Joseph D. DiBattista, Redpath Museum, 859 Sherbrooke St. West, Montréal, QC, H3A 2K6 Canada. Tel.: (514) 398 4086 ext. 0080; fax: (514) 398 3185; e-mail: firstname.lastname@example.org
Selection acting on large marine vertebrates may be qualitatively different from that acting on terrestrial or freshwater organisms, but logistical constraints have thus far precluded selection estimates for the former. We overcame these constraints by exhaustively sampling and repeatedly recapturing individuals in six cohorts of juvenile lemon sharks (450 age-0 and 255 age-1 fish) at an enclosed nursery site (Bimini, Bahamas). Data on individual size, condition factor, growth rate and inter-annual survival were used to test the ‘bigger is better’, ‘fatter is better’ and ‘faster is better’ hypotheses of life-history theory. For age-0 sharks, selection on all measured traits was weak, and generally acted against large size and high condition. For age-1 sharks, selection was much stronger, and consistently acted against large size and fast growth. These results suggest that selective pressures at Bimini may be constraining the evolution of large size and fast growth, an observation that fits well with the observed small size and low growth rate of juveniles at this site. Our results support those of some other recent studies in suggesting that bigger/fatter/faster is not always better, and may often be worse.
Studies estimating selection on quantitative traits have provided valuable insights into natural selection in the wild (reviews: Endler, 1986; Kingsolver et al., 2001; Hereford et al., 2004; Kingsolver & Pfennig, 2004). Nearly all such studies, however, have focused on terrestrial systems, and none of the above reviews includes a single study of a strictly marine species. Moreover, we are not aware of any study measuring selection on a large marine vertebrate – presumably because of the difficulty of obtaining replicated, large samples of individuals that can be tracked over long periods of time. And yet selection on such organisms may be qualitatively different, a point we consider in more detail below. The aim of our study was to estimate the strength and direction of natural selection acting on life history and morphology in a large marine vertebrate. We overcame the usual logistic constraints by intensively studying a localized, insular population of juvenile lemon sharks (Negaprion brevirostris).
We can see several reasons why selection in a marine environment might be qualitatively different from that in terrestrial or freshwater environments. First, the extensive three-dimensional nature of the ocean, as well as the often unpredictable distribution of resources (De Troch et al., 2006), may increase spatio-temporal variation in selection. Secondly, high gene flow in the ocean (Hare et al., 2005; Cowen et al., 2006) may increase maladaptation and thereby maintain strong local selection. This may be particularly true when juveniles reside at local nurseries for extended periods (Castro, 1993), allowing strong local selection despite highly vagile adults. This last prediction, however, may not hold for large marine vertebrates that exhibit strong philopatry, for example cetaceans, some elasmobranchs and pinnipeds (Palsbøll et al., 1995; Goodman, 1998; Feldheim et al., 2004; Hueter et al., 2005). Addressing these suppositions requires selection estimates for marine organisms.
Lemon sharks and the Bimini nursery site
The lemon shark is a large (maximum length: 3.4 m), placentally viviparous coastal species. It is found in the western Atlantic from New Jersey to Brazil, with relict populations along the West African coast as well as in the eastern Pacific between Baja California and Columbia (Compagno, 1984). In the western Atlantic, females give birth on a biennial reproductive cycle to between four and 18 juveniles (Feldheim et al., 2002). Adults provide no direct parental care (Pratt & Casey, 1990), and so juveniles forage independently in shallow, inshore nursery sites. At our study site, juveniles are highly site attached for at least 3 years and have daily home ranges of no more than a few hundred square metres (Morrissey & Gruber, 1993).
The lemon shark population at Bimini, Bahamas, has been intensively studied since 1995, when a yearly tagging and recapture programme was first implemented. The enclosed nature of the nursery lagoon allows for a remarkable sampling efficiency; with approximately 99% of newborn sharks captured each year, and a high proportion of these recaptured in subsequent years (Gruber et al., 2001). This exhaustive sampling allows the estimation of recapture probabilities, survivorship, and natural selection. Here, we estimate selection acting on body size, condition and growth rate.
Body condition (e.g. mass for a given length) is thought to be positively related to fitness; a hypothesis that might be called ‘fatter is better’. In particular, an animal in good condition should have more energy reserves, which should increase survival and reproductive output. This hypothesis is so widely accepted that measures of condition are commonly used as surrogates for fitness in studies of selection. And yet, the condition of an individual can be determined by a combination of environmental factors that include resource availability, habitat quality, and the abundance of predators (for review see Begon et al., 1996). Moreover, achieving high condition may involve many of the same constraints as achieving fast growth (see below). It therefore seems profitable to measure selection on body condition in natural populations.
