Genotype by environment interaction affectig growth and carcass traits in creole cattle raised across two contrasted fattening environments in Guadeloupe

HAL Id: hal-02819922

https://hal.inrae.fr/hal-02819922

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Genotype by environment interaction affectig growth

and carcass traits in creole cattle raised across two

contrasted fattening environments in Guadeloupe

Fabrizio Assenza

To cite this version:

Fabrizio Assenza. Genotype by environment interaction affectig growth and carcass traits in creole cattle raised across two contrasted fattening environments in Guadeloupe. Animal biology. 2010. �hal-02819922�

European Master in Animal Breeding and Genetics (EM-ABG)

Master Sciences et Technologies du Vivant

Spécialité : Sciences de l’Animal

Genotype by environment interaction

affecting growth and carcass traits in creole

cattle raised across two contrasted fattening

environments in Guadeloupe

Fabrizio Assenza

Supervisor : Michel Naves

TABLE OF CONTENT Introduction

• Creole cattle

• Farming systems in Guadeloupe

• Genotype by environment interaction

• Aim of this study

Materials and methods

• Experimental device

• Growth performance traits

• Carcass traits

• Statistical analysis

 Bivariate animal models within each environment  Bivariate character state models between environments  Bivariate random regression model between environments

Results and discussion

•Summary statistics

•Genetic parameters within environments

• Genetic parameters between environments

 Bivariate character state models between environments  Bivariate random regression model between environments

Conclusions Acknowledgments References

ABSTRACT

Creole cattle is the beef cattle farmed in Guadeloupe for meat production. Although its remarkable adaptation to the Guadeloupean environment, its production is not sufficient to fulfill the meat demand of the Guadeloupean population. A major reason for this is the lack of a structured breeding plan to improve its meat production and effort to design it is ongoing at INRA Antilles. Due to the variety of farming systems in Guadeloupe, genotype by environment interaction (GEI) affecting the growth and meat production traits could be a limiting factor and its significance must be estimated in order to design an appropriate breeding plan. We estimated GEI affecting growth traits (liveweight, average daily gain, yearling weight) and carcass traits (slaughter weight, hot carcass weight, muscles weight, adipose tissue weight, empty digestive tract weight) by implementing character state models whose response variables were the phenotypes recorded on groups of half-sibs split between two fattening systems: intensive system and pasture. We also estimated the heritabilities and genetic correlations between all traits within each environment by implementing bivariate mixed models. This study proofs that GEI affecting growth and meat production traits in Creole cattle is significant while their genetic parameters do not differ between environments.

Keywords: Genotype by environment interaction, Creole cattle, meat production traits,

SUMMARY

Bovine meat is a major source of high quality protein for human nutrition and many countries experience a gap between the demand for it and its production. Guadeloupe is one of those countries where the meat demand is fulfilled mainly by importation, although the local breed farmed for meat production (Creole cattle) features a remarkable adaptation to the farming conditions in this country. A major reason for this is the lack of a structured breeding plan to improve the production performance of Creole cattle and an effort to design it is ongoing at INRA Antilles. Breeding plans are designed to control livestock populations’ genetic make up in order for it to meet the human demand of food. This is done by selecting which animals will be mated and will transmit their DNA to the next generations. The selection of these animals is based on the ranking obtained for all the candidates according to an index which is estimated for each of them by recording the production performance of each animal and/or of its relatives from field data. This index represents the expected difference between the average production of the generation which the all candidates belong to and the average production of the offspring the selected candidates will give birth to. However, since the index is estimated from field data, it could be dependent on it and if the offspring is raised in an environment which differs very much from the environment within which the index was estimated, a re-ranking of the candidates could occur. If a significant re-ranking is neglected, the offspring’s performance is very likely to mismatch its expectation and finally the breeding plan won’t achieve the expected improvement of the population’s productivity. This could be the case in Guadeloupe, where the farming conditions vary widely from one farm to an other and could represent a limiting factor for the successful implementation of a breeding plan. The aim of this study was to estimate the magnitude of the re-ranking which could occur among the animals of a population under selection for the improvement of meat production when offspring is raised in two different fattening systems. To do this, we compared the ranking of the candidate animals obtained when their indexes are estimated from performance records collected within an intensive fattening system to their ranking obtained when their indexes are estimated from performance records collected in pasture. The results we obtained show that the re-ranking is significant, therefore: the design of the breeding plan for the improvement of meat production performance of Creole cattle must not neglect it.

