Modeling meat yield based on measurements of bodytraits in genetically improved giant freshwater prawn(GFP) Macrobrachium rosenbergii

Dinh Hung • Nguyen Hong Nguyen

Received: 17 May 2013 / Accepted: 13 August 2013 / Published online: 22 August 2013� Springer Science+Business Media Dordrecht 2013

Abstract A single-generation dataset consisting of 1,730 records from a selection pro-

gram for high growth rate in giant freshwater prawn (GFP, Macrobrachium rosenbergii)

was used to derive prediction equations for meat weight and meat yield. Models were

based on body traits [body weight, total length and abdominal width (AW)] and carcass

measurements (tail weight and exoskeleton-off weight). Lengths and width were adjusted

for the systematic effects of selection line, male morphotypes and female reproductive

status, and for the covariables of age at slaughter within sex and body weight. Body and

meat weights adjusted for the same effects (except body weight) were used to calculate

meat yield (expressed as percentage of tail weight/body weight and exoskeleton-off

weight/body weight). The edible meat weight and yield in this GFP population ranged from

12 to 15 g and 37 to 45 %, respectively. The simple (Pearson) correlation coefficients

between body traits (body weight, total length and AW) and meat weight were moderate to

very high and positive (0.75–0.94), but the correlations between body traits and meat yield

were negative (-0.47 to -0.74). There were strong linear positive relationships between

measurements of body traits and meat weight, whereas relationships of body traits with

meat yield were moderate and negative. Step-wise multiple regression analysis showed that

the best model to predict meat weight included all body traits, with a coefficient of

determination (R2) of 0.99 and a correlation between observed and predicted values of

meat weight of 0.99. The corresponding figures for meat yield were 0.91 and 0.95,

Nguyen Hong Nguyen is a Joint first author.

D. HungResearch Institute for Aquaculture N.2, 116 Nguyen Dinh Chieu Str, Dist 1, HCM City, Vietnam

D. HungScience and Engineering Faculty, Queensland University of Technology, Brisbane, QLD 4001,Australia

N. H. Nguyen (&)Faculty of Science, Health, Education and Engineering, University of the Sunshine Coast,Maroochydore, QLD 4558, Australiae-mail: nnguyen@usc.edu.au

123

Aquacult Int (2014) 22:619–631DOI 10.1007/s10499-013-9690-1

respectively. Body weight or length was the best predictor of meat weight, explaining

91–94 % of observed variance when it was fitted alone in the model. By contrast, tail width

explained a lower proportion (69–82 %) of total variance in the single trait models. It is

concluded that in practical breeding programs, improvement of meat weight can be easily

made through indirect selection for body trait combinations. The improvement of meat

yield, albeit being more difficult, is possible by genetic means, with 91 % of the variation

in the trait explained by the body and carcass traits examined in this study.

Keywords Selective breeding � Meat yield � Genetic improvement �Correlated response

Introduction

Carcass traits, especially meat weight and yield, are economically important in farmed

aquaculture species. The yield of edible meat in giant freshwater prawn (GFP) is generally

much lower than in marine shrimps (50–63 %) (Jory and Cabrera 2012) but greater than

other important crustaceans such as lobsters (36 %), crayfish (15 %) or crabs (15 %) (Silva

et al. 2010). In commercially cultured finfish, reported yields range from 25 to 35 % in

tilapia (Clement and Lovell 1994; Rutten et al. 2004; Pires et al. 2006; Nguyen et al. 2010),

43 % in Channel catfish (Argue et al. 2003), 58 % in Coho salmon (Neira et al. 2004),

64 % in rainbow trout (Kause et al. 2007) and 69 % in Atlantic salmon (Powell et al.

2008). Although meat yield is directly related to economic return of aquaculture enter-

prises, genetic improvement for this trait in aquatic animal species has been hampered by a

lack of efficient data recording and measurement methods. Both meat weight and yield

cannot be measured in live animals. Slaughter of the selection candidates is an option only

if gametes can be preserved. Alternatively, groups of relative have to be slaughtered,

considerably adding to the cost and complexity of the genetic evaluation. Some studies

have attempted to develop methods to predict meat weight and yield using ultrasound

imagery and body shape (Bosworth et al. 2001) or morphological screening in association

with software image analysis (Cibert et al. 1999). Both methods are promising, but their

practical application is limited largely due to the moderate accuracy of the prediction

models and their complexity in terms of data recording and image interpretation. In

selective breeding programs, body traits are usually available due to their ease of mea-

surement. Prediction of meat weight and yield based on body traits is therefore likely to be

practical, and if the prediction were accurate, there would be possibilities for indirect

selection on carcass traits without slaughter of the selection candidates or of their siblings.

