Genetics of century-old fish scales reveal population patterns of decline

Conservation scientists rarely have the information required to understand changes in abundance over more than a few decades, even for important species like Pacific salmon. Such lack of historical information can underestimate the magnitude of decline for depressed populations. We applied genetic tools to a unique collection of 100-year-old salmon scales to reveal declines of 56%–99% in wild sockeye populations across Canada’s second largest salmon watershed, the Skeena River. These analyses reveal century-long declines that are much greater than those based on modern era abundance data, which suggested that only 7 of 13 populations declined over the last five decades. Populations of larger-bodied fish have declined the most in abundance, likely because of size-selective commercial fisheries. Our findings illustrate how a deep historical perspective can expand our understanding of past abundances to a time before species incurred significant losses from fishing, and help inform conservation for diminished populations.


INTRODUCTION
Declines in population abundance of many of Earth's biota are outpacing species extinctions (Hughes, Daily, & Ehrlich, 1997), and disrupting ecosystems and human cultures. For example, loss of bison (Bison bison) populations decimated a cultural foundation of Indigenous peoples, and fundamentally re-shaped grasslands (Gaston & Fuller, 2008). Yet managers rarely have the historical information required to assess changes in abundance over more than a few decades, even for common species (Bonebrake, Christensen, Boggs, & Ehrlich, synchronous across broad regions due to large-scale climatic processes (Mantua, Hare, Zhang, Wallace, & Francis, 1997). Within each species, there are hundreds of unique, locally adapted populations; this diversity supports resilient fisheries and economies, as well as wildlife that feed heavily on salmon, such as bears . However, many salmon populations have declined due to over-exploitation and habitat degradation. Over 500 populations-predominantly in their southern range-have become extinct, with Chinook (O. tshawytscha) and sockeye (O. nerka) having experienced the highest proportional losses (Gustafson et al., 2007;Slaney, Hyatt, Northcote, & Fielden, 1996).
It has proven challenging to reconstruct population-specific abundances of salmon beyond the last several decades. For instance, data that contribute to extinction-risk evaluations for sockeye in Canada routinely begin after 1950 (COSEWIC, 2017), more than 70 years after the onset of commercial exploitation. While lake sediments have provided valuable insight into population dynamics over hundreds of years (e.g., Rogers et al., 2013), such paleolimnological tools are limited to systems where marine-derived nutrients drive primary productivity or nutrient budgets (Hobbs & Wolfe, 2007). Thus, there remains a need to quantify salmon abundances at a time when commercial fisheries began, which can be used by managers to measure the magnitude of population change in species like sockeye, which have long supported commercial fisheries.
We sought to characterize changes in abundance of sockeye populations in Canada over a century spanning the onset of industrial fishing to the present, and to test four leading hypotheses to explain the patterns observed: (1) fisheries selectivity, (2) habitat degradation, (3) population productivity, and (4) freshwater migration challenges. Here we combine modern genetic tools with century-old fish tissues to uncover declines of 56%-99% in wild sockeye populations from Canada's second largest salmon system, the Skeena River; had we used existing data (i.e., from the 1960s), only 7 of 13 populations show a decline. Fisheries selectivity of larger-bodied populations is the most probable driver of differences between populations. Our deep historical perspective can help inform status assessments and rebuilding discussions for diminished populations of this iconic species.