INTRODUCTION

Creole cattle. The Creole cattle of Guadeloupe originated from the bovines imported

into the Caribbean islands after the discovery of the American continent. Many breeds appear to have contributed to the genetic pool of Creole cattle (Naves, 2003). The exact dynamics of these importations are not easy to trace back because of the nature of the trades in that period, however it is agreed upon that the first breeds to be imported into the Caribbean islands came from Spain, Portugal and the Canary islands. Breeds from west Africa, India and American

continent were imported during the 19th _{century, as well as more importations from Europe}

occurred later during the 20th _{century. Further crossings also occurred between the Caribbean}

breeds themselves. Due to this complex history, genetic characterization of the Creole cattle of Guadeloupe by genetic markers and molecular markers resulted indeed in the conclusion that the Creole cattle of Guadeloupe represents an original genetic resource (Naves, 2003).

Farming systems in Guadeloupe. Creole cattle is farmed for meat production mainly

and in some contexts it is still used for drought power also. A special interest for breeding Creole cattle in Guadeloupe is due to its tolerance to the stressors featuring cattle farming in this island. Namely, the main limiting factors for meat production in Guadeloupe are: climatic conditions, parasites, vector-born diseases and farming conditions. All of these factors, but the farming conditions, will be described in more detail in the materials and methods section. The farming conditions in Guadeloupe features two main typologies: the mixed farming system and the specialized pasture system. The mixed farming system is the most diffused one and it features both animal and vegetal productions within the farm, with a great variety of management conditions. Traditionally, cattle are kept on pastures tethered with a chain fixed to the ground with a picket. They are moved once or twice a day, when they consumed all the grass available within the reach of their chain, and receive water. A few specialized pasture system also exist. It is diffused only among the biggest farms and features fenced fields where bovines are free to graze on savanna. Even when pastures are not sufficient to fulfill the food requirement of the animals due to adverse climatic conditions, complementation is seldom practiced. It is based mostly on sugar cane sub-products and rarely on concentrates. Creole cattle was shown to tolerate the stressors characterizing the Guadeloupean environment better than exotic cattle breeds and crossbreeds between Creole and exotic breeds. Indeed the attempts to import purebred highly producing cows from temperate regions normally fail, while crossbred (between Creole cattle and Limousine, for example) shown lower adaptation capacity to the guadeloupean environment then purebred Creole cattle. (Berbigier and Sophie,

1986; Bru et Al., 1987; Barré, 1997; Mahieu and Naves, 2008). However, even if Creole cattle is the most adapted and best performing breed in the Guadeloupean environment, its productivity is still not sufficient to fulfill the meat demand of the Guadeloupean population and they highly rely on meat importation. A major reason for this productive gap is that the Creole cattle population of Guadeloupe have never been object of any structured breeding plan. An effort to establish a breeding plan for the genetic improvement of Creole cattle population in Guadeloupe is ongoing at the experimental herd of INRA in Domaine de Gardel.

Genotype by environment interaction Due to the variety of farming systems in

Guadeloupe, a major limiting factor for the genetic gain of the Guadeloupean Creole cattle population could be the occurrence of genotype by environment interaction, between animals managed in grazing systems with forages only, or in intensive fattening systems. Statistically speaking, an interaction between two factors in a model is significant when their effects depends on each other, that is: the effect of one factor o the response variable depends on the level of the other and vice versa (Ott and Longnecker, 2001). In the context of genetic improvement of livestock this interaction is referred to as Genotype by environment interaction (GEI), that is: the effect of the environment on the phenotype depends on the genotype and vice versa (Lynch and Walsch, 1998). Hence, when GEI is significant each genotype reacts to the environmental change in a specific way and the effect of the environmental change on the estimated breeding value (EBV) is not the same for all the genotypes (figure 1). This causes a re-ranking of the reproducers in each environment and finally: if a breeding plan neglects a significant GEI affecting the traits included in the EBV and is applied to a population of animals which are raised across different environments, the genetic gain of the population will be very likely to mismatch its expectation (Falconer, 1952; Robertson, 1959).

Aim of this study. This study is a part of the effort to establish a structured breeding

plan for Creole cattle in Guadeloupe and focuses on the estimation of GEI affecting growth traits and meat production traits expressed within half sibs groups fattened either on pasture or in intensive fattening system. An estimate of the significance of GEI affecting meat production and growth traits of Creole cattle fattened across two contrasted environments is needed to decide whether or not GEI must be taken into account in the estimation of EBVs used to implement a breeding plan for Creole cattle in Guadeloupe.

Figure 1. Reaction norms of three genotypes in response to two environments.

