Heritable variation in swimming performance in Nile tilapia (Oreochromis niloticus) and negative genetic correlations with growth and harvest weight

Abstract

Nile tilapia is predominantly produced in smallholder ponds without aeration. We hypothesize that Nile tilapia with high oxygen uptake efficiency (O₂UE) may perform better under these conditions than Nile tilapia with low O₂UE. Critical swimming speed (U_crit, in cm s⁻¹) is a potential indicator for O₂UE. Our objectives were to estimate variance components for U_crit and fish size at swim testing early in life, and genetic correlations (r_g) between U_crit with harvest weight (HW) and daily growth coefficient (DGC) later after grow-out in a non-aerated pond. Substantial heritability was found for absolute U_crit (0.48). The estimated r_g between absolute U_crit and fish size at testing were all strong and positive (range 0.72–0.83). The estimated r_g between absolute U_crit and HW, and absolute U_crit and DGC were − 0.21 and − 0.63 respectively, indicating that fish with higher absolute U_crit had lower growth in the non-aerated pond as compared to fish with lower absolute U_crit. These results suggest a juvenile trade-off between swimming and growth performance where fish with high U_crit early in life show slower growth later under conditions of limited oxygen availability. We conclude that U_crit in Nile tilapia is heritable and can be used to predict growth performance.

Introduction

Nile tilapia (Oreochromis niloticus) is predominantly produced in smallholder tilapia ponds without aeration. In non-aerated ponds, dissolved oxygen (DO) drops below critical level (3 mg l⁻¹¹) during the night. Low DO in smallholder farms negatively affects Nile tilapia growth². It may be expected, therefore, that Nile tilapia with high oxygen uptake efficiency may grow better under these conditions than Nile tilapia with low oxygen uptake efficiency. As critical swimming speed (U_crit) may reflect the oxygen uptake efficiency, the hypothesis is that fish with high U_crit will grow better under conditions where oxygen is limiting.

A high throughput method to assess the individual variation in oxygen uptake efficiency is by subjecting fish to exhaustive exercise in a critical swimming challenge test. In this test, swimming speeds are incrementally increased at prescribed intervals until fish stop swimming and fatigue^3,4. Individual fish fatigue when swimming at a specific speed interval for a certain period, from which the U_crit³ can be determined. Recently we have developed and applied such tests for gilthead seabream (Sparus aurata) and Atlantic salmon (Salmo salar)⁵. Oxygen uptake is maximal at U_crit, although the anaerobic component by fast skeletal muscle increases when nearing U_crit⁶. Near U_crit, the metabolic demand for oxygen is becoming greater than can be provided by ventilatory and circulatory systems⁷. Fish that are able to consume more oxygen can swim faster, or reverse for the connection that we are interested in: faster swimming fish have higher oxygen uptake efficiency. Particularly for tilapia, the link between U_crit and maximal oxygen consumption may be strong because tilapia has a high U_crit (4.94 ± 0.45 BL s⁻¹ for ~ 15 cm fish) and a very high maximum metabolic rate⁸. Hence, U_crit could be a good indicator of oxygen uptake efficiency of individual tilapia.

The heredity of athletic performance has received considerable research attention in dog⁹, horse^10,11 and human¹². Genetic parameter estimates for swimming performance in fish are scarce, but suggest that swimming performance has a heritable component. Broad sense heritabilities (i.e. not corrected for dominance and epistatic interaction effects)¹³ of swimming performance were estimated by Garenc et al.¹⁴ in stickleback (Gasterosteus aculeatus), by Hurley and Schom¹⁵ in Atlantic salmon and by Nicoletto¹⁶ in guppy (Poecilia reticulata). More recently, Vandeputte et al.¹⁷ estimated the additive genetic variance component for relative U_crit (U_crit divided by standard length) in European sea bass (Dicentrarchus labrax) and found a heritability of 0.55, with a negative genetic correlation with body weight.