In fish, a number of studies have been carried out to develop prediction equations for fillet

weight and yield (Rutten et al. 2004; Pires et al. 2006; Sang et al. 2009). They were based

on different measurements of body traits. Irrespective of type and number of traits used in

the models, the general conclusion from these studies in fish is that fillet weight can be

effectively predicted from body traits (weight, length, width, depth). By contrast, predic-

tion equations for fillet (or meat) yield based on body traits give very low level of accuracy

The selection focus in the genetic improvement programs for GFP has so far been

almost exclusively on increased body weight (Hung et al. 2013). Weight at harvest has

been the main selection criterion but, other traits [total and abdominal lengths (ALs) and

abdominal width (AW)] are also routinely recorded. There is justification for the possible

620 Aquacult Int (2014) 22:619–631

123

inclusion of meat weight and yield to the GFP breeding objective because marketing

systems in major GFP producing countries are based on not only whole prawn live weight

but also meat (abdominal) weight or exoskeleton-off weight. In this study, we modeled

meat weight and yield based on three body traits (weight, total length and AW). The data

were collected from the third generation of the long-term selection program in the GFP

strain (Hung et al. 2013).

Materials and methods

The genetic selection lines

The origin of the GFP population is described in detail by Hung et al. (2013). The strain

was established in Vietnam from 81 full-sib families, each represented by 20–40 prawns

produced from the complete diallel cross involving three founder stocks (two native stocks

from Mekong and Dongnai rivers and Malaysian strain). The Malaysian strain was orig-

inated from wild stock at National Fry Production and Research Center, Malaysia, and was

imported to Vietnam in 2007. All the founder stocks were transferred to the National

Breeding Center for Southern Freshwater Aquaculture, under the Research Institute for

Aquaculture No. 2 in Vietnam, where they were reared to an average body weight of about

40–60 g before mating was initiated. In 2008, the progeny of the first spawning season

(G0) was produced in Vietnam, thus creating what we call the base population. Two lines

were formed with the 2008 progeny, one selected for high breeding value for live weight

(selection line) and another one selected for average breeding values (control line). A

combined between- and within-family selection was practiced in both lines. The average

proportion of selected animals was 5.1 % in females and 3.3 % in males. Note, however,

that selection was on breeding values for body weight but not by truncation (due to

inability to reproduce of some selected breeders, we had to resort to selecting lower-

ranking ones; also, the number of selected individuals contributed by each family was

restricted to avoid later inbreeding). On average, each generation was the progeny of

89–114 dams and 60–76 sires for the selection line, whereas they were the progeny of

17–42 pairs (full-sib families) for the control line. This design and experimental size were

followed throughout the course of the selection program (2008—present) during which the

data used for this study were collected.

Family production, rearing and performance testing

In each generation, production of progeny was conducted in separate breeding hapas

according to the mating design prepared for the selection (one male mated to two females)

and control (single pair mating) breeders. Larvae from each family were reared separately

in 70-l circular plastic containers (or in 1 m3 fiberglass tanks) at stocking densities of

30–40 nauplii per liter and fed with newly hatched brine shrimp nauplii (three times per

day) for the first 10 days followed by a combination of brine shrimp nauplii and egg

custard (chicken egg, high calcium milk powder, shrimp, squid flesh and fish oil) (Thanh

et al. 2009). Post-larvae (PL) were normally observed after 20–30 days in larval rearing

tanks, and metamorphosis into the PL stage occurred after 25–40 days. PL from each

family were reared separately in 1-m3 fiberglass tanks for 2 weeks at a stocking density of

1,000 PL per m3. They were fed with a commercial prawn pellet at starting feed size. After

about 2 weeks, families were subsequently transferred into a fine mesh hapa of 4 m2

Aquacult Int (2014) 22:619–631 621

123

submerged into an earthen pond at a stocking density of 150 individuals per m2. Hapas

were supplied with air from 9 pm to 6 am and were cleaned every 2 weeks to ensure good

water flow. PL were fed with a 40 % crude protein commercial prawn pellet (manufactured

by Uni-President Co. in Vietnam) and kept in hapas for 6–8 weeks until they reached a

suitable size for tagging (around 2 g). All juveniles in each family were tagged as a batch

using visible implant elastomer tags as described by Hung et al. 2012. After tagging,

approximately 60 juveniles (G0 generation) and 120 juveniles (G1–G3 generations) ran-

domly sampled from each family were released into one (G0 generation) or two common

earthen ponds (G1–G3 generations) of 3,500 m2 for grow-out. In the G2 generation, an

additional earthen pond of 800 m2 was stocked with 30 individuals from each family

produced in this generation, and at harvest, all individuals were slaughtered for assessment

of carcass weight. Grow-out stocking density was set at two individuals per m2. A detailed

description of husbandry and management practices is given in Hung et al. (2013).

