Limits...
Evolution of growth in Gulf of St Lawrence cod?

Heino M, Baulier L, Boukal DS, Dunlop ES, Eliassen S, Enberg K, Jørgensen C, Varpe O - Proc. Biol. Sci. (2008)

View Article: PubMed Central - PubMed

Affiliation: Institute of Marine Research, Nordnes, Bergen, Norway. mikko.heino@bio.uib.no

AUTOMATICALLY GENERATED EXCERPT
Please rate it.

First, as the evolving trait, they used length-at-age 4 years, an age at which cod are representatively sampled but have experienced little fishing mortality... Confounding demographic effects of size-selective fishing were therefore avoided... SSH's statistically favoured regression model was one including both S and Δd; they concluded that the data suggested an evolutionary response to fishing... Here, we comment on some caveats in the analysis by SSH... We do not challenge their novel approach, but question some key assumptions and the strength of their conclusions... An individual's length-at-age depends on the environment and at least three life-history traits: growth capacity, maturation schedule and reproductive investment... Data on southern Gulf of St Lawrence cod reveal that from 1990 to 1995, 35–60% of males and 10–50% females were mature at age 4; maturation data outside this 6-year window are unfortunately unreliable and cannot unravel temporal trends... To reduce the confounding effects of changes in reproductive investment or the proportion of mature fish, we thus recommend that lengths-at-ages before maturation be used as the evolving trait... For southern Gulf of St Lawrence cod, age 3 data are available for the entire time series, and the lower proportion of mature individuals at that age would reduce the confounding effects of reproduction... To test whether these assumptions hold, we added an intercept to SSH's favoured model (Δd+S); it was significantly different from zero (−0.98; p=0.03) and S became insignificant (p=0.34)... A model with Δd and an intercept C also has a lower AIC value than SSH's favoured model, and S is not significant for any other combination of variables... Ranked with AIC, SSH's favoured model is among the two best models, but only for the first window does it outperform the model where S is replaced by C... We look forward to seeing their methodology applied to other fish populations... However, it remains inconclusive as to whether fisheries have induced evolution of reduced growth capacity in the Gulf of St Lawrence cod.

Show MeSH

Related in: MedlinePlus

(a) Data from fig. 3 in Swain et al. (2007). (b) Ranking of models involving combinations of selection differential (S), change in density (Δd) and intercept (C) as explanatory variable(s) in sliding windows of 10 cohorts. ΔAICc is a small sample version of Akaike's information criterion of each model compared to the best model (the model with the lowest AICc of all the models considered). The  model, including only an intercept, is shown with a grey line. (c) Estimated regression coefficients in sliding windows of 10 cohorts (with 95% CIs) for selection differential and density effects in model 2 (ΔL4=βΔd+h2S) of Swain et al. (2007).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2376251&req=5

fig1: (a) Data from fig. 3 in Swain et al. (2007). (b) Ranking of models involving combinations of selection differential (S), change in density (Δd) and intercept (C) as explanatory variable(s) in sliding windows of 10 cohorts. ΔAICc is a small sample version of Akaike's information criterion of each model compared to the best model (the model with the lowest AICc of all the models considered). The model, including only an intercept, is shown with a grey line. (c) Estimated regression coefficients in sliding windows of 10 cohorts (with 95% CIs) for selection differential and density effects in model 2 (ΔL4=βΔd+h2S) of Swain et al. (2007).

Mentions: Our last point relates to the nature of the data that were available to SSH. The first and second half of the time series of S and ΔL4 differ qualitatively (figure 1a). To check robustness of the results presented by SSH, we estimated a range of alternative models for sliding windows of 10 successive cohorts (figure 1b; Δd and Δt are strongly correlated and we present only Δd as the results are similar). In the first sliding windows, a range of models can explain the data well (figure 1b). Ranked with AIC, SSH's favoured model is among the two best models, but only for the first window does it outperform the model where S is replaced by C. From the window beginning with cohort 1984, all models become non-significant and the observed change in length can best be interpreted as noise. At the same time, the estimated heritability dropped from between 0.5 and 0.7 to approximately 0 (figure 1c), which could be due to the erosion of additive genetic variance. However, it seems unlikely that such high levels of heritability could be purged in one to two cohorts and it does not explain why explanatory variables other than S lose significance at the same time (figure 1b).