We therefore aimed first to estimate variance components for swimming performance in Nile tilapia expressed as U_crit and to estimate the genetic correlation between U_crit and fish size at swim testing early in life. Next, tested fish were stocked in a non-aerated pond and grown to harvest weight, to determine the genetic correlations between U_crit early in life and harvest weight (HW) and daily growth coefficient (DGC) later in life.

Materials and methods

Ethics statement

This study utilised phenotypic data collected as part of the GIFT selective breeding program managed by WorldFish at the Aquaculture Extension Centre of the Malaysian Department of Fisheries at Jitra, Kedah State, Malaysia (6° 15′ 32° N; 100° 25′ 47° E). This study was approved by the internal WorldFish ethics committee. All fish in the GIFT breeding population are managed in accordance with the Guiding Principles of the Animal Care, Welfare and Ethics Policy of WorldFish.

Experimental fish

Nile tilapia of the Genetically Improved Farmed Tilapia (GIFT) strain from generation 18 was used in this experiment. The 60 full sib and half sib families were produced using 31 males and 58 females, of which two females were used twice with different males. The planned mating ratios were one male to at least two females. However, the successful matings were: 12 males each mated with one female (resulting in 12 full sib families), 12 males each mated with two females (resulting in 12 half sib groups equivalent to 24 full sib families), 4 males each mated with 3 females (four half sib groups equivalent to 12 full sib families) and 3 males each mated with 4 females (three half sib groups equivalent to 12 full sib families). Each full sib family was reared separately in a hapa (fine mesh net enclosure) set up in an earthen pond.

The image analysis was done as described previously by Mengistu et al.¹⁸. In total 1500 photographs were loaded into tpsUtil software¹⁹ and digitized for six landmarks using tpsDig 2.30²⁰. Landmarks 1 and 2 were on the 0 and 20 cm marks on the ruler which was photographed together with the fish for scaling. The landmarks 3 and 4 were used to measure standard length, the distance between the tip of the snout to the base of caudal fin. The landmarks 5 and 6 were the dorsal and ventral landmarks where the distance is maximum. These landmarks were used to calculate height, the maximum dorso-ventral distance (Fig. 1). To obtain the distance between the Cartesian coordinates, these landmarks were analysed in R software using Geomorph package version 3.0.7²¹ and the true distance in cm was computed based on the reference scale.

Figure 1

Nile tilapia with landmarks 1:6. Landmarks 1 and 2 marks a reference scale of 20 cm length, landmarks 3 and 4 represent the snout and base of the caudal fin, respectively, landmarks 5 and 6 were used to measure height (maximum dorso-ventral length) of the experimental fish.

Swim test experiment

The swim test was done in 30 working days, one swim test per day. Thirty to 35 relatively bigger fingerlings from 60 full sib families were selected, PIT tagged and housed in a tank. Three weeks after PIT tagging, 25 fish in a range from 5 to 10 cm standard length at swim testing (SLtest, in cm) from each of the 60 full sib families were measured using a ruler with a centimetre scale, weighed (Wtest, in g) and photographs were made, one day before the swim test. The SLtest and height at swim testing (Htest, in cm) of the fish used in our analysis were obtained from the photographs of each fish using image analysis. The number of fish tested per family was 25 and the number of fish per test was 50 fish. Therefore, we tested either 10 fish from 5 families or 5 fish from 10 families which resulted in all 25 fish from each family being tested in three consecutive days.

To determine the U_crit, a Brett type (rectangular oval shape raceway) swim flume of 230 cm length and 90 cm width with a water depth of 40 cm was used²². Water current was created using a Minn Kota Terrova 80 lbs propeller. The propeller has 10 speed settings, in this experiment speed levels from 2 to 10 were used. Supplementary Table 1 provides the flow speeds measured at each of the settings. As the assessment of U_crit requires all fish to fatigue, this experimental set-up could be applied for early life testing at small size and not for older and larger fish.