Fast growing individuals are generally thought to accrue the benefits of large size more rapidly (reviews: Arendt, 1997; Sogard, 1997). To the extent that bigger is better, we might therefore also expect that ‘faster is better’. Indeed, and even more so than for body condition, growth rates are often used as surrogates for fitness in studies of selection. And yet, a growing body of work indicates that fast growth may have attendant fitness costs (Arendt, 1997; Lankford et al., 2001; Biro et al., 2004; Sundström et al., 2005). As one example, the increased foraging effort required to achieve fast growth may increase the risk of predation (Biro et al., 2004; Sundström et al., 2005). As another, the consumption of a large meal may be detrimental to swimming ability and thus predator avoidance (Lankford et al., 2001). Other possible costs of fast growth include trade-offs with defence allocation, developmental stability, energy storage and immune system function (Arendt, 1997; Sogard, 1997). Direct estimates of selection on growth rate are needed to determine whether these costs are manifested in nature, and yet such estimates have been rare until recently (Hendry et al., 2003; McAdam & Boutin, 2003; Carlson et al., 2004).
We perform formal selection analyses using 705 juvenile lemon sharks in six cohorts (1995–2000), each of which was tracked for at least 4 years. From these data, we estimate the strength and direction of selection acting on size, condition and growth. Our findings are then interpreted with respect to the conventional ‘bigger is better’, ‘fatter is better’ and ‘faster is better’ hypotheses. We are reasonably confident that our results will reflect natural (rather than anthropogenic) selection because the population was not subject to heavy fishing pressure during the study period.
Study area and sampling
Bimini, Bahamas, is a mangrove-fringed chain of islands located 85 km directly east of Miami, Florida. The Bimini islands enclose a 21 km2 lagoon that is 0–120 cm deep at low tide and serves as a nursery for approximately 300 juvenile and sub-adult lemon sharks (Morrissey & Gruber, 1993). Each year since 1995, approximately 99% of the juveniles have been captured in two adjacent areas of the Bimini lagoon: North Sound and Sharkland (Gruber et al., 2001). Sampling always takes place between 23 May and 11 June (Table 1), just after pupping by adult females (S. Gruber, personal observation). During this time, newborn and juvenile sharks are captured in 180-m long and 2-m deep gill nets (Manire & Gruber, 1991). All sharks are weighed (kg), measured for precaudal length (PCL, tip of snout to precaudal pit in cm; Compagno, 1984), and tagged intramuscularly with an individually-coded passive integrated transponder tag. Tag number, PCL, and mass are recorded each subsequent time a shark is captured. A small piece of fin (2 mm2) is taken from every shark for subsequent DNA extraction. Genetic analyses of these samples have been used to characterize population genetics (Feldheim et al., 2001) and mating systems (Feldheim et al., 2002), and are used here to aid in determining the age of juveniles (see below).
Table 1. Sample sizes, sample dates, and mean trait values for age-0 and age-1 juvenile lemon sharks at the start (all fish) and end (surviving fish) of each selection interval*.
Data sets and traits
1995 (2–11 June)
1996 (23 May–10 June)
1997 (26 May–11 June)
1998 (26 May–10 June)
1999 (25 May–6 June)
2000 (21 May–8 June)
*Data are numbers of tagged fish in each of the three data sets at the start of each selection interval (starting n), the proportion of those fish that survived to age-1 or age-2, and trait means (mean ± 1 SEM) for fish at the start and end of each selection interval for one data set.
†Fish ages are the first year of life (age-0) and the second year of life (age-1).
