Enumeration potential of environmental DNA for Pacific salmon stock assessments

Abstract

The field of environmental DNA (eDNA) has advanced over the past decade, with multiple approaches available for a variety of sampling media and species. While using eDNA for the purpose of simply detecting species is becoming a routine process, the utility of eDNA to estimate species abundance is not well understood. Here, we quantify salmon environmental DNA upstream of a fish counting fence along with river velocity, and together, use these values to determine the correlation between the number of salmon passing by the fish fence daily with daily eDNA rates in water before, during, and after the salmon spawning season for four Pacific salmonids (Oncorhynchus gorbuscha, O. kisutch, O. tshawytscha, and O. nerka; pink, coho, chinook, and sockeye, respectively). Throughout the spawning season, approximately 182,000 salmon were counted passing through the fence, of which >98% were pink salmon. Pink salmon exhibited strong correlation between human counts (effect size = 0.65, SE = 0.040) to eDNA rates in the present study and exhibited day-to-day variation and a unimodal profile rising and falling with human fish counts. However, the salmon species observed in much lower numbers exhibited a much weaker correlation with eDNA levels higher during the pre-migratory period than during the migratory period for sockeye, coho, and chinook. Thus, for salmon spawning runs with less than ~1000 adults and daily counts of less than ~100, the juvenile and/or prior seasons eDNA signal appears to be indistinguishable from the adult spawning eDNA signal in our river system. However, for the large pink salmon run, eDNA rates appeared to reflect a local signal of salmon in space and time, essentially tracking these fish within days of passing through the eDNA sampling site.

INTRODUCTION

Environmental DNA (eDNA) methods have been rapidly evolving and developing over the last 10 years. Indeed, several approaches to utilizing eDNA to detect the presence of a target species are available, including a variety of sampling methods and DNA detection methods (e.g., Métris & Métris, 2023; Ruppert et al., 2019; Sansom & Sassoubre, 2017; Wesselmann et al., 2022; Zhu et al., 2023). In general, eDNA detection may be targeted, such as species-specific quantitative real-time polymerase chain reaction (qPCR)-based methods, metagenomic, or metabarcoding, which use combinations of PCR and next-generation sequencing (Acharya-Patel et al., 2023). The use of targeted qPCR to detect the presence of eDNA is gaining momentum because of its relative simplicity, lower cost, and publicly available assays for many species (e.g., chinook salmon, coho salmon, and cutthroat trout; Matter et al., 2018; Penaluna et al., 2021; Searcy et al., 2022; Shaffer, 2022).

While using eDNA for the purpose of simply detecting species is becoming a routine process, the utility of eDNA to estimate species abundance is not well understood. Several complex processes are hypothesized to influence the amount of eDNA captured in a sample such as eDNA organismal shedding rate, transport, decay, settling, and resuspension, and all of these are likely to vary under different environmental conditions (i.e., pH, UV exposure, temperature, substrate type, turbidity, microbial activity, etc.; Barnes & Turner, 2015; Carraro et al., 2018; Shogren et al., 2017; Strickler et al., 2015). Nonetheless, the current paradigm is that there is a positive correlation between the amount of eDNA and organism abundance in the environment, and therefore could be a useful tool for estimating population sizes and community composition (Yates et al., 2021). Indeed, several eDNA-based enumeration studies demonstrate a positive correlation between eDNA signal and fish abundance in mesocosm experiments, lakes, and streams (Doi et al., 2015; Lacoursière-Roussel, Côté, et al., 2016; Lawson Handley et al., 2019; Takahara et al., 2013; Wilcox et al., 2015; Yates et al., 2019). However, as reviewed by Yates et al. (2021), the strength of the correlation in natural systems varies across studies from strong (e.g., Doi et al., 2017; Levi et al., 2019) to moderate (e.g., Pilliod et al., 2013a; Wilcox et al., 2015) to weak or no correlation (e.g., Lacoursière-Roussel, Rosabal, & Bernatchez, 2016; Sepulveda et al., 2021; Spear et al., 2015).

Anadromous salmonids present a particularly unique opportunity to examine correlations between water eDNA concentration and fish abundance in natural systems by comparing human visual salmon counts in-stream or using fish enumeration facilities. Two recent studies exemplify the potential for using eDNA to estimate fish abundance of Pacific salmon species traveling along lotic environments (e.g., rivers and streams) to enter watersheds (Levi et al., 2019; Tillotson et al., 2018). Tillotson et al. (2018) reported that the sockeye eDNA concentrations roughly matched the temporal visual in-stream counts of adult sockeye salmon, although matched dead more closely than live, and in a unimodal fashion (i.e., increased as the spawning season progressed to a peak and then decreased as the spawning season gradually ended in Hansen Creek, southwest Alaska). A second study by Levi et al. (2019) demonstrated that river flow-corrected eDNA rate closely tracks daily numbers of returning sockeye and coho spawners and out-migrating sockeye smolts in Auke Creek (southeast Alaska). Although this represents considerable progress for developing eDNA as a tool for enumeration, whether these methods can be used for other Pacific salmon runs is unknown and requires additional studies comparing conventional fish counts to eDNA fish estimates.

The Kitwanga River Adult Salmon Enumeration Facility (KSEF, Figure 1) is an aluminum fence structure operated annually since 2003 by the Gitanyow Fisheries Authority (GFA) during the salmon migration period (July–October) and was the site selected for this study. The KSEF spans the entire Kitwanga River forcing upstream migrating fish into trap/camera boxes where they can be identified and enumerated to the species level, and a percentage of each run is biologically sampled. The Kitwanga Watershed is located within the Ministry of Forests, Skeena Stikine Forest District, and forestry development activity began in 1912 and by the mid-1960s, a selective harvesting silviculture system was replaced by clearcut harvesting systems. Although a Kitwanga Watershed Restoration Program was initiated in 2001 and several works have improved watershed health (i.e., road deactivation, riparian, and in-stream works), the impacts of land-use, fish harvest management practices, and natural environmental condition fluctuations cumulatively have resulted in reduced salmon populations in the system over the last Century (McElhany et al., 2000; Rabnett, 2005). Here, we quantify salmon eDNA in Kitwanga river samples upstream of the KSEF along with river velocity and together use these values to determine the correlation between the number of salmon passing by the fish fence on a daily basis with daily eDNA rates before, during and after the salmon spawning season for four Pacific salmonids (chinook, sockeye, coho, and pink). This study will assist in advancing methods for using eDNA to estimate fish abundance, and further assess its potential as a stand-alone stock assessment tool for the numerous streams currently lacking enumeration program.

FIGURE 1. Outline of British Columbia, the Skeena watershed (yellow area), the Skeena River (blue line), and the Kitwanga River (purple line). The zoomed-in inset is a satellite image showing the Kitwanga River Salmon Enumeration Facility (KSEF) hydrometric monitoring sites, the Water Survey of Canada's hydrometric site, and water collection site (yellow circle) of triplicate 1-L daily water samples used in subsequent qPCR assays to quantify eDNA copy number for pink, coho, chinook, and sockeye salmon. Blue arrow indicates direction of water flow.

METHODS AND MATERIALSConventional fish counts at the Kitwanga River Salmon Enumeration Facility

The KSEF fence is usually set-up and operational annually by July 10, a week before salmon begin showing up at the site. It is then operated until the end of October allowing fisheries staff to enumerate all upstream migrating salmon. For the remainder of the year, the river is left open and fish passage at the site is uninterrupted. When the KSEF is operational, trap boxes are installed on both sides of the river which allows for remote fish counting and identification of all fish, as well as biological sampling of a subset of fish (usually about 5% of the salmon are sampled annually). Most of the salmon migration passed the fence typically occurs in the morning and evenings, thus counting and identification by Gitanyow Fisheries Authority crews are performed twice per day at these times.