Feeding was stopped 24 h before the beginning of the swim testing. The fish were acclimatised for one hour in the swimming flume without flow. After acclimation, the propeller was turned on to induce swimming at the second setting. The time at each setting was fixed at 30 min and flow increments continued until all fish fatigued. At each setting, the average water flow velocity was recorded using a FP111 Global Water Flow Probe (FP111, Global Water, USA). The swim test could take maximally 4.5 h, with 9 propeller speed levels. A fish fatigued when it touched the back fence and could not be stimulated to continue swimming. Each fatigued fish was scooped out immediately and PIT tag number and time at fatigue were recorded.

The mean DO in the tank just before resuming the swim test was 5.6 ± 0.4 mg l⁻¹ (71.2% saturation), ranging from 4.9 to 6.4 mg l⁻¹, and during the swim test it was 7.6 ± 0.4 mg l⁻¹ (97.6% saturation), ranging from 6.4 to 8.8 mg l⁻¹. The mean water temperature in the tank just before resuming the swim test was 27.7 ± 0.6 °C, ranging from 26.5 to 28.5 °C, and during the swim test it was 28.3 ± 0.6 °C, ranging from 26.5 to 29.9 °C.

Calculation of critical swimming performance and surface area

Absolute and relative critical swimming speed (U_crit) was used as a measure of swimming performance and calculated according to Brett³:

$${\text{Absolute}}\;U_{{{\text{crit}}}} = {\text{U}}_{{ - 1}} + \left( {\frac{{\text{t}}}{{\Delta {\text{t}}}}} \right)\Delta {\text{U}}$$

(1)

$${\text{Relative}}\;U_{{{\text{crit}}}} = \left( {{\text{U}}_{{ - 1}} + \left( {\frac{{\text{t}}}{{\Delta {\text{t}}}}} \right)\Delta {\text{U}}} \right)/{\text{SLtest}}$$

(2)

where ${\text{U}}_{ - 1}$ is the highest velocity maintained for the prescribed period in ${\text{cm}}\;{\text{s}}^{{ - 1}}$, $\Delta {\text{U}}$ is velocity increment in ${\text{cm}}\;{\text{s}}^{{ - 1}}$, $t$ is time to fatigue at final velocity level in minutes, $\Delta {\text{t}}$ is the time each velocity level is maintained at (= 30 min) and ${\text{SLtest}}$ is standard length of fish at swim testing in cm. Figures were produced using Minitab software²³.

Surface area at swim testing (SAtest) of Nile tilapia is similar to the area of an ellipse and was calculated as:

$$SA = \frac{1}{4}\pi *SLtest*Htest$$

(3)

Grow-out in the non-aerated pond

Swim tested fish were stocked in a non-aerated pond for grow-out. The pond size was 500 ${\text{m}}^{2}$ and the stocking density was 3 fish per ${\text{m}}^{2}$. During the grow-out period, DO was above 5 mg ${\text{l}}^{ - 1}$ except from 9:00 p.m. to 9:00 a.m. when DO would drop below 3 mg ${\text{l}}^{ - 1}$. Fish were weighed and photographed before stocking into the non-aerated pond. The mean weight at cultivation start (Wstart) was 10.8 g and the coefficient of variation (CV) was 23.7. The fish were fed commercial feed at a rate of 3 to 5% of their body weight depending on their sizes, with the percentage of feed decreasing with size. The fish were harvested after 145 or 146 days of grow-out. Each fish was weighed at harvest. At harvest the sex of a random half of the fish (763 fish) were determined.

Daily growth coefficient (DGC)^24,25 was computed as:

$$DGC = \left[ {\frac{{\sqrt[3]{HW} - \sqrt[3]{Wstart}}}{time\, in\, days}} \right] \times 100$$

(4)

where HW is harvest weight and Wstart is stocking weight.