(A) No Genotype × environment interaction: all genotypes respond equally to the environmental change

(B) Scaling effect of genotype × environment interaction. (C) Re-ranking effect of Genotype × environment interaction.

(D) Combined scaling and re-ranking effect of Genotype × environment interaction

(1) genotype 1 shows very weak plasticity and is called a robust phenotype.

MATERIALS AND METHODS

Experimental device. The experimental device was located at Domaine de Gardel in

Grande-Terre, Guadeloupe, French West Indies, featuring the following general environmental conditions. Weather varying between two seasons: the dry season, from December to May, featuring temperatures between 22°C and 28°C; and the rain season, from June to November, featuring temperatures between 25°C and 31°C. The humidity is always 75% or higher. Parasitic pathogen agents, the most prevalent being strongyles (Toxocara

vitulorum) and ticks (Ambylomma variegatum, Boophilus spp). The latter is also the vector of Cowdria ruminantium, the etiologic agent of Cowdriosis, and predisposing factor for the

dermatophilosis caused by Dermatophilus congolensis. Phenotypic observations were recorded on several growth and carcass traits expressed by 33 half sibs groups of Creole calves. The calves were born in Creole cows herds managed outdoor and grazed in either natural savanna or improved pasture. After weaning, the calves were drenched against parasites, and followed an adaptation period of 40 days on average. The members of each group were then randomly assigned to receive one out of two fattening treatments in a roughly balanced fashion. All animals were Creole bovines and were produced either by artificial insemination or natural mating. The number of offspring per sire varied from 2 to 57 and the offspring were born from 138 dams in intensive system and 139 dams in pasture. The pedigree of the whole population included 961 animals and spanned 6 generations. The experimental device featured two fattening treatments: an intensive one (INT) and an extensive one (PAST). The former treatment featured keeping the animal in feedlots, sheltered from sun and precipitations and fed with unifeed plus a supplementation of concentrates. Namely the food provided in INT treatment was composed as follows: cropped grass ad libitum and concentrates (corn, middling, soybean meal, limestone, urea, minerals and vitamins). The concentrate was distributed using a automatic concentrate distributor, to a quantity tabulated according to the liveweight of each animal, in order to cover 60 % of the intakes capacity of the animal Whereas the latter featured free grazing on pasture as the only food source and no shelter either from sun nor precipitations. The stocking rate was between 1500 Kg/ha and 2500 Kg/ha for the pasture system. Acaricide treatment was given every two weeks.

Growth performance traits The traits analyzed to measure growth performance were:

live weight (LWGT), average daily gain (ADG) and yearling weight (YRWGT). The observations on LWGT were collected in a longitudinal dataset including repeated

observations on the weight of each animal at different ages. LWGT was used to compute ADG according to the following formula: _adg= (_lwgtnth _lwgt1 ) / (₋st _agenth _age1 )st _{. Where −} lwgtnth _{and lwgt1}st _{are the weight at the last weighting and at the first one, respectively; agen}th

and age1st _{are the ages, expressed in days, at the last weighting and at the first one,}

respectively. YRWGT was computed as follows: if the animal had no LWGT at 365 days, the average daily gain between the nearest observations on LWGT around 365 days was computed and the expected weight at 365 days was obtained according to it. If the first LWGT was after 365 days, the first two lwgts were used to compute the average daily gain and YRWGT was derived according to it.

Carcass traits Most of the males have been slaughtered, while the females were kept

or sold for reproduction,. Animals were sacrificed when they reached an acceptable commercial weight for the local market (around 320 Kg). Due to the difference in growth rate between INT and PAST, animals were slaughtered between 14 and 17 month in INT, and between 17 and 21 month in PAST. The traits measured on the carcasses were: slaughter weight (SLWGT), weight of the empty digestive tract (DTW), hot carcass weight (HCW), weight of the muscular tissue (MTW), weight of the adipose tissue (ATW) . SLWGT was the weight of the animal as measured just before being slaughtered and after one day of starving. DTW was the weight of the empty digestive tract. HCW was the weight of the carcass at the end of the slaughter chain. MTW was the weight of the skeletal muscular tissue in the carcass and was computed according to the following equation (Naves, 1996):

42.7591 0.7716* 1.8393* 35.4953* 11 48.3453 11 17.3082* 11 74.9204( / ) tmw hcw pat atr btr ldr hlt hll = − + − − + + + +

(r²=0.99). Where pat is the peritoneal adipose tissue; atr11, btr11 and ldr11 are the adipose tissue, the bone tissue and the muscle longissimus dorsi as measured at the 11th _rib,

respectively; hlt is the hind limb thickness; hll is the hind limb lenght. ATW was computed according to the following equation:

3.7591 0.1286* 0.6717* 31.2987* 11 53.1076* 11 48.3433* 11

atw= + lwt ckf + ldr − atr btr + −

(r²=0.93). Where ckf is the cold kidney fat.

