- Top of page
- Literature cited
Captive rearing often alters the phenotypes of organisms that are destined for release into the wild. Natural selection on these unnatural phenotypes could have important consequences for the utility of captive rearing as a restoration approach. We show that normal hatchery practices significantly advance the development of endangered Atlantic salmon (Salmo salar) fry by 30+ days. As a result, hatchery fry might be expected to face strong natural selection resulting from their developmental asynchrony. We investigated patterns of ontogenetic selection acting on hatchery produced salmon fry by experimentally manipulating fry development stage at stocking. Contrary to simple predictions, we found evidence for strong stabilizing selection on the ontogeny of unfed hatchery fry, with weaker evidence for positive directional selection on the ontogeny of fed fry. These selection patterns suggest a seasonally independent tradeoff between abiotic or biotic selection favoring advanced development and physiological selection linked to risk of starvation in unfed fry. We show, through a heuristic exercise, how such selection on ontogeny may exacerbate problems in restoration efforts by impairing fry productivity and reducing effective population sizes by 13–81%.
- Top of page
- Literature cited
The coordination of events in a life cycle is theoretically shaped by both ontogenetic and phenological evolution. In other words, individuals that express the optimal developmental condition (i.e. ontogeny) at the optimal time (i.e. phenology) are expected to have higher lifetime fitness and to thus be favored by natural selection. Although, ontogeny and phenology are often considered jointly, they may often reflect distinct adaptations that interact to shape life-history synchrony. For example, when in an organism’s development, and when seasonally, it undergoes a particular life cycle event, it may be affected by growth genes or circadian clock genes respectively. That said, time must to some degree constrain the apparent independence of these traits – a significant developmental delay may constrain whether, or how early, in a season some developmental processes might occur (Brannon 1987). Indeed, temporal linkages among life cycle events may mean that evolution of some ontogenetic and phenological traits are derived, in part, from constraints acting at other life stages (Stearns 1976; Moran 1994; Sinervo and Svensson 1998).
From these principles, one can hypothesize that natural or anthropogenic processes that disrupt the coordination of ontogeny or phenology may reduce the mean fitness of populations, cause population declines or drive further evolution (Bradshaw and Holzapfel 2001; Both et al. 2006). Here, we experimentally assess the strength and pattern of combined ontogenetic/phenological selection in the wild that results from anthropogenic alteration of ontogeny, and consider the potential significance of such selection for the recovery of endangered populations with partial captive propagation.
One common model of the evolution of life cycle synchrony is based on the premise that there is an optimal seasonal window within which individuals should undergo some key life cycle event. This is often referred to as the match–mismatch hypothesis; because individuals that fail to match some life cycle event to this window will suffer reduced performance and fitness (Cushing 1990; Frank and Leggett 1994). This model strongly emphasizes the importance of selection on seasonal timing (phenology), and does not clearly link to selection on ontogenetic variation. Indeed, tests of the match–mismatchhypothesis generally compare the performance of individuals that attain the same developmental state early or late in a season (Armstrong and Nislow 2006). By comparison, an orthogonal design would consider the relative performance of individuals of different developmental condition at the same seasonal time (i.e. controlling for phenology). Such a design would provide a test of ontogenetic selection. By analogy, to the match–mismatch hypothesis, we refer to this as the ‘ready-or-not hypothesis’ in recognition that an organism’s developmental ‘readiness’ may often influence its performance and fitness under prevailing environmental conditions. In real life, these two hypotheses are likely more complementary than mutually exclusive. Nonetheless, we believe that match–mismatch and ready-or-not can lead to qualitatively different expectations for the selective consequences of processes that disrupt normal ontogeny.
Population ecologists are often concerned with particular ontogenetic transitions, sometimes referred to as critical periods (Werner and Gilliam 1984), which have the potential to impose especially high mortality on populations. In this sense, critical periods can be a key factor of cohort strength (Elliott 1989, 1990; Sirois and Dodson 2000; Nislow et al. 2004). In many organisms, the period of transition from parental resources (e.g. endosperm, yolk, provisioning) to independent feeding is thought to represent a critical period. Because of the potentially high mortality experienced in this critical period, it is no surprise that many species conservation programs either seek to: (i) greatly increase the total number of juveniles available to enter this period, or (ii) greatly improve juvenile survival through this period (Brown and Day 2002). In fishes, partial captive propagation, where some or all members of a population are bred in captivity and their offspring are released back into the wild (e.g. hatcheries), is often employed to increase numbers of individuals that enter and survive such critical periods or to avert such critical periods altogether.