The Kitwanga River is a 59 km fifth-order stream (Beblow, 2021) that originates in the Hazelton Mountains and drains in a southerly direction into the Skeena River. The Kitwanga Watershed is fairly small when compared to other Skeena Tributaries, but it is very biologically rich supporting all five species of Pacific salmon (chinook, sockeye, pink, chum, and coho), steelhead, other salmonids, and various coarse fish. There is only one lake in the system, known as Gitanyow (Kitwancool) Lake. The KSEF is located approximately 4 km upstream of the confluence of the Skeena and Kitwanga Rivers and ~28 km downstream of Gitanyow Lake. Although spawning locations along the Kitwanga River have not been comprehensively studied for each salmon species, it is known that most chinook, pink, and chum salmon spawn at several locations between the KSEF and Gitanyow Lake. It is also known that coho spawn throughout the watershed including above Gitanyow Lake and in some of the larger tributaries in the system, while sockeye, at least in the last few decades are known to exclusively spawn along the lakeshore in Gitanyow Lake. As with all natal salmonid streams where spawning occurs, this results in waters that are also home to juvenile salmon as embryos hatch and develop into alevins, fry, and smolts prior to out-migration to ocean waters. Generally, it is typical to observe the following time periods for each salmon species in freshwater prior to oceanward migration: chinook, 0.5–2 years; sockeye, 1–3 years; coho, 0.5 to 1 year; pink, <0.5 year; and chum, <0.5 year. Thus, it is possible to observe the presence of chinook, sockeye, and coho juveniles year-round upstream and downstream of the KSEF, but less likely that pink and chum juveniles would be present.

Hydrological and water temperature data

The Kitwanga River is a nivo-pluvial stream, which means precipitation is dominated by snowfall (nival) in the winter and rainfall (pluvial) from spring through autumn. Nivo-pluvial streams have hydrographs with two peaks: snowmelt-driven spring freshet and rainfall-driven autumn high flows. To develop a clearer understanding of hydrology in the Kitwanga watershed, a monitoring station was established at the Kitwanga Adult Enumeration Facility (Figure 1). Hydrometric monitoring at the KSEF began in 2020. Ideally, each stage–discharge monitoring program receives between 5 and 8 field measurements annually, with a majority of those conducted during the open water season (May to October). This study developed a streamflow hydrograph from 2020 and 2021 based on 10 field measurements.

Measuring hydrology directly at the adult fence was complicated by the fact that the fence itself has a profound hydraulic effect on streamflow. To overcome these site constraints, two stations were installed on the KSEF itself: one immediately upstream of the fence and one immediately downstream of the fence (Figure 1). Each hydrometric station consisted of a barometer (Solinst Barologger, Solonist Canada Ltd., Georgetown, ON, Canada) and submerged pressure transducer (Solinst Levellogger, Solonist Canada Ltd.). Both barologgers and levelloggers record pressure and temperature every 10 min at identical intervals. Barometric pressure was used to correct the stage data (water pressure–air pressure) prior to stage–discharge calculations.

While stage was measured continuously, discharge was based on spot measurements conducted in the field and is not continuous data. To provide continuous stream discharge, it was necessary to develop a stage–discharge relationship between recorded stage data and measured discharge. This is referred to as the stage–discharge relationship or rating curve (ISO, 2020) and is used to convert continuous-stage data into a discharge time series or streamflow hydrograph. Manual river flow measurements were carried out at each site using one of two methods, depending on flow conditions. When the stream could be safely waded and flows were not overly turbulent, a handheld electromagnetic current velocity meter (Ott MF Pro; OTT HydroMet, Loveland, CO, US) was used. When the stream was too deep or swift to wade, or flows were highly turbulent, an Acoustic Doppler Current Profiler (RiverRay ADCP; Teledyne Marine RD Instruments, Poway, CA, US) was used. To establish and maintain vertical survey control at the hydrometric stations, a series (3) of semipermanent benchmarks were used to establish a local datum. During each stage–discharge measurement, a water level survey was performed to verify the water surface elevation within ±2 mm (British Columbia Ministry of Environment and Climate Change Strategy (BCMoE), 2018).

The rating curve (water stage–discharge relationship) was generated by a total of 13 stage–discharge measurements captured in 2020 and 2021 ranging between 4 and 60 m³/s. Because of the mid-channel bar which comes into play during high flows, it was discovered that during the high-flow seasons between mid-April and mid-July, the rating curve generated from data from two KSEF-managed hydrometric stations underestimates the overall flow rate of the river. As a result, the total number of stage–discharge measurements falls below 10–15, the recommended number of measurements to apply a robust relationship. Manning's formula was used to adjust the stream discharge relationship between the KSEF and Water Survey of Canada (WSC) hydrometric stations and generate the final hydrograph used to generate the hourly stream flow rates. The mean daily flow rate used in our eDNA data analyses (described below) was calculated by taking the mean flow rate measurements for 24 h beginning at 9:00 p.m. that day.

Collection of water samples for salmonid eDNA analyses

The study design included collecting triplicate Kitwanga River water samples (1 L volume each) daily. Sampling occurred ~20 days prior to, during, and ~20 days after up-migration of salmonids at the KSEF to capture a period of time that would result in low, high, and medium abundance of eDNA, and when conventional fish counts are simultaneously occurring. The water samples for this study that were subsequently processed to quantify salmonid eDNA were collected at a single sampling site on the Kitwanga River that was located 50 m upstream on river's left (when looking downstream) of the KSEF fish fence (Figure 1) where conventional counts were performed. This river water sampling site was located upstream of a small side channel that is typically dry during the majority of salmon migratory period. It should be noted that pink salmon are known to spawn in the section of river where water samples were collected. The triplicate 1-L river water samples were collected daily at the sampling site between 8:00 a.m. and 2:00 p.m. during the salmon pre- (June 21 to July 18, 2021) and post-migratory (October 28 to November 7, 2021) periods and between 8:00 p.m. and 9:00 p.m. during the salmon migratory period (between July 19, and September 20, 2021). All equipment (sample bottles, tools, etc.) in contact with each sample were cleaned daily by a minimum of 1-min soak in 50% bleach and 50% deionized water followed by a rinse with tap water or distilled water. The 50% bleach solution was prepared fresh daily. River water samples were collected in 1-L PYREX™ bottles (CatID: 06-414-1D, Corning Life Sciences—Axygen Inc., CA, USA) while standing on the shore downstream of the water collection site by submerging the 1-L bottle beneath the water surface with the opening facing upstream and filled to the 1 L mark. River water samples were then sealed with a cap, sprayed with 50% bleach, and placed in a cooler with ice. During each daily water sample collection event, one negative eDNA control was prepared after the collection of triplicate river water samples by pouring 1 L of commercially prepared distilled water into a 1-L PYREX collection bottle. This negative control distilled water sample was then handled and processed using the same methodology as the river water samples. The triplicate river water samples and single negative control were taken to a dry lab and filtered within 6 h of collection using a MasterFlex L/S® portable sampling (peristaltic) pump (CatID: RK-07571-02, Cole-Parmer Canada Company, Quebec Canada) through a Nalgene™ single-use, analytical, 0.45 μm cellulose nitrate filter funnel (CatID: 145-2045, Thermo Fisher Scientific, Waltham, MA, USA) attached to a 1.7-L Büchner flask (CatID: DS4101-2000, Thermo Fisher Scientific, Waltham, MA, USA). Pump and flasks were connected with size 24 MasterFlex tubing (CatID: HV-96410-24, Cole-Parmer Canada Company, Quebec Canada). Each filter was then removed from their sterile housing with forceps, folded in half, and placed into a paper envelope. If multiple filters were needed to process the entire 1-L water sample, each filter was stored in a separate envelope. For storage, two paper envelopes were sealed into a Ziplock bag along with one tablespoon of silica-dehydrating beads and placed in a freezer at −20°C. Samples were shipped on ice every 2 weeks from the Kitwanga dry lab to Simon Fraser University and stored at −20°C immediately upon receipt.

