- Top of page
- Model description
Worldwide depletion of fish stocks has led fisheries managers to become increasingly concerned about rebuilding and recovery planning. To succeed, factors affecting recovery dynamics need to be understood, including the role of fisheries-induced evolution. Here we investigate a stock’s response to fishing followed by a harvest moratorium by analyzing an individual-based evolutionary model parameterized for Atlantic cod Gadus morhua from its northern range, representative of long-lived, late-maturing species. The model allows evolution of life-history processes including maturation, reproduction, and growth. It also incorporates environmental variability, phenotypic plasticity, and density-dependent feedbacks. Fisheries-induced evolution affects recovery in several ways. The first decades of recovery were dominated by demographic and density-dependent processes. Biomass rebuilding was only lightly influenced by fisheries-induced evolution, whereas other stock characteristics such as maturation age, spawning stock biomass, and recruitment were substantially affected, recovering to new demographic equilibria below their preharvest levels. This is because genetic traits took thousands of years to evolve back to preharvest levels, indicating that natural selection driving recovery of these traits is weaker than fisheries-induced selection was. Our results strengthen the case for proactive management of fisheries-induced evolution, as the restoration of genetic traits altered by fishing is slow and may even be impractical.
- Top of page
- Model description
One quarter of the world’s fish stocks are overexploited, depleted, or recovering, according to the United Nations’ Food and Agriculture Organization (FAO 2009). Although biological extinctions are very rare (Dulvy et al. 2003; Swain and Chouinard 2008), several of these declines have lead to collapses of fishing activities to a state of ‘commercial extinction,’ where targeted fisheries are no longer commercially viable (Myers et al. 1996). Infamous examples of commercial extinctions of major fisheries targets include stocks of sardine Sardinops sagax off California and Japan in the late 1940s, the Peruvian and Chilean stocks of anchovy Engraulis ringens in 1972 (Csirke 1977; Murphy 1977), the Norwegian spring-spawning stock of herring Clupea harengus in 1968 (Toresen and Østvedt 2000), and the Newfoundland-Labrador stock of cod Gadus morhua through the 1980s (Hutchings 1996). Classic theory of fishing suggests rapid population recovery if fishing is ceased, but in practice recovery rates have been much slower than expected, and in some cases the expected recovery has not taken place at all (Hutchings 2000a).
Stock collapses have enormous social and economic costs (e.g., Haedrich and Hamilton 2000), and painful experiences have led politicians and fisheries management institutions to gradually shift focus from ‘how much to catch’ to ‘how to make sure there is something to catch’. For example, the United Nations 2002 World Summit on Sustainable Development declared that all fish stocks should be restored to levels that produce maximum sustainable yield by the year 2015. In the United States, the Magnuson-Stevens Fishery Conservation and Management Act of 1996 mandates that overfished stocks should be rebuilt within <10 years, unless some circumstance such as species biology dictates a longer time frame (Safina et al. 2005; Rosenberg et al. 2006).
To rebuild a stock successfully, it is crucial to have a good understanding of the factors that influence recovery dynamics and the associated timescales. Recovery can be interpreted as a reversal of processes involved in fishing a population down to low abundance. Hence, if we first understand what happens to fish stocks when we harvest them, it then becomes easier to address the subsequent question of how various harvest-induced changes are reverted if fishing is reduced or ceased. The dynamics a population exhibits in response to fishing involves changes at several levels of system organization. Below we review five of these. The classic theory of fishing considers the first two levels. The most obvious effect of fishing is a reduction in population abundance and biomass. This effect is represented in all fishery models. However, not all biomass is the same: at a second level, fishing changes the demographic composition of a stock toward a dominance of younger and smaller fish. A truncated age and size structure may have consequences for population dynamics, and has been shown to reduce reproductive potential (Marteinsdottir and Thorarinsson 1998; Murawski et al. 2001), increase variability in recruitment or population abundance (Longhurst 2002; Hsieh et al. 2006), and make a stock more vulnerable to environmental fluctuations (Ottersen et al. 2006; Hsieh et al. 2008).
At a third level, fishing may change the biotic environment, triggering changes in phenotypically plastic traits of individuals. Fishing may reduce intraspecific competition, thereby promoting phenotypic plasticity in the form of increased growth, which often leads to earlier maturation (Trippel 1995; Lorenzen and Enberg 2002; Kell and Bromley 2004). Fishing may also change interspecific interactions, an appreciation that has sparked a move toward the ecosystem approach to fisheries management (Pikitch et al. 2004; Francis et al. 2007).
