmental factors, such as flow rate and volume. Regardless, the simple approach we present in the present study could be used to try find the remaining populations of hundreds of other amphibians, globally, that are so urgently required to facilitate management.