Statistical analysis

Phenotypic and genetic parameters were estimated using ASReml version 4.1²⁶. The following animal model was used:

$${\varvec{y}} = {\mathbf{Xb}} + {\mathbf{Z}}_{1} {\mathbf{a}} + {\mathbf{Z}}_{2} {\mathbf{c}} + {\mathbf{e}}$$

(5)

where ${\varvec{y}}$ is a vector of either absolute U_crit, or relative U_crit in the univariate model, ${\mathbf{b}}$ is the vector of fixed effects, that is test day and sex fitted as class variable for relative U_crit while for absolute U_crit three different models were fitted with: (1) test day and sex fitted as class variables, (2) test day and sex as class variables and Wtest as a covariate and (3) test day and sex as class variables and SLtest as a covariate, sex was not significant in all the three models; therefore, sex was removed from the models; ${\mathbf{a}}$ is a vector of additive genetic effects, ${\mathbf{c}}$ is a vector of environmental effects common to full sibs (‘hapa effect’), and ${\mathbf{e}}$ a vector of residual effects. The X, Z₁ and Z₂ are design matrices assigning phenotypic values to the levels of fixed effect, additive genetic and common environmental effects, respectively. The effect of sex was also not significant when subset of the data with only 763 sexed fish was analysed.

Bivariate models were used to estimate the phenotypic and genetic correlations between absolute U_crit, and traits such as Wtest, SLtest, Htest, SAtest, HW and DGC. In the bivariate models test day and sex were fitted as a class variable for absolute U_crit, age at harvest was fitted as a covariate for HW and sex was fitted as a class variable for DGC. Common environmental effect was fitted as a random variable to all the traits except for DGC in the bivariate model absolute U_crit and DGC. The bivariate model with absolute U_crit and DGC did not converge when a common environmental effect was fitted as a random effect on both traits. The additive genetic effects were normally distributed as ${\text{N}} = \left( {\left[ {\begin{array}{*{20}c} 0 \\ 0 \\ \end{array} } \right],{ }{\mathbf{A}} \otimes \left[ {\begin{array}{*{20}c} {\sigma_{a,1}^{2} } & {r_{a,12} \sigma_{a,1} \sigma_{a,2} } \\ {r_{a,21} \sigma_{a,2} \sigma_{a,1} } & {\sigma_{a,2}^{2} } \\ \end{array} } \right]} \right)$, where ${\mathbf{A}}$ is the numerator genetic relationships matrix and ${\upsigma }_{{{\text{a}},1}}^{2}$ (${\upsigma }_{{{\text{a}},2}}^{2}$) being the additive genetic variance of trait 1(2) and $r_{{a,12\left( {21} \right)}}$ being the genetic correlation between trait 1 and 2. The pedigree depth was 18 generations, i.e. from the current generation G18 all the way back to the first generation of GIFT in WorldFish, Malaysia. The common environmental effects were normally distributed as ${\text{N}} = \left( {\left[ {\begin{array}{*{20}c} 0 \\ 0 \\ \end{array} } \right],\user2{ }{\mathbf{I}} \otimes \left[ {\begin{array}{*{20}c} {\sigma_{c,1}^{2} } & {r_{c12} \sigma_{c,1} \sigma_{c,2} } \\ {r_{c21} \sigma_{c,2} \sigma_{c,1} } & {\sigma_{c,2}^{2} } \\ \end{array} } \right]} \right)$, where I being an identity matrix and ${\upsigma }_{{{\text{c}},1}}^{2}$ (${\upsigma }_{{{\text{c}},2}}^{2}$) being the common environmental variance of trait 1(2) and $r_{{c,12\left( {21} \right)}}$ being the common environmental correlation between trait 1 and 2. The residual effects were normally distributed as ${\text{N}} = \left( {\left[ {\begin{array}{*{20}c} 0 \\ 0 \\ \end{array} } \right],\user2{ }{\mathbf{I}} \otimes { }\left[ {\begin{array}{*{20}c} {{\upsigma }_{{\text{e,1}}}^{{2}} } & {{\text{r}}_{{\text{e, 12}}} {\upsigma }_{{\text{e,1}}}^{{2}} {\upsigma }_{{\text{e,2}}}^{{2}} } \\ {{\text{r}}_{{\text{e,21}}} {\upsigma }_{{\text{e,2}}}^{{2}} {\upsigma }_{{\text{e, 1}}}^{{2}} } & {{\upsigma }_{{\text{e,2}}}^{{2}} } \\ \end{array} } \right]} \right)$, where ${\upsigma }_{{{\text{e}},1}}^{2}$ (${\upsigma }_{e,2}^{2}$) being the residual variance of trait 1(2) and $r_{{e,12\left( {21} \right)}}$ being the residual correlation between trait 1 and 2.