Environmental DNA isolation

Stringent prevention protocols were in place to prevent contamination of filters containing eDNA and reagents used during eDNA isolation procedures. DNA isolation procedures were performed in a HEPA-filtered and UV-irradiated biological safety cabinet (BioKlone A2, Design Filtration Microzone Inc, Ottowa, ON, Canada) sterilized with 50% bleach and 70% ethanol before each use. DNA isolations were performed in batches of 12 filters using a DNeasy Blood & Tissue kit (QIAGEN Inc., Mississauga, ON, Canada) according to the manufacturer's instructions with minor modifications. Specifically, to scale up the isolation to account for using the whole 47-mm-diameter filter, buffer ATL and proteinase K volumes were increased to 1120 and 80 μL, respectively, for the initial incubation. To dislodge DNA from the filter, the samples were then vortexed for 30 s once lysis buffers were added, and the initial incubation was increased to 48 h with buffers always covering the filters. Once incubation was completed, the filters and buffers were processed using a QIAshredder column (QIAGEN inc., Mississauga, ON, Canada) according to manufacturer's instructions with minor modifications. Specifically, loading 600 μL and half a filter at a time and an additional 30-min, 5000×g centrifugation was added after samples were processed in the QIAshredder columns to pellet sediment from the water samples collected on the filters was performed. In addition, the QIAshredder column buffers AL and ethanol volumes were increased to 600 μL and the final elution volume was 100 μL. For water samples that were processed using multiple filters, all filters for one water sample were processed using the same QIAshredder and DNeasy columns. In addition, each week of extractions also included an eDNA isolation blank using a unprocessed 0.45 μm sterile cellulose nitrate filter (47 mm diameter), the same filter type used to collect eDNA from river water samples. All eDNA isolations were completed within 1 month of filter arrival at Simon Fraser University and were performed chronologically, such that storage length did not vary more than 7 days between samples. Sample order was then randomized using a random number generator to blind the qPCR tests.

Design and validation of targeted eDNA qPCR assays

An eDNA assay for detection of coho salmon assay, eONKI4, was developed and validated previously as described in Hocking et al. (2022) and amplifies a 249 bp portion of the mt-nd1 gene. Herein, we present the design and validation of additional sensitive eDNA assays for specific detection of pink, sockeye, and chinook salmon (eONGO5, eONNE2, and eONTS5, respectively). We developed qPCR primers and TaqMan probes (Table 1) using mitochondrial gene sequences obtained from the National Center for Biotechnology Information database (https://www.ncbi.nlm.nih.gov). We collected gene sequences for all salmon as well as sympatric salmonids in the genus Oncorhynchus, and more distantly related taxa including humans. Using ClustalW (http://www.genome.jp/tools-bin/clustalw) to align these sequences, we determined where the sequences share identity. Using the output .aln file with BioEdit (Ibis Biosciences, Carlsbad, CA, USA) and Primer Premier Version 6 (Premier Biosoft, Palo Alto, CA, USA), we designed primers and probes for each salmon species targeting unique regions where there was limited identity to the other species in the alignment (Figures S1–S3). We designed multiple primer candidates and tested for sensitivity and specificity to the target DNA, and the successful primer/probe combinations are shown in Table 1 and their alignments are shown in Figures S1–S3. Primers and probes containing a 5′ FAM reporter dye and 3′ ZEN/Iowa Black FQ quencher were purchased from Integrated DNA Technologies (IDT; Coralville, IA, USA).

TABLE 1 Nucleotide sequences for the qPCR-based eDNA assays for pink (eONGO5), sockeye (eONNE2), and chinook (eONTS5) salmon comprised of primers and a probe. The amplicon sequence for the creation of the synthetic DNA sequence is included for each.

Assay	Gene	Sequence type	Sequence (5′ → 3′)
eONGO5	mt-nd1	Forward primer	CAAGCCTGGCAGTATATTCC
		Reverse primer	ACCAGTAGTCAGATGCTTTCC
		Probe	FAM-CATTATCACAGGAGGATTCACTCTTCAAAC-ZEN/Iowa Black FQ
		Amplicon	CAAGCCTGGCAGTATATTCCATCCTGGGGTCAGGGTGAGCCTCAAACTCTAAGTACGCTCTAATCGGAGCCCTCCGAGCAGTAGCACAAACCATCTCCTACGAAGTCAGCCTAGGATTAATCTTACTTAGCGTGATCATTATCACAGGAGGATTCACTCTTCAAACCTTCAACGTAGCCCAGGAAAGCATCTGACTACTGGT
eONNE2	mt-cyb	Forward primer	TGAGGACAAATATCCTTCTGG
		Reverse primer	GCAATGACGAAAGGGAAC
		Probe	FAM-TCCGTACGTCGGGGGCGCC-ZEN/Iowa Black FQ
		Amplicon	TGAGGACAAATATCCTTCTGGGGGGCCACTGTAATTACAAACCTTCTTTCCGCTGTTCCGTACGTCGGGGGCGCCCTGGTACAATGAATTTGAGGCGGATTCTCCGTTGACAACGCCACTTTGACACGATTTTTCGCCTTTCACTTCCTGTTCCCTTTCGTCATTGC
eONTS5	mt-nd1	Forward primer	ACTATTTATTAAAGAACCCGTC
		Reverse primer	GCTCAGCCTGATCCCAA
		Probe	FAM-TCCTGTTACAGATCTTAACCTAGGGGTAC-ZEN/Iowa Black FQ
		Amplicon	ACTATTTATTAAAGAACCCGTCCGACCTTCCACCTCTTCCCCCTTTCTATTCCTCGCCACACCCATACTTGCCCTTACACTTGCACTCACTCTTTGAGCCCCAATACCTATTCCTTATCCTGTTACAGATCTTAACCTAGGGGTACTATTTGTACTTGCACTATCTAGCCTANCCGTTTATTCTATCTTGGGATCAGGCTGAGC

Note: The assay details for eONKI4 are available in Hocking et al. (2022).

Abbreviations: FAM, reporter dye; ZEN/Iowa Black FQ, quencher.

We obtained specimens of target and sympatric species from grocery stores and DFO Chilliwack River Hatchery under Animal Use Protocol PPACC-1704 and isolated their total DNA. Genomic DNA isolations from each salmonid species tested in the present study for assay validation were isolated from fin clips. We also isolated human DNA from a HEK293 cell line (American Type Culture Collection (ATCC, Manassas, VA) Catalog number CRL-1573) under the University of Victoria Biosafety permit #96876-028. We used the DNeasy Blood and Tissue Kit (QIAGEN Inc., Mississauga, ON, Canada; Cat# 69506) to isolate total DNA from these samples by following manufacturer's protocol and eluted in 50 μL EB buffer. Extracted genomic DNA was used to determine the specificity and sensitivity of the designed primer sets using a CFX96 qPCR machine (Bio-Rad Laboratories, Hercules, CA, USA).

Each salmonid qPCR amplification reaction well contained 2 μL of an eDNA sample and 13 μL of qRT-PCR master mix for a final concentration of: 10 mM Tris–HCl (pH 8.3 at 20°C); 50 mM KCl; 3 mM MgCl₂; 0.01% Tween 20; 0.8% glycerol; 69.4 nM ROX (Life Technologies, Burlington, ON, Canada); 10.5 pmol of forward and reverse PCR primer; 1.5 pmol of TaqMan hydrolysis probe; 200 μM dNTPs (FroggaBio Inc., North York, ON, Canada); and 1 unit of Immolase DNA polymerase (FroggaBio). qPCR tests were performed under the following cycle conditions: 95°C for 10 min, followed by 50 cycles of 95°C for 15 s, 64°C for 30 s, and 72°C for 30 s.

To determine specificity, we first tested assays against 10 pg genomic DNA of target salmon, and other species listed in Tables S1–S3, with two technical replicates using a SYBR (Invitrogen, Carlsbad, CA, USA) qPCR assay and agarose gel visualization of the amplified product (amplicon) for the desired species and absence of the target amplicon in all non-target species. After confirming the expected amplicon, we combined the primers with the corresponding designed TaqMan hydrolysis probe candidates and tested on total DNA from all species in Tables S1–S3 (25 technical replicates). To confirm the ability of each assay to detect their respective target species DNA, we used genomic DNA from five individual specimens of each target species for validation. All qPCR runs were scored using a standardized cycle threshold of 50 relative fluorescence units. If amplification below the threshold was detected within 50 cycles (i.e., <50 C_t), then we scored that replicate as a positive hit.

