Declining Wild Salmon Populations in Relation to Parasites from Farm Salmon

See allHide authors and affiliations

Science  14 Dec 2007:
Vol. 318, Issue 5857, pp. 1772-1775
DOI: 10.1126/science.1148744

This article has a correction. Please see:


Rather than benefiting wild fish, industrial aquaculture may contribute to declines in ocean fisheries and ecosystems. Farm salmon are commonly infected with salmon lice (Lepeophtheirus salmonis), which are native ectoparasitic copepods. We show that recurrent louse infestations of wild juvenile pink salmon (Oncorhynchus gorbuscha), all associated with salmon farms, have depressed wild pink salmon populations and placed them on a trajectory toward rapid local extinction. The louse-induced mortality of pink salmon is commonly over 80% and exceeds previous fishing mortality. If outbreaks continue, then local extinction is certain, and a 99% collapse in pink salmon population abundance is expected in four salmon generations. These results suggest that salmon farms can cause parasite outbreaks that erode the capacity of a coastal ecosystem to support wild salmon populations.

The decline in ocean fisheries (1, 2) and rise in global demand for fish have driven the rapid growth of aquaculture (3, 4). Although aquaculture may augment fish supply (3), there are ecological risks, including competition and interbreeding of escaped farm fish with wild fish (5, 6), depletion of wild fish caught to feed farm fish (3, 4), and the spread of infection from farm fish to wild fish (7, 8). Disease threats of aquaculture to wild fish populations have long been contentious because of the uncertainty in impacts on those populations (912). We assess the impact of recurrent aquaculture-induced salmon lice (L. salmonis) infestations on wild pink salmon (O. gorbuscha) populations.

The salmon louse is a native marine ectoparasitic copepod of salmonids that feeds on surface tissues and causes stress, osmotic failure, viral or bacterial infection, and ultimately death (13). Lice are directly transmitted via planktonic nauplii and copepodids that can persist for several days. In areas without salmon farms, the prevalence of L. salmonis on juvenile pink salmon 2 to 3 months after marine emergence is low (<5%) (1416), because returning adult salmon are mostly offshore when juvenile salmon enter the sea (16, 17). Louse infestations of wild juvenile salmon have occurred throughout the Broughton Archipelago in Pacific Canada (Fig. 1) from 2001 to 2005 (7, 8, 14, 18, 19). There, salmon farms situated in inlets and channels near rivers can increase copepodid densities above background levels for more than 80 km of wild salmon migration routes or, equivalently, for the first 2.5 months of the wild salmon's marine life (8). In response to a pink salmon population collapse in 2002, a primary migration corridor was fallowed in 2003 (i.e., farm salmon were removed from aquaculture facilities in Tribune Channel through Fife Sound, but farms peripheral to this route remained active) (Fig. 1). For that salmon cohort, L. salmonis abundance declined (19), and pink salmon marine survival increased (20).

Fig. 1.

Study area in the Broughton Archipelago (boxed area in inset), depicting pink salmon populations from unexposed rivers (numbered circles) and exposed rivers (directly labeled within the lower rectangular frame). Inferred migration routes in the Broughton Archipelago are shown by the small arrows. Salmon farms are shown by black dots and sample sites by stars. Salmon farms south of Knight Inlet are not shown. Identities of the numbered (unexposed) rivers are provided in data set S1 (28).

To test for effects of lice on salmon population dynamics, we compiled Fisheries and Oceans Canada escapement data (the number of salmon per river), from 1970 to the present, for all pink salmon populations from rivers in the central coast of British Columbia, Canada (Fig. 1). There were 64 rivers whose salmon populations were not exposed to salmon farms and 7 rivers whose salmon populations must migrate past at least one salmon farm. Because pink salmon have a 2-year life cycle, there are distinct odd- and even-year lineages (21), which amount to 128 unexposed populations and 14 exposed populations. Rivers with substantial enhancement (e.g., spawning channels) were excluded because any increased salmon abundances in these rivers confound our estimates of natural changes in abundance. Unexposed populations had been and continue to be commercially fished. Exposed populations were commercially fished before the infestations, but the fishery remains closed since the onset of the infestations, when the data show a marked decline in productivity (Fig. 2 and fig. S1).

Fig. 2.

Time series of normalized population deviances {log[Ni(t)/mi], where Ni(t) is the population estimate for population i in year t and mi is the time-series mean abundance for population i} for 128 control populations of pink salmon (open gray circles) and 14 pink salmon populations exposed to salmon farms (black circles). The vertical dashed line marks the beginning of salmon aquaculture in the Broughton Archipelago. The vertical solid line marks the onset of louse infestations (and the commercial fishery closure) affecting the exposed populations. The arrow indicates data for exposed pink salmon cohorts that, as juveniles, experienced a fallowed migration corridor.

The analysis was based on the Ricker model (22), which is commonly used to model time-series data from density-dependent populations (2326), including pink salmon (24, 26), and provides robust estimates of population growth rates (24). The model is ni(t) = ni(t – 2)exp[r - bni(t – 2)], where ni(t) is the abundance of population i in year t, r is the population growth rate, and b determines density-dependent mortality. Upon log transformation to log[ni(t)/ni(t – 2)] = rbni (t – 2), the Ricker equation becomes a linear model with intercept r and slope b that can be estimated by linear regression and hierarchical mixed-effects modeling (23, 24, 27, 28). A preliminary model selection analysis did not support including random effects on r or b (fig. S2 and tables S1 and S2) (27). We therefore pooled data from multiple populations (27) and used linear regression to estimate parameters and parametric bootstrapping to construct 95% confidence intervals (CIs) on the parameter estimates (23). This allowed us to statistically compare parameters from pooled populations subjected to infestations, which is not possible with hierarchical mixed-effects models because there are only two data points per population during infestation years.