Heritability $\left( {{\text{h}}^{2} } \right)$ and the ratio of common environmental variance $({\text{c}}^{2}$) to phenotypic variance ($\sigma_{p}^{2} )$ of each trait was computed as ${\text{h}}^{2} = {\upsigma }_{{\text{a}}}^{2} /{\upsigma }_{{\text{p}}}^{2}$ and ${\text{c}}^{2} = {\upsigma }_{{\text{c}}}^{2} /{\upsigma }_{{\text{p}}}^{2}$, respectively. The significance of the random effects were tested using loglikelihood ratio test with one degree of freedom²⁷. To test whether the genetic correlation is larger than zero, a model without constraining the covariance was tested against a model where the covariance was constrained to zero. The full model, i.e. a model with both common environmental effects and additive genetic effects as random effects, was tested against a reduced model, i.e. a model with either only common environmental effect or additive genetic effects as a random effect. The common environmental variances were not significantly different from zero (P > 0.05) except for relative U_crit (P = 0.006). The most likely reason that the common environmental effect was not significant in most cases was because of the almost complete confounding of sire genetic, dam genetic and common environmental effects in the experiment. This reflected the fact that 24 of the males were mated to one or two females resulting in 12 families with no half sib families and 12 families with only one half sib family (40% of the total families), making genetic and common environmental effects difficult to disentangle. Although, common environmental effects were not significant for most traits, common environmental effects explained a substantial part of the phenotypic variance and were kept in the model, to prevent overestimation of the additive genetic variance. The loglikelihood for the bivariate model with absolute U_crit and DGC did not converge when common environmental effect was fitted as a random effect on both traits. Therefore, the common environmental effect was fitted as a random effect only on absolute U_crit in the bivariate model with absolute U_crit and DGC.

Results

Biometric data

In total 1500 fish were swim tested and stocked in the non-aerated pond. Out of the swim tested 1500 fish, the swimming performance data of seven fish were missing and resulted in 1493 U_crit records. The descriptive statistics for age at swim testing (Agetest), Wtest, SLtest, Htest, SAtest, weight at cultivation start, and HW and DGC later in life are presented in Table 1. Out of the stocked 1500 fish, ultimately 1199 were harvested which is equivalent to 79.9% survival.

Table 1 Number of fish (N), mean, standard deviation (SD), coefficient of variation (CV) and minimum and maximum values for critical swimming speed (U_crit), absolute and relative, age at swim testing (Agetest), body weight at swim testing (Wtest), standard length at swim testing (SLtest), and body height at swim testing (Htest), surface area at swim testing (SAtest), weight at cultivation start(Wstart), harvest weight (HW) and daily growth coefficient (DGC).

Swimming performance

The mean absolute U_crit and relative U_crit were 69.1 ± 5.5 cm s⁻¹ and 9.7 ± 0.9 SL s⁻¹, respectively (Table 1). Absolute U_crit and relative U_crit values showed normal distributions (Fig. 2). There was substantial variation in swimming performance between family means (Fig. 3), indicating existence of genetic variation.

Figure 2

Distribution of absolute U_crit (cm s⁻¹) and relative U_crit (SL s⁻¹).

Figure 3

Histogram of family average absolute U_crit (cm s⁻¹) for each of the 60 families.