Although captive propagation programs often seek to release juveniles into the wild at conducive times, captive rearing environments are often quite different from natural environments. Environmental differences may result in considerable disparities in ontogeny and phenology between captive reared individuals and their wild counterparts (Reisenbichler and Rubin 1999; Mackey et al. 2001). If we assume that natural populations are approximately ontogenetically and phenologically adapted, then we might presume that phenotypic asynchrony induced by artificial environments may increase mortality rates of captive individuals released into the wild. Likewise, anthropogenic asynchrony might alter which genotypes perform best in the wild, causing inadvertent selection with concomitant effects on genetic effective population sizes and adaptive diversity.
The objectives of the present study were threefold. First, we assess the degree to which captive propagation alters the ontogeny of endangered salmon fry and their phenology of exposure to stream environments in the wild. Second, we experimentally assess the strength and pattern of ontogenetic selection acting on captive bred salmon released into the wild during the critical period of transition from endogenous yolk to exogenous feeding (alevin to fry transition). In doing so, we directly assess the ready-or-not hypothesis and indirectly assess the match–mismatch hypothesis with respect to the fitness consequences of hatchery activities. Finally, we heuristically assess the potential for developmental asynchrony and natural selection on artificial phenotypes to confound conservation and restoration goals.
- Top of page
- Literature cited
As expected, hatchery reared fry experienced substantially advanced development compared to wild fish. Based on recorded temperatures, hatchery reared fish spawned in CBNFH would reach 100 DI on 5 May and naturally reared fish spawned in SB would reach 100 DI on 7 June. Conversely, CBNFH fish would reach well over 135 DI by 7 June, whereas naturally reared fish in SB would reach only 67 DI by 5 May (Fig. 1). Phenologically, fry stocking was thus estimated to have occurred 43 days earlier than predicted natural fry emergence, resulting in a dramatically premature exposure of fry to the free-flowing stream environment. Ontogenetically, variation in spawning date and thermal regimes at the hatchery also resulted in stocking fry at a wide range of DI. For the 5 years of data available on the Narraguagus salmon, fry are estimated to have been stocked at anywhere from 87 to 122 DI.
All recaptured fry had thermal marks on their otoliths, and we were able to classify 86% of the fish to a distinct manipulation group (Table 1). Only SB trials in 2006 and 2007 showed highly biased recapture patterns suggestive of strong selection (2006: χ2 1.8 < 0.01; 2007: χ2 1.10 < 0.001). Nonetheless, linear and quadratic selection coefficients provided evidence of varying modes of selection. In general, linear terms were positive in nearly all trials, albeit only marginally significant for the trial in KS (P = 0.078; β = 0.037), indicating selective mortality against individuals of lower DI. Several trials provided significant evidence for negative quadratic terms (SB 2006, SB 2007 and KS 2007; P ≤ 0.10; Table 2; Fig. 3), suggesting stabilizing selection against developmental extremes at stocking. Shorey Brook models had greater quadratic selection terms than models for Penobscot River sites. Indeed, AICc indicated that pure quadratic selection was a better fit (ΔAICc < 4.0) than models of pure linear or linear plus quadratic selection in both SB trials. Models containing quadratic model were not identified as a better fit for any of the Penobscot River trials (ΔAICc < 4.0; Table 3).
Table 2. Opportunity for selection (I), linear (β) and linear-quadratic (γ) selection coefficients for developmental index of Atlantic salmon (Salmo salar) fry at stocking in Maine.
|Stream||I||β (SE)||P||r2||γ (SE)||P||r2|
|SB 2006||0.277||−0.014 (0.068)||0.850||0.006||−0.091 (0.039)||0.063||0.536|
|SB 2007||0.199||0.011 (0.045)||0.812||0.007||−0.051 (0.022)||0.056||0.432|
|KS 2007||0.100||0.03 (0.018)||0.078||0.336||−0.012 (0.006)||0.096||0.565|
|KT 2007||0.086||0.031 (0.018)||0.113||0.283||0.002 (0.008)||0.727||0.297|
|AB 2007||0.184||0.013 (0.029)||0.663||0.025||−0.010 (0.012)||0.459||0.103|
Figure 3. Relative recapture rates by developmental group, and estimated linear (β) and linear-quadratic (γ) selection functions for Atlantic salmon (Salmo salar) fry in Maine.