METHODS
The Skeena watershed is composed of 31 sockeye Conservation Units (CU; Holtby & Ciruna, 2007), which generally are grouped into 13 population complexes (Figure 1). An industrial fishery for these fish began at the mouth of the Skeena in 1877, and a scale-collection program commenced in 1912 (Supporting Information [SI], section A). We sampled scales from 50 fish from the collection for each of 9 fishing weeks   1913, 1916, and 1918-1923, for a total of 3,400 scales. We digitally photographed one scale from each fish, and aged each scale by annuli counts. DNA was extracted from scales, genotyped at up to 12 microsatellite loci, and individuals assigned to population via genetic stock identification; here, population refers to one or more CUs that are genetically similar to each other, yet distinct from other CUs (Beacham, Cox-Rogers, MacConnachie, McIntosh, & Wallace, 2014; SI, section B). We reconstructed aggregate (all populations combined) total abundance (i.e., catch + spawning fish, which throughout we refer to as abundance) for Skeena sockeye during the entire history (1877-2018) of commercial fishing using three data sources: (1) reconstructed catch from canneries (Argue & Shepard, 2005) and exploitation derived from instantaneous fishing rates applied to the number of boats operating in each year (which varied annually; Shepard & Withler, 1958), and combine to provide total abundance (i.e., abundance = catch/exploitation rate) during 1877-1950, (2) total abundance during 1951-1969 from Wood (2008), and (3) total abundance (separated into "combined" abundance that includes production from three spawning channels at Babine Lake plus wild abundance, and "wild only" abundance) We then reconstructed annual historical (1913)(1914)(1915)(1916)(1917)(1918)(1919)(1920)(1921)(1922)(1923) abundance of sockeye at the population level in a four-step process: (1) We multiplied aggregate annual abundance detailed above by the proportion of Skeena-origin sockeye genetically identified from historical scales. (2) We multiplied annual abundance (Step 1) by weekly abundance proportions-randomly drawn from one of seven (2011-2017) years of data estimated at the Skeena Tyee Test Fishery. (3) We multiplied weekly abundance estimates (Step 2) by the proportions of individual Skeena-origin populations in those fishing weeks, then summed across weeks for each population. (4) We repeated steps 2-3 50,000 times to derive a median abundance estimate for each population for each historical year (SI, section C). These run-reconstructions are based on two assumptions: (1) migration-timing of sockeye has not changed from the historical period to present, and (2) populations were equally susceptible to capture by gill-net. We examined the sensitivity of our analysis to these assumptions by shifting weekly proportions of fish one week earlier (SI, section D), and applying gill-net selectivity corrections on population abundance estimates (and body-size and age; SI, section E) to account for this in our hypothesis tests below.
To estimate the magnitude of change in abundance of populations between historical (1913)(1914)(1915)(1916)(1917)(1918)(1919)(1920)(1921)(1922)(1923) and recent (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) periods, we fit a linear mixed-effects model to the population-specific abundances with a categorical predictor variable-time-period. Annual median historical abundance estimates (corrected for gill-net selectivity) were derived from our run-reconstructions, and recent estimates are reported in English, Peacock, Challenger, and Mochizuki (2016) and SI, section F. Each model, representing a hypothesis, included an interaction term with time-period to assess how well the model explains variation in the change in abundance among populations between time-periods. We normalized our response variable (abundance) for each population by dividing each historical and recent year's abundance estimate by the average historical abundance for the population, so that all "observations" across populations are comparable.
We considered four hypotheses to explain changes in population abundance over time; specifically, we hypothesized that larger declines in abundance have occurred for populations that: (1) had larger body sizes because gill-nets catch larger-sized fish, (2) are from more degraded habitat because of lost carrying capacity, (3) had older ages because latematuring populations tend to have lower productivity, and (4) had longer migrations because long-distance migrants face more in-river fisheries and changing hydrologic patterns. We compiled data to test our hypotheses: (1) average selectivity-corrected historical body-size (Size) as a proxy for fisheries selectivity, (2) human footprint (Footprint) as a measure of habitat degradation, (3) average selectivity-corrected historical age-at-maturity (Age) as a measure of productivity, and (4) freshwater migration distance (Distance) as a measure of the degree to which the migration is challenging (SI, section G). We used body-size and age data from populations (scales) with ≥ 90% genetic assignment probability. We standardized each covariate by subtracting the mean and dividing by its standard deviation (SD) to compare effect sizes directly; no variable was strongly correlated with another (i.e., correlation coefficient > .5; SI, section H). Finally, we included "population" as a random effect on the intercept in each model to account for nonindependence of observations within populations at the scale that the independent variables were measured.
We fit models describing each hypothesis to the data in a Bayesian framework. We used uniform priors (i.e., all parameter values are equally likely) for each model. We fit each model with a lognormal error distribution using four chains, each with 4,000 iterations (the first 1,000 calibrated the sampler), for a total of 12,000 posterior samples. We based our inference about the importance of the hypothesized drivers of change in abundance on three lines of evidence: (1) the relative variable importance of each hypothesis (i.e., WAIC model weights), (2) the posterior probability of each hypothesis based on marginal likelihood estimates (Burkner, 2017), and (3) the sign, magnitude, and uncertainty in the parameter estimates representing each hypothesis in SD units. We performed two sensitivity analyses to determine whether our findings were robust to our assumptions: (1) we re-ran each model using uncorrected historical abundance estimates, and uncorrected average body-size and age for each population, and (2) we ran 1,000 iterations of each model -each iteration using a random point estimate of selectivity-corrected annual historical abundance -to account for uncertainty in our historical estimates. These revealed that model results are robust to our assumptions (SI, section I). All analyses were performed in R (R Core Team, 2018) using the "boot," "brms," "lme4," and "purrr" packages.