Genetic parameters

Variances, heritability and the ratio of common environmental variance to the phenotypic variance (c²) effect for absolute and relative U_crit are presented in Table 2. The heritability for absolute U_crit was 0.48 ± 0.17 when Wtest or SLtest was not fitted in the model as a covariate. The heritability for absolute U_crit was 0.41 ± 0.16 when SLtest was fitted as a covariate and 0.44 ± 0.16 when Wtest was fitted as a covariate. The heritability for relative U_crit (0.15 ± 0.13) was low. The common environmental effect explained a small proportion of the phenotypic variance (0.02 to 0.04) for absolute U_crit, while the contribution was substantial for the phenotypic variance of relative U_crit (0.13). The analyses with absolute U_crit with Wtest or SLtest in the model as a covariate showed that U_crit contained considerable heritable variation even when corrected for body size.

Table 2 Additive genetic variance (${\sigma }_{a}^{2}$), common environmental variance (${\sigma }_{c}^{2}$), phenotypic variance (${\sigma }_{p}^{2}$), heritability and common environmental effect (${c}^{2}$) of absolute and relative critical swimming speed (U_crit).

The additive genetic variance contributed a significant proportion to the phenotypic variance of absolute U_crit (P = 0.000) and absolute U_crit when either Wtest (P < 0.001) or SLtest (P = 0.001) was fitted in the model as covariate, while the contribution was not significant for relative U_crit (P = 0.175). The contribution of common environmental effect to the phenotypic variance of absolute U_crit (P = 0.584) and for absolute U_crit when either Wtest (P = 0.384) or SLtest were fitted as covariates (P = 1.000) were not significant, while the contribution to the phenotypic variance of relative U_crit (P = 0.007) was significant.

The estimated genetic correlations (r_g) and phenotypic correlations (r_p) between absolute U_crit and Wtest, SLtest, Htest, SAtest, HW and DGC are presented in Table 3. The genetic correlations were significant (P < 0.05) except for the genetic correlation between U_crit early in life and HW later in life (P = 0.507) based on likelihood ratio test²⁷. The estimated r_g and r_p correlations between absolute U_crit and Wtest were 0.78 and 0.44, respectively. The less than one r_g between absolute U_crit and Wtest indicates the presence of genetic variance in absolute U_crit that is not explained by Wtest. Genetic and phenotypic correlations with the other size measurements SLtest, Htest and SAtest were very similar. Fish with higher absolute U_crit had lower HW and DGC after grow-out in a non-aerated pond. The estimated r_g and r_p between absolute U_crit and HW were − 0.21 and − 0.04, respectively and the estimated r_g and r_p between absolute U_crit and DGC were − 0.63 and − 0.24, respectively. The negative genetic correlations between U_crit and HW and between absolute U_crit and DGC indicate that fish with higher absolute U_crit perform less in terms of HW and DGC compared to fish with lower absolute U_crit.

Table 3 Genetic and phenotypic correlations between absolute critical swimming speed (U_crit) and body weight at swim testing (Wtest), standard length at swim testing (SLtest) height at swim testing (Htest), surface area at swim testing (SAtest), harvest weight (HW) and daily growth coefficient (DGC).

Discussion

Our objectives were to estimate variance components for swimming performance in Nile tilapia, assessed as critical swimming speed (U_crit) early in life, and to estimate the genetic correlation between U_crit and body size early in life, and harvest weight (HW) and Daily Growth Coefficient (DGC) later in life after a grow-out period in a non-aerated pond. For the first time, we show with a large-scale experiment that swimming performance is heritable in Nile tilapia, and that the genetic correlation with harvest weight is strongly negative, even when corrected for body size at testing. Heritabilities, the genetic correlations, methodology and the practical application of a swimming performance test in breeding programs are discussed.

This study shows the existence of heritable variation in critical swimming performance with a moderate heritability of 0.41–0.48. Our heritability estimate for U_crit early in life is in the same range as reported previously for other species and for similar traits (for summary see Table 4). Of the four studies that estimated genetic parameters for swimming performance in fish, only the study that assessed the burst swimming performance trait is not comparable with U_crit in our study¹⁴. Our heritability estimate for relative U_crit (0.15) was not significantly different from zero, which is different from the heritability of 0.55 for relative maximum swimming speed in European sea bass Dicentrarchus labrax¹⁷. The difference in heritability of relative U_crit might be due to a species specific difference, particularly reflecting the high or long body shape of tilapia and seabass, respectively.