We compared parameter estimates among three groups: unexposed populations, exposed preinfestation populations, and exposed populations during infestations (excluding the fallow year). The groups did not differ in b, and so we reanalyzed the data with b fixed among the three groups. Unexposed populations did not differ from exposed preinfestation populations in growth rate (unexposed populations: r = 0.62, 95% CI: 0.55 to 0.69; exposed preinfestation populations: r = 0.68, 95% CI: 0.46 to 0.90). The growth rate of exposed populations during the infestations was significantly lower and significantly negative (r = –1.17, 95% CI: –1.71 to –0.59; Fig. 3), meaning that if infestations are sustained, then local extinction is certain (29). Population viability analysis (28, 29) revealed the mean time to 99% population collapse is 3.9 generations, with the 95% CI from 3.7 to 4.2. During two generations of infestations, some exposed populations have declined to <1%, whereas others have exceeded their historical abundance. We initially excluded the fallow data, because they contain only 1 year of observations and correspond to a nonrandom management action. By fixing b = 0.64, as estimated above, and estimating r from the remaining seven data points, we found the growth rate of fallow populations was significantly increased (r = 2.50, 95% CI: 1.28 to 3.62). The maximum reproductive rate for pink salmon is r* = 1.2 (24). Fishing mortality probably reduced r for unexposed and exposed preinfestation populations. The depressed growth rate of exposed salmon populations during the infestations indicates that previous fishing mortality (now ceased) has been greatly exceeded by louse-induced mortality.

Fig. 3.

Fits of the log-transformed Ricker model to escapement data for unexposed populations (A), exposed populations before infestations (B), and exposed populations during the infestations (C), and a comparison of the log-transformed Ricker model for the three groups in panels (A) to (C) (D). The intercept (growth rate) is lower for the exposed population during the infestations than for exposed populations before the infestations and the unexposed populations.

To estimate the mortality of pink salmon caused by lice, we extended the Ricker model to directly accommodate louse data collected from exposed populations during the infestations (14, 18, 19, 28). We constrained the model by fixing b = 0.64 and by requiring r = r* = 1.2, because there was no fishing mortality. Louse-induced mortality is represented by multiplying by exp[–aPi(t – 1)], where P is the mean abundance of motile (adult and preadult) lice per juvenile salmon from population i that spawned in year t. We log-transformed the model to log[ni(t)/ni(t – 2) = rbni(t – 2) – aPi t] and used linear regression to estimate a. The term exp[–aPi(t – 1)] significantly improved the fit of the model (t = –5.019, df = 33, P = 1.74× 10–5; fig. S3), and results remained strong when the data were restricted by averaging populations and excluding some population groups (P < 0.005 for all groups; table S3). The parameter a corresponds to the rate of parasite-induced host mortality multiplied by the time that juvenile salmon are exposed to the parasites, a = αT. The exposure time, T, is about 2 months (based on the migration speed of juvenile pink salmon through the archipelago), and the value of α has been estimated at 0.022 (motile lice × day)–1 (based on survival experiments of naturally infected juvenile pink salmon) (8). Dividing the estimated a = 0.89 (95% credible intervals are from 0.46 to 1.34) by 60 days reveals an excellent correspondence between these two independent estimates of pathogenicity (a/60 = 0.015, with 95% credible estimates from 0.0077 to 0.022). Using a hierarchical Bayesian simulation (28) that represents uncertainty in the model fit as well as in the distribution of r* (12), we found the estimated mortality of pink salmon, 1 – exp[–aPi(t – 1), caused by lice ranged from 16% to over 97% and was commonly over 80% (Table 1). The lowest mortality comes from fallow populations when louse abundance was nevertheless elevated, possibly resulting from transmission from active farms outside the fallowed corridor (7, 19, 20).

Table 1.

Mean abundances, P, of motile L. salmonis on juvenile pink salmon and estimated parasite-induced host mortality, M (with upper and lower bounds of the 95% credible interval in parentheses), for exposed populations during infestations.

View this table:

These results provide strong empirical evidence that salmon farm–induced L. salmonis infestations of juvenile pink salmon have depressed wild pink salmon populations and may lead to their local extinction. However, this parasite threat may not exist at low farm salmon abundances; the delay between the onset of salmon aquaculture in 1987 and louse infestations in 2001 (Fig. 2) may be explained by farm fish abundance crossing a host density threshold above which outbreak conditions occur (30). It is unlikely that another factor is responsible: The increased growth rate in response to fallowing rules out other factors that could have affected exposed, but not unexposed, populations. The results rely on extensive spatial replication to compensate for short time series in infestation years. The time to reach sufficient temporal replication to support hierarchical mixed-effects modeling, say 10 generations (which equals 20 years), greatly exceeds the predicted time to extinction. That is, there is a major risk associated with waiting for large data sets to accumulate before implementing conservation policy. Industrial aquaculture is rapidly expanding to new species, regions, and habitats (31), which can create parasite outbreaks that contribute to the decline of ocean fisheries and ecosystems.

Supporting Online Material

Materials and Methods

Figs. S1 to S3

Tables S1 to S3

Dataset S1

References and Notes

View Abstract

Navigate This Article