To generate standard curves, a fivefold serial dilution series of synthetic double-stranded amplicon (gBlocks®) was tested from 62,500 to 0.032 copies/reaction for each assay plus no template controls (NTCs) using the procedures described previously (Hocking et al., 2022; Lesperance et al., 2021; Matthias et al., 2021; Robinson et al., 2022). Briefly, for >100 copies/reaction, eight technical replicates were run, and for concentrations ≤100 copies/reaction, including NTCs, 24 technical replicates were run.

Two standard curves were generated for each assay: one for higher copy number and the other for low copy number. The first method uses a linear regression of continuous data for samples with 100% hits (Klymus et al., 2019). The limit of quantification for this continuous data (LOQ_continuous) is defined as the lowest copy number per reaction where ≥95% technical replicates amplified and serves as the breakpoint for continuous and binomial data (Lesperance et al., 2021). For samples, where there were less than 100% hits, a binomial Poisson distribution method was used to generate a standard curve for low copy number reactions (Lesperance et al., 2021). This “eLowQuant” method (Lesperance et al., 2021) was performed using publicly available R code (https://github.com/mlespera/eLowQuant) to calculate the limit of detection (LOD) and LOQ for each assay when there are n = 8 technical replicates (Table 2). The limit of blank (LOB) was zero (no intercept model) (Hocking et al., 2022) as no amplification was observed in the NTCs in all cases in these experiments.

TABLE 2 Limit of detection (LOD), limits of quantitation (LOQ) for the four assays used in the present study as calculated from the Binomial data using eLowQuant, and a no intercept model for n = 8 technical replicates where LOB = 0 (Lesperance et al., 2021).

Target species	Assay name	Binomial data						Continuous data
		LOD (c/rxna)	LOD 95% CI lower	LOD 95% CI upper	LOQ (c/rxn)	LOQ 95% CI lower	LOQ 95% CI upper	LOQcontinuousb (c/rxn)	Slope	Y-intercept	R2 value	% efficiency
Pink salmon	eONGO5	2.5	1.8	4	9.4	6.8	15.2	100	−3.6	43.1	0.9997	88
Coho salmon	eONKI4	0.4	0.3	0.7	1.6	1.2	2.5	20	−3.7	38.8	0.9967	87
Sockeye salmon	eONNE2	0.7	0.5	1.1	2.5	1.8	4.2	4	−3.4	39.0	0.9989	96
Chinook salmon	eONTS5	0.5	0.4	0.9	2	1.4	3.3	4	−4.1	42.6	0.9993	76

Note: The linear regression results were generated using C_t values at concentrations at the LOQ_continuous and above. The validation data for the eONKI4 are available in Hocking et al. (2022).

Abbreviation: CI, 95% confidence interval.

a

Copies/reaction.

b

Lowest copy number with ≥95% hits.

Environmental DNA qPCR analysis

All qPCR reagent preparations and loading of eDNA samples into qPCR assay wells were performed in a HEPA-filtered and UV-irradiated biological safety cabinet (BioKlone A2, Design Filtration Microzone Inc, Ottawa, ON, Canada) sterilized with 50% bleach and 70% ethanol before each use. All qPCR assays were performed on a Bio-Rad CFX384™ Real-Time PCR Detection System and CFX Manager™ Software in Hard-Shell 384-well plates (Bio-Rad Laboratories, Hercules, CA, USA).

Each eDNA sample was first tested for the presence of plant eDNA using the IntegritE-DNA™ quality control test (Hobbs et al., 2019; Veldhoen et al., 2016). The objective of this initial quality control step is to eliminate mistaken assignment of a negative result due to the presence of inhibitors or degradation of DNA in the eDNA sample. If ≤2/4 technical replicates had a quantification cycle (C_q) of <30, then the sample failed the IntegritE-DNA™ test; these samples were diluted 1:1 in molecular grade water and tested again, and any that still failed were omitted from subsequent testing. If >2/4 technical replicates had a C_q of <30, then the sample passed and these samples proceeded to salmonid qPCR assays.

All salmonid qPCR assays were performed with eight technical replicates of each eDNA sample, and each plate included four 1200 copies/μL gBlock® positive controls that were physically isolated from eDNA samples by 19 NTCs. gBlocks® were resuspended and diluted in 100 ng tRNA/μL TE buffer (QIAGEN Inc., Mississauga, ON, Canada). In addition, four intra-assay positive control wells were tested in the lower right quadrant of each 384-well plate separated by six negative controls. The intra-assay control was comprised of one known positive eDNA field-collected sample located on the right lower quadrant of the 384-well plate to determine if any degradation of reagents or DNA occurred. Standard curve and positive control wells were loaded last in a separate fume hood (HH Hawkins Ltd., Surrey, BC, Canada) to prevent contamination of the sample preparation area. Each assay was run under the conditions described in Section 2.5.

The C_q values for the eight technical replicates per water sample derived from the Bio-Rad CFX384™ Real-Time PCR Detection System and CFX Manager™ were converted to estimated copy number ($\widehat{S}$) per eDNA sample using the standard curves described in Section 2.5.

For water samples where there were eight of eight hits, calculation of estimated copy number/reaction was achieved using linear regression using the parameters given in Table 2. The mean $\widehat{S}$ was obtained and then multiplied by the dilution factor of 50 to obtain the copy number of each individual eDNA water sample in copies/L. Standard deviations were calculated for scenarios when eight of eight technical replicates exhibited C_q values and those that exceeded a standard deviation (SD) of 0.5 were omitted according to similar limits set by Lacoursière-Roussel, Côté, et al. (2016) and Karlen et al. (2007). Details of these numbers are presented in Section 3.4. The dilution factor was calculated by the reduction of a full 1-L water sample onto a single filter, which is eluted into 100 μL, of which only 2 μL is run in each well.

For samples where there were less than 100% hits, the eLowQuant code (Lesperance et al., 2021) was used to calculate $\widehat{S}$ and 95% confidence intervals (CI) for each technical replicate. The copy number per reaction was multiplied by a dilution factor to obtain the copy number of each individual eDNA water sample in copies/L.

Correlational analysis of conventional salmonid counts to salmonid eDNA copy number

Mean daily eDNA rates of the triplicate water samples collected daily that underwent eDNA isolation procedures were calculated by taking the average eDNA copy number (i.e., $\widehat{S}$, copies) of these three samples obtained in the qPCR assays for each species. Mean flow-adjusted eDNA rates for each species in copies/L · ft³/s were calculated by multiplying the mean daily streamflow rate with mean daily $\widehat{S}$.

We ran quasi-Poisson generalized linear models in R version 4.2.0. Specifically, we modeled daily fish counts as a function of mean daily flow-adjusted eDNA (log transformed) using a quasi-Poisson GLM with a log-link function.

RESULTSeDNA assay performance results

The best assays chosen for detection of pink, sockeye, and chinook salmon were named eONGO5, eONNE2, and eONTS5, respectively (Table 2). The full characterization of the assay for coho salmon, eONKI4, has been already published by Hocking et al. (2022), including specificity and sensitivity information, and data for this are shown in Table 2. Each assay demonstrated complete specificity to their target species when tested against target and sympatric specimen DNA (Tables S1–S3). The LOD of eONGO5 is 2.5 copies/reaction [95% confidence interval (CI): 1.8–4], and the LOQ was 9.4 copies/reaction (95% CI: 6.8–15.2; Table 2). The LOQ_continuous of eONGO5 is 100 copies/reaction. For technical replicates with >95% detections where a linear regression of C_t values relative to copy number is appropriate, the amplification efficiency of eONGO5 is 88% with an equation of the line of: y = −3.64x + 43.1 (Table 2; Figure S4). The LOD of eONNE2 is 0.7 copies/reaction (95% CI: 0.5–1.1), and the LOQ was 2.5 copies/reaction (95% CI: 1.8–4.2; Table 2). The LOQ_continuous of eONNE2 is 4 copies/reaction. For technical replicates with >95% detections where a linear regression of C_t values relative to copy number is appropriate, the amplification efficiency of eONNE2 is 96% with an equation of the line of: y = −3.4x + 39 (Table 2; Figure S5). The LOD of eONTS5 is 0.5 copies/reaction (95% CI: 0.4–0.9), and the LOQ was 2 copies/reaction (95% CI: 1.4–3.3; Table 2). The LOQ_continuous of eONTS5 is 4 copies/reaction. For technical replicates with >95% detections where a linear regression of C_t values relative to copy number of eONTS5 is appropriate, the amplification efficiency of eONTS5 is 76% with an equation of the line of y = −4.1x + 42.6 (Table 2; Figure S6). Sequence data of all three assays are specified in Table 1, and alignments of assay components with sympatric species are shown in Figures S1–S3.