At a fifth level, because of several of the processes above, harvesting can lead to changes in ecosystem structure, potentially causing ecological regime shifts and alternative stable states (Jackson et al. 2001; Scheffer et al. 2005). Frank et al. (2005) have suggested that removal of top predators from the ecosystem may lead to cascading effects that affect the whole ecosystem. Such trophic cascades may have long-lasting consequences, and also influence recovery processes of harvested populations (Frank et al. 2005).
The extents to which overfishing affects stock biomass, population structure, phenotypic plasticity, adaptive evolution, or ecosystem structure have important implications for recovery. In particular, the failure of some fish stocks to recover after their collapse (Hutchings 2000a; Hutchings and Reynolds 2004) has raised concerns as to whether fisheries-induced evolution is contributing to this lack of recovery (Hutchings 2004, 2005; Walsh et al. 2006). For example, before the infamous collapse of the northern cod off Newfoundland and Labrador, marked changes occurred in key life-history traits such as maturation schedule (Olsen et al. 2004). Around the same time, other cod stocks in the region showed changes in growth that are interpreted as being indicative of evolution (Swain et al. 2007, 2008).
The design and implementation of a rebuilding plan depend on the nature of the changes in population-level and individual-level characteristics that occurred through fishing, and on the degree to which these changes are anticipated to be reversible. Our aim in this study is to investigate the rebuilding of stocks following a period of exploitation, focusing on the role of fisheries-induced evolution and using a model that includes four of the five biological response levels described above (excluding ecosystem-level effects). The life history of our model population resembles that of a slow-growing and late-maturing fish species such as Atlantic cod Gadus morhua in the northern parts of its range. By comparing the evolutionary model to a model version in which evolution is ‘turned off,’ both during the exploitation phase and throughout the subsequent recovery, we address how fisheries-induced evolution affects the biological dynamics and examine implications for recovery. Focal questions are: How will adaptive evolution change the expected rate and extent of recovery? And, when fisheries management is oblivious of fisheries-induced evolution, what errors will be made?
- Top of page
- Model description
We use an individual-based model that combines the quantitative genetics of evolving life-history traits with individual-level ecological processes of growth, survival, and reproduction and with population-level ecological processes such as density dependence and environmental variability. Our modeling methodology extends earlier individual-based evolutionary models (Holland 1992; Huse et al. 1999) and falls within the framework of eco-genetic modeling (Dunlop et al. 2007, 2009; Höök and Wang 2009; Okamoto et al. 2009).
Briefly described, each individual carries genetic traits affecting its life history through growth rate, maturation schedule, and reproductive investment. An individual’s genetic traits are expressed imperfectly, to allow for the chance environmental variation of phenotypic traits around genetic traits. The individual’s genetic traits together with its environment thus determine its life history of growth and reproduction, and its risk of dying from natural causes or fishing. Population dynamics emerge when many such interacting individuals are coupled through a shared environment. The environmental influences are manifested as density dependence and stochastic fluctuations acting on recruitment and growth conditions, and fitness emerges as the success of individuals to grow, survive, mate, and reproduce—thus producing offspring that transmit their genetic traits to future generations. We model only female life histories. However, the ‘females’ in our model reproduce sexually and mate with each other, and can therefore be considered as hermaphrodites. The model thus includes sexual reproduction and evolves toward evolutionarily stable strategies for the female sex; it is not meant to predict how evolution might differentially affect males.
A model in which genotypes are inherited and selection acts on phenotypes requires careful consideration of the noise processes affecting the link from genotypes to phenotypes. Without environmental variation confounding this link, the coupling of genotypes to selection pressures would be too strong, the quantitative traits would have unrealistically high heritabilities, and the speed of evolution would be exaggerated. Adding environmental variation obscures the link from genotypes to phenotypes, and thus to selection pressures, and helps bring heritabilities down to levels often observed in nature (0.2–0.3 for life-history traits; Gjedrem 1983; Carlson and Seamons 2008). In living organisms, the ‘noise’ in the correlation between genotypes and selection pressures is a conglomerate of different processes, most of which are not fully understood. We have chosen to infuse such environmental variability at different stages through the following noise components:
Inheritance of genetic traits. For each genetic trait, an offspring’s trait value typically deviates from the mid-parental value, reflecting the effects of mutation, segregation, and recombination.