Statistical analysis

The dataset was analyzed by using, SAS®_{, ASREML}® _{(Gilmour at Al, 1995) and MATLAB}®

software. Data editing and exploration was performed by using SAS® _{software, the linear}

models only, the genetic parameters were computed by using MATLAB® _{software. Several}

linear mixed models were implemented: bivariate animals models to estimate the genetic correlations between traits within each fattening system; bivariate character states models (Falconer, 1952) to estimate the genotype by environment interaction affecting each trait expressed across the two fattening systems; a bivariate random regression model between environments (Kirkpatrik and Heckman, 1989) in order to estimate the genetic parameters of LWGT and the significance of GEI affecting it.

Bivariate animal models within each environment The bivariate animal models read as

follows: µ µ 1 2   ₊                                        1 1 1 1 1 1 2 2 2 2 2 2 y X 0 b Z 0 u e = + + 0 X 0 Z y b u e

Where y1 is the vector collecting observations on one trait expressed within one fattening

environment and y2 is the vector collecting the observations on a different trait expressed

within the same fattening environment. μ1 and μ2 are the means of trait one and two,

respectively. b1 andb2 are the vectors of fixed effects for trait one and two, respectively; the

fixed effects included in the model were a class factor coding for the contemporary groups and the following covariables: weaning age (WAGE), weaning weight (WWGT) and, for the

carcass traits only, age at slaughter (SLAGE). X1 and X2 are the incidence matrices connecting

each observation to its set of fixed effects. u1 andu2 are the vectors of breeding values for trait

one and two, respectively. Z1 and Z2 are the incidence matrices connecting each observation

to its set of random effects. The only random effect fitted in the model was the animal effect.

e1 ande2 are the vectors of residuals for trait one and two, respectively. The variance structure

of all these models was:

11 12 21 22 11 12 21 22 0 0 0 0 var 0 0 0 0 g g g g r r r r ⊗ ⊗    _{⊗ ⊗}   _{  } =  _{  }  _{  }   1 2 A A y A A y

Where: g11 and g22 are the genetic variances for trait one and two, respectively. g12 is the

genetic covariance between the two traits. A is the genetic relationships matrix between the animals in the pedigree. r11 and r22 are the residual variances of trait one and two, respectively.

r12 is the residual covariance between the two traits.

The heritability of each trait was computed as h2 = σ g2/ p2σ. Where h is the narrow sense 2 heritability of the trait, σ g2 is the estimate of its genetic variance obtained by restricted

maximum likelihood and σ 2p is the estimate of its phenotypic variance obtained by the same procedure. The genetic correlations between two traits expressed within the same fattening

environment were computed as _{1,2 1,2 /} 2_{1 *} 2₂

g g g g

r = σ σ . Where: σ r is the genetic correlation g1,2

between trait 1 and trait . σ g1,2is the genetic covariance between them. 1

2

g

σ and σ g2₂ are the genetic variances of trait one and two, respectively (Mrode, 2005).

Bivariate character state models between environments The character state models

implemented to estimate GEI read as follows:

µ µ 1 2   ₊                                        1 1 1 1 1 1 2 2 2 2 2 2 y X 0 b Z 0 u e = + + 0 X 0 Z y b u e

Where y1 is the vector collecting the observations on one trait expressed within INT and y2 is

the vector collecting the observations on the same traits expressed within PAST. μ1 and μ2 are

the means of the trait for INT and for PAST, respectively. b1 andb2 are the vectors of fixed

effects in INT and in PAST, respectively; the fixed effects included in the model were: a class factor (coding for the contemporary groups), wage, WWGT and, for the carcass traits only,

slage. X1 and X2 are the incidence matrices connecting each observation to its set of fixed

effects. u1 andu2 are the vectors of breeding values for INT and PAST, respectively. Z1 and Z2

are the incidence matrices connecting each observation to its random effects. The only random effect fitted in the model was the animal effect. e1 ande2 are the vectors of residuals in

INT and PAST, respectively. Using the same notation as for the variance structure on the bivariate animal models, the variance structure of the character state models reads as follows:

11 12 21 22 11 22 0 0 0 0 var 0 0 0 0 0 0 g g g g r r ⊗ ⊗    _{⊗ ⊗}   _{  } =  _{  }  _{  }   1 2 A A y A A y