Evolution of growth in Gulf of St Lawrence cod?

Heino M, Baulier L, Boukal DS, Dunlop ES, Eliassen S, Enberg K, Jørgensen C, Varpe O - Proc. Biol. Sci. (2008)

(a) Data from fig. 3 in Swain et al. (2007). (b) Ranking of models involving combinations of selection differential (S), change in density (Δd) and intercept (C) as explanatory variable(s) in sliding windows of 10 cohorts. ΔAICc is a small sample version of Akaike's information criterion of each model compared to the best model (the model with the lowest AICc of all the models considered). The  model, including only an intercept, is shown with a grey line. (c) Estimated regression coefficients in sliding windows of 10 cohorts (with 95% CIs) for selection differential and density effects in model 2 (ΔL4=βΔd+h2S) of Swain et al. (2007).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2376251&req=5

fig1: (a) Data from fig. 3 in Swain et al. (2007). (b) Ranking of models involving combinations of selection differential (S), change in density (Δd) and intercept (C) as explanatory variable(s) in sliding windows of 10 cohorts. ΔAICc is a small sample version of Akaike's information criterion of each model compared to the best model (the model with the lowest AICc of all the models considered). The model, including only an intercept, is shown with a grey line. (c) Estimated regression coefficients in sliding windows of 10 cohorts (with 95% CIs) for selection differential and density effects in model 2 (ΔL4=βΔd+h2S) of Swain et al. (2007).
Mentions: Our last point relates to the nature of the data that were available to SSH. The first and second half of the time series of S and ΔL4 differ qualitatively (figure 1a). To check robustness of the results presented by SSH, we estimated a range of alternative models for sliding windows of 10 successive cohorts (figure 1b; Δd and Δt are strongly correlated and we present only Δd as the results are similar). In the first sliding windows, a range of models can explain the data well (figure 1b). Ranked with AIC, SSH's favoured model is among the two best models, but only for the first window does it outperform the model where S is replaced by C. From the window beginning with cohort 1984, all models become non-significant and the observed change in length can best be interpreted as noise. At the same time, the estimated heritability dropped from between 0.5 and 0.7 to approximately 0 (figure 1c), which could be due to the erosion of additive genetic variance. However, it seems unlikely that such high levels of heritability could be purged in one to two cohorts and it does not explain why explanatory variables other than S lose significance at the same time (figure 1b).

View Article: PubMed Central - PubMed

Affiliation: Institute of Marine Research, Nordnes, Bergen, Norway. mikko.heino@bio.uib.no

AUTOMATICALLY GENERATED EXCERPT
Please rate it.

First, as the evolving trait, they used length-at-age 4 years, an age at which cod are representatively sampled but have experienced little fishing mortality... Confounding demographic effects of size-selective fishing were therefore avoided... SSH's statistically favoured regression model was one including both S and Δd; they concluded that the data suggested an evolutionary response to fishing... Here, we comment on some caveats in the analysis by SSH... We do not challenge their novel approach, but question some key assumptions and the strength of their conclusions... An individual's length-at-age depends on the environment and at least three life-history traits: growth capacity, maturation schedule and reproductive investment... Data on southern Gulf of St Lawrence cod reveal that from 1990 to 1995, 35–60% of males and 10–50% females were mature at age 4; maturation data outside this 6-year window are unfortunately unreliable and cannot unravel temporal trends... To reduce the confounding effects of changes in reproductive investment or the proportion of mature fish, we thus recommend that lengths-at-ages before maturation be used as the evolving trait... For southern Gulf of St Lawrence cod, age 3 data are available for the entire time series, and the lower proportion of mature individuals at that age would reduce the confounding effects of reproduction... To test whether these assumptions hold, we added an intercept to SSH's favoured model (Δd+S); it was significantly different from zero (−0.98; p=0.03) and S became insignificant (p=0.34)... A model with Δd and an intercept C also has a lower AIC value than SSH's favoured model, and S is not significant for any other combination of variables... Ranked with AIC, SSH's favoured model is among the two best models, but only for the first window does it outperform the model where S is replaced by C... We look forward to seeing their methodology applied to other fish populations... However, it remains inconclusive as to whether fisheries have induced evolution of reduced growth capacity in the Gulf of St Lawrence cod.

Show MeSH
Related in: MedlinePlus