All age-0 fish†
48.25 ± 1.7
48.73 ± 1.66
48.36 ± 1.60
48.78 ± 2.03
48.21 ± 1.67
48.31 ± 2.15
1.28 ± 0.28
1.38 ± 0.17
1.35 ± 0.17
1.43 ± 0.21
1.32 ± 0.17
1.27 ± 0.17
1.14 ± 0.21
1.19 ± 0.086
1.19 ± 0.087
1.23 ± 0.095
1.18 ± 0.092
1.13 ± 0.086
Surviving age-0 fish
48.31 ± 1.86
48.46 ± 1.39
48.15 ± 1.70
48.96 ± 1.89
48.14 ± 1.64
48.38 ± 2.17
1.27 ± 0.26
1.34 ± 0.16
1.33 ± 0.18
1.42 ± 0.20
1.32 ± 0.16
1.27 ± 0.14
1.13 ± 0.20
1.18 ± 0.090
1.19 ± 0.085
1.22 ± 0.096
1.19 ± 0.098
1.12 ± 0.095
All age-1 fish without growth data
54.24 ± 3.33
54.02 ± 3.31
53.83 ± 3.25
56.25 ± 2.45
54.19 ± 2.63
1.77 ± 0.39
1.79 ± 0.41
1.80 ± 0.41
2.02 ± 0.31
1.68 ± 0.29
1.095 ± 0.088
1.12 ± 0.10
1.14 ± 0.11
1.14 ± 0.089
1.055 ± 0.12
Surviving age-1 fish without growth data
53.95 ± 3.42
52.74 ± 1.87
52.59 ± 2.26
55.58 ± 2.12
53.30 ± 2.78
1.74 ± 0.40
1.64 ± 0.27
1.65 ± 0.29
1.96 ± 0.27
1.60 ± 0.29
1.10 ± 0.095
1.12 ± 0.12
1.13 ± 0.097
1.14 ± 0.10
1.054 ± 0.086
All age-1 fish with growth data
Growth (cm per year)
6.18 ± 2.88
5.59 ± 2.72
6.25 ± 3.12
7.33 ± 2.19
6.11 ± 2.72
Surviving age-1 fish with growth data
Growth (cm per year)
5.94 ± 2.95
4.80 ± 2.22
4.90 ± 1.64
6.89 ± 2.32
5.067 ± 1.95
The following methods allowed us to confidently assign ages to 91% of the sharks caught between 1995 and 2000; the remainder were excluded from analyses.
Umbilical status has been used to identify newborns (age-0 or ‘young of the year’) since 1997. The umbilical scar is open at birth and then slowly closes during the first few months of life, enabling age-0 sharks to be unambiguously identified at the time of sampling. Age-0 sharks identified in this way were always <52 cm in body length (PCL), and so any shark smaller than this size in the other years (1995 and 1996) was also considered to be of this age (Barker et al., 2005). A few sharks were not caught at age-0 but rather at age-1 or age-2. For these, we determined the year of birth (and therefore age) based on length measurements and an average annual growth range at Bimini of 5.2–7.1 cm (Barker et al., 2005). This method allowed us to narrow the year of birth for a given shark to at most two different years, which we then discriminated between based on microsatellite analyses of family relationships. Specifically, we matched individuals of unknown age to their siblings of known age (for more details see Feldheim et al., 2004).
Recapture probabilities, emigration and survival
Like most other studies, our estimates of selection during a particular interval (one year to the next) were based on whether individuals were recaptured (assumed to have survived: absolute fitness = 1) or not recaptured (assumed to have died: absolute fitness = 0). The reliability of this approach depends on recapturing most of the individuals that were indeed still alive (Letcher et al., 2005). We therefore needed to ensure that we had a high probability of recapturing fish that remained in the study area and that emigration was relatively rare and unbiased with respect to phenotype.
The first of these properties was assessed by using the program mark (White & Burnham, 1999), to estimate annual probabilities of recapture (P, probability of catching individuals that were alive and present in the study area) and apparent survival (φ, probability that individuals were alive and present in the study area). Individual capture histories (captured = 1 or not captured = 0, in each year) were generated for each tagged shark over a 4-year period following its initial capture. Data for age-0 and age-1 juveniles were analysed separately, but all cohorts from each age class were included in the same analysis.
The Cormack–Jolly–Seber model was used as a global starting point for the survival analyses (Lebreton et al., 1992), and four discrete models were tested: (1) survival varies with both age and cohort, (2) survival varies with age but not cohort, (3) survival varies with cohort but not age and (4) survival does not vary with age or cohort. For all of these models, we followed the general convention of allowing recapture probabilities to vary in the most complex way; among all ages and cohorts (e.g. Carlson & Letcher, 2003). Akaike's Information Criterion (AIC) was used to compare the four candidate models, with the best model being that with the lowest AIC score and therefore the highest AIC weight (Burnham & Anderson, 2002). If the best model had an AIC weight >0.8, it alone was used to estimate the probability of recapture and apparent survival. If no single model had an AIC weight >0.8, estimates were averaged across the set of models that had the highest weights (Burnham & Anderson, 2002).
The second property of concern (emigration) was addressed in two ways. First, we performed the above mark analyses including or excluding the few sharks (22 age-0 and 15 age-1) that were initially tagged in the nursery area but later recaptured during opportunistic sampling at other sites around the island. Secondly, we determined whether these emigrants differed phenotypically from non-emigrants by comparing their size and condition when captured earlier at the same age in the nursery site.