RESULTS
The total number of wild adult sockeye returning to the Skeena River in recent years is 75% lower than during historical times. The arithmetic mean annual abundance of wild fish in recent (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) years is 469,000 (range = 175,000-700,000), compared to 1,783,000 (range = 710,000-3,200,000) one century ago; production of sockeye from Babine Lake spawning channels has largely offset the long-term decline in wild fish (Figure 2).
Of 3,400 scales analyzed, 69% (2,354) were successfully assigned to a population, and 85% of these fish originated in the Skeena River. Within the Skeena, 13 populations were identified; all have declined in abundance over the last century (median = -82%), some by more than 90% (Table 1). For example, Motase and Sustut historically averaged ∼40,000 and ∼42,000 sockeye, respectively, while recently they average ∼600 and ∼1,600, respectively (Figure 3). Other populations, such as Bear (-56%) and Ecstall (-61%), have experienced smaller proportional declines. However, comparisons within the modern era (1960s to present), suggest that only 7 of 13 populations have declined (median = -6%). Population diversity also has shifted between time periods, such that the proportion of all wild fish returning to the Skeena River that are not from the Babine population has declined by 1/4 (from 33% one century ago, to 25% in recent years). Spawning channels on Babine Lake now produce roughly 70% of all sockeye returning to the Skeena River.
We found the strongest statistical support for the hypothesis that fisheries selectivity drives differences in the magnitude of decline among populations (Table 2). Declines were greatest for populations with larger body sizes; there was a 30% difference (96% compared to 66%) in rates of decline between populations with the largest and smallest body-size, respectively. For each SD unit increase in body-size (i.e., 1.8 cm), there was a predicted 7.4% greater decline in abundance, compared to 5.4% and 5.0% for each SD increase in age and migration distance, respectively. Rates of decline also tended to be greater in populations with older ages, and longer migrations, whereas rates of decline tended to be lower in populations with greater habitat degradation (Figure 4).

DISCUSSION
Using modern molecular tools and century-old fish scales we show that wild sockeye salmon populations in Canada have declined more over the last century than are currently recognized. Specifically, our findings indicate that: (1) wild populations in the Skeena watershed have declined in abundance by between 56% and 99% since 1913; such diminishment is much greater than that revealed by recent (1960 to present) population data and (2) the populations that declined most had larger body-size, suggesting that fisheries selectivity may have contributed to variation in declines among populations. We acknowledge that salmon populations exhibit large variation in abundance over centuries, even in the absence of human impacts (Rogers et al., 2013). Regardless of the underlying mechanisms driving the change, the utility of our historical perspective is that it shifts modern era abundance data for Skeena sockeye back to a time before they incurred significant losses from commercial fisheries, and expands our T A B L E 1 Skeena River sockeye salmon populations identified in genetic analyses, total (catch + escapement) abundance estimates and genetic proportions of populations during the historical (1913)(1914)(1915)(1916)(1917)(1918)(1919)(1920)(1921)(1922)(1923) and recent (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) periods, the percent change in historical abundance compared to the recent period, and population metrics used in hypothesis tests (note: Bear and Slamgeesh populations were omitted in modeling due to low sample size) Population understanding of the production potential of populations that can inform status evaluations and rebuilding discussions. Applying modern genetic techniques to ancient tissue is an emerging and powerful tool for understanding historical con-texts for conservation, such as verifying population extinctions (Iwamoto, Myers, & Gustafson, 2012) or loss in diversity (Thompson et al., 2019). Our work utilized such genetic tools to build upon previous studies that used century-old salmon cannery data and reported similar declines for multiple species across the Northeast Pacific (Gresh, Lichatowich, & Schoonmaker, 2000). Salmon are fundamental components of ecosystems, and such loss of abundance may decouple interlinked systems dependent on salmon in several ways. First, diminished salmon returns have reduced nutrient subsidies that once were delivered to rivers throughout the watershed (Doughty et al., 2016), with potential impacts on wildlife that rely on salmon . Second, the decline in abundance across all nonenhanced populations has undermined Indigenous peoples' food security throughout the rest of this vast watershed; many of these communities no longer catch sockeye due to low local abundance (Gottesfeld & Rabnett, 2008). Fewer salmon generally are associated with lower Indigenous fishery catches (Peterman, 1980), and diminished population diversity of salmon likely will compromise the stability of fisheries (Nesbitt & Moore, 2016).