Hydrological and water temperature results

During the field measurements, a mid-channel bar was discovered along the discharge transect as waters receded in the summer. The presence of this bar presented complex channel geometry and hydraulics, which could be challenging for the development of a streamflow hydrograph; however, mid-way through the Kitwanga River eDNA Project, the WSC (Figure 1) established a hydrometric monitoring station ~75 m downstream which provided a hydrograph comparison. Thus, Manning's formula was used along with a discharge–discharge relationship between KSEF and WSC station data to adjust the KSEF streamflow and derive the final hydrograph (Figure 2a). In total, 10 stage–discharge measurements were captured in 2021 with three additional measurements captured in 2020. These were collected between July and October in both years and represent flows between 141 and 2119 ft³/s (4–60 m³/s). However, outside of the main peaks in stream flow/stage–discharge that occurred during the salmon pre-migration period (i.e., prior to Julian date ~200), no major stage–discharge peaks were reported during the salmon migration and post-migration period (Figure 2a). Indeed, 60% of the stage–discharge measurements were in the moderate-flow range from 211 to 424 ft³/s (5.97–12.0 m³/s, Figure 2a), with an overall rating curve R² of 0.91 (data not shown).

FIGURE 2. Daily averaged (a) hydrological streamflow (cubic feet per second, cfs) and (b) water temperature (°C) data obtained from all hydrometric sites in the Kitwanga River outlined in Figure 1. Measurements of pressure and temperature were recorded at 10-min intervals and were converted to flow rate using a rating curve. Daily average flow rate and temperature were calculated by taking the mean during a 24-h period from 9:00 p.m. that day.

During the migratory period encompassing all species (July 24 to September 20, Julian days 205–263), streamflow was relatively stable ranging from 217 to 379 ft³/s (6.14–10.7 m³/s), and water temperatures ranged from 7.8 to 17.2°C (Figure 2b).

Conventional salmonid counts at KSEF

Details of the total counts of salmon collected by fisheries staff in the present study at the KSEF in 2021 are summarized in Table 3. In 2021, the KSEF was operated from July 12 until September 20. Of all salmon that returned in 2021, 98.8% of them were adult pink salmon (179,967 counted in total). Of the remaining 2130 non-pink salmonids counted, 49.9% were coho (1063 adults), 31.2% were chinook (544 adults and 120 jacks), 9.58% were sockeye (204 adults), and 9.34% were chum (199 adults). The KSEF was operating on June 21, 2021 (Julian day 172), and captured a pre-migration period for all salmon investigated that ended on various dates when the first salmon species were counted, which also defined the beginning of the up-migration period (Table 3). Due to heavy rains and flooding, the fence had to be decommissioned/opened on September 21, 2021 (Julian day 264), thus no human counts occurred from this date on. However, based on historical records, >90% of the chum and pink salmon runs were completed by September 21. For chinook, the post-migration period was evident by September 11 based on no chinook crossing the KSEF from September 10 to September 20.

TABLE 3 Total salmon counts at the Kitwanga River Salmon Enumeration Facility on the Kitwanga River while in operation from July 12 to September 20, 2021.

Species	Total salmon counts KSEF	Pre-migration period (Julian date)	Up-migration start date (Julian date)	Post-migration period (Julian date)
Pink	179,967	June 21 to July 26 (172–207)	July 27 (208)	October 28 to November 7 (301–311)
Coho	1063	June 21 to August 11 (172–223)	August 12 (224)	October 28 to November 7 (301–311)
Chinook	664 (544 adults, 120 jacks)	June 21 to July 23 (172–204)	July 24 (205)	September 10 to September 20 and October 28 to November 7 (253–264 and 301–311)
Sockeye	204	June 21 to July 30 (172–211)	July 31 (212)	October 28 to November 7 (301–311)
Chum	199	June 21 to August 5 (172–217)	August 6 (218)	October 28 to November 7 (301–311)

Note: The up-migration start date refers to the first day a salmon species was counted at the fish fence and the post-migration period refers to the day after the last siting of a species crossing the fence. Due to heavy rains, the fish fence was not operating after September 21, thus no human fish counts were performed from this date forward. Based on historical records, >90% of the coho, sockeye, and pink salmon runs were completed by September 21. For chinook, the post-migration period was evident by September 11 based on no chinook crossing the KSEF from September 10 to September 20.

Correlation between salmon counts and eDNA

To quantify salmonid eDNA in the Kitwanga River upstream of the KSEF where fish were counted as they passed through the fence, three 1-L water samples and one field blank were collected daily and isolated eDNA from each water sample was tested first with the IntegritE-DNA™ test. Of all 103 sampled days, 2 full days of samples were omitted due to failing the IntegritE-DNA™ test. The water samples for one of these days were filtered 24 h after collection due to a power outage, exceeding the water collection to filtration period of 6 h or less that was used for all other water samples. The reason for the second sample that failed the IntegritE-DNA™ test is unknown.

The remaining samples were tested with the four fish qPCR assays using eight technical replicates (Figures 3–6). For continuous data where there were 8/8 hits, the SD range and percentage of technical replicates removed out of 309 water samples for each species are as follows: pink salmon 0–0.62, 6.51%; coho 0–1.11, 11.77%; chinook, 0–1.76, 5.34%; and sockeye, 0–0.58, 7.93% (Table 4). The precision of the three 1-L water samples is best described by the median and mean coefficient of variation (CV) of 309 water samples evaluated for each species and is as follows: pink 0.207 and 0.354; coho 0.424 and 0.487; chinook 0.862 and 0.755; and sockeye 0.755 and 0.717. Thus, precision in eDNA concentrations in triplicate water samples collected daily at one location was highest for the most abundant salmon species passing through the KSEF (i.e., 179,167 adults), and precision between triplicates decreased as the number of individuals for a given species decreased. All field blank water samples and the eight technical replicates per sample tested in qPCR assays did not have detectable eDNA for any of the salmon species investigated.

FIGURE 3. Comparison of Kitwanga River Salmon Enumeration Facility pink salmon counts to salmonid eDNA copy number in triplicate 1-L water samples collected daily upstream of the fish fence at this facility. These plots describe data collected from June 21 to September 20, 2021, and from October 28 to November 7, 2021, for pink salmon, including (a) daily human-performed adult fish counts at fish fence; (b) non-water flow-adjusted qPCR estimated eDNA copy number (Ŝ, copies/L); and (c) the flow-adjusted estimated qPCR eDNA copy number derived by multiplying the mean daily water flow rate by eDNA copy number of each triplicate daily water sample (copies/L · ft3/s). Error terms represent standard error of the mean of eight qPCR technical replicates for each water sample.

TABLE 4 Summary of standard deviation range (SD) of 309 water samples tested in eDNA qPCR assays (eight technical replicates per sample) for four salmonids, % of replicates omitted from final eDNA quantification analyses, and the precision of the three 1-L water samples collected daily based on the coefficient of variation (CV) observed for each species.

Salmonid species	SD range between 8 technical replicates per water sample	% Replicates removed (of 2472)	Mean CV between 309 triplicate daily 1-L water samples	Median CV between 309 triplicate daily 1-L water samples
Pink	0–0.62	6.51	0.207	0.354
Coho	0–1.11	11.77	0.424	0.487
Chinook	0–1.76	5.34	0.862	0.755
Sockeye	0–0.58	7.93	0.755	0.717

Note: For continuous data where there were eight of eight hits, the SD range and percentage of technical replicates removed out of 309 water samples for each species are presented. For these eight technical replicates per sample in each qPCR assay, the criteria for acceptable precision was a standard deviation (SD) <0.5, thus replicate(s) with the most extreme values were removed one by one until this criterion was satisfied. For samples where there were less than eight of eight hits, the eLowQuant code (Lesperance et al., 2021) was used to calculate copy number and 95% confidence intervals (CI) for all eight technical replicates.