We estimated selection acting on four traits: body length, body mass, relative condition factor and growth rate. Body length (PCL) and body mass were measured directly on individual fish at the time of capture. Relative condition factor was calculated as 10 000 × mass × PCL−b, where b is the slope of the regression line of log10 mass on log10 PCL for the entire data set (b = 2.999, r = 0.99). Fish that are heavier (lighter) than expected for their length have a higher (lower) relative condition factor (Schulte-Hostedde et al., 2005). Growth rate was calculated as the change in body length between subsequent years so as to consider changes in structural size.
Selection on length, mass, and condition factor was estimated by relating these traits for individuals at the start of an interval (year i) to whether or not they survived to the end of that interval (year i + 1). Growth rate, however, can only be estimated for fish captured at both the beginning and end of an interval (i.e. those that survive). Selection on growth was therefore estimated by relating the change in length between year i and year i + 1 (here age-0 to age-1) to survival between year i + 1 and year i + 2 (here age-1 to age-2). Because growth rate data were only available for a subset of the fish, estimates of selection on all other traits excluded data for growth rate, thereby maximizing sample size. Selection on growth rate was then estimated by adding this trait to the analyses. These approaches parallel those used by Hendry et al. (2003).
Selection analyses benefit from standardization of both phenotypic trait values and fitness (Lande & Arnold, 1983; Janzen & Stern, 1998). Our trait values were standardized to a mean of zero and a standard deviation of unity (based on fish at the start of each selection interval) within each combination of cohort and age class (Lande & Arnold, 1983; Janzen & Stern, 1998). Fitness was then estimated as survival across the interval (i.e. from age-0 to age-1 or from age-1 to age-2). Any tagged sharks that were captured at the end of a given interval, or in any subsequent year, were known to have survived through that interval (absolute fitness = 1). Any tagged sharks not recaptured at the end of a given interval, or in any subsequent year, were assumed to have died (absolute fitness = 0). Further justification for this latter assumption is provided below. Relative fitness was then determined for each shark by dividing its absolute fitness over an interval (0 or 1), by the mean fitness of all individuals for that combination of cohort/age/interval.
Selection was estimated using standard procedures (Lande & Arnold, 1983; Schluter, 1988; Brodie et al., 1995; Janzen & Stern, 1998; Hereford et al., 2004). First, simple regressions of relative fitness on each trait alone were used to estimate selection differentials (i). Secondly, multiple regressions that included all traits were used to estimate selection gradients (β), which represent selection acting on each trait independent of correlations with the other traits. These regressions excluded body mass because it was too highly correlated with body length (r = 0.99; see Mitchell-Olds & Shaw, 1987). Thirdly, multiple regressions that included two variables (a trait and its squared values) were used to estimate univariate quadratic (nonlinear) selection differentials. Fourthly, multiple regressions that included all traits, as well as all squared and cross-product terms for those traits, were used to estimate univariate and bivariate quadratic selection gradients.
All regressions were logistic in form because of the binary response variable (i.e. fitness = 0 or 1), but here we present coefficients after conversion to their linear equivalents (Janzen & Stern, 1998). The resulting variance-standardized selection coefficients represent the number of standard deviations that selection changes the mean trait value within a generation (Kingsolver et al., 2001). We also calculated mean-standardized selection coefficients (Hereford et al., 2004), which represent the increase in relative fitness for a proportional change in the trait mean (with a coefficient of one theoretically indicative of selection on fitness itself). Thus, a mean-standardized coefficient of 0.50 indicates that a 100% change in the mean of a trait would lead to a 50% increase in fitness.
Our six cohorts allowed a combined analysis that can (1) evaluate temporal variation in selection for a given age and; (2) generate more precise estimates of average selection on a given trait. Specifically, we re-ran the above regressions after including ‘cohort’ as a random factor. Interactions between cohort and coefficients for a given trait reveal the amount of temporal variation in selection on that trait. Coefficients for a given trait without the interaction term in the model, then provide the best estimate of selection on a trait averaged over the six cohorts.
Finally, we used univariate cubic splines (Schluter, 1988; glmswin1.0 spline program, Schluter, 2000) to visualize the form of selection acting on each trait for each combination of cohort and age. To facilitate the interpretation of these fitness surfaces, we used raw trait data and absolute fitness rather than standardized values. We used a binomial model, as well as smoothing parameters (λ) that minimized prediction error and best revealed the general trends: length, λ = 2; mass, λ = −2; relative condition factor, λ = −2; and growth, λ = 2. Multivariate visualizations were not necessary because few traits were considered and these were not strongly correlated (after excluding mass). Further, nearly all bivariate quadratic coefficients were nonsignificant and univariate interpretations were straightforward.