Finally, the disproportionate loss in abundance from largerbodied, older-aged, populations may have eroded the biocomplexity and stability of the Skeena sockeye aggregate by homogenizing size-and age-at-maturity. Loss in population diversity-and associated life-histories-can destabilize salmon stock complexes by reducing the range of potential responses to varying environmental conditions (Moore, Yeakel, Peard, Lough, & Beere, 2014;Schindler et al., 2010). Fisheries selectivity of larger-bodied populations was the most probable driver of differences in rates of decline among populations over the last century, though age-at-maturity and migration distance also were associated with declines. Empirical comparisons among species show that large bodysize and late-maturity are important predictors of fish vulnerability to fisheries (Hutchings, Myers, Garcia, Lucifora, & Kuparinen, 2012;Reynolds, Dulvy, Goodwin, & Hutchings, 2005). Within species, gill-net fisheries can induce differential selection pressure such that large-bodied populations undergo higher mortality rates compared to smallbodied populations; a mechanism thought to have contributed to the disproportionate decline in sockeye from large-bodied beach-spawning populations in Alaska (Hamon, Foote, Hilborn, & Rogers, 2000). Indeed, differential catchability of gill-nets was the hypothesized driver of disproportionate declines in non-Babine populations in the Skeena prior to 1970 (Ricker & Smith, 1975). While selection pressure for large-bodied fish likely has weakened since 1950 due to the change towards less-selective gear (e.g., seine-nets; Kendall & Quinn, 2012), our results suggest that there may be a legacy effect of historical gill-net fisheries on Skeena sockeye populations. Body-size can be associated with vulnerability to other stressors, such as predators or temperature. Furthermore, as our results suggest, multiple factors (including ageat-maturity and migration distance) likely have contributed to large-scale declines in sockeye abundance and associated changes in population diversity. Surprisingly, habitat degradation failed to explain rates of decline. This may be because: (i) human footprint is a coarse measure of habitat loss (e.g., it excludes a "forest removal" score), and may be confounded by other factors that we did not consider or (ii) the Skeena watershed is relatively intact (Gottesfeld & Rabnett, 2008), and habitat has not been degraded sufficiently to impact salmon production.
Modern conservation policies for salmon (e.g., Canada's Policy for the Conservation of Wild Pacific Salmon; WSP) aim to maintain and rebuild distinct populations (DFO, 2005). The WSP requires the classification of extinction risk for populations based on metrics, like abundance trends, and such assessments are most reliable when the extent of decline is measured using long-term data (d'Eon-Eggertson, Dulvy, & Peterman, 2015). For sockeye populations in Canada, historical data begin after 1950 (COSEWIC, 2017). Our results identify larger declines in abundance than have previously been documented in this region (Price, English, Rosenberger, Mac-Duffee, & Reynolds, 2017). While it is difficult to set recovery targets when salmon productivity and abundances are naturally variable (Mantua et al., 1997;Rogers et al., 2013), our historical abundance estimates extend current time-series data to help inform assessments of population status, and contribute to rebuilding plan discussions.
Declines in wild Skeena sockeye abundance has been accompanied by an increase in production at the Babine Lake spawning channels. Fisheries that catch a mixture of wild and enhanced populations can complicate the conservation of diversity (Walters, Lichatowich, Peterman, & Reynolds, 2008;Wood, 2008). Given that annual commercial fishery decisions for Skeena sockeye are based on aggregate abundance targets, channel-enhanced production greatly increases the potential for overfishing of vulnerable wild populations (Walters et al., 2008). While the shared decline in all wild populations that we report may indicate persistent overfishing during the period leading up to the 1950s (Pritchard, 1948;Ricker & Smith, 1975), this historical period may simply have been more favorable for salmon compared to the unproductive modern era. Indeed, oceanic-scale processes appear to have driven broad-scale changes in sockeye abundance across much of western North America over the last several decades (Peterman & Dorner, 2012).
Historical perspectives for exploited species are critical to understand the extent of decline in depressed populations; naturally, the better we understand the past, the more informed our decisions towards recovery will be (Bonebrake et al., 2010). Our analyses demonstrate that wild sockeye salmon in Canada are far more diminished than previously realized, and our results should help inform status evaluations and rebuilding plan discussions for depleted populations by expanding our understanding of their production potential. More generally, our study shows how genetic analyses of historical samples can provide insight into centennial-scale changes in population abundance and diversity.