We used the genetic correlation between the same trait expressed across INT and PAST as measure of GEI. This parameter was computed as:

, ,

2 2

/ *

INT PASTt INT PAST INT PAST

g g g g

r = σ σ . Where σ

,

INT PAST

g

r _{is the genetic correlation between the observations on the same trait expressed across}

INT and PAST. σ 2g_INT and

2

PAST

g

σ are the genetic variances of the trait expressed within INT and

PAST, respectively. σ gINT PAST, is the genetic covariance between the observations on the same

other measures of GEI. The second method relied on the two vectors of EBVs obtained for each trait analyzed by the character state model: one vector collected the EBVs of all animals estimated from the phenotypic observations on the trait expressed within INT, whereas the other vector collected the EBVs of the same animals estimated from the phenotypic observations on the same trait expressed in PAST. The correlation between the elements of these two vectors was computed for each trait as a measure of GEI. This correlation was

computed as: _{, , /} 2 _* 2

INT PAST INT PAST INT PAST

EBV EBV EBV EBV

r = σ σ . Where σ rEBVINT PAST, is the correlation

between the EBVs of all animals estimated from the observations in INT and PAST. ,

INT PAST

EBV

σ is the covariance between the EBVs estimated in INT and PAST. σ EBV2 _INT and

2

PAST

EBV

σ are the variances of the EBVs estimated within INT and PAST, respectively. As a

third measure of GEI we estimated the r2 _{adjusted of the following regression model}

implemented by proc reg of SAS® _software:

i i i

EBVi = µ EBVp+ e . Where + EBVi and i

i

EBVp are the EBVs of animal i estimated from the observations on a trait expressed within

INT and PAST, respectively. µ is the mean of the EBVs estimated within INT. e is the i

residual (Mrode, 2005).

Bivariate random regression model between environments lwght was analyzed by

implementation of a bivariate random regression models with Legendre polynomials of order 2 for the fixed and animal effect, and Legendre polynomials of order 1 for the individual permanent environmental effects. Its expression in matrix notation reads as follows:

1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2                                           y X 0 b Z 0 a Q 0 pe e = + + + 0 X 0 Z 0 Q y b a pe e

Where y1 and y2 are the vectors collecting observations on LWGT in INT and in PAST,

respectively. X1 and X2 are the incidence matrices relating each weighting to its set of fixed

regression coefficients for trait one and two, respectively. b1 and b2 are the column vectors of

fixed regression coefficients for trait one and two, respectively. The fixed effects fitted in this model were: the contemporary group (defined as the group of animals weighted on the same day, of the same gender, fattened within the same fattening system and which entered the

experiment on the same year), the weaning age and the weaning weight. Z1 and Z2 are the

incidence matrices relating each observation to its random regression coefficient for animal

effect for trait one and trait two, respectively. a1 and a2 are the column vectors containing the

Q2 are the index matrices relating each observation to its random regression coefficient for

permanent environmental effect for trait one and two, respectively. pe1 and pe2 are the column

vectors of random regressions coefficients for permanent environmental effects for trait one and two respectively, respectively. The variance structure of these model was:

1 2 var e e ⊗ ⊗    _{⊗ ⊗}     ⊗    =     _⊗             11 12 21 22 1 1 2 2 G A G A 0 0 0 0 G A G A 0 0 0 0 y 0 0 P I 0 0 0 0 0 0 P I 0 0 y 0 0 0 0 0 0 0 0 0 0

Where G11 and G22 are the 3 by 3 matrices containing the random regression coefficients

modeling the genetic variance function of trait one and two, respectively. G21 is the 3 by 3

matrix containing the random regression coefficients modeling the genetic covariance

function between trait one and two , respectively. G12 is the transpose of G21. A is the 961 by

961 genetic relationships matrix. P1 and P2 are the 2 by 2 matrices containing the random

regression coefficients modeling the variance of permanent environmental effects for trait one

and two respectively. e1 and e2 are the residual variances of trait one and two respectively. The

genetic variance of trait i at age j was computed as σ =i2jj t G t' . Where j ii j

2 1jj

i

σ is the genetic

variance of trait i at age j, tj is jth the row vector of the Nage by 3 (Nage is the number of ages

at which weightings were performed) Legendre polynomials matrix. t’j is the transpose of tj.