Phenotypic expression of genetic traits. At birth and for each genetic trait, the individual’s expressed phenotypic trait deviates from its inherited genetic trait, reflecting micro-environmental variation as well as chance effects of epistatis and dominance.
Population-level inter-annual variation in growth conditions. This reflects temporal fluctuations in the abiotic and biotic environment, for example, in temperature or resource availability.
Individual-level inter-annual variation in growth conditions. This reflects variability between individuals because of chance events, for example, in resource acquisition or environmental exposure.
Population-level variation around recruitment function. This reflects a stochastic element of population dynamics and affects growth conditions through density dependence.
Demographic stochasticity in mortality. An individual’s probability of dying from natural causes or harvesting is determined by its growth strategy and size, but whether it actually dies is a random event.
Demographic stochasticity in mating. Similarly, an individual’s probability of being a parent depends on its gonad size, but random parents are drawn according to this probability.
We have used normally distributed random variables for (i), (ii), and (iv). For (iii) and (v), lognormal distributions were used. A lognormal distribution was chosen for (iii) because ecological data is often characterized by distributions with a long tail (Hilborn and Mangel 1997) and for (v) following several studies that consider this process lognormal in nature (Hennemuth et al. 1980; Caputi 1988; Fogarty 1993a,b; Myers et al. 1995).
Genetic traits and their expression
Each individual possesses four inherited quantitative genetic traits: a growth coefficient g that affect its resource acquisition, two traits that specify its maturation schedule through the slope s and intercept y of a linear probabilistic maturation reaction norm (PMRN; Heino et al. 2002a; Dieckmann and Heino 2007) and a trait r that quantifies its reproductive investment in terms of its gonado-somatic index (gonad mass/somatic mass) and thus governs resource allocation to reproduction from maturation onward. The population’s distribution of these genetic traits determines their additive genetic variances.
During expression, phenotypic traits (denoted by G, S, Y, and R) are subject to different noise processes as described above. Here and in the following, ɛ with different subscripts denotes a random number drawn from a normal distribution with mean 1 and given standard deviation (Table 1). The phenotypic trait S(i) for individual i’s genetic PMRN slope s(i) is thus
where ɛs(i) describes the expression noise and is drawn once per lifetime (the argument i in ɛs(i) indicates that different values are drawn for each individual). Analogously, we have and (see Table 2 for units).
Table 2. Model variables.
|Individual length||L(i, t)||cm||[3, 5, 7, 9]|
|Individual weight||W(i, t)||kg|||
|Individual fecundity||Q(i, t)||–|||
|Offspring number||n0(i, t)||–|||
|Genotypic growth coefficient||g(i)||[2, 8]|
|Genotypic PMRN intercept||y(i)||cm|| |
|Genotypic PMRN slope||s(i)||cm·year−1|||
|Genotypic gonado-somatic index||r(i)||–|| |
|Phenotypic growth coefficient||G(i, t)||[2, 5]|
|Phenotypic PMRN intercept||Y(i)||cm|| |
|Phenotypic PMRN slope||S(i)||cm·year−1|||
|Phenotypic gonado-somatic index||R(i)||–||[5, 6]|
|Instantaneous predation mortality rate||mp||year−1||[7, 8]|
|Instantaneous foraging mortality rate||mf||year−1|||
|Maximum instantaneous harvest rate||fmax||year−1|||
|Instantaneous harvest mortality rate||F(i, t)||year−1|||
|Total instantaneous mortality rate||Z(i, t)||year−1|| |
An individual’s phenotype consists of its expressed genetic traits S(i), Y(i), R(i), and G(i, t), its length L(i, t), its age A(i), and its maturity status. The time step in our model is 1 year. The annual modeling cycle can be divided into four processes: (1) sexual maturation, (2) growth of soma and gonads, (3) natural and fishing mortality, and (4) reproduction and inheritance.
