We provide a novel method to improve the use of natural tagging approaches for subpopulation discrimination and source-origin identification in aquatic and terrestrial animals with a passive dispersive phase. Our method integrates observed site-referenced biological information on individuals in mixed populations with a particle-tracking model to retrace likely dispersal histories prior to capture (i.e., particle backtracking). To illustrate and test our approach, we focus on western Lake Erie’s yellow perch (Perca flavescens) population during 2006–2007, using microsatellite DNA and otolith microchemistry from larvae and juveniles as natural tags. Particle backtracking showed that not all larvae collected near a presumed hatching location may have originated there, owing to passive drift during the larval stage that was influenced by strong river- and wind-driven water circulation. Re-assigning larvae to their most probable hatching site (based on probabilistic dispersal trajectories from the particle backtracking model) improved the use of genetics and otolith microchemistry to discriminate among local breeding subpopulations. This enhancement, in turn, altered (and likely improved) the estimated contributions of each breeding subpopulation to the mixed population of juvenile recruits. Our findings indicate that particle backtracking can complement existing tools used to identify the origin of individuals in mixed populations, especially in flow-dominated systems.
Citation: Fraker ME, Anderson EJ, Brodnik RM, Carreon-Martinez L, DeVanna KM, Fryer BJ, et al. (2015) Particle Backtracking Improves Breeding Subpopulation Discrimination and Natal-Source Identification in Mixed Populations. PLoS ONE 10(3): e0120752. https://doi.org/10.1371/journal.pone.0120752
Academic Editor: Chih-hao Hsieh, National Taiwan University, TAIWAN
Received: July 28, 2014; Accepted: January 26, 2015; Published: March 23, 2015
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication
Data Availability: Data are available from Dryad under the DOI 10.5061/dryad.14625.
Funding: Funding for this research was provided by the Great Lakes Fishery Commission's Fisheries Research Program (http://www.glfc.org; to SAL) and the National Oceanic and Atmospheric Administration's Great Lakes Environmental Research Laboratory (http://www.glerl.noaa.gov; to SAL), with other in-kind support provided by the Ohio Department of Natural Resources–Division of Wildlife (http://wildlife.ohiodnr.gov). Funding for Open Access was provided by The Ohio State University Open Access Fund. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Identifying dispersal patterns and the degree of connectivity among local breeding subpopulations, as well as the natal source(s) of dispersed individuals in a mixed population, can improve our ability to understand the structure and dynamics of aquatic and terrestrial populations [1–2]. In turn, such understanding can have applied benefits. Most obvious has been the enhancement of species conservation strategies wherein knowledge of dispersal pathways and subpopulation connectivity has guided the placement of dispersal corridors and the design of reserve networks [3–4]. Knowledge of the compositional makeup of a mixed population also can improve the effectiveness of resource management [5–6] by allowing for the development of harvest allocation strategies that protect important breeding subpopulations  and by identifying underperforming subpopulations that might be in need of rehabilitation . Likewise, the identification of key or unexpected sources of individuals to a mixed population can help with the control or mitigation of native pests, invasive species, or disease [8–11].
To enhance our ability to address questions related to dispersal, population connectivity, and population structure and dynamics, much effort has been spent developing tools that can discriminate among local breeding subpopulations [12–13]. This need is paramount because mixing of local breeding subpopulations is common during one or more life stages for many species, including aquatic and terrestrial plants , invertebrate animals , and vertebrates  (but see also ). While artificial tags continue to offer a viable means to discriminate among breeding subpopulations in many systems, technological advances made during the past two decades have increased the use of natural tags .
To be a reliable natural marker, it needs to be consistently found in individuals originating from a particular location, must vary among breeding sources, and must remain unchanged during ontogeny. Otherwise, the link to the natal site would be lost [12–13]. Several types of natural tags have been explored as breeding subpopulation markers, including body structures (typically “hard parts”) that can record chemical or isotopic signatures from the environment in which an individual resides [13,17–20] and various genetic markers [13,19–20].
While the approaches described above have been used in solo and combination with varying levels of success, each is potentially limited by the fundamental assumption that the individuals used to develop functions that can discriminate among local breeding subpopulations originated at their collection site. For some organisms, this assumption may be quite realistic (e.g., organisms that breed in an upstream river that is far removed from the downstream mixing zone). For others, the possibility exists that an individual collected in one breeding location may not have originated there, but instead passively or actively dispersed to its collection location from a different source (breeding) location. Such a possibility seems especially plausible in situations in which a local breeding population reproduces in or near a mixing zone, or in cases where the early life stages of a species are highly susceptible to passive dispersal by physical forces such as wind or water circulation (e.g., some fishes, aquatic and terrestrial invertebrate animals, and plants). In these systems, failure to account for pre-collection dispersal could lead to sub-par discrimination abilities that might suggest a technique is not useful (when it actually might be). Or worse, the imprecise or inaccurate discrimination functions developed might provide misleading ecological insights that then drive the development of inappropriate management or conservation strategies.
Given these concerns, we suggest complementing natural tagging approaches with information on pre-collection dispersal, which can be obtained from particle backtracking (hindcasting) approaches wherein hydrodynamic or atmospheric models are used to recreate probabilistic dispersal histories prior to collection . Combining particle backtracking with natural tagging seems especially useful when considering that hydrodynamic and atmospheric models are becoming increasingly accurate and have been calibrated and validated for many ecosystems [22–24]. By integrating these approaches, the source-origin of individuals that are used to develop the breeding subpopulation discrimination and classification functions can be confirmed or revised as needed, thus eliminating the assumption of natural tagging approaches that the collection site of young individuals is equivalent to the natal (source-origin) site [18, 25–27].
Particle backtracking involves using water or wind circulation computer simulation models to retrace the most likely path a dispersed individual has taken [11, 21, 28]. Backtracking is currently most useful when individuals can be considered to be passively dispersed particles (e.g., small propagules such as eggs, seeds, or small larvae), as individual behavior can strongly influence dispersal and often is difficult to accurately include in a model [29–30]. Theoretically, when dispersal histories are combined with information from a natural marker and other relevant biological information (e.g., particle age), the reliability of initial assignment should improve, potentially leading to an increased ability to discriminate among breeding subpopulations and determine the natal origin of individuals in a mixed population.
Here, we provide an example of how particle backtracking simulations can improve the use of two commonly used natural markers—otolith micro-elemental composition and microsatellite DNA [31–32]—to discriminate among yellow perch (Perca flavescens) breeding subpopulations and determine their relative contribution of recruits to the age-0 juvenile stage of in the western basin of Lake Erie (USA-Canada), a system with strong river- and wind-driven water currents . Because our primary intention is to illustrate the potential of backtracking to improve discrimination and assignment capabilities rather than to precisely delineate yellow perch subpopulation structure, we used intentionally broad subpopulation assignments (i.e., north shore and south shore). Towards this end, we first use larval yellow perch collection location to develop functions to discriminate between our two geographically distinct breeding subpopulations . We then use these functions to assign (unknown-origin) juveniles from the mixed, open-lake population to one of the two breeding (hatching) locations. Finally, we show how combining probabilistic backtracking trajectories of larval yellow perch (particle) dispersal prior to their collection with knowledge of larval age (i.e., days post-hatch from otoliths) can be used to revise initial hatching locations that, in turn, 1) improve the consistency of our breeding subpopulation discrimination functions that are based on larvae and 2) alter (and thereby seemingly enhance) estimations of the relative contribution of individuals from each breeding subpopulation to the to the mixed population of (age-0) juveniles in late summer, which is a strong predictor of future population size at the time individuals recruit to the fishery at age-2 [34–35]. In so doing, we use a range of hatch-location assignment thresholds to test the sensitivity of both natural tagging approaches.
Materials and Methods
Study species and system
Lake Erie (USA-Canada) is a part of the Laurentian Great Lakes system, and is warmer, shallower (mean depth of western basin = 7.4 m), and more productive than the other Great Lakes . The western basin is a hydrodynamically active system, characterized by inflows from the Detroit and Maumee Rivers , large-scale circulation primarily driven by the Detroit River inflow and basin-wide winds , and the formation of a turbid river plume driven by the Maumee River inflow .
The yellow perch population in Lake Erie’s western basin contributes to economically important recreational and commercial fisheries  and appears to be supported by multiple discrete local breeding subpopulations [18, 27, 38–39]. Yellow perch larvae hatch at ∼5 mm total length (TL), are provided no parental care, and spend 30–35 d in the water column feeding on zooplankton before becoming demersal as juveniles at 20–25 mm total length (TL) [34, 40–41]. While the mixing of western basin north-shore and south-shore breeding subpopulations (i.e., potential stocks) is known to occur during the juvenile stage [27, 42], mixing of larvae has not been investigated. However, because yellow perch larvae are weak swimmers until ∼9.5 mm TL , we suspect that mixing of breeding subpopulations occurs during the larval stage, owing to the observed mixing of the Detroit River (north shore) and Maumee River (south shore) water masses during the larval production period [37, 44].