Table 4 Summary of the studies that estimated heritability for different swimming performance traits, genetic phenotypic correlation between swimming performance and body weight, and between swimming performance and body length.

Species specific differences also exist in the relation between U_crit and body size. Absolute U_crit was genetically strongly correlated with body weight at swim testing (0.78). This is higher than the estimated genetic correlation between swimming stamina and body weight in Atlantic salmon (0.23)¹⁵. The genetic correlation between absolute U_crit and standard length (0.83) was also different from the estimated r_g between swimming stamina and fork length in Atlantic salmon (− 0.14)¹⁵.

To the best of our knowledge there are no studies on r_g between absolute U_crit and traits such as Htest, SAtest, HW and DGC with which to compare our results. In our study, the r_g estimates between absolute U_crit and Htest, between absolute U_crit and SAtest were 0.72 and 0.83, respectively. These strong genetic correlations between absolute U_crit and SLtest and Wtest early in life show that larger fish swim faster in absolute terms.

The estimated r_g values between absolute U_crit and HW and absolute U_crit and DGC were − 0.21 and − 0.63, respectively, meaning that fish with high U_crit at testing had lower growth rate (DGC) and harvest weight (HW) later in life. These negative genetic correlations do not support our hypothesis that Nile tilapia with higher U_crit, reflecting higher oxygen uptake efficiency, are those that perform better in terms of weight increase in non-aerated ponds where hypoxia is frequent. Instead, the negative r_g shows that fish with higher U_crit early in life show less body weight increase later in life. These data do not provide insight on fish body shape and composition at slaughter size. For example, it may be that fish with higher U_crit are the leaner fish later as compared to fish with lower U_crit. Fish with lower U_crit may be heavier but not necessarily have more fillet mass. Results of a U_crit test in Gilthead seabream (Sparus aurata), also a high bodied fish, showed that the (residual) U_crit was negatively correlated with fillet mass suggesting that fast swimmers build lower fillet mass later in life⁵. A plausible explanation for our results may be the existence of a juvenile trade-off between swimming and growth performance where fish with high U_crit early in life show slower growth later. Young juveniles may choose to either swim fast or grow fast, which may represent, for instance, two anti-predator strategies: to be able to escape predators or to become too large to be eaten rapidly. Studies have shown that a trade-off between growth rate and locomotor performance can exist²⁸, for instance during accelerated growth²⁹ which can negatively influence muscle cellularity and development^30,31. Indeed, fast-growing growth hormone (GH) transgenic carp³² had lower critical swimming performance than non-transgenic controls. Fast-growing GH transgenic salmon had similar critical swimming speeds than non-transgenic controls but were also able to consume considerable more oxygen³³ and may thus have compensated for lower critical swimming performance.

In our study, 1,493 fish were used to estimate genetic parameters. The mating ratio used to produce the experimental fish was 1 male to 1 – 4 females, which gave full sib and half sib families. The previous studies that estimated genetic parameters used a much lower number of fish (range 96–129) as compared to our study and estimated broad sense heritability using full sib families (Table 4)^15,16. The much larger sample size gave a much higher precision of estimates of narrow-sense heritability. Furthermore, broad-sense heritability estimates are biased estimates of narrow-sense heritabilities, because broad-sense heritabilities contain non-additive genetic variation due to dominance and epistasis that is not heritable from parent to offspring and may contain common environmental effects, because in such full-sib designs estimation of common environmental effects is not feasible³⁴. Narrow-sense heritability, however, is the ratio of additive genetic variance to phenotypic variance¹³ and therefore a better indication of the proportion of genetic variation that is transmitted to the next generation. In our study, we used half sib families that enabled us to estimate a narrow sense heritability. Similarly, Vandeputte et al.¹⁷ estimated narrow sense heritability using half sib families based on 547 fish. The main difference in the swimming performance trait between our study and Vandeputte et al.¹⁷ was that these authors did not include the last water velocity level that the fish did not fully complete. Besides the species difference mentioned earlier, also the number of fish and the way the swimming performance was calculated could provide additional explanation for the difference in the parameter estimates between our study and Vandeputte et al.¹⁷.