Similarly, the genetic covariance between trait one and two at each age was computed as:

12jj

σ = t G t' . While the genetic covariance of trait i between age j and k was computed as: j 12 j

jk

i

σ = t G t' . From these point estimates of genetic parameters, we estimated the j 12 k

heritabilities of each trait at each age and the genetic correlations between traits at each age as well as the genetic correlation of each trait at different ages by applying the same formulas

described for the animal models. The EBVs of animal i at age j for trait k (EBVijk) were

computed as EBVijk = u t' . Where uik j ik is the 1 by 3 row vector of random regression

coefficients of animal i for trait k and t’j is the 3 by 1 column vector of Legendre polynomials

for age j. The correlation between the EBV for trait one and the EBV for trait two of all animals was computed at each age using the same formula described for the character state model. (Mrode, 2005)

RESULTS AND DISCUSSION

Summary statistics Observations on growth performance traits were available on each

of the 718 animals (376 on INT and 342 on PAST), while observations on carcass traits were available on 293 animals only (171 in INT and 122 in PAST). The total number of observations on LWGT was 13163, which was the sum of 6819 observations on LWGT of 376 animals within INT and 6344 observations on LWGT of 342 animals within PAST. The number of observations on LWGT at each age are plotted in Figure 2, which shows the parabolic trend of the plot with downward concavity. The number of observations on LWGT was low until 300 days and then decreased again after 500 days, consequently the graphs of its parameter estimates were considered more reliable between 300 and 500 days. Due to the resources available for this research, both the number of observations on each trait and the number of animals included in the pedigree were small. Indeed the standard errors of all the estimated genetic parameters were big, singularities occurred in the average information matrices of some models and their convergence was not easy to be reached (Hill and Thompson, 1978; Wang et Al., 1992; Reverter and Kaiser, 1997). Further, while the number of observations per fattening treatment, sex and year was quite balanced, the number of offspring per sire was not. However, since all genetic parameters were estimated by restricted maximum likelihood, this should not cause too much bias on their estimates (Harville, 1977; Swallow and Monahan, 1984; Wang et Al., 1991; Liu et Al., 1997).

Table 1 summarizes the observed mean of each trait within each fattening system. The means of YRWGT, ADG, HCW and ATW were higher in INT than in PAST. The means of SLWGT, MTW, weaning weight and weaning age did not differ between the two fattening systems. The means of DTW, btw and age at slaughter were higher in PAST than in INT. Due to the higher nutritional level of INT treatment with respect to PAST treatment, the mean of the traits related to growth and tissue synthesis was higher in INT than in PAST. SLWGT did not differ between treaments because the animals were slaughtered after having reached a commercial weight around 300 Kg, indeed SLAGE was higher in INT than in PAST. While WAGE and WWGT did not differ between treatments because the animals were randomly assigned to receive one of them. We expected MTW to be higher in INT that in PAST as well, but the analysis did not confirm this expectation. DTW was higher in PAST than in INT, we expected this result because the fiber content of the food used was higher in PAST than INT and it induces the digestive tract to increase its capacity. (Kay, 1989; Hoffman, 1989).

INT PAST mean se n mean se n p YRWGT 243.56 48.70 376 201.57 36.16 342 *** ADG 682.84 166.32 376 421.04 148.37 342 *** SLWGT 319.99 38.23 171 315.71 44.39 122 N.S. HCW 180.01 24.25 171 172.15 28.87 122 ** MTW 121.12 19.34 171 118.19 24.60 122 N.S. ATW 28.83 6.26 171 23.88 4.88 122 *** DTW 16.39 2.11 171 17.79 2.30 122 *** WAGE 211.50 16.30 376 211.31 16.85 342 N.S. WWGT 156.35 29.81 376 158.54 27.69 342 N.S. SLAGE 15.65 1.59 171 19.51 2.01 122 ***

Table 1. Summary statistics of the traits analyzed. The table reports the mean, standard error

(se) and number of observations (n) obtained for each trait in each system. p is the p value of the two sided t test for the difference between the mean observed in innt and the mean observed in PAST for each trait (*** is p<0.0001; **is 0.0001<p<0.025; N.S. is p>0.025). YRWGT is yearling weight. ADG is average daily gain. SLWGT is slaughter weight. HCW is hot carcass weight. MTW is the weight of the muscles in the carcass. ATW is the weight of the adipose tissue in the carcass. DTW is the weight of the empty digestive tract. WAGE is the weaning age. WWGT is the weaning weight. SLAGE is the age at slaughter.