We model the maturation process using PMRNs (Fig. 1, Heino et al. 2002a; reviewed in Dieckmann and Heino 2007). Whether an immature individual is likely to mature in a given year probabilistically depends on its age, size, and its PMRN. We assume a linear PMRN with constant width. The PMRN is thus determined by the phenotypic intercept Y(i) and slope S(i). The probability to mature is described by the logistic regression:
Figure 1. Functions used in the model and examples of emergent population dynamics and life histories at the stochastic equilibrium before harvesting. (A) The maturation process is governed by a probabilistic maturation reaction norm (PMRN), with the slope and intercept (with the vertical axis) of the Lp50 curve (at which the maturation probability reaches 50%; continuous black line) serving as evolving traits, and with the PMRN width w between the Lp25 and Lp75 curves (dashed lines) being kept constant. The average PMRN is shown together with the average growth curve (gray curve). (B) Individual growth is density-dependent, so that increased population biomass decreases the growth rate linearly. The continuous black line shows the corresponding dimensionless factor (Eqns 2 and 14), with a thick part indicating the range of encountered biomasses over the 100 simulated years. (C) The number of recruits (at age 1 year) depends on the population’s total egg production. The continuous black curve shows the expected deterministic recruitment values, while gray dots show stochastic recruitment values. (D) A sample time series of the population-level inter-annual environmental noise that influences the resource intake of all individuals. (E) Growth curves and annual fecundities of three individuals subject to the environmental variability highlighted in gray in panel D. (F) A sample time series of population biomass (black line) and number of recruits (gray line).
Download figure to PowerPoint
where L(i, t) is the length of individual i in year t, and Lp50(i, t) is the length at 50% maturation probability, calculated as . The parameter δ is determined by the PMRN width w as
with , where pl and pu specify the lower and upper probability bounds, respectively, chosen for defining the PMRN width (in our model, we choose quartiles, pl = 0.25 and pu = 0.75; Fig. 1A). Maturation is modeled as a stochastic process of Bernoulli trials, and takes place if a number randomly drawn from a uniform distribution between 0 and 1 is smaller than p(i, t).
Growth of soma and gonads
Before maturation, R(i) = 0, as juveniles use all acquired resources for somatic growth. Equation (5) implies a maximum possible gonado-somatic index of , at which R(i) hence is capped. Individual fecundity Q is given by gonad weight divided by the weight We of a single egg,
The amount of available resources is variable, so that under unfavorable resource conditions, mature individuals may not have enough resources to grow as well as to reproduce. In such cases, individuals will prioritize reproduction over growth.
Natural and fishing mortality
Natural mortality consists of three components: (i) size-independent mortality m0 because of, for example, diseases and parasites; (ii) size-dependent predation mortality mp during activities other than foraging (e.g., resting, migrating, and hiding); and (iii) size-dependent predation mortality mf related to foraging (C. Jørgensen and Ø. Fiksen, unpublished data). We base the size dependence of mortality on observations in marine systems showing that mortality scales with length as with an allometric exponent of about d2 = 0.75 (Peterson and Wroblewski 1984; Brown et al. 2004). Because length can change substantially over 1 year, we use the average length in a year to determine the instantaneous rate of predation mortality,
While smaller individuals are generally more vulnerable to predators, all individuals can accept a higher foraging mortality to achieve a higher resource intake and, consequently, a higher growth rate (Walters and Juanes 1993; Biro et al. 2006; Biro and Post 2008; C. Jørgensen and Ø. Fiksen, unpublished data). In our model, the higher resource intake enabled by higher risk-exposure thus implies higher foraging mortality,
The foraging mortality scales with the overall size-dependent predation mortality mp, and is thus higher for smaller fish, which is in line with observations of juvenile fish spending much of their time hiding from predation and trying to minimize the risk associated with foraging (Walters and Juanes 1993 and references therein). This risk associated with foraging can depend, for example, on the total time spent foraging, which, when increased, results in higher encounter rates with predators.
The total instantaneous mortality rate is Z(i, t) = m0 + mf(i, t) + mp(i, t) + F(i, t), and individual i’s resultant annual probability of dying is P(i, t) = 1−e−Z(i,t)·year, which the model again realized through Bernoulli trials. The parameter values chosen (Table 3) produce natural mortality rates that are comparable with estimates from field studies and that are used in assessment work of Atlantic cod (Sinclair 2001; ICES 2003), giving a total instantaneous natural mortality rate of m0 + mf(i, t) + mp(i, t) ≈ 0.25·year−1 for an individual of the average age and size at maturation.