Of the salmonids investigated in the present study, ~180,000 pink salmon were enumerated at the KSEF, and the daily eDNA rates and flow-corrected eDNA rates appeared to reflect this local signal closely, but this was not the case for sockeye, coho, and chinook (Figures 3–6). For these latter species, the number of fish that crossed the fish fence was much lower ranging from ~200 to 1100. The noted stable streamflow is also reflected in the unadjusted eDNA rates and flow-corrected eDNA rates exhibiting a very similar pattern over time for each salmonid species investigated (compare Figure 2a to Figures 3–6).

For all salmon species investigated, eDNA was detected prior to each species crossing the fish fence. In the case of pink salmon, there were only three instances on July 28, 29, and 30 (Julian dates 209, 210, and 211) where detectable levels of pink salmon were observed in eDNA samples prior to the counting of five pink salmon at the fish fence on August 1 (Julian day 213; Figure 3). However, each of the three water samples collected on these 3 days when tested in qPCR assays using eight technical replicates per water sample exhibited detectable eDNA in ≤2/8 technical replicates, indicating very low, albeit detectable rates of eDNA using the eLowQuant eDNA estimation methodology. For comparison, the instances whereby detection of eDNA in triplicate daily water samples occurred before fish crossed the fence was 39, 27, and 33 days for coho, chinook, and sockeye. Detection of salmon eDNA prior to them being conventionally counted could be due to juveniles of these species present upstream of the KSEF in the Kitwanga River. In particular, for coho, the flow-adjusted eDNA rates were measurable in water samples collected during the entire pre-migratory period (Julian date 173 to 223; mean of 16,549 copies · ft³/s) and were similar to those measured during the migratory period except for September 9 (Julian date 252) where values for all three water samples exceeded 180,000 copies · ft³/s (Julian dates 224–263; Figure 4). Further support of coho juveniles present in the Kitwanga River upstream of the KSEF is evident based on eDNA detected at high levels for this species during the entire post-migration period. GFA technicians have also captured coho juveniles upstream from the KSEF, using baited gee traps and a rotary screw trap.

FIGURE 4. Comparison of Kitwanga River Salmon Enumeration Facility coho salmon counts to salmonid eDNA copy number in triplicate 1-L water samples collected daily upstream of the fish fence at this facility. These plots describe data collected from June 21 to September 20, 2021, and from October 28 to November 7, 2021, for coho salmon, including (a) daily human-performed adult fish counts at fish fence; (b) non-water flow-adjusted qPCR estimated eDNA copy number (Ŝ, copies/L); and (c) the flow-adjusted estimated qPCR eDNA copy number derived by multiplying the mean daily water flow rate by eDNA copy number of each triplicate daily water sample (copies/L · ft3/s). Error terms represent standard error of the mean of eight qPCR technical replicates for each water sample.

Chinook salmon was the only species to have its entire migratory period captured since human counts dropped to zero by September 10 (Julian date 253) and KSEF was in operation until September 21 (Figure 5). Chinook eDNA was detected 3 days into the pre-migration period (i.e., on June 23, Julian date 174) through to the start of the up-migration period (i.e., July 24, Julian date 205). However, a steep spike in eDNA rate is evident in the pre-migration period beginning on July 15 (Julian date 196), which was 8 days before the first chinook crossed the fence. Chinook eDNA remained detectable throughout the up-migration period and in 15 out 28 of the post-migration period water samples, albeit closer to the LOD in this latter period, and likely represents juvenile chinook.

FIGURE 5. Comparison of Kitwanga River Salmon Enumeration Facility chinook salmon counts to salmonid eDNA copy number in triplicate 1-L water samples collected daily upstream of the fish fence at this facility. These plots describe data collected from June 21 to September 20, 2021, and from October 28 to November 7, 2021, for chinook salmon, including (a) daily human-performed adult fish counts at fish fence; (b) non-water flow-adjusted qPCR estimated eDNA copy number (Ŝ, copies/L); and (c) the flow-adjusted estimated qPCR eDNA copy number derived by multiplying the mean daily water flow rate by eDNA copy number of each triplicate daily water sample (copies/L · ft3/s). Error terms represent standard error of the mean of eight qPCR technical replicates for each water sample.

The total number of sockeye that crossed the fish fence at the KSEF was the lowest of the species we measured, with just 204 adults (Figure 6). Sockeye eDNA rates were highest between July 13 and August 3 (Julian days 194 and 215) which encompasses most of the pre-migratory period which was June 21 to July 30 (Julian days 172–212), and 3 days of the migratory period that started July 31 (Julian date 212; Figure 6b). However, the highest eDNA rates fell on July 30 (Julian day 211), the day before the first fish was counted at the KSEF. There is also a very large spike in human counts near the final day of fish fence operation on September 19 (Julian day 263) that does not coincide with high eDNA measurements around this date. The post-migratory period is like chinook, where there is a very low detection of fish, with 13 of 28 water samples collected during this period exhibiting no eDNA (Figure 6a,b). Again, this data set supports the hypothesis that sockeye juveniles were present in the Kitwanga River upstream or near the eDNA sampling site. Interestingly, the peak of the eDNA copy number for sockeye on July 30 (Julian day 211) was 66,064 copies/L, which is more than 10-fold higher than those obtained for chinook and coho despite larger numbers of these two latter species counted. This may be evidence of different eDNA shedding rates between these species.

FIGURE 6. Comparison of Kitwanga River Salmon Enumeration Facility sockeye salmon counts to salmonid eDNA copy number in triplicate 1-L water samples collected daily upstream of the fish fence at this facility. These plots describe data collected from June 21 to September 20, 2021, and from October 28 to November 7, 2021, for sockeye salmon, including (a) daily human-performed adult fish counts at fish fence; (b) non-water flow-adjusted qPCR estimated eDNA copy number (Ŝ, copies/L); and (c) the flow-adjusted estimated qPCR eDNA copy number derived by multiplying the mean daily water flow rate by eDNA copy number of each triplicate daily water sample (copies/L · ft3/s). Error terms represent standard error of the mean of eight qPCR technical replicates for each water sample.

To test for statistical associations between observed fish counts and eDNA measurements, we ran generalized linear models in R version 4.2.0. We modeled daily fish counts as a function of mean daily flow-adjusted eDNA (log transformed) using a quasi-Poisson GLM with a log-link function. We found significant correlations for pink salmon (effect size = 0.762, SE = 0.046; $p<2{\mathrm{e}}^{-16},R^2=0.7738$), but weaker correlations for coho, chinook, and sockeye salmon ($\begin{array}{c}\text{effect size}=-0.343,p=0.114,R^2=0.01926;\text{effect size}=0.355,\hfill \end{array}$$\begin{array}{c}p=\mathrm{0.0.0363},R^2=0.07143;\text{effect size}=-0.0155p=0.885,R^2=0.05294\hfill \end{array}$, respectively; Table 5; Figure 7).

TABLE 5 Estimated effect size for each of two models across four species.

Species	Quasi-Poisson estimate SE		p-value	Estimate	Linear SE	p-value
Pink	0.762	0.046	<0.001	0.463	0.037	<0.001
Coho	−0.343	0.114	0.313	−0.077	0.055	0.162
Chinook	0.355	0.099	0.0363	0.164	0.040	<0.001
Sockeye	−0.0155	0.064	0.885	−0.013	0.023	0.576

Note: In both cases, flow-corrected eDNA scores (predictors) were log transformed. For linear models, species counts (response variable) were also log transformed. Models include data points from pre-, up-, and post-migration periods. Significant effects are bolded.

FIGURE 7. Results of the general linear model (red line)-relating log-transformed daily human fish counts and log-transformed water flow-adjusted estimated eDNA copy number (Ŝ) for (a) pink Salmon (p[less than]2.87e−06,R2=0.7738); (b) coho Salmon (p=0.313,R2=0.01926); (c) chinook Salmon (p=0.0363,R2=0.07143); and (d) sockeye salmon (p=0.855,R2=0.05294). Each eDNA copy number per liter estimate (Ŝ) is the mean Ŝ of all three triplicate water samples measured on that day. Data were x+1 transformed to ensure data with a value of 0 were included. The eDNA copy number of each water sample was derived using qPCR (eight technical replicates per sample) and was multiplied by mean daily water flow rate (water flow rate averaged over a 24-h period from 9:00 p.m. to 9:00 p.m.).