Following a grow-out period of approximately 100 days, all the test prawns were

harvested. All the prawns were then stocked in conditioning cages installed in a different

pond at a very low density for 1–2 days before slaughter. Approximately half an hour

before slaughter, the prawns were collected from the conditioning cages and were killed by

cold shock using ice. They were then washed and dried, after which individual prawn were

weighed using a digital scale (BW) with a precision of 0.1 g. Standard length (BL), AL and

AW were also measured with a ruler to the nearest 0.1 mm. The tag number, sex and

morphotypes of each prawn were recorded. Immediately after data recording and mea-

surements of body traits, head and exoskeleton of individual prawns were manually

removed following our standard procedure to record tail weight (AWT) and exoskeleton-

off weight (SOW), respectively. In a similar way, the telson of individual prawns were also

taken off and weighed. Meat yield was calculated as (AWT/BW) 9 100 (MY1) and

(SOW/BW) 9 100 (MY2).

Statistical analysis

Statistical analyses were carried out on 1,730 carcass and performance records extracted

from the full pedigree of 18,387 prawns collected over three generations of selection on the

GFP strain. We conducted exploratory analyses to detect possible errors which could occur

during the data collection. Only few prawns which had meat yield out of the biological

range were removed from the dataset. Some obvious errors in data entry were also cor-

rected in the database. We further examined characteristics and distribution of the data for

all traits, using PROC UNIVARIATE in SAS (SAS Inc 1997). The data ranges of all traits

were generally within three phenotypic standard deviations. The sample statistics for

skewness and kurtosis values were close to zero indicating that the data were normally (or

approximately normally) distributed. Transformations (e.g., arcsine, natural logarithm or

square root) did not improve the distribution of the data especially for meat yield traits.

Therefore, all analyses were performed on original data for all traits.

After checking and editing, there were 1,730 prawns with both carcass and performance

records. We used a general linear model (GLM procedure in SAS) to investigate systematic

effects associated with traits. The model included the effects of selection line (high growth

and control), male morphotypes and female reproductive status and their first order

interactions. All these effects were statistically significant for all traits (P \ 0.01–0.001).

The first order interactions between line and sex were marginally significant (P = 0.059).

The age at harvest (also at slaughter) with sex was fitted as a linear covariate in the final

model.

622 Aquacult Int (2014) 22:619–631

123

Based on the above model, the data were corrected for systematic effects before mul-

tiple regression analyses had been conducted. All the traits studied were adjusted for the

systematic effects as described above, and they were assigned the same abbreviation.

Before conducting regression analyses, we also derived simple correlations among

traits, using PROC CORR in SAS. The Pearson’s correlations were based on the original

data, whereas the residual correlations were calculated from the data after the adjustments.

Stepwise forward or backward linear regression analyses were then carried out to inves-

tigate fitness of statistics tests and to derive prediction equations based on all possible

combinations of body traits for meat weight and yield. The assumptions for validity of

regression analyses (linearity between response and independent variables, normality of

model errors and collinearity) were tested and found generally satisfactory. A simple

scatter plot of body weight and meat weight showed linear relationship between the two

traits (Fig. 1a). The relationship between body weight and meat yield was negative, but

there was no evidence of nonlinearity (Fig. 1b). Since there were high correlations among

body traits, we investigated multicollinearity of these independent variables using a Var-

iance Inflation Factor (VIF). The independent variables were generally orthogonal since

the VIF value was small. A practical rule is that a VIF greater than 10 suggests multi-

collinearity (Kaps and Lamberson 2004). Finally, we also determined the best model based

on conceptual predictive criterion (Cp) and Akaike’s information criterion (AIC). The VIF,

Cp and AIC options were included in the model statement of the SAS codes (SAS Inc

1997).

Results

Body and carcass characteristics

Simple statistics for measurements of body and carcass traits are presented in Table 1. At

slaughter, the average body weight of GFP was about 34 g, with corresponding tail and

exoskeleton-off weights of 15 and 12 g and meat yield of 46 and 37 %, respectively. As

observed in several studies for other species, the coefficients of variation (CV) for body

0 20 40 60 80 100

510

1520

2530

35

Body Weight, g

Abd

omin

al w

eigh

t, g

0 20 40 60 80 100

2030

4050

60

Body Weight, g

Mea

t yie

ld, %

Fig. 1 a Linear positive relationship between abdominal meat weight (AWT) and body weight (BW).b Inverse relationship between meat yield (AWT/BW) and body weight (BW)

Aquacult Int (2014) 22:619–631 623

123

weight and meat weights were much greater than for other traits. By contrast, CV for meat

yield was low (around 15 %).