We followed the fates of 450 age-0 and 255 age-1 sharks for at least 4 years after their birth. Although yearly sample sizes (Table 1) are less than ideal for estimating selection (Hersch & Phillips, 2004), these are the first estimates for a large marine vertebrate. Further, they will not be strongly influenced by sampling error because we captured nearly 99% of all newborn sharks (Gruber et al., 2001) and essentially all of the subsequent survivors (see below). Moreover, we were able to combine the six cohorts and thereby estimate selection coefficients with much larger sample sizes.
Recapture and survival probabilities
Each cohort was sampled in five separate years, allowing an analysis of recapture and apparent survival probabilities across four different ages (Fig. 1a,b). Only the first three of these yielded informative estimates, however, because few fish were captured after age-4 (Fig. 1a,b). Yearly recapture probabilities were generally high up to age-3 (0.67–0.85; Fig. 1a,b), but dropped at age-4 (0.20–0.48), presumably because older juveniles were finally leaving the nursery site (B. Franks, unpublished data). Of the four possible survival models starting at age-0 (Table 2), two received much stronger support than the others: one in which survival varied with age but not cohort (AIC weight = 0.3903) and one in which survival varied with age and cohort (AIC weight = 0.3543). Averaging parameter estimates for these two models yielded survival estimates that ranged from 48% to 70% (Fig. 1b). Of the four possible survival models starting at age-1, two models again received much stronger support than the others: one in which survival varied with age but not cohort (AIC weight = 0.59159) and one in which survival did not vary with age or cohort (AIC weight = 0.30394). Averaging parameter estimates for these two models yielded survival estimates that ranged from 43% to 85% (Fig. 1b).
Table 2. Model selection for estimating apparent survival in age-0 and age-1 juvenile lemon sharks.
Number of parameters
Four discrete survival models were tested: (1) survival does not vary with age or cohort (phi(.)), (2) survival varies with cohort but not age (phi(cohort)), (3) survival varies with age but not cohort (phi(age)) and (4) survival varies with age and cohort (phi(age+cohort)).
*Indicates these models were used for inference.
All age-0 fish
All age-1 fish
Our selection estimates were likely robust to some potential biases. First, we had a very high probability of recapturing a tagged fish that was alive and present at the study site. Combining the yearly recapture probabilities of the four sampling events for each cohort, we calculate that the probability of subsequently recapturing an age-0 fish that was alive and present at the site when it reached age-1 was 0.99 (0.92 for an age-1 fish that reached age-2). Secondly, emigration did not confound our selection estimates. Only 22 age-0 and 15 age-1 fish initially tagged in the nursery area were later recaptured in opportunistic sampling at other sites around the island (3.3–7.6 km from the main nursery). Yearly survival and recapture probabilities did not differ between analyses that included or excluded these fish (paired sample t-test comparing estimates with and without emigrants, P > 0.17). When sampled at age-0, future emigrants and residents did not differ in size or condition (Student's t-test: length, P = 0.162; mass, P = 0.181; relative condition factor, P = 0.745). When sampled at age-1, future emigrants and residents did not differ in condition (Student's t-test: P = 0.066) or mass (Student's t-test: P = 0.175), but emigrants were slightly smaller (PCL = 53.02 ± 0.81 mm) than residents (54.51 ± 0.20 mm; student's t-test, P = 0.044).Thus, the only observed difference between residents and emigrants would act in opposition to our inferred selection (see below).
Selection differentials and gradients were closely correlated (r = 0.931, P < 0.001) across trait/age/cohort combinations and were always of the same sign (save one, Table 3). We therefore do not separately discuss differentials and gradients, but rather refer to them collectively as ‘selection coefficients’.
Table 3. Linear (directional) selection coefficients acting on the length, mass, relative condition factor and growth rate of age-0 and age-1 juvenile lemon sharks.
Year of sampling
Coefficients to the left of the slash are variance-standardized, those to the right are mean-standardized.
*P < 0.05, **P < 0.01.
†Fish ages are the first year of life (age-0) and the second year of life (age-1).
‡Regressions include growth-rate data.
§Coefficients in this column are summary estimates of overall selection (i.e. all cohorts combined).