Larval yellow perch were collected weekly during late April through June 2006 and 2007 at up to 12 sites near the north shore (NS) and south shore (SS) of Lake Erie’s western basin (Fig. 1A, see also ). For the purpose of this study, a line at latitude 41.81° N (approximately the northernmost extent of the Maumee River plume) was used to divide the NS and SS spawning grounds. All larvae were collected with paired bongo nets (1 m diameter; 500 μm mesh and one with 1000 μm mesh) towed ∼2 m from the bottom of the lake to the surface (∼8 min tows). All larvae were euthanized with a lethal dose of tricaine methanesulfonate, then preserved in 100% ethanol until identification and processing. In total, 242 larvae were captured during 2006 (n = 151 in NS; n = 91 in SS) and a total of 364 larvae were captured during 2007 (n = 283 in NS; n = 81 in SS).
The dotted horizontal line in (a) marks the boundary between the north shore (NS) and south shore (SS).
Bottom-trawling surveys (10.7 m headrope, 13 mm cod-end liner, 3 km h−1 tow speed) for juvenile (age-0) yellow perch were conducted by the Ohio Department of Natural Resources-Division of Wildlife and Ontario Ministry of Natural Resources during late August, 2006 and 2007, at 71 and 73 sites, respectively (Fig. 1B, see also ), with trawl-site selection based on a stratified (by depth), random design . Juveniles were euthanized with a lethal dose of tricaine methanesulfonate and kept frozen until processing. We processed 119 and 167 juveniles during 2006 and 2007, respectively, with the number of individuals processed from each site determined proportionally from trawl catch per unit effort data.
During 2007, water samples were also collected at each site from 1–2 m depth for trace-elemental analysis. The samples were filtered through 0.45 μm nylon filters, acidified with nitric acid (1% of the total volume of water; 0.5 mL acid to 20 mL water), and analyzed using ICP-MS.
Larval and juvenile yellow perch were collected from public areas (Fig. 1) under the authority and with the assistance of the Ohio Department of Natural Resources-Division of Wildlife (Sandusky, OH) and the Ontario Ministry of Natural Resources (Wheatley, ON). No protected species were sampled. All work was approved by The Ohio State University’s Institutional Animal Care and Use Committee (Protocol number 2008A0056).
Larval otolith extraction and hatch date determination
Otoliths were used to estimate the hatch date of each larva collected in each breeding area. Briefly, both sagittal otoliths were removed from larvae with glass probes under a Class 100 clean hood, with the otolith used for aging being mounted to a glass microscope slide with Crystalbond 509 thermoplastic cement (Structure Probe, Inc., West Chester, PA) and otolith used in micro-elemental analysis mounted to a petrographic glass slide using double-sided tape. All glassware were acid-washed prior to use, with all otoliths sonicated and further cleaned in ultra-pure deionized water (see [18, 27] for more details on the processing and cleaning process).
To determine the hatch date of each larva, we first identified the hatch check  and then counted post-hatch daily rings to the longest otolith edge, using ImagePro imaging software (Media Cybernetics, Inc., Rockville, MD) and a Nikon E200 compound microscope (100x and 50x magnification, oil immersion, Nikon Inc., Melville, NY). Otolith ages from larvae < 25 d old were determined from a single count, as previous research conducted with Lake Erie yellow perch showed that single ring counts are reliable for yellow perch of less than this age . For larvae > 25 d of age, at least one additional blind count was conducted, with additional counts being performed as needed . With knowledge of the each larva’s post-hatch age and day of collection, we could determine hatch date.
Predicting larval dispersal through hydrodynamic backtracking
Water circulation model.
A hydrodynamic model was used to simulate the currents and temperatures in Lake Erie, providing the three-dimensional physical predictions necessary for tracking individual particles (larvae). Our model is based on the Princeton Ocean Model platform , which solves the hydrostatic, three-dimensional primitive equations in a second-order finite difference framework. The model used the Smagorinsky parameterization for the horizontal diffusion (coefficient of 0.1) and the Mellor-Yamada level 2.5 turbulence closure scheme in the vertical direction. Our model was calibrated and validated for Lake Erie using eight National Ocean Service (NOS) water level gauges along the shoreline, current measurements in each basin, and lake surface temperature in each basin . It currently runs in real-time as part of the NOAA Great Lakes Coastal Forecasting System (GLCFS; ), where hourly observations of wind speed/direction, cloud cover, air temperature, and dew point temperature are used to compute three-dimensional currents, temperature, water level, and waves on a 2 km resolution unstructured grid (21 vertical sigma levels).
Hourly output from the real-time GLCFS was used to drive a Lagrangian particle transport model to simulate passive trajectories of individual larva in western Lake Erie. The latitude and longitude where each larva was collected (see above) and its post-hatch age from otolith ring counts  were used to set the particle-tracking model’s initial parameters. The particle-tracking model used a 2nd-order Lagrangian scheme  to simulate passive, neutrally-buoyant particle movement in three dimensions. We assumed the larvae to be passive, neutrally-buoyant particles based on their small size, weak swimming ability, and positive photacticity [40, 43]. The Smagorinksy parameterization was used for horizontal diffusion (coefficient of 0.005), based on previous calibrations , and a random-walk approach was used for vertical diffusion (0.0005 m2 s−1).
In each simulation, a group of particles (n = 5,000) was initiated at the capture location of each larva and spread over a 5 m radius; the pathways taken by these particles were used to create a probability distribution for the pre-capture dispersal path of the larva captured at that location. Water-current uncertainties and variability were accounted for by the calibrated diffusion coefficients/schemes, as described above, as well as the defined particle patch. Simulations were performed backward in time, starting at the time of capture and proceeding back to the estimated hatching day. Daily location and temperature were recorded for each particle, delineating the backtracked pathway from capture to hatch location for each larva. In addition, a grid-based probability of daily larva location was computed for each modeled grid cell by accounting for the percentage of 5,000 particles present within the cell at any given time. In this sense, the spatial distribution of a larva’s location and the point of highest probability could be determined for each day (S1 Appendix).
With knowledge of the probability that a larva was in any given cell on any given day since hatching, we determined the most likely hatching location (NS or SS) by summing the probabilities of potential hatch cells in each location (e.g., NS probabilities were summed for cells above 41.81° N). Doing so allowed us to test how certainty in hatch location (60, 70, 80, and 90% certainty thresholds) could influence subpopulation discrimination abilities and juvenile classifications (using methods describe below). Output from these analyses also were compared to the “Best Estimate”, which used the single most probable hatch cell to assign larvae to the NS or SS without consideration of a certainty threshold.
Breeding subpopulation discrimination and juvenile classification
For both the genetics and otolith microchemistry described below, we first conducted all analyses without using backtracking (i.e., we assumed larvae originated in their capture location). These initial larval assignments served as our null condition. We then repeated all analyses, using backtracking predictions to revise initial larval hatching locations where appropriate (per above). For genetic analyses, we first used larvae of all size classes in analyses. Afterwards, we repeated all analyses with a subset of larvae that only included individuals < 8 mm TL, which are likely to be primarily passively dispersed [43, 48]. We also only included herein loci in Hardy-Weinberg equilibrium (7 of the 12 loci, see below), with analyses that used data from all 12 loci being found in S4 Appendix. For otolith microchemical analyses, we included all larvae in analyses due to smaller initial sample sizes.
Microsatellite extraction and genotyping.
DNA was recovered from tissue samples using a plate-based extraction method . Extracted larval DNA samples were re-suspended in 50 μL of Tris–EDTA buffer (10 mM Tris, 1.0 mM EDTA, pH 8.0), whereas juvenile DNA was re-suspended in 100 μL of the same buffer.
Each individual was genotyped at a total of 12 microsatellite loci . PCR amplification was performed in 25 μL reactions with the following components: 1.5 μL of template DNA; 2.5 μL 10x PCR buffer; 2.5 μL of MgCl2 (25 mM); 0.3 μL of dNTPs (50 μM of each); 0.2 μL (0.5 μM) of dye-labeled primer; 0.2 μL (0.5 μM) of the reverse primer; and 0.10 U Taq polymerase (Applied Biosystems, Foster City, CA). PCR conditions were as follows: initial denaturation at 94°C for 2 min, followed by 35 to 40 cycles of denaturing at 94°C for 15 s; annealing at various temperatures for 30 s (following ); extension at 72°C for 30 s; and a final extension of 72°C for 10 min. Microsatellite allele sizes were determined using a LI-COR 4300 DNA analyzer (Lincoln, NE) and scored using GeneImage IR 4.05 (Scanalytics Inc., Rockville, MD).