Genetic parameters within environments The heritability estimates, the genetic

correlations and phenotypic correlations between traits within the same fattening system are given in table 2 together with their standard errors. The difference between the genetic correlations obtained in INT and the cprresponding values obtained in PAST spanned from -0.53 to 0.44 and difference between the phenotypic correlations spanned from -0.22 to 0.18. The standard errors of the genetic correlations spanned from 0.08 to 0.61 in INT and from 0.09 to 0.91 in PAST. While the standard errors of the phenotypic correlations spanned from 0.01 to 0.10 in INT and from 0.01 to 0.10 in PAST. The estimates of the genetic parameters obtained from the bivariate models including MTW and DTW, MTW and YRWGT, MTW and HCW, in PAST as response variables were not considered very reliable because the models converged after many iterations and required the constraining of their Average information matrices to be positive definite. This result was probably related to the heritabilities obtained for those traits, which were lower in PAST than in INT (Hill and

Thompson, 1978). The models including HCW and SLWGT, DTW and MTW (in both

systems) and the models including ATW and DTW, ATW and SLWGT (in PAST) as response variables gave the same problems although their heritability estimates in PAST were similar or higher than the corresponding values in INT. The phenotypic correlations of the models which required the constraining of the average information matrices were considered as indicators of the corresponding genetic correlations (Lynch and Walsh, 1998; Koots and Gibson, 1998). The estimates of genetic correlations between YRWGT, ADG, SLWGT and HCW were in accordance with the estimates reported in previous similar studies (Barwick et

Al., 2006; Bouquet et Al., 2010) and were all high and positive. Given the standard errors

associated with the genetic correlation estimates, most the values obtained in INT were considered equal to the corresponding estimates obtained in PAST. The largest differences between the values estimated in INT and in PAST were observed between the genetic and phenotypic correlations of DTW with the other traits, which were higher in PAST than in INT. This result together with the result we obtained for the average of DTW in PAST and the literature referring to the effect of diet on the development of the digestive tract of ruminants (Kay, 1989; Hoffman, 1989) suggests the hypothesis that the overall development of the digestive tract plays an important role for meat production performance in pasture. It is worth to note that the genetic correlations between ATW and the other traits are positive in both systems. These results suggests two hypotheses. The efficiency of the digestive system of Creole cattle such that it can extract from low energy foodstuffs enough resources to be allocated simultaneously both to production traits and to its own energy deposits (De Jong and

Van Noordwijk, 1994). This efficiency could be related to the higher weight of the digestive tract observed in pasture respect to its weight in intensive fattening system, the genetic correlation of DTW with growth and carcass traits are indeed higher in PAST than in INT. However, no literature was found concerning the genetic parameters of tdw.

INT YRWGT ±se ADG ±se DTW ±se SLWGT ±se HCW ±se MTW ±se ATW ±se YRWGT 0.30 ±0.12 0.81 ±0.16 0.01 ±0.47 0.77 ±0.21 0.79 ±0.19 0.60 ±0.23 0.60 ±0.26 ADG 0.57 ±0.04 0.24 ±0.11 0.62 ±0.57 0.93 ±0.13 0.95 ±0.11 0.83 ±0.14 0.52 ±0.28 DTW 0.33 ±0.08 0.48 ±0.06 0.15 ±0.16 0.47 ±0.46 0.14 ±0.59 0.00 ±0.58* 0.63 ±0.43 SLWGT 0.64 ±0.05 0.72 ±0.04 0.62 ±0.05 0.33 ±0.20 0.87 ±0.09* 0.75 ±0.16 0.70 ±0.22 HCW 0.71 ±0.04 0.79 ±0.03 0.47 ±0.07 0.93 ±0.01 0.39 ±0.21 0.97 ±0.03 0.35 ±0.34 MTW 0.65 ±0.05 0.74 ±0.03 0.38 ±0.07 0.86 ±0.02 0.96 ±0.01 0.49 ±0.22 0.09 ±0.37 ATW 0.50 ±0.06 0.54 ±0.06 0.54 ±0.06 0.67 ±0.05 0.58 ±0.06 0.35 ±0.08 0.44 ±0.21