Differentials: age-0 fish†
Gradients: age-0 fish
Differentials: age-1 fish
Gradients: age-1 fish
Linear (directional) selection was variable across cohorts and ages, but some clear patterns emerged (Table 3). For age-0 sharks, selection coefficients for mass and relative condition factor were negative in five of six cohorts (Table 3), whereas the coefficients for length showed no consistency (Table 3; Fig. 2a). When cohorts were analysed together in a single anova model, temporal variation in selection was not significant for length (F5,431 = 0.885, P = 0.491), mass (F5,431 = 0.757, P = 0.581), or condition factor (F5,431 = 1.139, P = 0.339). However, all combined selection coefficients were negative in sign, although none were significant (i.e. P > 0.05, see Table 3). Selection, thus, generally (but not always) favoured lighter fish, both in an absolute sense (mass) and relative to body length (condition factor), a conclusion supported by the cubic splines (Fig. 2c,e). However, the variation associated with many of the cohort-specific and overall selection estimates was so large that they were almost never significant at α = 0.05 (Table 3). The safest interpretation then may simply be that selection does not favour larger fish.
Selection on age-1 sharks was roughly similar to that on age-0 sharks, but much stronger. Length, mass, and growth rate showed negative coefficients in all cohorts, and 10 of the 25 estimates were significant at α = 0.05. Cubic splines confirm the interpretation that selection strongly and consistently favoured small size (Fig. 2b,d) and slow growth (Fig. 2g). Selection on relative condition factor varied dramatically in both sign and magnitude for age-1 sharks, and was never significant (Table 3; Fig. 2f). When all cohorts were analysed together in a single anova model, temporal variation was not significant for any trait (length: F4,240 = 0.936, P = 0.444; mass: F4,240 = 0.905, P = 0.461; condition factor: F4,240 = 0.274, P = 0.895; growth: F4,168 = 0.221, P = 0.926), whereas combined selection coefficients were significant for all traits (save condition factor, see Table 3).
Quadratic (nonlinear) selection did not act in a consistent fashion on any of the traits. Univariate quadratic coefficients were variable in both sign and magnitude (Table A1), but all of the significant ones were negative. This suggests that selection may be stabilizing overall and is at least not disruptive, a pattern confirmed by the cubic splines (Fig. 2). Bivariate quadratic coefficients were also variable and rarely significant (Table A1).
We examined selection acting on juveniles of a large marine vertebrate. No study has previously accomplished this task, presumably because of the difficulty in finding populations where adequate numbers of site-attached individuals can be tagged and recaptured over multiple years. We overcame these limitations through a long-term mark-recapture study of a lemon shark population, where essentially all newborns could be captured, and where nearly all then remained resident for at least 3 years. Analyses using the program mark showed that we would rarely fail to recapture individuals who remained alive in the study site. We also confirmed that emigration from the site did not drive the inferred patterns of selection. Survival probabilities for the first two years of life ranged from 50% to 59%, similar to estimates based on mark–depletion methods at this site (Gruber et al., 2001). This nontrivial mortality rate, which may be the result of predation, starvation, or disease, suggests that significant viability selection could act on this population.
Our first major conclusion is that bigger is not better for juvenile lemon sharks at Bimini. Instead, selection generally favours smaller size, particularly between age-1 and age-2 (Table 3; Fig. 2b,d). This finding conflicts with the conventional wisdom that large size confers considerable fitness benefits (Sogard, 1997; Blanckenhorn, 2000; Kingsolver & Pfennig, 2004). And yet, a growing number of recent studies have documented selection against large size and fast growth (Quinn et al., 2001; Sinclair et al., 2002; Carlson et al., 2004). Moreover, a number of plausible hypotheses can be advanced for why bigger may not be better. We will later consider these hypotheses in relation to our study population (see below).
Our second major conclusion is that ‘fatter’ is not better; directional selection does not act in a consistent fashion on the relative condition factor of juvenile lemon sharks at Bimini. This variation may be the result of between-year fluctuations in environmental conditions, which are known to occur at this site (S. Gruber, personal observation). Indeed, selection was also somewhat variable on size and growth, albeit to a lesser degree. This variation could conceivably reflect the stochastic nature of selection, which is unlikely to be constant in space or time (e.g. Blanckenhorn et al., 1999; Jann et al., 2000; Przybylo et al., 2000; Kinnison & Hendry, 2001), although a temporal analysis of our selection estimates does not support this idea. Thus, although recent studies suggest that bursts of strong directional selection are often separated by periods of reversal or stasis (Hoekstra et al., 2001; Grant & Grant, 2002), it remains to be seen whether year-to-year variability in selection is the rule rather than the exception.