Relationship among body and carcass measurements

The measurements of body traits (body weight, total length and abdominal width) were

highly correlated (P \ 0.001) (Table 2). The correlation coefficients of body traits with

meat weight were high and positive (almost close to one for some pairs of traits), whereas

the relationships of body traits with meat yield were negative. Body weight had a higher

correlation with meat weight than with meat yield. Correlations between meat weight and

yield were of moderate magnitude. In most cases, the coefficients of simple Pearson

correlations were lower than those of residual ones because the latter was calculated after

adjusting for the systematic effects as described in Sect. ‘‘Statistical analysis.’’ Both the

Pearson’s and residual correlations indicate that body traits are strongly associated with

both meat weight and yield.

Prediction equations for meat weight and yield

Properties of regression analysis for the prediction of meat weight (ABW and SOW) and yield

(MY1 and MY2) are shown in Tables 3 and 4. Models based on body weight alone or in a

combination with other traits accounted for a majority of the variation in meat weight, as

much as 91 % (R2 = 0.91). Weight and length were better predictors of meat weight than

width. Two trait combinations of length or width with weight improved prediction models for

meat weight. The coefficient of determination (R2) or root mean square error (RMSE) values

increased slightly from 0.91 to 0.98. The full model including all three traits merely added

very little value to the prediction of meat weight (R2 remained unchanged or changed very

trivially). In all cases, the coefficients for body weight were positive, indicating that any

improvement in body weight would lead to an increase in meat weight.

For meat yield (AWT/BW and SOW/BW), prediction equations based on body traits

were generally achieved with moderate level of accuracy (Table 4). The greatest variation

in meat yield was explained by body weight when it was fitted alone or combined with any

other traits in the model. Tail width alone was a poor predictor of meat yield, but it made a

significant contribution to prediction models together with body weight and length. In

comparison with meat weight, inclusion of other characters with body weight significantly

improved accuracy of the model prediction. However, the full model including all body

traits explained from 86 to 91 % of total variation, suggesting that a fraction of variation in

meat yield was still unaccounted for in the prediction equations based on body traits. The

partial regression coefficients for body weight were all negative.

Table 1 Number of records (N),mean, SD and the coefficient ofvariation (CV, %) for body traitsand carcass measurements

BW Body weight, BL bodylength, AW abdominal width,AWT abdominal weight, SOWexoskeleton-off weight, MY1meat yield (AWT/BW) 9 100and MY2 meat yield (SOW/BW) 9 100

Traits Unit N Mean SD CV

BW g 1,730 34.3 18.5 53.9

BL cm 1,730 8.8 1.5 16.9

AW g 1,730 1.6 0.3 17.5

AWT g 1,730 14.8 6.2 41.7

SOW g 1,730 11.9 5.2 43.7

MY1 % 1,730 45.7 6.6 14.4

MY2 % 1,730 36.7 5.6 15.4

624 Aquacult Int (2014) 22:619–631

123

Model validation

Other diagnostic statistics rather than R2 and RMSE for choice of optimal prediction

models of meat weight and yield are presented in Tables 5 and 6, respectively. The best fit

of model when the Cp approaches number of parameters model and AIC is smallest. For all

traits studied, the results are consistent with Tables 3 and 4, based on R2 and RMSE. The

full model including all body traits had the smallest Cp and AIC and the greatest corre-

lations (r) between actual and predicted values, and therefore was deemed to be the best.

Ranking of model with different trait combinations for both meat weight and yield is also

given in Tables 6 and 7.

Table 2 Simple Pearson’s (above) and residual (below the diagonal) correlations between body and carcasstraits

Traits BW BL AW AWT SOW FY1 FY2

BW 0.92 0.75 0.94 0.94 -0.74 -0.68

BL 0.95 0.86 0.94 0.94 -0.62 -0.58

AW 0.79 0.92 0.85 0.79 -0.40 -0.55

AWT 0.96 0.97 0.91 -0.61 -0.47

SOW 0.97 0.98 0.83 0.97 0.67

FY1 -0.84 -0.72 -0.45 -0.66 -0.77

FY2 -0.82 -0.72 -0.73 -0.79 -0.69 0.65

Table 3 Prediction equations for meat weight (AWT and SOW) based on body traits

Trait combination Eq. Model property Partial regression coefficients

R2 MSE Intercept BW BL AW

For AWT

BW 1 0.91 1.48 3.98 0.32

BL 2 0.94 1.20 -19.51 3.89

AW 3 0.82 2.12 -18.2 20.09

BW ? BL 4 0.96 1.05 -11.59 0.12 2.53

BW ? AW 5 0.98 0.79 -7.11 0.21 8.97

BL ? AW 6 0.94 1.20 -19.78 3.62 1.62

BW ? BL ? AW 7 0.98 0.78 -5.29 0.24 -0.62 10.59

For SOW

BW 1 0.94 1.04 2.59 0.27

BL 2 0.95 0.95 -17.31 3.31

AW 3 0.69 2.37 -13.42 15.61

BW ? BL 4 0.97 0.70 -8.79 0.13 1.85

BW ? AW 5 0.95 0.93 -1.44 0.23 3.26

BL ? AW 6 0.98 0.56 -15.84 4.77 -8.76

BW ? BL ? AW 7 0.99 0.53 -12.94 0.05 3.92 -6.97

MSE root mean square error and trait abbreviations in Table 1

Aquacult Int (2014) 22:619–631 625

123

Prediction of bias

Arbitrarily we classified meat weight and yield into three different categories, each with