We also considered analyzing the pre-, post-, and up-migration periods analyzed on their own, in order to reduce the impact of high eDNA readings during periods with low fish counts such as in the coho post-migratory period. However, analyses on either the pre- or post-migration period, alone, turn out to be not sensible, as there is only a single non-zero $\mathit{\text{count}}$, even though eDNA measurements were taken (Counts ~ Flow-Adjusted eDNA).

Findings when considering only the up-migration period (Table S4, all conventional fish count data points vs. up-migration period only day shift $n=0$) are not dramatically changed for any species. Coho salmon showed a minor increase in effect size (a β-estimate of −0.433 compared to −0.1506) but the trend remains negative. Effect sizes in quasi-Poisson GLMs for pink, chinook, and sockeye fall from β = 0.7616 to β = 0.5978, β = 0.3548 to β = 0.09291, and β = −0.0155 to β = −0.0591, respectively. Effect sizes in the linear models with only up-migration period data points also show only a small change, going from β = 0.463 to β = 0.512, β = −0.0770 to β = +0.0685, β = 0.164 to β = 0.121, and β = −0.0127 to β = −0.0509 in pink, coho, chinook, and sockeye, respectively.

Removing the pre- and post-migratory periods changes effect sizes of the models depending on the level of eDNA signal observed in these periods. If the pre- and post-migratory periods contain many positive eDNA signals, but no observed fish, then removing them improves the predictive power of the model. Conversely, if these periods without fish comprise mostly null or background eDNA signals, removing them will decrease the model's predictive power (because their inclusion increases the strength of the association between counts and eDNA measures). This raises an interesting question, namely how many days of pre- and post- migration should be included?

Visually comparing the flow-adjusted eDNA rates and conventional fish counts reveals obvious peaks in both sets of data (Figures 3–6); however, these peaks are not perfectly synchronous. For example, for pink salmon, fish counts peaked prior to peak eDNA rates, early in the up-migration period. However, for coho, a peak in the eDNA rate occurs on Julian day 252 while the fish counts peak later on Julian day 256. The former observation suggests that there may be a lag between eDNA rates and fish counts, such that eDNA rates taken 1 or more days after a fish count is observed may best reflect that count. It is also possible that the eDNA collected on a given day may better correspond to the cumulative sum of fish counts or some other composite metric. To examine this, we considered two potential additional analyses. First, we considered grouping several days of collection data together and using the total or mean of these grouped measurements in our models (i.e., take the mean or total across Julian days 200–203 for both eDNA rate and fish counts). However, this option has several disadvantages: (1) a reduction in total sample number due to the binning of samples, which results in a loss of statistical power, and (2) a change in the original goal of our model which is to evaluate the suitability of eDNA as a proxy for daily fish counts. Thus, we opted not to take this approach. Alternatively, we considered shifting the flow-adjusted eDNA rate by n days (forward or backward) relative to counts. This still allows us to assess our main question, namely whether eDNA measurements correlate with daily up-migration fish counts, but possibly for counts taken 1 or more days after the fish pass the fish fence (i.e., fish count compared to eDNA rates from −3 ≤ n ≤ 0 days prior). One would not expect eDNA rates to be predictive of upcoming fish counts, but this could also be tested. We examined this second approach using both quasi-Poisson generalized linear models and linear models to quantify correlations between eDNA rates collected from 3 days before ($n=-3$) to 3 days after ($n=3$) the conventional fish counts (Table S4 for reference). To assess which of the day-shifted models provided the best model fit, we compared AIC values (Table S4 for reference). Here, we considered up-migration period only because including extended pre- or post-migration periods potentially biases the model, as described above. For all species except chinook, models with shifts largely agree on which time lag has the strongest predictive power. Indeed, it appears that a lag of 2 or 3 days improves model fit, albeit not dramatically. The 2-day lag for the pink salmon linear model, for example, shows an AIC reduction of almost 10 and improved R². While there can be some improvements to the model data fitness when adjusting for time lag as shown in the pink quasi-Poisson generalized linear model, coho and chinook models show improvements in p-value and β-estimate but remain statistically different. Sockeye also shows higher statistical significance, but the β-estimate remains negative. Interestingly, we did find statistically significant models with n = −3 and n = −2 time lags for chinook, with a larger β-estimate, but the AIC values suggest that these models are a poorer fit. Given the strong temporal autocorrelation present in our data, it is expected that patterns should be similar with both forward and backward time shifts. Ultimately, while shifting time scales by 2–3 days may change p-values and effect sizes slightly, the only species with strong correlations between eDNA rates and conventional fish counts is pink salmon in the present study.

DISCUSSION

The results of the present study showed that eDNA rates measured once daily were predictive of human salmon counts for the salmon species with large numbers of up-migrating adults (i.e., pink ~180,000), but exhibited little correlation for salmon runs with ~1000 or less individuals and daily counts of less than ~100 individuals (i.e., coho, chinook, and sockeye). Similarly, triplicate water samples appear to have acceptable precision for large salmon runs, but higher replication may be necessary for smaller salmon runs (i.e., ~1000 or less individuals). Undoubtedly, the large numbers of adult up-migrating pink salmon in the river relative to the numbers of the smaller coho, chinook, and sockeye salmon runs was a key driver of the higher precision between replicate daily water samples and higher accuracy in terms of correlation of eDNA rates to fish counts. However, one cannot rule out other confounding factors that may have influenced the daily eDNA rates of each species measured in the water samples collected at one location in this system. For example, the presence and distance of juvenile rearing sites and adult spawning sites relative to the eDNA collection site within the Kitwanga River are not fully understood but are known to vary across the species investigated, as does the duration of up-migration prior to spawning and timing of post-spawning death. These types of confounding factors related to distance between the eDNA shedding site and organism status (juvenile, live/dead adult ratios) have been shown to affect eDNA release rate, transport, degradation, and sediment settling and resuspension rates in previous studies (Andruszkiewicz et al., 2017; Barnes & Turner, 2015; Sassoubre et al., 2016; Yates et al., 2021). Nonetheless, these data demonstrate the importance of quantifying eDNA levels prior to adult spawning surveys to estimate the contribution of juvenile or residual eDNA from previous seasons versus adult spawner eDNA. In addition, for pink salmon, the daily eDNA rates were not consistently detected in qPCR assays until ~1000 fish crossed the fish fence during a 1-week period and then correlated with a gradual increase and an eventual decrease in fish abundance as the pink salmon run proceeded and then ended. Thus, for the large pink salmon run, eDNA rates appeared to reflect a local signal of salmon in space and time, essentially tracking these fish within days of passing through the eDNA sampling site.

The evidence for the presence of juvenile coho, chinook, and sockeye was strong based on measurable eDNA in water samples collected during the pre-migratory periods in this study, which likely reflects the freshwater residence time (>0.5 years) of juveniles for these species. However, for pink salmon, there were only three instances whereby eDNA analyses detected pink salmon near the LOD of the qPCR assay during the pre-migratory period, which corroborates the typically short river residence time (<0.5 years) of pink salmon juveniles. Nevertheless, it is also possible that some of this pre-migratory detection of eDNA for all species examined may have been due to late outward migrating fry, residual egg casings in the previous seasons' redds, resuspension of eDNA deposited into sediments from previous seasons or deposition of salmon upstream by birds or other animals. However, a similar field study reported the positive detection of sockeye eDNA prior to visual observations of adult spawners in Hansen Creek, southwest Alaska (Tillotson et al., 2018). Indeed, in the present study, the eDNA levels were higher during the pre-migratory period than during the migratory period for sockeye, coho, and chinook, and for coho, the post-migratory eDNA levels were also higher. Thus, for salmon spawning runs with less than ~1000 adults and daily counts less than ~100, the juvenile and/or prior seasons' eDNA signal appears to be indistinguishable from the adult spawning eDNA signal. We hypothesize that juveniles were main contributors to these pre-migratory period eDNA levels for these three species which could be examined in future studies by the inclusion of juvenile population surveys. Ultimately, the present study demonstrates that quantifying eDNA prior to Pacific salmon adult up-migration periods in a water body is critical in distinguishing the contribution of non-adult eDNA signals during adult enumeration studies.