Strong selection against large size and fast growth should cause the evolution of smaller size and slower growth. Lemon sharks are too long lived (i.e. >25 years) for us to see if this sort of evolutionary change is indeed taking place at Bimini. We can, however, examine the outcome of this selection by comparison with another surveyed population (Marquesas Key, Florida; Barker et al., 2005). Compared with Marquesas, Bimini sharks are smaller at age (length: 54 cm vs. 74 cm at age-1) and grow much slower (6 cm vs. 20 cm between age-0 and age-1). Although we cannot be certain that this difference is genetic, the observed selection against large size and fast growth is at least consistent with the apparent evolution of small size and slow growth of Bimini sharks.
Selection in a large marine vertebrate
Nearly all previous estimates of selection have been for terrestrial or freshwater organisms, and yet it may be qualitatively different for large marine vertebrates. Here we compare strengths of selection in our study to those documented in other taxa (from Kingsolver et al., 2001) and to theoretical predictions (Hereford et al., 2004). Based on variance-standardized coefficients, selection acting on size-related traits for age-0 lemon sharks is weak (median absolute value = 0.08) in relation to other taxa (median = 0.16). For age-1 sharks, however, selection acting on length (median = 0.31) and growth (median = 0.22) was relatively strong. Indeed, these latter values fall into the 77th and 65th percentile for all taxa combined (Kingsolver et al., 2001). These differences are even greater when comparisons were restricted to estimates based on viability selection (86th and 77th percentile).
For mean-standardized coefficients, a suggested benchmark for strong selection is unity, because this value should correspond to the strength of selection on fitness itself (Hereford et al., 2004). Our mean-standardized coefficients have a median absolute value of 0.32 (range = 0.0089–22.24), suggesting that the typical strength of selection acting on Bimini sharks was at least a third as strong as selection on fitness itself, and comparable with other taxa (Hereford et al., 2004). Some of our estimates, however, are so much greater than unity that they call into question the utility of such comparisons. One problem is that absolute values are biased upward because of sampling error (Hereford et al., 2004). Other possible explanations include: (1) a bias caused by considering only one component of fitness or; (2) environmentally induced covariance between traits and fitness (Hereford et al., 2004).
Thus, the only available estimates for a large marine vertebrate suggest that selection is not weaker than for other taxa, and may even be stronger. We suggest that high dispersal in the marine realm maintains strong selection by preventing full adaptation to local conditions. Indeed, dispersal rates appear to be high for lemon sharks, as population structure at neutral markers is generally lacking on the scale of thousands of kilometres (Feldheim et al., 2001). The juveniles, however, remain in their local nursery sites for several years (Morrissey & Gruber, 1993). Selection may therefore be fine-grained, whereas evolutionary responses are coarse-grained.
Mechanisms of selection
Regression coefficients in and of themselves do not provide guidance as to the specific cause of apparent selection (e.g. sampling bias, emigration, predation, starvation, or disease). So much is known about our study site, however, that we can at least make some informed speculations. One possibility is that selection may not be acting at all, but that our estimates are biased because of preferential capture of particular individuals during sampling. Gillnets can be selective for smaller sharks (Carlson & Cortés, 2003), which may cause apparent selection on size-related traits. This seems unlikely at Bimini, however, because we catch nearly all of the fish within the nursery (Gruber et al., 2001), and because we routinely catch larger sharks using the same nets at a different nursery site (Marquesas Key, Florida).
A second possibility is size-selective emigration (Kingsolver & Smith, 1995; Letcher et al., 2005); larger, faster-growing sharks may be more likely to emigrate from our study site. This potential bias seems unlikely in our study because emigration by age-0 or age-1 sharks is rare at Bimini: (1) telemetry reveals high site fidelity and limited movement (Morrissey & Gruber, 1993); (2) displaced lemon sharks return to their original home ranges (Edrén & Gruber, 2005) and; (3) age-0 sharks are rarely captured outside of the nursery area (Gruber et al., 2001). Moreover, age-1 emigrants were actually smaller than residents, a difference that would act in opposition to our inference of selection against large size.
A third possibility is that selection for small size and slow growth at Bimini is a function of low resource availability. On the one hand, this seems unlikely because the Bimini lagoon is not resource limited, especially with respect to the primary prey item of the lemon shark (i.e. yellowfin mojarra, Gerres cinereus; Newman & Gruber, 2002). Moreover, we find no relationship across years between shark density (range: 70–105 individuals km−2) and selection coefficients for any trait (all r2 < 0.544; all P > 0.155). We have also never witnessed aggressive interactions among sharks in the Bimini nursery area (Gruber, 1982), suggesting a lack of interference competition. On the other hand, the nursery area does suffer from wide swings in ecological conditions, with exceedingly high and variable temperatures (due to shallow water and frequent rain showers), low nutrients and wide salinity fluctuations (S. Gruber, personal observation). Selection for small size and slow growth may be related to episodic variation in resource availability driven by these ecological factors.