low, medium and high scales (Table 7). This aimed to assess possible bias of the prediction

Table 4 Prediction equations for meat yield based on body traits

Traits Eq. Model property Partial regression coefficients

R2 MSE Intercept BW BL AW

For FY1

BW 1 0.70 2.83 55.46 -0.29

BL 2 0.52 3.60 72.04 -2.98

AW 3 0.20 4.65 62.60 -10.32

BW ? BL 4 0.75 2.61 38.92 -0.49 2.69

BW ? AW 5 0.82 2.19 39.53 -0.44 12.88

BL ? AW 6 0.84 2.09 66.42 -8.59 33.56

BW ? BL ? AW 7 0.86 1.94 54.50 -0.20 -5.10 26.18

For FY2

BW 1 0.68 2.36 44.28 -0.22

BL 2 0.55 2.76 58.22 -2.44

AW 3 0.54 2.82 58.47 -13.30

BW ? BL 4 0.69 2.32 38.18 -0.30 0.99

BW ? AW 5 0.69 2.29 49.21 -0.18 -3.98

BL ? AW 6 0.57 2.72 59.15 -1.52 -5.52

BW ? BL ? AW 7 0.91 1.28 20.76 -0.63 9.70 -29.28

Table 5 Validation statistics (Conceptual predictive criterion, Cp and Akaike’s Information Criterion,AIC) and correlation (r) between actual and predicted values for the prediction equations for meat weight

Traits Eq. Cp AIC Correlation Rank

For AWT

BW 1 4508 1357 0.96 5

BL 2 2401 2503 0.97 6

AW 3 11055 2599 0.97 7

BW ? BL 4 1398 162 0.98 3

BW ? AW 5 44 -821 0.98 2

BL ? AW 6 2346 621 0.97 4

BW ? BL ? AW 7 4 -861 0.99 1

For SOW

BW 1 4995 2991 0.97 6

BL 2 3891 -191 0.98 5

AW 3 33619 2991 0.83 7

BW ? BL 4 1334 -1241 0.99 3

BW ? AW 5 3700 -249 0.98 4

BL ? AW 6 209 -2034 0.99 2

BW ? BL ? AW 7 4 -2228 0.99 1

626 Aquacult Int (2014) 22:619–631

123

equations based on predicted values, absolute residual means, and correlations between

actual and predicted values. With one or two exceptions, the correlations between actual

and predicted values were very high (close to one) for three different categories within

each trait (Table 7). Absolute residual means in these groups were almost zero. These

results indicate that the full equations provide little biased prediction for meat weight and

yield groups. It is, however, necessary to validate these equations in future dataset. In this

way, mean bias between the predicted equation and experimental data can be calculated.

Table 6 Validation statistics(Conceptual predictive criterion,Cp and Akaike’s InformationCriterion, AIC) and correlation(r) between actual and predictedvalues for the prediction equa-tions for meat yield

Traits Eq. Cp AIC Correlation Rank

For FY1

BW 1 1924 3605 0.84 5

BL 2 4167 4134 0.72 6

AW 3 8102 5319 0.45 7

BW ? BL 4 1365 3319 0.87 4

BW ? AW 5 456 2715 0.91 3

BL ? AW 6 255 2548 0.87 2

BW ? BL ? AW 7 4 2313 0.93 1

For FY2

BW 1 4134 2967 0.82 4

BL 2 6341 3520 0.74 6

AW 3 6667 3589 0.73 7

BW ? BL 4 3959 2916 0.83 3

BW ? AW 5 3810 2870 0.83 2

BL ? AW 6 6097 3468 0.75 5

BW ? BL ? AW 7 4 856 0.95 1

Table 7 Relationship between actual and predicted values of the full model (BW ? BL ? AW) for dif-ferent categories of meat weight and yield