Critical swimming speed can be calculated in four different ways: as absolute U_crit, with or without Wtest or SLtest as covariate in the model, as relative U_crit, or as residual U_crit which is the difference in U_crit of an individual fish with the predicted value on basis of its length⁵. Analysing absolute U_crit without a covariate for either Wtest or SLtest, has the highest additive genetic variance, but part of that genetic variance is due to genes affecting body size. The use of fish with similar body weight at similar SLtest is practically difficult as the variation is considerable; in our experiment the Wtest was from 4.8 to 20.1 g for fish from 5.4 to 10 cm SL. Therefore, it is important to account for Wtest or SLtest in the analysis to be able to estimate heritable variation in U_crit independent of body size.

Relative U_crit is a ratio of U_crit to SLtest for which the estimated heritability was not significantly different from zero in our study. Relative U_crit is a ratio trait and therefore the genetic variance becomes a complex function of absolute U_crit and SLtest. Ratio traits are generally not recommended in animal breeding³⁵. For instance, the heritability of a ratio trait cannot be used to predict the genetic change for the ratio trait³⁶. Therefore, we recommend using the absolute U_crit and to fit either Wtest or SLtest as a covariate in a model when estimating heritability. Such an analysis shows the existence of heritable variation in U_crit beyond body size.

The less than unity genetic correlation between absolute U_crit and Wtest indicates the presence of genetic variation in U_crit, independently of Wtest. A genetic correlation of unity between two traits means that the two traits are controlled by the same genes while a genetic correlation of less than unity indicates that there are additional genes that are not common for the two traits and only control one of the two traits. The negative r_g between absolute U_crit and HW, and between U_crit and DGC, clearly indicates that selection for high harvest weight will favour faster growing animals with lower U_crit. Whether this is desirable needs to be determined. One can speculate that under conditions of hypoxia, as frequently encountered in non-aerated ponds or ponds with algal blooms, smaller, more active fish will have a higher chance of survival. In optimal management conditions, however, growth rate can be further increased by including U_crit at testing in the breeding goal, next to harvest weight. Fish with higher U_crit may also be more resilient: swimming exercise improves physiological fitness; cardiovascular and respiratory performance, and increases mitochondrial densities and muscle tissue capillarization³⁷. Also the immune system capacity appears to be linked to swimming performance as Castro et al.³⁸ found 21 virus-responsive genes with significantly higher transcript abundance in phenotypically poor swimmers as compared to good swimmers in Atlantic salmon.

In conclusion, including absolute U_crit in a breeding goal in addition to HW and DGC could be beneficial if the aim is to select for fitter fish, especially in environments where oxygen is limiting. Absolute U_crit can be measured at an early stage on the selection candidates themselves, high throughput and non-invasively although size of the tested fish may be restricted due to difficulty in reaching sufficiently high flow speed. However, selection on U_crit with 10% selection intensity for the highest values of U_crit could lead to a 19% reduction in mean harvest weight of the offspring, compared to direct selection on harvest weight. In practice we recommend a two-stage selection scheme, where selection in the first stage is on retaining 90% of the fittest fish in terms of U_crit, followed by a second stage selection on harvest weight. This study showed for the first time the existence of significant additive genetic variance for critical swimming speed in Nile tilapia. Favourable r_g between U_crit and traits such as Wtest, SLtest, Htest and SAtest early in life were found. The main finding demonstrated a negative r_g between U_crit and HW later in life, and between U_crit and DGC later in life. Including U_crit in the breeding goal may help to improve resilience of Nile tilapia.