Reproduction and inheritance
The number n0(i, t) of offspring produced by parent i in year t is proportional to that parent’s share of total population fecundity,
Here, the sum extends over the entire mature population and N0(t) is the total number of recruits in year t as determined by Beverton–Holt recruitment as explained under ‘density-dependent recruitment’ below (Eqn 13). The parent produces each offspring with a randomly selected partner. The partner is chosen with a probability proportional to its gonad size. The use of multiple partners is motivated by the many marine fish that are batch spawners. For example, Atlantic cod can produce 20+ batches within a month (Kjesbu et al. 1996) and there is thus a high probability that the offspring of one female are sired by several partners.
In our model, trait inheritance follows quantitative genetics theory (Roughgarden 1979; Falconer and Mackay 1996; Lynch and Walsh 1998), as life-history traits are usually highly polygenic quantitative characters determined by many loci (Roff 1992; Conner and Hartl 2004). An offspring inherits the genetic traits from its parents, and we randomly draw the offspring’s trait values from a normal distribution with a mean given by the trait’s mid-parental value. Thus, the offspring o of parents i and j will have its genetic traits, here specified for the growth trait, determined by
where θg is randomly drawn from a normal distribution with zero mean that reflects the effects of mutation, segregation, and recombination of the underlying loci; its standard deviation is specified for each trait separately (Table 1) and corresponds to a coefficient of variation of 7.1% in the population prior to fishing, implying a constant mutation-segregation-recombination kernel (Roughgarden 1979). The emergent heritability for age at maturation is around 0.2 at equilibrium before fishing, a conservative value within the range typically observed for life-history traits in general (Gjedrem 1983; Law 2000; Carlson and Seamons 2008) and for the proportion of mature 2-year-old Atlantic cod in particular (Kolstad et al. 2006).
Conspecific density may affect the growth of fish through resource competition (Lorenzen and Enberg 2002). Realized growth is thus influenced also by population biomass (Fig. 1B). We express the effect of density dependence through the factor D(t) that influences growth multiplicatively (Eqn 2),
where d1 specifies the strength of density dependence, B(t) is the total biomass of all individuals aged 1 year or older in year t, and we choose so that the time average of D(t) equaled 1 at the stochastic equilibrium before harvesting.
Model parameterization and model runs
The model was parameterized to describe a population resembling Atlantic cod Gadus morhua in the northern part of its range. We expect this parameterization to be roughly representative also of other slow-growing and late-maturing fish species. Where available, parameters for cod were taken from the literature (Table 3). However, some parameters are unknown and cannot readily be estimated from available data. Thus, the unknown parameters were chosen following a pattern-oriented modeling strategy, which ensures that the emergent model properties qualitatively and quantitatively resemble the observed natural patterns (Grimm et al. 2005). To achieve this, the model was initially run with likely parameter values and its output compared with data available in the literature and in stock-assessment reports. Parameters responsible for discrepancies were adjusted. This was repeated until the modeled patterns—such as growth curves, age and size distributions, natural mortality levels, and fecundity measures—resembled the natural patterns observed for Atlantic cod. All model parameters are listed in Tables 1 and 3. Figure 1D shows sample time series of the population-level inter-annual noise affecting resource acquisition. The corresponding sample growth trajectories and annual fecundities are shown in Fig. 1E, while Fig. 1F shows population biomass B(t) and the number N0(t) of recruits. The average PMRN of the preharvest population is shown in Fig. 1A.
Before harvesting was started, the modeled population was allowed to reach a stochastic evolutionary and ecological equilibrium, so that all genetic traits and the correlations among them had converged close to an evolutionarily stable strategy. This was achieved by running the model for several hundred thousand years before saving a population from which the simulations were started. During each of 50 replicate runs with different random seeds, we let the model again equilibrate for 100 years before harvesting was started. In individual-based models, evolution is less influenced by genetic drift and individual-level stochasticity when a population’s size is large. Because the populations we used were large (around 220 000 individuals), genetic drift and other forms of historical evolutionary contingency are less relevant for our results than they would be for smaller populations. Ideally one might have wanted to start simulation from several populations, but because of the long computing time needed for reaching the stochastic evolutionary equilibrium, we created only one such population. In all figures except for Fig. 2 the harvest period was 100 years. There was no fishing during the subsequent recovery period.
Figure 2. Decrease and subsequent recovery of population biomass (of individuals aged 1 year and older) in dependence on harvest duration (increasing from left to right) and instantaneous harvest rate (increasing from top to bottom). Black curves: evolutionary model; gray curves: nonevolutionary model; gray shading: harvesting period. For the highest harvest rate and the longest harvest duration (lower right corner), the nonevolutionary population contains <100 individuals at its lowest population size, so that increasing the harvest rate further would lead to its extinction.
