Leveraging environmental DNA (eDNA) to optimize targeted removal of invasive fishes

Figures & data

Figure 1. Study area within the San Luis National Wildlife Refuge (Merced County, California, USA) containing sites sampled during the 2022 pilot study and the 2023 main study. For the purposes of this map, sampled means visited for eDNA collection, trapping, or both.

Table 1. Primer and probe sequences, optimal final concentrations, and annealing temperature for P. dabryanus + M. mizolepis (PD_MM) and M. anguillicaudatus (MA) assays.

Target	Gene	Assay	Primer and probe (5’ to 3’)	Primer/probe concentration (final)	Annealing temperature (°C)
Large-scale loach (Paramisgurnus dabryanus) + Muddy loach (Misgurnus mizolepis)	COI	PD_MM_FWD	GGTGCCTGAGCCGGAAT	900 nM	60
		PD_MM_REV	AGCTCAGCACGAATCAAAAGG	900 nM
		PD_MM	6FAM-TCGGAACGGCTCTC-MGBNFQ	200 nM
Pond loach (Misgurnus anguillicaudatus)	COI	MA_FWD	TGGTGGATTTGGTAATTGACTTGT	900 nM	60
		MA_REV	ACGCGGAAATGCCATGTC	900 nM
		MA	6FAM-CCTTTAATGATTGGAGCAC-MBGNFQ	140 nM

Table 2. Species tested for cross-reactivity with the M. anguillicaudatus and P. dabryanus + M. mizolepis assays.

Scientific name	Common name
Misgurnus anguillicaudatus	Pond Loach
Misgurnus mizolepis	Fine-Scale Loach
Paramisgurnus dabryanus	Large-Scale Loach
Alosa sapidissima	American Shad
Catostomus occidentalis	Sacramento Sucker
Dorosoma petenense	Threadfin Shad
Gasterosteus aculeatus	Threespine Stickleback
Hypomesus nipponensis	Wakasagi Smelt
Hysterocarpus traskii	Tule Perch
Lavinia exilicauda	Hitch
Lepomis macrochirus	Bluegill Sunfish
Lepomis microlophus	Redear Sunfish
Micropterus dolomieu	Smallmouth Bass
Micropterus salmoides	Largemouth Bass
Morone saxatilis	Striped Bass
Oncorhynchus mykiss	Rainbow Trout/steelhead
Oncorhynchus tshawytscha	Chinook Salmon
Pogonichthys macrolepidotus	Sacramento Splittail
Pomoxis nigromaculatus	Black Crappie
Tridentiger bifasciatus	Shimofuri Goby
Hemiramphus brasiliensis*	Ballyhoo*

All DNA used in testing was derived from tissue samples collected from voucher specimens. The DNA used for testing was normalized to 2 ng·μL⁻¹ (± 0.2 ng·μL⁻¹) for testing. *Hemiramphus brasiliensis was tested for cross-reactivity because it was used during the fixed DNA source experiment; this species is not present in the Sacramento San-Joaquin Delta watershed.

Table 3. Assay performance with tenfold serial dilutions of each target.

Species	Assay	Slope	Y-Intercept	Efficiency	R^2	LOD (ng·μL^-1)	LOQ (ng·μL^-1) (CV)
Paramisgurnus dabryanus	PD_MM	−3.41	40.04	96.39%	> 0.99	1.15·10^-5	1.15·10^-4 (0.28)
Misgurnus mizolepis	PD_MM	−3.32	40.84	100.05%	0.99	2.04·10^-5	2.04·10^-4 (0.27)
Misgurnus anguillicaudatus	MA	−3.39	41.23	97.13%	0.99	1.94·10^-5	1.94·10^-4 (0.28)

Slope, y-intercept and R² are calculated based on the linear relationship of Cq at each dilution step for each target. The limit of detection (LOD) is the lowest standard concentration of template DNA that produced at least 95% positive replicates. The limit of quantification (LOQ) is the lowest standard concentration that could be quantified with a coefficient of variance (CV) value below 0.35.

Figure 2. Box plot summarizing the distribution of total lengths (TLs) of loaches caught during the pilot study and the main study. Weeks 2, 4, and 6 of trapping were preceded by eDNA sampling during Weeks 1, 3, and 5. Weeks 7–8 comprised of only trapping using the remaining eDNA positive sites that were not trapped during week 6. One minnow trap per site was used during the pilot, while ten minnow traps per site were used for the main. The open circle points represent total length outliers. The bold black line in each box represents the median length and the whiskers extend 1.5 times the interquartile range.

Table 4. The 2022 field effort for the pilot study.