Larval population genetic structure.
First, Fisher’s exact tests for Hardy-Weinberg Equilibrium (HWE) were performed using Arlequin (v220.127.116.11; ). The α-level of each test was adjusted by dividing by the number of tests conducted (i.e., 12 tests, one per locus). Loci were considered unreliable and removed from subsequent analyses, if consistent violations of HWE occurred at the same locus across both subpopulations or within a subpopulation during both years. Second, tests for linkage disequilibrium were run for all pairs of loci in all larval groups (and age-0 juveniles) using Genepop (version 4.0.7; ). Third, FST estimates were calculated to assess genetic differentiation within a year between NS and SS subpopulations using Genepop (version 4.0.7, ; following ). Lastly, we used discriminant analysis of principal components (DAPC) in the R package ‘adegenet’ (version 1.3–9.2; [54–55]) to explore genetic structure (spatial data based on capture or hatch location was not included in our analysis). DAPC first transforms the data using principal components analysis, then uses k-means clustering to define clusters that maximize the variation among groups (based on the lowest Bayesian Information Criterion).
In describing larval population genetic structure, we were not concerned with identifying population substructure per se. Instead, we sought to identify whether at least a minimal level of structure existed, and then using this weak structure to develop assignment and discrimination functions. Previous studies have demonstrated that the most common assignment tests are robust to low differentiation among subpopulations (e.g., low FST) and violations of assumptions (e.g., HWE; [56–57]).
Larval self-assignment with microsatellites.
We used GENECLASS 2.0  to conduct a rank-based self-assignment genotype test during both years  that was based on microsatellite information from larvae from both subpopulations. It uses a bootstrapping approach wherein each individual larva is removed from the analysis (one at a time) and subsequently treated as an “unknown” larva that is then assigned to NS or SS subpopulations, based on the genetics of all other individuals . By determining the percent of larvae successfully assigned back to their NS or SS collection location (i.e., null larval assignment, no backtracking) or hatch site (i.e., after backtracking revision), as well as by exploring posterior probabilities of assignment for each individual, reliability in assignments for each subpopulation could be assessed.
Juvenile classification with microsatellites.
Microsatellite information from both larvae and juveniles was used to quantify both breeding subpopulation’s relative contribution of juvenile recruits to the mixed population during both years. To do so, we used GENECLASS 2.0 to assign each unknown-origin juvenile to either the NS or SS breeding subpopulation. These determinations were made using the null larval assignments (i.e., not revising larval hatch locations using backtracking) and revised larval assignments (i.e., accounting for pre-capture dispersal using backtracking). Our analysis consisted of a two-step procedure . First, we performed a Bayesian assignment  with Monte Carlo re-sampling using Paetkau et al.’s  simulation algorithm (10,000 simulated individuals at an assignment threshold of P = 0.05). The Bayesian analysis allowed us to exclude unknown-origin juveniles with less than a 30% of belonging to either one of our focal breeding subpopulations (NS or SS), as the potential for contributions from other local breeding subpopulations exists in Lake Erie . Second, we classified remaining juveniles to the NS or SS, using a rank-based genotype assignment (frequency method; ). In so doing, we considered successful ranked-based assignments to be those with probability of 70% or higher of belonging to one of the two breeding subpopulations (hence, the second group assignment probability would be 30% or lower). Failed assignments (i.e., unknown origin) were those with likelihood between 30% and 70%.
We used laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) to quantify trace-metal concentrations in each otolith not used for hatch date determination. An ICP-MS (model X7, Thermo Elemental, Franklin, MA) with a Continuum Surelite I solid-state ND-YAG laser (wavelength = 266 nm, maximum power = 20 mJ, pulse rate = 20 Hz, pulse rate = 4–6 ns) was used to measure isotopic masses, which then were used to determine lithium (Li), magnesium (Mg), manganese (Mn), zinc (Zn), strontium (Sr), barium (Ba), and lead (Pb) concentrations (see [18, 27] for more details). Calcium (Ca) was used as an internal standard, to correct for ablation-yield differences, and the occurrence of mass 120 (measured as 120Sn, tin isotope) in samples was used as a contamination indicator . A glass standard (NIST 610) was analyzed twice before and after every 16 samples to correct for drift and estimate the precision (coefficient of variation) of the instrument between runs. The argon carrier gas was analyzed for 60 s before every sample to determine instrument background levels and to estimate the limits of detection (LOD) of every sample. For larvae, ablated transects spanned from the outer edge of the otolith, through the core, and to the opposite outer edge. We integrated the entire otolith except for 5–10 s at the outer edge to avoid tape contamination. For juveniles, we ablated the portion of the otolith that captured the larval period based on the mean otolith radius length of 28 d old larvae caught from two breeding locations. We integrated and analyzed this larval portion before, through, and after the core . Transect data were integrated using ICP-MS PlasmaLab software (Thermo Electron, Waltham, MA) to quantify mean elemental concentrations.
To be included in final analyses, concentrations had to be above the LOD for 90% of the samples within a breeding area and the glass standard coefficient of variation had to be < 10.5% . For 2006 and 2007, Sr and Ba met our LOD criteria. Elements that were significantly related to otolith radius based on the results of an analysis of covariance (ANCOVA) were detrended to remove the effect of fish size using the slope of the relationship between the elemental concentration and the otolith radius [18, 31].
Larval self-assignment and juvenile classification using otolith microchemistry.
We used linear discriminant function analysis (LDA), quadratic discriminant function analysis (QDA), random forest analysis (RF), and neural network analysis (NN) in R  to test for self-assignment consistency within each year. The LDA and QDA analyses used jackknife procedures to classify larvae. RF and NN use machine-learning algorithms. Based on the near-identical results across all methods for larval self-assignment, we used only LDA in R to quantify the contribution of both breeding subpopulations to the mixed population of unknown-origin juvenile recruits within each year . We used similar cutoffs as with the genetic assignments to assign juveniles (i.e., likelihoods > 70% were successfully assigned to a subpopulation, whereas likelihoods < 70% were considered to be failed assignments). For both the larval self-assignment tests and the juvenile assignments, used the same null larval capture-location assignments (no backtracking) and backtracking-revised hatching-location assignments as in the microsatellite analyses.
Hatch location determination using backtracking
Backtracking data from both years led us to revise the initial assignment of numerous larvae, mostly shifting the larval hatching location from the SS to the NS (S1 Appendix). During 2006, 151 of 242 larvae (62%) were captured in the NS, with the remaining 91 larvae collected in the SS. Reenactment of pre-capture dispersal pathways by backtracking, however, suggested that the hatching location of > 50 larvae should have been the NS instead of the SS (Best Assignment, using all larvae: 84% NS). Including the < 8 mm TL cutoff and more stringent certainty criteria regarding the confidence in our backtracked hatch locations further increased the percentage of larvae predicted to have hatched in the NS, but also simultaneously reduced the number of individuals included in subsequent analyses (no backtracking: 64% NS, total n = 190; Best assignment: 88% NS, n = 165; 60% certainty: 87% NS, n = 165; 70% certainty: 88% NS, n = 163; 80% certainty: 89% NS, n = 160; 90% certainty: 91% NS, n = 149). During 2007, 283 of 364 larvae (78%) were captured in the NS, with the remaining 81 larvae collected in the SS. Consideration of backtracked dispersal pathways prior to capture again shifted the initial assignment of hatch location more strongly toward larvae originating from the NS (Best assignment including all larvae: 88% NS). Including the < 8 mm TL cutoff and increasing our level of confidence in the backtracking data again increased the percentage of larvae predicted to have originated in the NS and reduced the number of larvae that could be used to build functions to assign juveniles (no backtracking: 84% NS, total n = 247; Best assignment: 91% NS, n = 242; 60% certainty: 91% NS, n = 240; 70% certainty: 91% NS, n = 240; 80% certainty: 93% NS, n = 236; 90% certainty: 98% NS, n = 224).
Larval population genetic structure
For larvae captured during 2006, two of the NS loci and five of the SS loci (out of 12 for each subpopulation) deviated significantly from HWE. One locus deviated significantly from HWE across both subpopulations during 2006 (Table 1). During 2007, 6 of the 12 NS loci and 5 of the 12 SS loci significantly deviated from HWE in larvae from both the NS and SS. Five loci deviated significantly from HWE across both subpopulations during 2007 (Table 1). Because 5 of 12 loci (loci YP78, YP96, YP60, YP65, and YP49) significantly deviated from HWE in either both years within a subpopulation or in both subpopulations within a year, we removed these loci from subsequent analyses, unless noted otherwise. No evidence of linkage disequilibrium was found in either year. See S2 Appendix for genetic diversity indices based on larvae < 8 mm TL and including only the seven loci in HWE.