Download figure to PowerPoint
To assess the consequences of evolutionary change for recovery and rebuilding, we created a nonevolutionary version of our model in which the genetic traits were not allowed to evolve. For the initial comparison in Fig. 2, we did this simply by giving each individual born into the nonevolving population the genetic traits of a random individual that was alive at the time when fishing was initiated. We first constructed a library of about 200 000 individuals assembled in the last year of the long stabilization period without fishing; then, during fishing and the subsequent recovery, offspring were assigned genetic traits from this library. In this way, the distribution of genetic traits at birth was prevented from evolving under fishing. Nevertheless, the distribution of genetic traits later in life could change under fishing, through the differential vulnerability of genotypes to fishing. The distribution of phenotypic traits later in life changed under fishing for two further reasons: through the effects of fishing on density and thus on the phenotypic expression of the density-dependent growth coefficient, and through the knock-on effects of density-dependent growth and recruitment altered by fishing on a stock’s length structure and thus on the differential vulnerability of phenotypes to fishing.
A related challenge arises because the ecological conditions of the evolutionary model differ from those in the nonevolutionary model as soon as the populations are fished. This prevents the comparison above from satisfactorily isolating the effects of evolutionary changes on the recovery process, because age, size, and maturity distributions, and hence density-dependent feedbacks on growth and recruitment, differ among populations starting their recovery from different initial biomasses. To better isolate the effects of evolution on recovery (in Figs 4 and 5, as opposed to Fig. 2), we therefore used four steps to scale the stock biomass of the nonevolving population at the beginning of the recovery period to the corresponding level of the evolving population, while ensuring the former population’s demographic and genetic composition matches its adjusted biomass. First, we determined the evolving population’s biomass B(T) at the time T at which fishing was stopped. Second, we fished the nonevolving population with the same harvest rate and identified the year τ just before it reached the evolving population’s target biomass B(T). Third, we determined the nonevolving population’s biomasses BN(τ) and BN(τ + 1) and separately stored all individuals it contained in those 2 years. Fourth, we randomly drew individuals from these two stored populations—complete with genetic traits and other individual traits such as length, age, and maturity status—until the population assembled in this way reached the target population biomass B(T). During this assembly, the probability that an individual was drawn from the population at time τ was
while the remainder was drawn from the population at time τ + 1. In this way, we separately constructed a new nonevolving population for 50 replicate model runs.
Figure 4. Comparison of recovery times in the evolutionary and nonevolutionary models after harvesting for 100 years, when recovery was initiated from the same population biomass in both models (by scaling up the biomass in the nonevolutionary model to that of the evolutionary model). Black curves: evolutionary model; gray curves: nonevolutionary model; thin curves: fmax = 0.3 year−1; thick curves: fmax = 0.7 year−1; dashed lines: preharvest levels. (A) Population biomass; (B) spawning stock biomass; (C) eggs per spawner; (D) recruits (aged 1 year) per spawner; (E) age at maturation; (F) length at maturation; (G) length increment at age 2 years; (H) lengths at ages 3 years (below) and 10 years (above). The preharvest average population biomass, spawning stock biomass, and eggs per spawner are scaled to 1.
Download figure to PowerPoint
Figure 5. Recovery times to 50%, 70%, and 90% of preharvest levels for different instantaneous harvest rates fmax. (A), (B) Population biomass; (C), (D) number of recruits; (E), (F) age at maturation; (A), (C), (E) evolutionary model; (B), (D), (F) nonevolutionary model.
Download figure to PowerPoint
- Top of page
- Model description
To quantify the impact of evolution on recovery, we compared responses in the biomasses of evolving and nonevolving populations to fishing and a subsequent moratorium (Fig. 2), investigated the corresponding evolution of genetic traits (Fig. 3), and analyzed the differential dynamics of evolving and nonevolving populations that were of equal biomass at the recovery’s start (Fig. 4), with a special focus on identifying differences in recovery times (Fig. 5).