Download figure to PowerPoint
Table 3. Models of mode of selection on Atlantic Salmon (Salmo salar) fry (ΔAICc < 4.0 used for model selection). Asterisk indicates model is significant at P < 0.01.
|Stream||Linear||Quadratic||Linear-quadratic||Best model fit|
|KS 2007||10.206*||11.298||11.977*||All similar|
|KT 2007||9.485||12.745||15.299||Linear, quadratic|
|AB 2007||20.154||19.580||25.315||Linear, quadratic|
Heuristically, we found that selection against the hatchery-produced phenotypes could theoretically result in substantial demographic and genetic costs (Fig. 4). Postselection, we estimate that stocked fry could suffer as much as a 24–81% reduction in overall survival (assuming hard selection). In addition, the nonrandom survival of offspring from various spawning groups increased variation in offspring production among adult salmon, and thus greatly reduced the estimated genetic effective population size contributing to the restoration effort (Table 3). Specifically, the average idealized Ne between 2003 and 2007 was 167.7 individuals (27.1 SD). After applying our selection function, average Ne dropped to 63.4 individuals (30.3 SD). This represents over a 2.5-fold reduction from the theoretical optimum (Table 4).
Figure 4. Percentage of Atlantic salmon fry (Salmo salar) stocked into the Narraguagus River (black bars) based on developmental index (2003–2007) and percentage of original numbers stocked (grey bars) remaining after applying an average selection function from 2006 to 2007.
Download figure to PowerPoint
Table 4. Estimated effects of ontogenetic selection on fry abundance and effective population sizes for entire Narraguagus River population (2003–2007). These heuristic estimates assume a single pattern of hard selection throughout the Narraguagus system.
|Year||Total no. fry stocked||Average DI||% Fry reduction||Estimated effective population size|
- Top of page
- Literature cited
We found that captive propagation greatly advances the development of hatchery salmon fry relative to expectations for their wild counterparts. We also found evidence for selective mortality on the development stage of hatchery fry following release into the wild. However, contrary to simple predictions that selection would be strongly directional, due either to fry asynchrony with an optimal seasonal window (match–mismatch) or a general advantage for larger and more advanced fry (ready-or-not), we found stronger evidence for stabilizing selection (significant in two of five trials). That said, the general trend in the linear components of selection was positive in nearly all trials (statistically significant in one). We suggest that these findings for hatchery-stocked fish generally support a modified version of the ready-or-not hypothesis. In the remainder of this discussion, we consider the proximal basis for these patterns of selection, how they might compare with selection on wild fry, and the potential implications of such selection for salmon recovery in Maine.
We have demonstrated a wide disparity in the phenology of fry experiencing natural and hatchery rearing regimes (Fig. 1). Hatchery fry are stocked, and enter the stream environment, roughly a month earlier on average than is expected for their wild counterparts. Early May is typically a time of high and variable stream flows and variable temperatures, whereas June is dominated by more constant stream flow and temperature with an added factor of abundant food resources in the form of dramatic hatches of stream invertebrates. Also significant, we found that variation in hatchery and stocking practices can result in fry being stocked at anywhere from 87 to 122 DI. Although some of these fry may be able to seek shelter back in the gravel, abrupt exposure to stream environments would be expected to have very different consequences across such a large developmental range.
In both years of this study, fry released into SB faced strong stabilizing selective pressures over a relatively small window of development (Table 2). Although selective optima differed modestly between years (DI = 109 in 2006 and DI = 105 in 2007), observed patterns of selection were generally very similar. The slight difference in the optimum fry size between years may relate to higher than average flows during the spring of 2006. Larger fry may be better equipped to cope with such turbulent conditions. Flows in 2007 were more typical for both SB and the Penobscot. Regardless of these slight differences between years, our estimates of quadratic selection for SB fry fall close to the median of other estimates of quadratic selection values in the wild for a wide range of taxa (γ = −0.091 for 2006 = 40th percentile and γ = −0.051 for 2007 = 55th percentile; Kingsolver et al. 2001).
Penobscot fry do not appear to have encountered as strong of a stabilizing selection regime as SB fry. Rather, trends in the three Penobscot groups showed a greater linear component favoring larger fry. Shorey Brook (Narraguagus) and the Penobscot River are geographically disjunct and their fry originate from different parental sources; hence, divergent adaptations or parental rearing regimes may account for some of these differences in patterns of selection. However, we believe the most parsimonious explanation for different patterns of selection is simply that Penobscot fry were fed before stocking, whereas SB (Narraguagus) fry were not (consistent with usual hatchery practices). Affording developmentally advanced fry a chance to feed before stocking theoretically increases the probability of a successful transition from endogenous to exogenous resources (Nislow et al. 2004) and reduces the risk that advanced fry will starve when stocked into the wild. Under such conditions, larger and more advanced fry may fare better in predatory and competitive interactions than less advanced fry. Letcher and Terrick (2001) found that developmentally accelerated, but unfed, fry showed poor survival relative to fed fry or fry that were not fed but stocked at a lesser DI. Indeed, based on lab experiments they inferred that unfed fry were likely approaching starvation within a few days of stocking.