Pairwise FST values, based only on the seven loci, indicated only weak genetic differentiation between the NS and SS subpopulations during 2006 and 2007 (Table 2). Using backtracking information to improve the certainty of the larval hatch location caused a ∼3-fold (Best; i.e., > 50% certainty in hatching location) to > 5-fold (90% certainty in hatching location) increase in the FST value (relative to the null condition; i.e., no backtracking) during 2006, although a similar increase was not evident during 2007 (Table 2). Conducting the same analyses with only larvae < 8 mm TL (i.e., those most likely to drift passively) caused FST values to generally increase during both years relative to when all larvae were included in the analysis. Further, increasing confidence in the hatching location of larvae (those < 8 mm TL) with backtracking generally led to a 3-fold to 4-fold increase in the FST value over the null condition (Table 2).
DAPC also provided evidence for genetic differentiation, using seven microsatellite loci. Near equal support for two to five genetic clusters was found during 2006 using the full set of larvae (S3 Appendix). Using only larvae that were < 8 mm TL and met the 90% hatch-location certainty criterion gave similar results (S3 Appendix). For the full set of 2007 larvae, DAPC found the strongest support for four clusters, with three to seven clusters similarly supported (S3 Appendix). Using only 2007 larvae that were < 8 mm TL and with a 90% hatch location certainty gave similar results (S3 Appendix). Because our interest was in discriminating between subpopulations from the SS and NS for this application, we plotted two clusters for each dataset against the first discriminant function to illustrate the improved consistency of each cluster after using the backtracking procedure to revise initial group membership (i.e., we compared cluster assignments of all larvae to assignments of larvae for which hatch location was most certain; S3 Appendix).
Larval self-assignment accuracy (microsatellites)
Using backtracking to correct for pre-capture dispersal and only using small (< 8 mm TL) larvae that are likely to drift passively improved larval self-assignment accuracy. During 2006, self-assignment success of larvae increased from ∼73% without backtracking (i.e., initial hatching-location assignment based on larval capture location) to 85% to 90% with consideration of backtracking at all levels of larval hatching-location certainty when no size cutoff was used (Fig. 2A, black bars). Implementing the 8 mm TL size cutoff tended to increase larval assignment success by a small percentage, with >90% self-assignment success being achieved when the most stringent (90%) backtracking certainty was used (Fig. 2A, white bars). During 2007 (Fig. 2B), the pattern was similar, although backtracking offered more modest gains (< 12%) in assignment success (from ∼65% success to ∼78% success). However, when the 8 mm TL size cutoff was implemented, leaving only larvae that are likely to drift passively , self-assignment accuracy rose to >80% for most of the lower levels of certainty in larval hatching location, but increased to 97% when the 90% certainty criterion in backtracked hatching location was implemented (Fig. 2B).
For sample sizes, see Table 2. Confidence in hatching locations is as follows: None = null assignments based on capture location (i.e., no backtracking used); Best = larvae assigned to single most likely hatching location (SS or NS) based on backtracking; and 60, 70, 80, 90 = larvae assigned to hatching location after backtracking revision with 60, 70, 80, 90% levels of certainty in hatching origin, respectively.
Classification of unknown-origin juveniles (microsatellites)
Improved classification accuracies resulting from consideration of corrected larval hatching locations (via backtracking) and implementation of an 8 mm TL size cutoff (see Fig. 2) both improved our ability to determine the origins of juvenile recruits captured in a mixed population during August 2006 and 2007 and altered each breeding subpopulation’s predicted contribution to it. Some individuals (0% to 20%, depending on the analysis) likely did not emanate from either of our breeding populations, and therefore were excluded from analysis (Fig. 3). However, of those that were viable for classification, the percentage of failed assignments decreased by ∼2-fold with consideration of both backtracking and the 8 mm TL cutoff during both years, with the reduction in failed assignments being far greater when the 90% certainty level in backtracked hatch location was used (Fig. 3). In addition, during both years, the percentage of juveniles classified to NS tended increased, with the percentage assigned to the SS concomitantly decreasing. During 2006, the percentage of juveniles predicted to emanate in the NS increased by ∼20% to ∼30% with consideration of backtracked hatching location, whereas the increase on the order of ∼10% during 2007.
A total of N = 119 and N = 167 juveniles were analyzed during 2006 and 2007. Juveniles with < 30% likelihood of originating within a population were “excluded” (i.e., they were not included in the analysis); juveniles with a 30 to 70% likelihood were considered “failed,” (i.e., we had little confidence in their hatching-location assignment); and juveniles with a probability > 70% were assigned to the NS or SS breeding population with high confidence. Certainty in hatching locations of larvae used to develop classification functions was as follows: None = null assignments based on capture location (i.e., no backtracking used); Best = larvae assigned to single most likely hatching location (SS or NS) based on backtracking; and 60, 70, 80, 90 = larvae assigned to hatching location after backtracking revision with 60, 70, 80, 90% levels of certainty in hatching origin, respectively.
Larval self-assignment accuracy (otolith microchemistry)
Mean otolith Sr concentrations in larvae differed strongly between the NS and SS breeding subpopulations during both years (2006: SS: 847±216 μg g−1, NS: 443±55 μg g−1; 2007: SS: 843±239 μg g−1, NS: 554±103 μg g−1), which reflected differences in water chemistry (Sr:Ca ratios) between breeding locations (2006: SS: 0.0021, NS: 0.0011; 2007: SS: 0.0021, NS: 0.0014). Mean Ba concentrations in both larval yellow perch otoliths and the ambient water also differed between breeding locations during both years (2006: SS: 30±24 μg g−1; NS: 99±37 μg g−1; 2007: SS: 50±26 μg g−1; NS: 125±213 μg g−1; Ba:Ca ratios: 2006: SS: 0.000075; NS: 0.00025; 2007: SS: 0.00012; NS: 0.00031 Ba:Ca). During 2007, however, Ba exhibited a significant interaction between subpopulation and otolith radius (F1,111 = 30.25, P < 0.001) and could not used for discrimination. Because otolith size was significantly related to Sr concentrations in both years and Ba in 2006 (2006 Ba: F1,36 = 16.24, P < 0.001; 2006 Sr: F1,36 = 128.14, P < 0.001; 2007 Sr: F1,111 = 128.14, P < 0.001), we detrended the data using slope estimates of −0.340 (2006 Ba), −0.00134 (2006 Sr), and −0.00161 (2007 Sr).
Self-assignment accuracy of larvae, based otolith micro-elemental concentrations, was high during both years, with the value of backtracking differing between years. Owing to otolith Sr being a near-perfect discriminator in 2006, larval self-assignment success across all multivariate tests was high (> 98%) even without consideration of backtracking, and remained similarly high with each backtracking scenario (Fig. 4A). During 2007, assignment success across all multivariate tests was slightly lower (83–87%) without consideration of backtracking results, but increased to > 95% (up to 98%) with the most stringent backtracking criterion (90% certainly in larval hatch location, Fig. 4B).
A total of N = 47 and N = 71 larvae were analyzed in 2006 and 2007. Confidence in hatching locations is as follows: None = null assignments based on capture location (i.e., no backtracking used); Best = larvae assigned to single most likely hatching location (SS or NS) based on backtracking; and 60, 70, 80, 90 = larvae assigned to hatching location after backtracking revision with 60, 70, 80, 90% levels of certainty in hatching origin, respectively.
Classification of unknown-origin juveniles (otolith microchemistry)
Similar to the assignments of juveniles based on microsatellite information (see Fig. 3), backtracking showed potential to improve the use of otolith microchemistry for identifying the hatching locations of juvenile recruits. While backtracking did not reduce the percentage of failed assignments (i.e., < 70% classification accuracy) during 2006 (Fig. 5A), the percentage of failed assignments declined from ∼ 25% to < 5% during 2007 (Fig. 5B). In addition, consideration of backtracked hatch locations of larvae used to develop classification functions altered the relative contributions of juveniles from both subpopulations during both years. The proportion of the total fish assigned to the NS increased in both years relative to the null (no backtracking) scenario. During 2006, this increase was >50% for the Best and 90% backtracking-certainty scenarios (Fig. 5A). This increase in individuals assigned to NS corresponded to a decrease in those assigned to SS. During 2007, the percentage of juveniles predicted to originate in the NS doubled with consideration of backtracking (from ∼20% without it to ∼40% with it; Fig. 5B). However, during 2007, the percentage of the total fish assigned to SS remained fairly constant (∼45%). Rather, during 2007, the increase in individuals assigned to the NS corresponded to a decrease in the percentage of failed assignments (from ∼25% to < 5%; Fig. 5B).