In the present study, a significant positive correlation was observed between fish counts and eDNA counts during the pre-, up-, and post-migration periods for pink salmon only. Interestingly, despite the lack of correlation between conventional counts and eDNA rates for coho, chinook, and sockeye, daily eDNA rates tended to vary day to day during the pre-migratory and/or migratory period and did not appear to accumulate over time. This supports the results of several studies conducted on rivers that eDNA can be carried downstream but generally does not accumulate at downriver sites (Jerde et al., 2016; Pilliod et al., 2013b; Tillotson et al., 2018). Indeed, in the present study, a high accumulation of eDNA was not evident toward the end of the migratory period for pink salmon, or for the smaller coho, chinook, and sockeye runs. Furthermore, during the post-migratory period, the accumulation of eDNA for all salmon species investigated appears to be unlikely since the levels of eDNA were close to the LOD for all salmon species, except for coho. For coho, eDNA levels during the post-migratory period were similar to those observed during the pre- and up-migration periods. In this case, the higher post-migratory eDNA signal for coho may be due to this species' spawning period continuing relatively later (i.e., to the end of October) than the other species investigated (i.e., mid-September to mid-October), such that dead or live adults were prevalent at the onset of the post-migratory period for coho in addition to the higher probability of juvenile coho presence. We did not count dead salmon in the present study, but it is reasonable to hypothesize that the further into the migratory season the greater likelihood that natural post-spawning deaths would contribute to eDNA concentration in the river.

The pink salmon eDNA not only exhibited strong correlation to human counts (effect size = 0.447, SE = 0.046; $p<2{\mathrm{e}}^{-16},R^2=0.7738$) in the present study but also exhibited day-to-day variation and a unimodal profile rising and falling with human fish counts, which corroborates the two other salmon eDNA enumeration studies published to date. Tillotson et al. (2018) conducted a similar study measuring eDNA concentration every 24 h in a smaller, slower-flowing stream (i.e., ~3.5 to 17.7 ft³/s and ~0.10 to 0.50 m³/s) in Alaska, USA, but did not adjust eDNA levels to streamflows, used in-stream visual surveys, and counted total live and dead sockeye salmon during a spawning season that peaked at 2286 individuals. In addition, Tillotson et al. (2018) removed the dead sockeye from the stream after counting daily, which was not performed in the present study. Tillotson et al. (2018) reported that the sockeye eDNA concentrations roughly matched the temporal visual counts, although matched dead more closely than live, and in a unimodal fashion that increased as the spawning season progressed to a peak and then decreased as the spawning season gradually ended. Another coho and sockeye enumeration study by Levi et al. (2019) conducted human counts using the same methods as the present study (i.e., human counts at a permanent fish enumeration structure and no removal of dead fish) and quantified mean streamflow-adjusted eDNA levels. Levi et al. (2019) also demonstrated that during the up-migration period in Auke Creek (north of Juneau, Alaska), daily eDNA streamflow-adjusted rates closely tracked daily spawners and attenuated toward the end of the spawning season. The stream flow in Auke Creek was approximately one-tenth lower than the Kitwanga River flows measured in the present study, thus lower stream flows likely contributed to the stronger correlations between eDNA and human counts with even fewer individual fish in Auke Creek (i.e., >400 sockeye and > 100 coho) observed by Levi et al. (2019) compared to the present study (i.e., ~1000 pink salmon counted strongly correlated with eDNA signal). Ultimately, Tillotson et al. (2018), Levi et al. (2019), and the findings of the present study suggest a rapid production and degradation or transport of eDNA in a lotic system over a 24-h period, rather than the alternative hypothesis of accumulating and persistent eDNA over a longer timeframe. Thus, the eDNA profile in these salmon enumeration studies appears to have mirrored a local signal of salmon, tracking these fish within days of passing through the eDNA sampling site.

In addition to the large number of pink salmon up-migrating in the Kitwanga River relative to coho, chinook, and sockeye, the high correlation between pink salmon conventional counts and eDNA rates near the KSEF may also be influenced by distance to each species' spawning sites. Indeed, based on previous studies on Pacific salmonids in the Kitwanga River, pink and chum salmon have been shown to spawn in the lowest reaches of the mainstem closest to the KSEF and eDNA collection site (< ~5 km) as well as upstream ~20 km or greater along with coho, chinook, and sockeye. Clearly, the sheer abundance of pink salmon (i.e., >180 times) relative to coho, chinook, and sockeye makes it difficult to discern the impacts of spawning location on eDNA rates, but it is possible that the proximity of the eDNA water collection site to the pink salmon spawning sites may have inflated the pink salmon eDNA counts. Previous studies of eDNA in lotic environments have found a wide range of eDNA transport distances (e.g., <50 m up to ~12 km; Deiner & Altermatt, 2014; Jane et al., 2015; Pilliod et al., 2013b) and degradation rates (i.e., 50% degradation within 100 m in Wilcox et al., 2015). The study by Tillotson et al. (2018) is also of relevance, whereby daily visual in-stream counts of adult sockeye salmon were conducted along with simultaneous eDNA quantification in water (not flow rate adjusted) at various distances from the spawning site/eDNA source in Hansen Creek, Alaska, during the spawning season. Interestingly, Tillotson et al. (2018) reported no degradation of eDNA at ~30 m downstream of an eDNA source, but 50% degradation of sockeye eDNA at ~1.5 km downstream of an eDNA source in this lotic system with a streamflow rate ranging from 3.5 to 17.7 ft³/s (0.10 to 0.50 m³/s). Compared to the present study with much higher streamflow (211 to 424 ft³/s and 5.97 to 12.0 m³/s for 60% of stage–discharge measurements), this may imply faster transport of eDNA in the Kitwanga River. However, it is also possible that this was offset by a higher eDNA degradation rate due to the longer eDNA transport distances between the eDNA collection site and spawning sites for coho, chinook, and sockeye of ~20 km compared to the closer pink salmon spawning sites of <5 km. Future studies incorporating an estimate of salmonid eDNA transport and degradation rates between an eDNA collection site and spawning location are needed. Furthermore, determining the influence of genetic material release at spawning sites from live salmon (i.e., eDNA shedding rates for each species), dead salmon, and salmon gametes/eggs on live adult salmon eDNA levels downstream at a particular eDNA collection site are necessary next steps to further develop eDNA protocols for salmonid enumeration purposes.

Overall, the results of the present study generally concur with previous salmon enumeration studies comparing eDNA levels to human counts using similar sampling regimes in two other salmon-bearing waterbodies. However, we report little correlation between eDNA and human counts at a permanent fish fence for salmon runs with ~1000 or less individuals and daily counts of less than ~100 individuals (i.e., coho, chinook, and sockeye), while studies in Alaska by Levi et al. (2019) found strong correlations for up-migrating adult coho and sockeye runs in this size range as did Tillotson et al. (2018) for a sockeye salmon run of intermediate size (2286 adult fish). However, Tillotson et al. (2018) did examine other factors than simply the presence of live fish and reported that location of fish within the stream, live/dead ratio, and water temperature did affect eDNA rates over space and time. For the salmonids investigated in the present study in the Kitwanga River, these types of factors may account for some of the lack of correlation in eDNA rates to fish numbers in the smaller runs. In particular, understanding the distance of juvenile rearing sites and adult spawning sites relative to the eDNA collection site within the Kitwanga River may be necessary variables for statistical modeling and more accurate enumeration of salmon runs in this system. Collectively, these findings provide an important step toward the quantitative applications of eDNA for salmon enumeration, an area of strong potential for developing both the methods and monitoring objectives.

ACKNOWLEDGMENTS

We are greatly appreciative of the feedback and input from Mark Louie D. Lopez over the course of the editorial process. We thank Genome British Columbia (SIP#024) for funding this work.

FUNDING INFORMATION

Sector Innovation Program, Genome British Columbia, Grant/Award Number: SIP024.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

qPCR dataset is available on the Federated Research Data Repository at the link https://doi.org/10.20383/102.0727 or by email at [email protected].