Another factor that might have influenced apparently divergent modes of selection was that Penobscot River fry were stocked at a nominally higher density (50 vs 100 per 100 m2 suitable habitat), which could theoretically place a greater premium on the competitive abilities of larger fry than in SB. However, we think this is less likely as an explanation given that our stocking densities are relatively low for this species, and probably overestimate the differences in functional densities actually experienced by fry. Nonetheless, future experiments can, and should, be designed to test these alternative hypotheses independent of river source and stocking site.
Although hatchery stocking of fry may be viewed as a manipulation of fry ontogeny and phenology in general, it may also be considered a model for ontogenetic selection on the stage at which fry undergo emergence from the gravel. To the degree that stocking mimics forced emergence into the stream environment, our results may reflect that there is moderate to strong selection for emerging at a relatively advanced stage, albeit this is at times balanced by a risk (e.g. starvation) for fry that delay emergence too late. Fry that developmentally delay emergence, or that are stocked unfed at a late stage, may have expended most of their yolk reserves, and may thus have difficulties learning to feed properly, hastening them to a point-of-no-return (Elliott 1989, 1990). Artificial redd experiments have been used to study selection on date (i.e. phenology) and size of fry under more natural conditions of emergence in the wild (Einum and Fleming 2000), but not selection on their ontogenetic stage at emergence.
Phenologically speaking, fry that emerge too early in the spring are often suggested to suffer severe costs (Elliott 1990; Kennedy et al. 2008). Stream-spawned fish that seasonally delay emergence from the gravel are protected from high flows (Erman and Ligon 1988) and predation (Brännäs 1995), and may enter the stream environment at a time closer to peak spring food abundance. Stream conditions postemergence can be a potent selective force on fry emergence timing (Einum and Fleming 2000). In our study system, stocked fry were developmentally advanced relative to their wild counterparts and normal seasonal phenology. However, we did not find support for the match–mismatch hypothesis prediction that selection would predominately favor stocked fry with lower DI, despite the fact that such fry should be in closer synchrony with normal seasonal phenology (and more similar to wild fry in this regard). Admittedly, this constitutes only an indirect test of the match–mismatch hypothesis, as we did not hold fry ontogeny constant while varying release timing. Nonetheless, a tendency for intermediate or advanced fry to be favored at any given point in time is more consistent with the ready-or-not perspective that developmental preparedness is particularly important to fry performance. This finding is consistent with the experimental findings of Einum and Fleming (2000) in which early emerging fry were favored over later emerging fry, albeit that study involved delaying emergence, rather than advancing it as hatcheries do in the present study.
This is not to say that there is not selection for an optimal seasonal window for wild or hatchery fry to enter the natural environment, indeed, other studies have found evidence to support such a window (Letcher and Terrick 2001; Jones et al. 2003; Kennedy et al. 2008). For that matter, it is quite possible that the very early phenology of stocked fry in this system places a premium on more advanced DI, and is in part responsible for the moderate to strong selection we observed. Overall, our findings merely imply that variation in fitness (i.e. selection) associated with ontogenetic preparedness (i.e. ready-or-not) at emergence/stocking (e.g. competitive ability, swimming ability etc) is likely greater than that associated with the precision of phenological synchrony (i.e. match–mismatch), particularly under the prevailingly advanced developmental and stocking conditions associated with hatchery rearing.
Implications for restoration
Within the last couple of decades, managers have seen an increase in evidence that captive propagation can change the genetic features of populations and inadvertently compromise their fitness and sustainability in the wild (reviewed in Reisenbichler and Rubin 1999; Araki et al. 2008). However, the specific selective mechanisms that underlie inadvertent domestication and fitness reductions are poorly understood (Araki et al. 2008). Most hypotheses have tended to focus on the role of inadvertent artificial selection occurring within the period of captive propagation (Reisenbichler and Rubin 1999; McLean et al. 2005; Araki et al. 2008). Very often, the prescription to minimize or counter these effects in most species management plans is to release individuals into the wild as early as possible to maximize exposure to natural patterns of selection (National Marine Fisheries Service and U.S. Fish and Wildlife Service 2005). Our findings highlight a different and potentially confounding source of potential inadvertent selection and fitness loss – natural selection on artificial phenotypes.