A total of N = 98 juveniles were analyzed each year. Juveniles with < 70% likelihood of originating within a population were consider “failed” and juveniles with a probability > 70% were assigned to NS or SS. Certainty in hatching locations of larvae used to develop classification functions was as follows: None = null assignments based on capture location (i.e., no backtracking used); Best = larvae assigned to single most likely hatching location (SS or NS) based on backtracking; and 60, 70, 80, 90 = larvae assigned to hatching location after backtracking revision with 60, 70, 80, 90% levels of certainty in hatching origin, respectively.
Our results demonstrate that the use of particle backtracking approaches that account for dispersal history prior to collection can improve the ability of natural tagging approaches to discriminate among breeding subpopulations (i.e., potential stocks). In addition, the general backtracking approach used herein is valuable in that it allows for detection and characterization of population structure despite physical mixing of individuals at a young age (e.g., in the case of western Lake Erie, transport of larvae from one hatching location to another by strong river- and wind-driven currents), which none of the classical population genetics methods can do without an unmixed control sample to train the analysis to identify strays. For both natural tagging approaches used herein (i.e., microsatellites and otolith micro-elemental composition), revision of larval hatch locations via particle backtracking simulations improved our ability to develop reliable classification functions breeding subpopulation discrimination. In our example, the use of backtracking improved the accuracy of assignment of larval yellow perch to their proper breeding subpopulation in western Lake Erie (NS or SS) by as much as 43% over the null condition (i.e., use of larval capture location without backtracking; Table 3). In addition, larval self-assignment accuracies tended to increase with increasing confidence in the backtracked pre-capture dispersal trajectories. Using backtracking to correct the hatch locations of larvae used to develop classification functions also generally caused FST values measured between breeding subpopulations to increase, suggesting that this method could assist efforts to quantify genetic structure in weakly differentiated stocks. An improved ability to discriminate between breeding subpopulations (using genetics or otolith microchemistry) would, in turn, be expected to benefit efforts to correctly identify the source origin of older individuals found in a mixed population. In our western Lake Erie application, the use of backtracking to revise some of the larval hatch locations caused the percentage of failed juvenile recruit classifications to decrease by half or greater (relative to the null, no-backtracking scenario; Table 3) and also caused us to revise the relative subpopulation contribution of juvenile recruits from the NS versus SS. The increase in the contribution of juvenile recruits from the NS with consideration of backtracking was as much as 45% of the total number of individuals (a 4x increase over initial estimated contributions from this subpopulation when backtracking was not used). As we explain below, changes as substantial might not only lead to a new understanding of population structuring and dynamics in an ecosystem, but also simultaneously benefit conservation or management efforts.
Application of particle backtracking to Lake Erie yellow perch
Specific to western Lake Erie yellow perch, contributions of juvenile recruits from the NS and SS regions have been previously documented in western Lake Erie using genetics  and otolith microchemistry , respectively. However, like many studies conducted in other aquatic ecosystems, these studies did not account for potential dispersal prior to larvae being collected for development of discrimination functions. Our consideration of passive dispersal during the larval stage (using backtracking) suggests that the NS subpopulation contributes more larvae and juvenile recruits to the open-lake population than has been previously reported (e.g.,  as inclusion of backtracking increased the contribution of juvenile recruits from the NS by up to 60% and 108% during 2006 and 2007, respectively. Though direct comparisons to  would not be appropriate because that investigation compared relative recruitment rates of larvae residing inside versus outside of an open-lake river plume (Maumee River) that is generally only found in the SS of western Lake Erie and focused on larval habitat use rather than natal origin, our findings do suggest that a large proportion of the individuals that used this river plume as nursery habitat during the larval stage actually originated in the NS (49% and 47% of the larvae captured in 2006 and 2007, respectively). We suspect that many of these larvae were passively advected by strong Detroit River water currents from the NS into the SS prior to capture for discrimination purposes (the mean travel time is less than 21 hours; ), most likely at an early age when they are expected to have weak swimming abilities .
Recognition of the heightened importance of the NS to juvenile recruit production most certainly would be of interest to Lake Erie researchers and fisheries management agencies. Ecologists, for example, might be interested in learning more about the mechanisms that allow for maintenance of this weak genetic differentiation. While such a discussion is beyond the scope of this study, natal homing behavior and kin recognition/selective mating provide two reasonable hypotheses [66–67]. By contrast, Lake Erie agencies might wonder whether the NS larvae and subsequent juvenile recruits are originating from outside of Lake Erie proper, given that the current management plan for Lake Erie yellow perch does not consider the Detroit River – Lake St. Clair corridor as a contributing subpopulation . The possibility certainly exists that the majority of NS larvae are produced outside of Lake Erie proper, which would have a profound effect on predictions of Lake Erie agency-derived yellow perch population size that in part depends on recruitment being dictated by in-lake spawning stock biomass . Further, research conducted with Pacific salmonines has demonstrated that maintenance of a diverse “portfolio” of breeding subpopulations (i.e., stock diversity) can be critical to ensuring population viability during changing ecosystem conditions . Given that Lake Erie has experienced large-scale ecosystem change during recent decades, owing to such factors as altered nutrient inputs [69, 70], invasive species , and climate change [72–73], we recommend investigations that use a backtracking approach in combination with one or more natural tagging approaches to determine whether larvae collected in the Detroit River – Lake St. Clair corridor can be discriminated from NS and SS larvae. If discrimination is possible, quantification of whether this outside source contributes recruits to the juvenile population, which is a strong predictor of future recruitment to the fishery at age-2 , seems a logical next step.
Application of particle backtracking to other populations
Lake Erie yellow perch have life-history characteristics that are similar to many other freshwater and marine organisms. Such traits include the production of small propagules (i.e., eggs, larvae) that are subject to passive dispersal – for at least a period of time—by physical forces such as water circulation . For this reason, hydrodynamic models have been used extensively to describe the dispersal trajectories of eggs and/or larvae in other fish populations, both freshwater  and marine , as well as aquatic invertebrates, including mollusks , crustaceans , and corals . While this modeling historically has consisted of forecasting the post-spawning dispersal trajectories of eggs or larvae, efforts to hindcast dispersal histories of eggs or larvae prior to capture from a known collection (sampling) location have increased in recent years [21, 80].
Although our discussion thus far has focused primarily on aquatic organisms, we see great potential for our combined backtracking-natural tagging approach to also benefit our understanding of and the management of the dynamics of terrestrial organisms. Because many terrestrial plant and invertebrate animal species also have life stages that are vulnerable to passive dispersal by physical processes such as wind, atmospheric models have been used in a similar way as water circulation models to describe dispersal trajectories of a variety of terrestrial organisms. Admittedly, the bulk of this research has been conducted with terrestrial plants ; however, the application to understanding the dispersal dynamics of both invertebrate animals [10, 81–82] and microorganisms  is readily apparent. For example, Guichard et al.  used atmospheric models to hindcast moth dispersal patterns.
Relatedly, our backtracking approach also can help to identify which discrimination approaches are likely to be most useful in a given system. If backtracking results, for example, indicate that individuals from different, breeding (source) populations have been mixed for a significant amount of time prior to capture such that they had experienced similar natal or post-hatching environments, discrimination based on tools that depend on these environmental differences (e.g., otolith microchemical approaches that rely on differences in elemental or isotopic composition; approaches that use parasites or contaminants as discriminators) may need to be complemented or substituted with other approaches that do not depend on environmental differences among breeding locations (e.g., genetic approaches).
Backtracking approaches: caveats and research needs
While our analyses conducted with Lake Erie yellow perch highlight the value of using a backtracking model to account for dispersal history in breeding-subpopulation discrimination studies, this approach is not a panacea. As we discuss below, many important assumptions must be made when developing the backtracking model and the use of this technique can increase the need for heightened sample collections.
A key decision that will need to be made is whether to include behavior (i.e., active movement) in the backtracking model. In our application, we did not include any larval behavior and assumed that larvae drifted passively. While this assumption of passive dispersal appears valid for the sizes of yellow perch larvae used in our primary analyses (< 8 mm TL; ), its validity would be expected to decline with increasing fish size as active movement behavior has been shown to increase with ontogeny and body size, owing to in a large part a heightened ability of individuals to swim horizontally with more power [43, 84]. Thus, incorporation of active movement behavior in predictive models of dispersal may become necessary at times. Indeed, previous research conducted on both invertebrates  and fishes [30, 48, 86] has shown that even slight active horizontal or vertical movement during the larval stage can affect dispersal patterns. Such an effect might underlie the reduced discrimination ability between NS and SS yellow perch breeding subpopulations when all sizes of larvae were used in our study relative to when only those larvae < 8 mm TL were used. Given this consideration, the use of backtracking approaches that assume passive dispersal may not be valid for some species or only during very brief periods of time, thus requiring behavior (active movement) to be included in the backtracking model.