Traits Category n PV AR r

AWT Low (\10 g) 393 6.9 0.05 0.99

Medium (10–15 g) 862 10.9 -0.03 0.78

High ([15 g) 475 17.6 0.01 0.84

SOW Low (\10 g) 173 4.1 0.03 0.98

Medium (10–15 g) 860 13.3 3.55 0.91

High ([15 g) 697 19.3 4.20 0.94

FY1 Low (\35 g) 75 38.0 -0.46 0.93

Medium (35–50 g) 1320 44.8 0.09 0.95

High ([50 g) 335 50.8 -0.24 0.38

FY2 Low (\35 g) 673 32.3 -0.07 0.66

Medium (35–40 g) 776 38.6 0.04 0.97

High ([40 g) 281 41.7 0.05 0.95

n number of records, PV predicted values, AR absolute residuals and r correlation between actual andpredicted values

Aquacult Int (2014) 22:619–631 627

123

Discussion

The meat yield of the GFP population in our study (37–46 %) falls in the high range

reported for GFP in the literature, from 25 to 48 % (Smith et al. 1980; Sahu et al. 2012).

However, note that the edible meat yield depends on populations, culture environments,

slaughter body weight, definition of calculations and age of the prawn. It is thus hard to

make a rigorous comparison because our present study carried out under farm environ-

ments compared with other experiments mostly conducted under well-controlled condi-

tions. Due to biological characteristics of giant freshwater prawn, the meat yield of this

species is remarkably lower than that in marine shrimps (56–63 %) (Jory and Cabrera

2012) or other fishes such as rainbow trout, Coho or Atlantic salmon (43–67 %) (e.g.,

Kause et al. 2002; Neira et al. 2004; Powell et al. 2008). This is mainly due to the relative

big head and claws in relation to other parts of the body of the prawn. Dressing percentage

with head and without head in our GFP population was 54.1 % and 37–46 %, respectively.

Further, a large difference between meat yield with and without skin, e.g., 46 versus 37 %

has also been observed in GFP, which is similar to the reported data for common carp

(Kocour et al. 2007). In this study, we report both skin and skinless meat yield which is

substantially higher than the average value of the same species in the literature (Smith et al.

1980).

This is the first dataset used to derive prediction equations for meat weight and yield in

giant freshwater prawn. Results reported here indicate that prediction equations based on

measurements of body traits can give very high degree of accuracy for meat weight but

only moderate level of accuracy for meat yield. Among body traits, live weight and body

length are the most important predictors which explained majority of total variation in

prediction models for meat weight. This suggests that in practical breeding programs for

crustaceans where improvement of meat weight is targeted, body weight or length can be

used alternatively as the sole selection criterion. Ideally, recording all four body traits

would improve accuracy and precision of prediction equations. When all measurements are

not available, a combination of body weight and length seems to be satisfactory for the

prediction of both meat weight and yield. For the latter trait (meat yield), body weight and

tail width are the most significant contributors explaining from 92 to 97 % of total vari-

ation across models. The addition of body length added very little improvement to pre-

diction models for both measures of meat yield, with increase in R2 value of only

0.01–0.05. The results obtained in our study indicate that tail width is not the effective

selection criterion for overall performance improvement in breeding programs.

The ability of prediction models is generally dependent on two main factors: i) number

of parameters and ii) accuracy of measurements. We found that the three body traits (body

weight, total length and abdominal width) were adequate for the prediction of meat weight.

The R2 value (0.94) achieved in our study was almost the same as that of Rutten et al.

(2004), who used the same four traits in addition to corrected length, calculated as a

difference between standard length and head length in tilapia. The model developed by

Pires et al. (2006) including six traits (body weight, head length, head height, teal width

and teal height) gave R2 of 0.82. It is impossible to make a rigorous comparison among

studies because growth characteristics and relationships between response and independent

variables are population specific. Also note that these studies were carried out on Nile

tilapia whose biological characteristics could be different from giant freshwater prawn and

hence comparison should be avoided. Rutten et al. (2004) also demonstrated that use of

separate regression coefficients for each strain gave an improvement of the explanatory

values of the prediction models for both fillet weight and yield. In terms of accuracy of

628 Aquacult Int (2014) 22:619–631

123

measurements for body traits, there is very little room for further improvement, based on

this work. We weighed the prawn using electronic scale with high precision (nearest 0.1 g)

and measured body length and tail width using standard rulers.

In addition, the use of the three body traits (weight, length and width) is expected to

result in unbiased prediction of meat weight. Although correlations among body traits and

meat weight were very high, they are statistically independent of each other. In other

words, there is no multicollinearity between the traits as determined from VIF statistic (see

Sect. ‘‘Family production, rearing and performance testing).’’ Additional measurements of

biometrical traits can increase accuracy of prediction models; however, linear dependence

among traits can result in biased regression coefficients. Explanatory variables need to be

tested before they are included in regression models to analyze meat weight and yield.