A fourth possibility is predation pressure. Individual fish that achieve large size and fast growth presumably forage more frequently and in riskier situations, which increases predation risk (Martel & Dill, 1995; Mangel & Stamps, 2001; Biro et al., 2003; Biro et al., 2004; Brown & Braithwaite, 2004). Juvenile lemon sharks are susceptible to both inter- and intra-specific predation (S. Gruber, unpublished data), particularly from larger, sub-adult lemon sharks. Most feeding by juvenile lemon sharks takes place in or near the mangrove roots, which afford protection from predation (S. Gruber, personal observation). Fast growth, however, may also require foraging away from the mangroves. In fact, some juveniles stray outside the nursery area from time to time (Morrissey & Gruber, 1993), which will increase their exposure to predators. Predation may also favour smaller size and slower growth if faster growing fish are less adept at escaping predators (Billerbeck et al., 2001; Lankford et al., 2001) or if larger prey are preferred by predators (see Sogard, 1997).
A fifth possibility is that selection favouring small size and slow growth may be offset by selection favouring large size and fast growth at some other life stage (see Schluter et al., 1991). Indeed, selection is known to vary with age in other fish species (Hendry et al., 2003; Carlson et al., 2004; Zabel & Achord, 2004), and this was also the case in our study (Table 3). Opposing selection seems inevitable at some point; otherwise this population would be forever evolving a smaller size. We suggest that size and growth in Bimini sharks reflects a balance between opposing selection pressures acting during different life history stages – and that this balance is different from other populations. In particular, selection against large size and fast growth in young juveniles may be stronger at Bimini than elsewhere, leading to a smaller equilibrium body size. Indeed, we have already noted that size-at-age and growth rates are lower at Bimini than at Marquesas (Barker et al., 2005), the only other site where lemon sharks have been intensively studied.
Although studies of natural selection are logistically difficult for large marine organisms, we were able to generate robust estimates through an intensive, long-term, mark–recapture study of a lemon shark nursery site at Bimini, Bahamas. Our results suggest that selection at this site may play an important role in the evolution of size-related traits. We found strong directional selection against large size and fast growth, which fits with the small size and slow growth of sharks at this site. And yet, body size and growth in this population may still be greater than the optimum – otherwise selection should largely be absent. Partial maladaptation that maintains selection could be the result of high gene flow from other nursery sites where selection favours different phenotypes (i.e. larger size and faster growth). The specific selection pressures at Bimini have not been confirmed but may relate to increased predation on individuals that take more risks during foraging, or to other environmental characteristics at this site.
Our findings further challenge the conventional ‘bigger is better’, ‘fatter is better’ and ‘faster is better’ hypotheses. This conclusion is particularly interesting given that lemon sharks are large in general. We suggest that this paradox may be partly resolved by considering opposing selection pressures, which could theoretically favour larger size and faster growth later in life. Monitoring a population through only a portion of their life history can give an incomplete picture of selection, and so future work should aim to study selection at other life stages. Only then can we confirm whether selection on large marine vertebrates differs qualitatively from that for other organisms.
This study was funded in part by a Natural Sciences and Engineering Research Council of Canada postgraduate fellowship to J.D.D., as well as by grants from the Company of Biologists, the Canadian Society of Zoology, the Bimini Biological Field Station, The National Fish and Wildlife Foundation, Florida Sea Grant, National Geographic Society, University of Illinois at Chicago Research Foundation, and PADI's Project AWARE. Thanks to W. Blanckenhorn for helpful comments on an earlier version of this manuscript. S. Carlson provided invaluable suggestions and technical assistance with the program mark. X. Thibert-Plante and B. McGill provided assistance with probability calculations. We also thank Rose Mann and Lacey Hoover for their kind efforts to secure private funding, and we are indebted to the Hoover Foundation, and Drs Tadashi and Toshi Fujino, for generous private support. We gratefully acknowledge the following corporate support: Mario Aiello, Davey Marine; the late Dan Schaad of Mercury Division, Brunswick Corporation; The Carolina Skiff corporation; Digital Angel Corporation (Destron), especially Sean Casey; Andrea Obrian, Bimini Island Air; and Cathy Bosch of Pelican Products. This research was carried out under a permit from the Department of Fisheries of the Commonwealth of the Bahamas (Michael Braynan, director).