Captive rearing environments can significantly alter the phenotypes that are expressed by particular genotypes. Because these changes result from phenotypic plasticity, and are not necessarily heritable, they are usually not perceived as a long-term threat to the genetic health of populations. This perception might be shortsighted. Upon release, ‘artificial’ phenotypes induced by captive environments may face moderate to strong natural selection, as detected in this study. To the extent that artificial propagation disrupts normal phenotype/genotype relationships, exposure of these phenotypes to otherwise normal modes of natural selection could significantly and permanently alter underlying genetic distributions.
Given that (i) the relative timing and DI of fry stocking can be a largely arbitrary product of fry rearing, and (ii) that all of the offspring of particular males or females spawned in captivity usually experience the same date and DI at stocking; it seems very feasible that natural selection on artificial phenotypes may often favor or disfavor various genotypes in ways that bear little similarity to their performance under natural reproduction. Deleterious evolution similarly seems likely given that traits linked to ontogeny and phenology, such as adult spawning time, egg size, and emergence date are quite heritable (Quinn et al. 2000; Kinnison et al. 2001; Carlson and Seamons 2008). Moreover, in program’s like Maine’s, broodstock derives from stocked fry that are recollected from the wild, potentially compounding deleterious effects on adaptive genetic variation generation to generation. If hatchery operations consistently associate certain genotypes and artificial stocking phenotypes (e.g. early, middle or late spawners) then the natural selection, such as we observed, could ultimately favor the evolution of semi-domesticated genotypes adapted to a combination of early hatchery development and later wild rearing. Insidiously, such domestication may appear as an improvement in the performance of hatchery released fish over time, while masking fitness declines among the fully wild component of the population.
Importantly, this hypothetical outcome is relevant to more aspects of the phenotypes of salmon, and other species, than just phenology and ontogeny. Phenology and ontogeny are convenient traits for demonstrating dramatic phenotypic effects of captive propagation in salmon, but ample evidence shows that hatchery rearing influences phenotypic expression of numerous morphological (Fleming et al. 1994), behavioral (Fleming et al. 1996) and physiological (Fleming et al. 2002) characters in Atlantic salmon and other species (reviewed in Snyder et al. 1996). Further studies like the present one are needed to assess the scope for natural selection on these aspects of artificial phenotypes, and the associated severity of domestication.
Notably, even if the hatchery were largely random with respect to which genotypes experience a given DI in a given year, we have heuristically shown that natural selection on artificial phenotypes is still expected to increase variation in reproductive success and substantially reduce Ne relative to theoretical expectations. This is consistent with the theoretical effects of selection on Ne in general (Nunney and Elam 1994). This genetic cost would be in addition to a potentially significant demographic cost (e.g. 24–82%) of fry abundance. Admittedly, our heuristic demonstration of such effects is crude at best. We extrapolate selection at one site in two years to an entire drainage, extrapolate selective mortality beyond the bounds of our empirical data, and assume hard selection wherein frequency and density-dependent effects are not factored. However, these concerns can be addressed to a degree. We and others have found evidence to support that selection patterns on juvenile salmon can be relatively consistent within drainages in this region (Good et al. 2001; M. Bailey, unpublished data), somewhat justifying our extrapolation to the entire Narraguagus system. Likewise, hatchery experience suggests that unfed fry stocked at extremely low or high DI probably do not survive well in the wild. Although some of the selection that we quantified at particular sites may arise through competition of fry of different DI (i.e. soft selection; Wallace 1975), stocking densities were again notably modest in our selection trials. Moreover, current hatchery practices usually dictate that fry stocked at a given site are of uniform DI, which would largely negate local soft selection on DI and favor hard selection among stocking groups/sites. Regardless, our heuristic analyses should be considered as suggestive of potential demographic and genetic costs and not demonstrative.
How might the potential deleterious consequences of natural selection on artificial phenotypes be mitigated? One obvious measure in the present case would be to spawn adults, incubate eggs and larvae, and stock fry under a regime that more precisely mimics the natural ontogeny and phenology of various genotypes. This would likely take considerable knowledge of natural salmon systems, and a major revision of normal hatchery operations. Alternative rearing systems, such as spawning channels, stream-side incubator, or artificial redds might all aid such an objective. Alternatively, managers might partly mitigate the effects of natural selection on artificial phenotypes by substantially increasing phenotypic DI variation within families, while reducing mean DI variation among families. By spreading the phenotypes of all families out over a similar DI range, the variance in parental reproductive success due to selection on DI would be substantially reduced. Interestingly, such bet hedging may occur to some degree in nature due to the tendency for individuals to naturally spawn over a period of days to weeks (breeding ecology reviewed in Fleming and Reynolds 2004; Fleming 1996).