Figure 3. Evolutionary dynamics of mean genetic trait values during harvesting (gray shading) and during a subsequent moratorium. Dynamics are shown for three different instantaneous harvest rates: fmax = 0.1 year−1 (light-gray curves), fmax = 0.5 year−1 (dark-gray curves), and fmax = 0.9 year−1 (black curves); thinner outer curves indicate ranges of ±1 SD. (A) Probabilistic maturation reaction norm (PMRN) intercept; (B) PMRN slope; (C) gonado-somatic index; (D) growth coefficient.
Download figure to PowerPoint
We found that increased harvest, in terms of intensity or duration, magnified the evolutionary response of the harvested stock. This can be seen by comparing time series of total biomass for the evolving population with its hypothetical nonevolving counterpart (Fig. 2). The evolutionary response involved adaptations in the genetic traits that allowed the evolving population to better withstand fishing during the harvested phase; these adaptations persisted after harvesting had ceased. Since the evolving population adapted to the harvest, its biomass began to rebound after the initial drop, whereas the nonevolving population declined monotonically as long as it was harvested.
One consequence of fisheries-induced evolution is that populations that adapted to the new mortality regime could withstand considerably higher fishing pressures than if evolution was not occurring. While the hypothetical nonevolving populations went extinct for harvest rates exceeding 0.5 year−1, the evolving populations acquired a capacity for withstanding considerably higher harvest rates. However, even though fisheries-induced evolution appeared to make populations more resistant to extinction, the flip side of adaptation was observed when the harvest pressure was removed. Although starting recovery from a higher biomass, the evolving populations could not fully recover to the preharvest level within the recovery period of 350 years shown in Fig. 2. This lag effect was the more pronounced the more intense and prolonged a population’s exposure to fishing had been.
The capacity of evolving populations to withstand higher harvest pressure resulted from evolution of their genetic traits. The most prominent evolutionary change took place in the maturation schedule, and in the PMRN intercept in particular (Fig. 3A). The PMRN also evolved to become more steeply inclined during harvesting (Fig. 3B), the gonado-somatic index evolved to higher values (meaning that individuals invested progressively more of their resources into reproduction; Fig. 3C), and the growth coefficient evolved to lower values (Fig. 3D). The higher the harvest pressure, the larger the magnitude of the evolutionary response. In all genetic traits, recovery was slower than the preceding harvest-induced changes. Since the heritabilities of genetic traits did not change significantly over time, this implies that the natural selection pressures reverting the genetic traits during the moratorium were weaker than the preceding fisheries-induced selection pressures.
Harvesting changed a range of stock characteristics in the evolving and nonevolving populations, including those commonly used for quantifying biomass, recruitment, maturation, and growth (Fig. 4). Although the genetic growth coefficient evolved to lower values (Fig. 3D), the phenotypic growth rate increased because of relaxed density regulation (Fig. 4G). This finding underscores the importance of considering phenotypic plasticity and density dependence when modeling fisheries-induced evolution, or when interpreting observed empirical changes in the light of fisheries-induced evolution, as phenotypic plasticity and density dependence can both mask and exaggerate fisheries-induced changes in a stock’s genetic composition.
As evolution changes a population’s genetic traits, it also affects population dynamics and biomass variation (Fig. 4). A naïve comparison of evolving and nonevolving populations, such as in Fig. 2, thus cannot separate the effects of evolution from the effects of phenotypic plasticity and density dependence, because the recovery processes of the evolving and nonevolving populations start from different population biomasses after harvesting. We eliminated this confounding factor by rescaling the nonevolving population’s biomass to the same level observed for the evolving population at the end of the harvesting period (see ‘Nonevolutionary model’ under ‘Methods’ above). Initiating the recovery of evolving and nonevolving populations from the same biomass revealed that fisheries-induced evolution had little influence on the recovery of population biomass during the moratorium’s first 15 years (Fig. 4A). However, after this initial period, the evolving population took hundreds of years to fully recover to its preharvest biomass level. In contrast, the hypothetical nonevolving populations were fully recovered within about 50 years.
The recovery of some stock characteristics was faster for populations that had undergone fisheries-induced evolution. For example, spawning stock biomass increased more rapidly during the moratorium, because individuals were maturing earlier as a result of fisheries-induced evolution. Likewise, when harvesting ceased, spawning stock biomass in the evolving population exceeded its preharvest level, because a larger part of the population was mature and thus contributed to the spawning stock (Fig. 4B). At the same time, the number of eggs per spawner and the number of recruits per spawner decreased considerably due to fisheries-induced evolution, because mature individuals were on average smaller (Fig. 4C,D). Furthermore, the recovery of these metrics was extremely slow, and there was little short-term recovery. In contrast, the nonevolving populations showed a demographic increase in the number of eggs and recruits per spawner at the beginning of the recovery period, because the number of old and large individuals increased when harvesting was ceased. With time, density dependence began to kick in and reproduction thus fell to preharvest levels. The initial demographic increase and subsequent density-dependent decrease in reproductive traits were also present, albeit less pronounced, in the evolving populations.