Week	Dates	Site type	# of sites sampled for eDNA	# of sites positive for eDNA	# of sites trapped	# of sites where loach collected	# of traps with loach	# of loach removed
1	3/14–3/18	Pond	4	0	17	0	0	0
2	3/21–3/25	Pond	4	0	21	0	0	0
3	3/28–4/1	Pond	0	0	24	0	0	0
1	3/14–3/18	Canal	4	2	14	1	1	1
2	3/21–3/25	Canal	6	2	24	3	3	4
3	3/28–4/1	Canal	10	3	27	5	5	13
Totals			28	7	127	9	9	18

Weeks 1–3 comprised of randomly trapping an approximately equal number of pond and canal habitats. A random subset of those sites was also sampled for eDNA. One baited minnow trap was set per site.

Table 5. Sensitivity, specificity, and related rates comparing minnow traps (‘predictor’) to eDNA (‘truth’) for the pilot study.

Metric	Estimate	Lower 95% confidence limit	Upper 95% confidence limit
Accuracy	0.79	0.60	0.90
Sensitivity (true positive rate)	0.14	0.06	0.31
Sensitivity (true negative rate)	1.00	0.88	1.00
Positive predictive value	1.00	0.88	1.00
Negative predictive value	0.78	0.60	0.89
False negative rate	0.86	0.69	0.94
False positive rate	0.00	0.00	0.12
False discovery rate	0.00	0.00	0.12
False omission rate	0.22	0.11	0.40
F1 score	0.25	NA	NA
Matthews correlation coefficient	0.33	NA	NA
True positive (count)	1.00	NA	NA
False positive (count)	0.00	NA	NA
False negative (count)	5.00	NA	NA
True negative (count)	21.00	NA	NA

The ‘NA’ means not applicable, confidence intervals do not exist for the metrics indicated.

Figure 3. Quantitative PCR detections of E. mordax (circle) and H. brasiliensis (square) within the small and large canal during initial sampling (a and b, respectively) and 24h after the removal of the proxy fishes from the small and large canal (c and d, respectively). Each point above 0 ng·μL⁻¹ indicates that target eDNA was detected and represents a positive technical replicate.

Figure 4. Quantitative PCR detections of P. dabryanus/M. mizolepis within the small and large canal during initial sampling (a and b, respectively) and 24h later within the small and large canal (c and d, respectively). Each point above 0 ng·μL⁻¹ indicates that target eDNA was detected and represents a positive technical replicate.

Figure 5. Artemis model output of the estimated effect size of target and distance on ln[eDNA] at the small canal location (Cq ∼ distance + target + (1|FilterID).

Figure 6. Probability of detecting a small biomass (circle) and large biomass (square) in the small canal if it is present using the effect sizes of each variable as calculated by the model with three filter replicates. Points indicate mean probability of detection and bars indicate the 95% confidence intervals of that estimate.

Table 6. The 2023 field effort for the main study.

Week(s)	Dates	Site type	# of sites sampled for eDNA	# of sites positive for eDNA	# of sites trapped	# of sites where loach collected	# of traps with loach	# of loach removed
1–2	3/11–3/22	Canal	73	25	13	12	37	45
3–4	4/22–5/3	Canal	92	62	12	8	30	66
5–6	5/6–5/17	Canal	85	52	19	18	95	377
7	5/23–5/26	Canal	NA	NA	18	16	56	95
8	5/30–6/2	Canal	NA	NA	15	14	41	75
Totals			250	139	77	68	259	658

During each week, sites were sampled for eDNA or trapped over a five-day period. Weeks 1, 3, and 5 comprised of eDNA sampling while Weeks 2, 4, and 6 comprised of trapping eDNA-positive sites. Weeks 7–8 were extended trapping efforts that utilized the remaining 23 eDNA-positive sites from week 6 that were not trapped. Ten baited minnow traps were set per site. The ‘NA’ means not applicable.

Figure 7. Expected mean catch per minnow trap per day (24 h or 1440 min), as predicted by the best models for each study, considering both the count and zero portions of the hurdle models. Error bars indicate the 95% confidence intervals.

Table 7. Comparison of models for mean catch in the pilot study and the eDNA study.

Model family	Pilot df	Pilot AIC	eDNA df	eDNA AIC
Poisson	1	94.9	6	960.4
Negative binomial	2	80.2	7	473.7
Truncated Poisson	2	78.6	6	960.1
Truncated negative binomial	3	80.6	7	469.6

Bold values indicate the model with the lowest AIC values for each study.

Data availability statement

All data that supports the findings of this study are available in Dryad at https://doi.org/10.5061/dryad.n8pk0p342. Statistical code is available at https://github.com/USFWS/LFWO-Loach-eDNA.