A related consideration when using backtracking approaches concerns the legitimacy of the backtracking model itself. In terms of constructing the backtracking model, many important decisions must be made with respect on how to deal with vertical diffusion, account for turbulent effects, and how stochasticity is included in the model, among many other considerations. Such considerations can be critically important. For example, how the stochasticity due to turbulence is accounted for in the backtracking model can lead to strongly differing predictions . Predicting advection in nearshore habitat where larvae may be most common can be particularly uncertain [87–88]. While much research continues to be conducted in this area, and manuals have been developed for developing robust predictive models of particle transport in aquatic ecosystems , proper calibration and validation of the physical model must be conducted, if it is to provide useful data [28, 89]. Our backtracking simulations conducted during 2006 and 2007 follow known hydrodynamic phenomena in the western basin [33, 44], which can explain the subsequent reassignment of larvae origination. Wind-driven currents in the western basin dictate the transport of Detroit and Maumee River waters, including the extent, flushing time, and degree of mixing. In other sysems, many quality hydrodynamic and atmospheric models already have been developed for non-ecological applications, which could readily be applied for ecological studies (e.g., the hydrodynamics model used here was developed for another purpose, and adapted to model the transport of larval fish). In turn, interdisciplinary collaborations between physical modelers and ecologists most certainly would save both time and money by allowing a pairing of expertise and providing cost-effective ways to estimate spatially and temporally explicit environmental parameters that are difficult to measure via non-modeling methods .
While we showed that backtracking can improve the performance of natural tags for breeding subpopulation discrimination, this benefit did not come without a cost. Specifically, we learned that, as more stringent criteria were implemented to increase confidence in larval hatching locations, the number of individuals included in the analyses declined. For example, during 2006, our sample sizes used in genetic analyses declined from 68 individuals in the SS subpopulation when backtracking was not considered to 21, 20, 19, and 8 individuals when confidence in our backtracking estimate larval hatching locations was increased to 60%, 70%, 80%, and 90%, respectively. As a result of this tradeoff, achieving high confidence in discrimination ability may leave insufficient numbers of individuals to perform robust (powerful) assignments of individuals to their source origin. Indeed, previous research has demonstrated that minimal sample sizes exist in the (training) data used to develop classification functions, with these samples sizes varying among natural tagging techniques. For example, a minimum of 15 individuals from each breeding subpopulation has been suggested for the use of statolith microchemical approaches . Likewise, a minimum of sample size of 25–30 individuals per breeding population has been suggested for the use of genetics as a natural tag . As a result, more effort may be required during the initial collection phase to ensure that enough individuals remain in the analyses, once initial assignments to a breeding subpopulation are corrected via backtracking. Unfortunately, the number of additional individuals that would need to be collected at each breeding location will vary as a function of the degree of mixing of individuals among subpopulations, which will be driven in large part by stochastic water movement (or wind) in relation to propagule (e.g., seeds/eggs, larvae) production from each breeding subpopulation that mixes. Because we find it unlikely that one could estimate a priori how many individuals at a specific breeding location emanated from another breeding location, we recommend the collection of as many individuals at a given location that is practical. For situations in which such analyses are likely to be conducted annually, we are optimistic that a target number of individuals needed from each breeding subpopulation would emerge with time.
Despite the growing use of particle backtracking and natural tagging approaches in both aquatic and terrestrial ecosystems, we are unaware of any previous study that has coupled these approaches in a way that we have done here. Given the improved discrimination capabilities that resulted from our use of particle backtracking in conjunction with microsatellite and otolith microchemical information in a western Lake Erie fish population, we strongly believe that an integrated particle backtracking-natural tagging approach that accounts for pre-collection dispersal history holds great potential to enhance our ability to address a wide range of ecological questions that applicable to many ecosystem types. Such applications include: 1) determination of metapopulation and genetic structure/dynamics; 2) identification of recruitment mechanisms and how they vary among breeding subpopulations; 3) exploration of the degree of connectivity among populations; 4) assessment of whether local breeding subpopulations have evolved a life history in response to predictable physical features ; and 5) quantification of which local breeding or nursery area(s) disproportionately contribute(s) recruits to the broader population. Such advancement in our understanding of the natural world undoubtedly could be used to benefit management and conservation efforts. For example, identification of important breeding subpopulations or nursery areas could lead to altered population-specific harvest quotas (e.g., increased or decreased harvest, depending on whether the focal species was valued or considered a nuisance) or area-specific protection measures (e.g., no-take zones; ). By contrast, breeding subpopulations or areas that are found to not be contributing recruits in a manner that is on par with production at earlier life stages might require the establishment of rehabilitation efforts . Given all of these potential benefits to science, management, and conservation, we strongly encourage the use of integrated particle backtracking-natural tagging approaches, as well as research geared towards overcoming some of their known limitations.
S1 Appendix. Larval dispersal plots based on backtracking model.
We thank A. Drelich, T. Bambakidis, J. Foster, A. Giuliano, H. Gunder, J. Kim, K. Mabrey, G. Milanowski, C. Rae, K. Soloway, K. Wang, and S. Upton for assistance in the collection and processing of samples, and Z. Yang and J. Barrett for assistance with LA-ICP-MS analysis. We thank Captain T. Joyce and crew of the R/V Laurentian, Captain M. Taetsch and crew of the R/V Cyclops (NOAA-GLERL), Captain C. MacDonald and crew of the R/V Keenosay (OMNR), and Captain M. Turner and E. Weimer and crew of the R/V Explorer (ODNR-DOW) for providing assistance with field collections. We thank the OMNR (Wheatley, ON) and the ODNR-DOW (Sandusky, OH) for providing age-0 juveniles. We thank T. Johnson, K. Pangle, and J. Tyson for their assistance with multiple aspects of this project. This manuscript benefitted from comments provided by four anonymous reviewers. This is GLERL contribution # 1742.
Conceived and designed the experiments: SAL. Performed the experiments: JMR BJF SAL EJA MEF LC DDH. Analyzed the data: MEF RMB KMD. Contributed reagents/materials/analysis tools: SAL DDH BJF LC EJA. Wrote the paper: MEF SAL RMB KMD EJA LC BJF DDH JMR.
- 1. Turchin P. Quantitative analysis of movement: measuring and modeling population redistribution in plants and animals. 1st ed. Sunderland: Sinauer Associates; 1998.
- 2. Kool JT, Moilanen A, Treml EA. Population connectivity: recent advances and new perspectives. Landsc Ecol 2013;28: 165–185.
- 3. Cowen RC, Sponaugle S. Larval dispersal and marine population connectivity. Ann Rev Mar Sci 2009;1: 443–466. pmid:21141044
- 4. McConkey KR, Prasad S, Corlett RT, Campos-Arceiz A, Brodie JF, Rogers H, et al. Seed dispersal in changing landscapes. Biol Conserv 2012;146: 1–13.
- 5. Hilborn R, Quinn TP, Schindler DE, Rogers DE. Biocomplexity and fisheries sustainability. PNAS USA 2003;100: 6564–6568. pmid:12743372
- 6. Soons MB, Heil GW, Nathan R, Katul GG. Determinants of long-distance seed dispersal by wind in grasslands. Ecology 2004;85: 3056–3068.
- 7. Allendorf FW, England PR, Luikart G, Ritchie PA, Ryman N. Genetic effects of harvest on wild animal populations. TREE 2008;23: 327–337. pmid:18439706
- 8. Kim KS, Jones GD, Westbrook JK, Sappington TW. Multidisciplinary fingerprints: forensic reconstruction of an insect reinvasion. J Roy Soc Interface 2010;7: 677–686.
- 9. Hand CP, Ludsin SA, Fryer BJ, Marsden JE. Statolith microchemistry as a technique for discriminating among Great Lakes sea lamprey (Petromyzon marinus) spawning tributaries. Can J Fish Aquat Sci 2008;65: 1153–1164.
- 10. Eagles D, Deveson T, Walker PJ, Zalucki MP, Durr P. Evaluation of long-distance dispersal of Culicoides midges into northern Australia using a migration model. Med Vet Entomol 2012;26: 334–340. pmid:22211884
- 11. Guichard S, Krtiticos DJ, Leriche A, Kean JM, Worner SP. Individual-based modelling of moth dispersal to improve biosecurity incursion response. J Appl Ecol 2012;49: 287–296.