Based on the results of our present study together with the literature review, any combi-

nation of body traits gives moderate degree of accuracy in prediction models for meat

yield. Hence, it is more difficult to improve meat yield than meat weight through indirect

selection for body traits. Until now, there has been in general a lack of efficient methods of

measurement and selection for meat yield. Strategies to improve meat yield by applying

selection for smaller size (weight) of head or lower volume of intestinal organs are not

encouraged because this goes against biological nature of the species. There may also be

unfavorable or negative impact on fitness-related traits or on other biological character-

istics of the prawn. These areas have not been well known in aquatic animals. With a study

in rainbow trout, Kause et al. (2007) proposed that selection for increased eviscerated

weight can lead to correlated increase in fillet weight and yield. However, measurements of

eviscerated weight still require slaughter of siblings of breeding candidates. Furthermore,

our results showed that the correlations between body or abdominal weight and meat yield

in GFP were different from those estimated in fish, i.e., moderate to high and negative. We

therefore propose a desired gain selection index approach to improve body or meat weight

without concomitant changes in meat yield. The estimates of genetic correlations between

these traits estimated in this population (Hung et al. 2013) are useful source of information

to model alternative selection strategies for simultaneous improvement of meat weight and

yield in GFP.

In summary, the prediction equations for meat weight developed in the present study

can be effectively applied in the selection program for this GFP population as indicated by

the very high level of accuracy of models. However, extrapolation of the equations to other

populations should be with cautions unless they are independently tested. Under com-

mercial context, the equations can be used by the processing industry as an approximation

for product grading and payment purposes. In a future study, we will validate these

prediction equations for meat weights based on the same body traits in the GFP since the

relationships among body and carcass traits can be changed with the ongoing selection

program. The equations would have the greatest degree of fit (or minimum bias) if a

regression of predicted on observed values results in an intercept not different from zero

and a slope not different from one (MacNeil 1983). A bias between the predicted and

actual equations can also be calculated, and accuracy of the equations can be known from

validation statistics.

Our study also demonstrates that there are several factors affecting meat weight and

yield (Sect. ‘‘Statistical analysis’’). Among systematic effects, male morphotypes and

female reproductive status within sex explained a very large proportion of variation in the

model (53 % of total variation), followed by line and age at slaughter. The other possible

effects may have been involved such as pre-slaughter handling, transport and lairage

conditions, fasting times, killing methods as well as post-slaughter management. However,

Aquacult Int (2014) 22:619–631 629

123

they were not parts of the present experiment although these effects along with environ-

mental factors on meat yield and meat quality have been well documented in several

terrestrial animal species (e.g., Warriss 2003; Wiklund and Malmfors 2004). For farmed

aquaculture species, studies in this area have been mostly carried out in salmon (Ras-

mussen 2001) or other species (reviewed by Poli et al. 2005). There are also several other

factors impacting meat yield of the animal. Obviously this area of research in GFP

deserves further studies.

Conclusion

This study showed that body traits are the effective predictors of meat weight. By contrast,

the accuracy of prediction equations for meat yield based on measurements of body traits

was moderate. It is concluded that the application of prediction models that successfully

predict meat weight in breeding animals will enable the acceleration of genetic progress in

the nucleus herds and commercial production.

Acknowledgments The authors thank Vu, T. N., Ky, T. L. and Nga, T. K. N. for their valuable technicalassistance in both the laboratory and field trials. This work was supported by the Ministry of Agriculture andRural Development (MARD) in Vietnam and partially by the WorldFish Center in Malaysia through the‘‘Family-based selective breeding program on GFP in Vietnam.’’ The Australian government AusAIDprogram provided Hung Dinh with an ALA award to undertake his PhD research at Queensland Universityof Technology (QUT).

References

Argue BJ, Liu Z, Dunham RA (2003) Dress-out and meat yields of channel catfish, Ictalurus punctatus, bluecatfish, Ictalurus furcatus, and their F1, F2 and backcross hybrids. Aquaculture 228:81–90

Bosworth BG, Holland M, Brazil BL (2001) Evaluation of ultrasound imagery and body shape to predictcarcass and meat yield in farm-raised catfish. J Anim Sci 79:1483–1490

Cibert C, Fermon Y, Vallod D, Meunier FJ (1999) Morphological screening of carp Cyprinus carpio:relationship between morphology and meat yield. Aquat Living Resour 12:1–10

Clement S, Lovell RT (1994) Comparison of processing yield and nutrient composition of cultured Nilegiant freshwater prawn (Oreochromis niloticus) and channel catfish (Ictalurus punctatus). Aquaculture119:299–310

Hung D, Coman G, Hurwood D, Mather P (2012) Experimental assessment of the utility of VIE tags in astock improvement program for giant freshwater prawn (Macrobrachium rosenbergii) in Vietnam.Aquac Res 43:1471–1479