The average age at maturation is one of the stock characteristics that responded most strongly to fishing. The reason for this is threefold: first, evolution of the PMRN increases the probability of maturation for smaller individuals at younger ages; second, released density dependence increases growth and thus allows fish to reach sizes at which they are more likely to mature early in life; and third, population-level averages of age at maturation are based on live fish and therefore are biased toward early maturation, because harvest removes fish that otherwise would have matured late in life. In Fig. 4, the evolving populations exhibited all three effects, whereas the nonevolving populations underwent only the latter two effects (Fig. 4E). Even though the average age at maturation decreased in the nonevolving populations, the decline was smaller and the recovery to the preharvest level was considerably quicker (Fig. 4E).
The average length at maturation decreased dramatically because of fisheries-induced evolution and recovered very slowly thereafter, whereas in the nonevolving populations it not only decreased less during fishing, but also rebounded to its preharvest level within just a few years (Fig. 4F). For a PMRN with significantly negative slope, one would expect that faster growth leads to earlier maturation at larger size when everything else is unchanged. In the nonevolving populations, however, (i) average maturation length at the end of the fishing period lies below its preharvest value even though individuals grow faster during the fishing period and (ii) average maturation length increases during the early phase of recovery even though growth rates go down concomitantly. Both points seemingly contradict expectations based on the PMRN slope. The contradiction is only apparent, as late- and large-maturing individuals were more exposed to fishing, resulting in a bias toward fish that are smaller at maturation. When fishing pressure was released, this bias vanished, and the average length at maturation quickly increased. Thereafter, the average length at maturation slowly decreased again together with the decreasing average growth rate and thus in line with expectations based on the PMRN slope.
The average length increment of 2-year-old individuals was only minimally influenced by fisheries-induced evolution (Fig. 4G). This early age was chosen to avoid the confounding effects resulting from maturation evolution and the associated changes in resource allocation to gonads. Although the growth coefficient evolved to lower values during harvesting (Fig. 3D), the length increment of 2-year-old individuals increased slightly and decreased relatively quickly to the preharvest level as population density increased. This weak effect of growth evolution on body length can also be observed at other ages: length at age 3 years decreased little because of evolution, and although length at age 10 years showed a dramatic response, this was mainly because of maturation evolution (Fig. 4H). In the nonevolving populations, the average length at age 10 years increased during the harvesting period because of the release of density dependence acting on growth. In contrast, fisheries-induced evolution led to a smaller length at age 10 years, because individuals evolved to mature earlier in life, so that their growth slowed down as part of their resources were used for reproduction (Fig. 1E). The recovery of the length at age 10 years was very slow in the evolved populations, reflecting the slow evolutionary recovery of the maturation schedule.
Increased harvest pressure led to longer recovery times. We quantified this as the time it took a population to reach 50%, 70%, and 90% of its preharvest levels of biomass, number of recruits, and average age at maturation (Fig. 5). Biomass recovery to 50% and 70% of the preharvest level was not strongly influenced by fisheries-induced evolution, whereas recovery to 90% of the preharvest level took roughly 10 times longer when the population had adapted to fishing (Fig. 5A,B). Recruitment, on the other hand, recovered faster when the population had evolved: the fish matured earlier and made higher reproductive investments, and the resultant number of recruits was also larger. As a consequence, the recovery time to preharvest levels of recruitment was shorter for evolved populations than for nonevolved populations (Fig. 5C,D). Recovery of the average age at maturation was strongly affected by fisheries-induced evolutionary changes (Fig. 5E,F). Recovery to 70% of the preharvest maturation age took more than 300 years when harvest rates had exceeded 0.5 year−1, and recovery to 90% of the preharvest maturation age could take more than a thousand years (Fig. 5E). In the hypothetical nonevolving populations, age at maturation recovered to 90% of its preharvest level within <40 years (Fig. 5F), and it never fell below 80% of its preharvest level (not shown).