- 12. Thorrold SR, Jones GP, Hellberg ME, Burton RS, Swearer SE, Neigel JE, et al. Quantifying larval retention and connectivity in marine populations with artificial and natural markers. Bull Mar Sci 2002;70 (Suppl): 291–308.
- 13. Gillanders BM. Tools for studying biological marine ecosystem interactions-natural and artificial tags. In: Nagelkerken I, editor. Ecological connectivity among tropical coastal ecosystems. New York: Springer; 2009. pp. 457–492.
- 14. Wilson MF, Traveset A. The ecology of seed dispersal. In: Fenner M, editor. Seeds: the ecology of regeneration in plant communities, 2nd edition. Wallingford: CAB International; 2000. pp. 85–110.
- 15. Bilton DT, Freeland JR, Okamura B. Dispersal in freshwater invertebrates. Ann Rev Ecol Evol Syst 2001;32: 159:181.
- 16. James MK, Armsworth PR, Mason LB, Bode L. The structure of reef fish metapopulations: modelling larval dispersal and retention patterns. Proc Roy Soc Lond B 2002;269: 2079–2086.
- 17. Hobson KA. Tracing origins and migrations of wildlife using stable isotopes: a review. Oecologia 1999;120: 314–326.
- 18. Ludsin SA, Fryer BJ, Gagnon JE. Comparison of solution-based versus laser-ablation ICPMS for analysis of larval fish otoliths. Trans Amer Fish Soc 2006;135: 218–231.
- 19. Miller JA, Bellinger MR, Golden JT, Fujishin L, Banks MA. Integration of natural and artificial markers in a mixed stock analysis of Chinook salmon (Onchorynchus tshawytscha). Fish Res 2010;102: 152–159.
- 20. Berry O, England P, Marriott RJ, Burridge CP, Newman SJ. Understanding age-specific dispersal in fishes through hydrodynamic modelling, genetic simulations, and microsatellite DNA analysis. Mol Ecol 2012;9: 2145–2159. pmid:22417082
- 21. Christensen A, Daewel U, Jensen H, Mosegaard H, St. John M, Schrum C. Hydrodynamic backtracking of fish larvae by individual-based modeling. Mar Ecol Prog Ser 2007;347: 221–232.
- 22. Beletsky D, Schwab DJ. Modeling circulation and thermal structure in Lake Michigan: annual cycle and interannual variability. J Geophys Res 2001;106: 19745–19771.
- 23. Nathan R, Sapir N, Trakhtenbrot A, Katul GG, Bohrer G, Otte M, et al. Long-distance biological transport processes through the air: can nature's complexity be unfolded in-silico? Divers Distrib 1005;11: 131–137.
- 24. Westbrook JK, Eyster RS, Allen CT. A model for long-distance dispersal of boll weevils (Coleoptera: Curculionidae). Int J Biometeorol 2011;55: 585–593. pmid:20835731
- 25. Kimura R, Secor DH, Houde ED, Piccoli PM. Up-estuary dispersal of young-of-the-year bay anchovy Anchoa mitchilli in the Chesapeake Bay: inference from microprobe analysis of strontium in otoliths. Mar Ecol Prog Ser 2000;208: 217–227.
- 26. Hedges KJ. Use of calcified structures for stock discrimination in Great Lakes walleye (Stizostedion vitreum). M.Sc. Thesis, The University of Windsor. 2002. Available: http://scholar.uwindsor.ca/etd/4472/
- 27. Reichert JM, Fryer BJ, Pangle KL, Johnson TB, Tyson JT, Drelich AB, et al. River-plume use during the pelagic larval stage benefits recruitment of a lentic fish. Can J Fish Aquat Sci 2010;67: 987–1004.
- 28. Thygesen UH. How to reverse time in stochastic particle tracking models. J Mar Syst 2011;88: 159–168.
- 29. North EW, Gallego A, Petigas P. Manual of recommended practices for modeling physical-biological interactions during early fish life. ICES Report 295. Copenhagen: ICES; 2009.
- 30. Willis J. Modelling swimming aquatic animals in hydrodynamic models. Ecol Mod 2011;222: 3869–3887.
- 31. Campana SE. Chemistry and composition of fish otoliths: pathways, mechanisms and applications. Mar Ecol Prog Ser 1999;188: 263–297.
- 32. Selkoe KA, Toonen RJ. Microsatellites for ecologists: a practical guide to using and evaluating microsatellite markers. Ecol Lett 2006;9: 615–629. pmid:16643306
- 33. Beletsky D, Hawley N, Rao YR. Modeling summer circulation and thermal structure in Lake Erie. J Geophys Res: Oceans 2013;118: 6238–6252.
- 34. Ludsin SA. Exploration of spatiotemporal patterns in recruitment and community organization of Lake Erie fishes: a multiscale, mechanistic approach. PhD dissertation, The Ohio State University. 2000.
- 35. Yellow Perch Task Group. Report of the Lake Erie Yellow Perch Task Group. Report to the Standing Technical Committee, Lake Erie Committee, Great Lakes Fishery Commission, Mar-2013. Ann Arbor: Great Lakes Fishery Commission; 2013.
- 36. Fuller K, Shear H, Wittig J. The Great Lakes: an environmental atlas and resource book. Chicago: Environment Canada and U.S. Environmental Protection Agency; 1995.
- 37. Baker DB. The sources and transport of bioavailable phosphorus to Lake Erie, Final Report: Part 2. US EPA/GLNPO ID: GL 00E75401-1. Chicago: US EPA; 2011.
- 38. Sepulveda-Villet OJ, Stepien CA. Fine-scale population genetic structure of the yellow perch Perca flavescens in Lake Erie. Can J Fish Aquat Sci 2011;68: 1435–1453.
- 39. Kocovsky PM, Sullivan TJ, Knight CT, Stepien CA. Genetic and morphometric differences demonstrate fine-scale population substructure of the yellow perch Perca flavescens: need for redefined management units. J Fish Biol 2013;82: 2015–2030. pmid:23731149
- 40. Craig J. The biology of perch and related fishes. Portland: Timber Press; 1987.
- 41. Gopalan G, Culver DA, Wu L, Trauben BK. The effect of recent ecosystem changes on the recruitment of young-of-year fish in western Lake Erie. Can J Fish Aquat Sci 1998;55: 2572–2579.
- 42. Carreon-Martinez, L. Larval yellow perch (Perca flavescens) and predation in the Detroit and Maumee River plumes in the western basin of Lake Erie. PhD dissertation, University of Windsor. 2012.
- 43. Houde ED. Sustained swimming ability of larvae of walleye (Stizostedion vitreum vitreum) and yellow perch (Perca flavescens). J Fish Res Board Can 1969;26: 1647–1659.
- 44. Michalak AM, Anderson EJ, Beletsky D, Boland S, Bosch NS, Bridgeman TB, et al. Record-setting algal bloom in Lake Erie caused by agricultural and meteorological trends consistent with expected future conditions. PNAS USA 2013;110: 6448–6452. pmid:23576718
- 45. Blumberg AF, Mellor GL. A description of a three-dimensional coastal ocean circulationmodel. In: Heaps N, editor. Three-dimensional coastal ocean models. Washington DC: American Geophysical Union; 1987. pp. 1–6.
- 46. Schwab DJ, Bedford KW. Initial implementation of the Great Lakes Forecasting System: a real-time system for predicting lake circulation and thermal structure. Water Poll Res J Can 1994;29: 203–220.
- 47. Bennett JR, Clites AH. Accuracy of trajectory calculation in a finite-difference circulation model. J Comput Phys 1987;68: 272–282.
- 48. Humphrey S, Zhao Y, Higgs D. The effects of water currents on walleye (Sander vitreus) eggs and larvae and implications for the early survival of walleye in Lake Erie. Can J Fish Aquat Sci 2012;69: 1959–1967.
- 49. Elphinstone MS, Hinten GN, Anderson MJ, Nock CJ. An inexpensive and high-throughput procedure to extract and purify total genomic DNA for population studies. Mol Ecol Notes 2003;3: 317–320.
- 50. Li L, Wang HP, Givens C, Czesny S, Brown B. Primer note: Isolation and characterization of microsatellites in yellow perch (Perca flavescens). Molec Ecol Notes 2007;7: 600–603.
- 51. Excoffier L, Laval G, Schneider S. Arlequin ver. 3.0: An integrated software package for population genetics data analysis. Evol Bioinforma Online 2005;1: 47–50.
- 52. Rousset F. Genepop'007: A complete reimplementation of the Genepop software for Windows and Linux. Molec Ecol Resources 2008;8: 103–106. pmid:21585727
- 53. Weir BS, Cockerham CC. Estimating F-statistics for the analysis of population structure. Evolution 1984;38: 1358–1370.