Hung D, Nguyen NH, Ponzoni RW, Hurwood DA, Mather PB (2013) Quantitative genetic parameters ofbody and carcass traits in freshwater prawn Macrobrachium rosenbergii. Aquaculture404–405:122–129

Jory D, Cabrera T (2012) Chapter 19: Marine Shrimp. In: Lucas JS, Southgate PC (eds) Aquaculture:Farming Aquatic Animals and Plants, 2nd edn. Wiley-Blackwell, Oxford, p 648

Kaps M, Lamberson, W (2004) Biostatistics for Animal Science. CABI Publishing, Wallingford, UKKause A, Ritola O, Paananen T, Mantysaari E, Eskelinen U (2002) Coupling body weight and its com-

position: a quantitative genetic analysis in rainbow trout. Aquaculture 211:65–79Kause A, Paananen T, Ritola O, Koskinen H (2007) Direct and indirect selection of visceral lipid weight,

meat weight, and fillet percentage in a rainbow trout breeding program. J Anim Sci 85:3218–3227Kocour M, Mauger S, Rodina M, Gela D, Linhart O, Vandeputte M (2007) Heritability estimates for

processing and quality traits in common carp (Cyprinus carpio L) using a molecular pedigree.Aquaculture 270:43–50

MacNeil MD (1983) Choice of a prediction equation and the use of the selected equation in subsequentexperimentation. J Anim Sci 57:1328–1336

630 Aquacult Int (2014) 22:619–631

123

Neira R, Lhorente J, Arande C, Diaz N, Bustos E, Alert A (2004) Studies on carcass quality traits in twopopulations of Coho salmon (Oncorhynchus kisutch): phenotypic and genetic parameters. Aquaculture241:117–131

Nguyen HN, Ponzoni RW, Abu-Bakar KR, Hamzah A, Khaw HL, Yee HY (2010) Correlated responses infillet weight and yield to selection for high growth in genetically improved farmed tilapia (GIFT strain)Oreochromis niloticus. Aquaculture 305:1–5

Pires AV, Fonseca Junior AM, Silva LHS Kuster, Araujo LDS, Fonseca CV, Pereira R, Silva FF (2006)Modeling meat weight and yield on body traits in Nile tilapia (Oreochromis niloticus L) 8th WorldCongress on Genetics Applied to Livestock Production, August 13-18, 2006. Belo Horizonte, MG

Poli BM, Parisi G, Scappini F, Zampacavallo G (2005) Fish welfare and quality as affected by pre-slaughterand slaughter management. Aquacult Int 13:29–49

Powell J, White I, Guy D, Brotherstone S (2008) Genetic parameters of production traits in Atlantic salmon(Salmo salar). Aquaculture 274:225–231

Rasmussen RS (2001) Quality of farmed salmonids with emphasis on proximate composition, yield andsensory characteristics. Aquac Res 32:767–786

Rutten MJM, Bovenhuis H, Komen H (2004) Modeling fillet traits based on body measurements in threeNile tilapia strains (Oreochromis niloticus L). Aquaculture 231:113–122

Sahu BB, Pilli BR, Lalrinsanga PL, Samal RP, Meher PK, Kanaujia DR, Sahu AK, Jayasankar P (2012) Acomparison of commercial characteristics and yield partitioning between Macrobrachium rosenbergii(De Man 1879) and Macrobrachium malcolmsonii (Milne Edward 1894). Am-Eurasian J Scientific Res7:82–85

Sang NV, Thomassen M, Klemetsdal G, Gjøen HM (2009) Prediction of meat weight, meat yield, and filletfat for live river catfish (Pangasianodon hypophthalmus). Aquaculture 288:166–171

SAS/STAT software (1997) Release 61 Cary, NC, USASilva JL, Philips H, Madrid RMM (2010) Post-harvest handling and processing. In: New MB, Valenti WC,

Tidwell JH, D’Abramo LR, Kutty MN (eds) Freshwater Prawns; Biology and Farming. Wiley-Blackwell, Oxford, pp 400–419

Smith TIJ, Waltz W, Sandifer PA (1980) Processing yields for Malaysian prawns and the implications. ProcWorld Maric Soc 11:557–569

Thanh NM, Ponzoni RW, Nguyen NH, Vu NT, Barnes A, Mather PB (2009) Evaluation of growth per-formance in a diallel cross of three strains of giant freshwater prawn (Macrobrachium rosenbergii) inVietnam. Aquaculture 287:75–83

Warriss PD (2003) Optimal lairage times and conditions for slaughter pigs: a review. Vet Rec 153:170–176Wiklund E, Malmfors G (2004) The effects of pre-slaughter handling on reindeer meat quality—a review.

Animal Breed Abstr 72:1N–6N

Aquacult Int (2014) 22:619–631 631

123