- 54. Jombart T. adegenet: a R package for the multivariate analysis of genetic markers. Bioinforma 2008;24: 1403–1405.
- 55. Jombart T, Devillard S, Balloux F. Discriminant analysis of principal components: a new method for the analysis of genetically structured populations. BMC Genet 2010;11: 94.
- 56. Hauser L, Seamons TR, Dauer M, Naish KA, Quinn TP. An empirical verification of population assignment methods by marking and parentage data: hatchery and wild steelhead (Oncorhyncus mykiss) in Forks Creek, Washington, USA. Mol Ecol 2006;15: 3157–3173. pmid:16968262
- 57. Bradbury IR, Campana SE, Bentzen P. Estimating contemporary early life-history dispersal in an estuarine fish: integrating molecular and otolith elemental approaches. Mol Ecol 2008;17: 1438–1450. pmid:18321254
- 58. Piry S, Alapetite A, Cornuet JM, Paetkau D, Baudouin L, Estoup A. GENECLASS2: A software for genetic assignment and first-generation migrant detection. J Hered 2004;95: 536–539. pmid:15475402
- 59. Paetkau D, Calvert W, Stirling I, Strobeck C. Microsatellite analysis of population structure in Canadian polar bears. Molec Ecol 1995;4: 347–354.
- 60. Beneteau CL, Mandrak NE, Heath DD. The effects of river barriers and range expansion on the population genetic structure of and stability in greenside darter. Conserv Genet 2009;10: 477–487.
- 61. Rannala B, Mountain JL. Detecting immigration by using multilocus genotypes. PNAS USA 1997;94: 9197–9221. pmid:9256459
- 62. Paetkau D, Slade R, Burden M, Estoup A. Genetic assignment methods for the direct, real-time estimation of migration rate: a simulation-based exploration of accuracy and power. Mol Ecol 2004;13: 55–65. pmid:14653788
- 63. Mercier L, Darnaude AM, Bruguier O, Vasconcelos RP, Cabral HN, Costa MJ, et al. Selecting statistical models and variable combinations for optimal classification using otolith microchemistry. Ecol Appl 2011;21: 1352–1364. pmid:21774435
- 64. Cadrin SX, Friedland KD, Waldman JR. Stock identification Methods: applications in fishery science. Burlington: Elsevier; 2005.
- 65. Dereki JA. Travel times in the Great Lakes Connecting Channels. NOAA Great Lakes Environmental Research Laboratory Technical Report No. 370. Ann Arbor: NOAA; 1983.
- 66. Gerlach G, Schardt U, Eckmann R, Meyer A. Kin structured subpopulations of Eurasian perch (Perca fluviatilis). Heredity 2001;86: 213–221. pmid:11380667
- 67. Stepien CA, Murphy DJ, Lohner RN, Sepulveda-Villet OJ, Haponski AE. Signatures of vicariance, postglacial dispersal, and spawning philopatry: population genetics of the walleye, Sander vitreus. Molec Ecol 2009;18: 3411–3428. pmid:19659479
- 68. Schindler DE, Hilborn R, Chasco B, Boatright CP, Rogers LA, Quinn TP, et al. Population diversity and the portfolio effect in an exploited species. Nature 2010;465: 609–612. pmid:20520713
- 69. Ludsin SA, Kershner M, Blocksom K, Knight R, Stein R. Life after death in Lake Erie: Nutrient controls drive fish species richness, rehabilitation. Ecol Appl 2001;11: 731–746.
- 70. Scavia D, Allan JD, Arend KK, Bartell S, Beletsky D, Bosch NS, et al. Assessing and addressing the re-eutrophication of Lake Erie: Central Basin Hypoxia. J Great Lakes Res 2014;40: 226–246.
- 71. Ricciardi A, MacIsaac HJ. Recent mass invasion of the North American Great Lakes by Ponto-Caspian species. TREE 2000;15: 62–65. pmid:10652557
- 72. Jones ML, Shuter BJ, Zhao Y, Stockwell JD. Forecasting effects of climate change on Great Lakes fisheries: models that link habitat supply to population dynamics can help. Can J Fish Aquat Sci 2006;63: 457–468.
- 73. Wang J, Bai X, Hu H, Clites A, Colton M, Lofgren B. Temporal and spatial variability of Great Lakes ice cover, 1973–2010. J Climate 2012;25: 1318–1329.
- 74. Ludsin SA, DeVanna KM, Smith RE. Physical-biological coupling and the challenge of understanding fish recruitment in freshwater lakes. Can J Fish Aquat Sci 2014:71: 775–994.
- 75. Zhao Y, Jones ML, Shuter BJ, Roseman EF. A biophysical model of Lake Erie walleye (Sander vitreus) explains interannual variation in recruitment. Can J Fish Aquat Sci 2009;66: 114–125.
- 76. Baumann H, Hinrichsen H-H, Voss R, Stepputtis D, Grygiel W, Clausen LW, et al. Linking growth to environmental histories in central Baltic young-of-the-year sprat, Sprattus sprattus: an approach based on otolith microstructure analysis and hydrodynamic modelling. Fish Oceanogr 2006;15: 465–476.
- 77. Broekhuizen N, Lundquist CJ, Hadfield MG, Brown SN. Dispersal of oyster (Ostrea chilensis) larvae in Tasman Bay inferred using a verified particle tracking model that incorporates larval behavior. J Shellfish Res 2011;30: 643–658.
- 78. Pedersen OP, Aschan M, Rasmussen T, Tande KS, Slagstad D. Larval dispersal and mother populations of Pandalus borealis investigated by a Lagrangian particle-tracking model. Fish Res 2003;65: 173–190.
- 79. Tay YC, Todd PA, Rosshaug PS, Chou LM. Simulating the transport of broadcast coral larvae among the Southern Islands of Singapore. Aquat Biol 2012;15: 283–297.
- 80. Hinrichsen HH, Dickey-Collas M, Hurat M, Peck MA, Vikebo FB. Evaluating the suitability of coupled biophysical models for fishery management. ICES J Mar Sci 2011;68: 1478–1487.
- 81. Nathan R, Katul GG, Bohrer G, Kuparinen A, Soons MB, Thompson SE, et al. Mechanistic models of seed dispersal by wind. Theoret Ecol 2011;4: 113–132.
- 82. Thomas CFG, Brain P, Jepson PC. Aerial activity of linyphiid spiders: modeling dispersal distances from meteorology and behaviour. J Appl Ecol 2003;40: 912–927.
- 83. Smith DJ, Timonen HJ, Jaffe DA, Griffin DW, Birmile MN, Perry KD, et al. Intercontinental dispersal of bacteria and archaea by transpacific winds. Appl Environ Microbiol 2013;79: 1134–1139. pmid:23220959
- 84. Leis JM. Ontogeny of behaviour in larvae of marine demersal fishes. Ichthyol Res 2010;57: 325–342.
- 85. Tilburg CE, Seay JE, Bishop TD, Miller HL, Meile C. Distribution and retention of Petrolisthes armatus in a coastal plain estuary: the role of vertical movement in larval transport. Estuar Coast Shelf Sci 2010;80: 260–266.
- 86. Leis JM. Behaviour as input for modelling dispersal of fish larvae: behaviour, biogeography, hydrodynamics, ontogeny, physiology, and phylogeny meet hydrography. Mar Ecol Prog Ser 2007;347: 185–193.
- 87. Nickols KJ, Gaylord B, Largier JL. The coastal boundary layer: predictable current structure decreases alongshore transport and alters scales of dispersal. Mar Ecol Prog Ser 2010;464: 17−35.
- 88. Romero L, Uchiyama Y, Ohlmann JC, McWilliams JC, Siegel DA. Simulations of Nearshore Particle-Pair Dispersion in Southern California. J Phys Oceanogr 2013;43: 1862–1879.
- 89. Miller TJ. Contribution of individual-based coupled physical-biological models to understanding recruitment in marine fish populations Mar Ecol Prog Ser 2007;347: 127–138.
- 90. Hale ML, Burg TM, Steeves TE. Sampling for microsatellite-based population genetic studies: 25 to 30 individuals per population is enough to accurately estimate allele frequencies. PLoS ONE 2012;7: e45170. pmid:22984627
- 91. Sinclair M. Marine populations. An essay on population, regulation, and speciation. Seattle: University of Washington Press; 1988.
- 92. Lester SE, Halpern BS, Grorud-Colvert K, Lubchenco J, Ruttenberg BI, Gaines SD, et al. Biological effects within no-take marine reserves: a global synthesis. Mar Ecol Prog Ser 2009;384: 33–46.