PAST YRWGT ±se ADG ±se DTW ±se SLWGT ±se HCW ±se MTW ±se ATW ±se YRWGT 0.21 ±0.12 0.88 ±0.15 0.52 ±0.30 0.75 ±0.23 0.91 ±0.17 0.96 ±0.00* 0.27 ±0.38 ADG 0.60 ±0.04 0.35 ±0.13 0.93 ±0.17 0.87 ±0.17 0.94 ±0.23 1.00 ±0.92 0.94 ±0.19 DTW 0.44 ±0.07 0.58 ±0.06 0.85 ±0.23 0.86 ±0.18 0.67 ±0.28 0.40 ±0.22* 0.92 ±0.16 SLWGT 0.64 ±0.06 0.74 ±0.04 0.68 ±0.06 0.39 ±0.25 0.92 ±0.09* 0.76 ±0.28 1.00 ±0.00* HCW 0.74 ±0.04 0.77 ±0.04 0.55 ±0.07 0.94 ±0.01 0.28 ±0.23 0.96 ±0.07* 0.67 ±0.42 MTW 0.71 ±0.04 0.71 ±0.04 0.48 ±0.08 0.88 ±0.02 0.98 ±0.00 0.17 ±0.19 0.41 ±0.66 ATW 0.32 ±0.08 0.56 ±0.06 0.54 ±0.08 0.59 ±0.06 0.45 ±0.08 0.27 ±0.09 0.52 ±0.25

Table 2. Genetic correlations (above diagonal) and phenotypic correlations (below diagonal) between the traits analyzed. The heritability of

each trait is on the bold diagonal. The values obtained in intensive fattening system are in the upper table (INT), the values obtained in pasture are in the lower table (PAST). YRWGT is yearling weight. ADG is average daily gain. DTW is the weight of the empty digestive tract. SLWGT is slaughter weight. HCW is hot carcass weight. MTW is the weight of the muscles in the carcass. ATW is the weight of the adipose tissue in the carcass. * means that the model required the constrain of the average information matrix and converged after many iterations.

Genetic parameters between environments

Bivariate character state models between environments. The results from the character state

models are summarized in table 3. While the heritabilities obtained from this models were included in the computation of the average heritabilities shown in the diagonal of table 2. The heritabilities estimates spanned from 0.15 to 0.49 in INT and from 0.17 to 0.85 in PAST. Their standard errors spanned from 0.11 to 0.22 in INT and from 0.12 to 0.25 in PAST. The estimates of the heritabilities obtained for YRWGT, ADG SLWGT, HCW and ATW in this study were similar to the values reported on previous studies performed within the tropics for the same traits or related traits (Marshall, 1994; G.B. Mourão et Al., 2007; Boligon et Al, 2009;).

The genetic correlation between any trait expressed within INT and the same trait expressed in PAST were all lower than 0.8 except for DTW. The standard errors of these correlations spanned from 0.26 to 0.43. Genetic correlations estimates below 0.8 were considered as indicators of significant GEI (Robertson, 1959). The genetic correlation between MTW expressed in INT and MTW expressed in PAST was 0.84, which is slightly higher than the threshold of significance we defined. This result could be related to the fact that the average of MTW in PAST did not differ from the average of MTW in INT. However, given its high correlation with traits affected by significant GEI, this value was considered indicator of significant GEI. The results obtained for YRWGT, ADG, SLWGT, HCW, and MTW agreed with previous literature about GEI affecting growth and carcass traits in the tropics (Alencar

et Al., 2005; Texeira et Al, 2006; Barwick et Al., 2009). GEI affecting DTW was not

significant, this result proofs the effect of the fattening environment on the weight of DTW is the same for all the genotypes. No literature was found about this.

The correlations between the EBVs of all animals for any trait expressed in INT and the EBVs of all animals for the same trait expressed in PAST were higher than the corresponding genetic correlations and followed the same trend, but their standard errors were much smaller. The R2_{-adjusted of the regressions spanned from 0.27 to 0.99 and supported the results}

YRW GT ± se ADG ± se DTW ± se SLW GT ± se HCW ± se MTW ± se ATW ± se

Cov 49 ±31.03 889 ±988 1.36 ±0.71 229 ±167.11 81.1 ±61.91 69.34 ±42.80 4.54 ±4.32

rg 0.59 ±0.35 0.31 ±0.34 0.93 ±0.40 0.69 ±0.39 0.66 ±0.41 0.85 ±0.43 0.43 ±0.38

rBV 0.81 ±0.01 0.52 ±0.02 0.99 ±0.00 0.83 ±0.01 0.82 ±0.01 0.97 ±0.00 0.62 ±0.02

Reg 0.66 0.27 0.99 0.69 0.68 0.94 0.38

Table 3. This table shows the genetic covariance (Cov) and the genetic correlation (rg) between each trait expressed within the intensive fattening

system and the same trait expressed within pasture. rBV is the correlation between the breeding values of all the animals for each trait estimated in

intensive fattening system (INT) and their EBVs estimated in pasture (PAST). Reg is the r2 _{adjusted of the regression model where the EBV of}

each animal estimated in INT is regressed on its EBV estimated in PAST. YRWGT is yearling weight. ADG is average daily gain. DTW is the weight of the empty digestive tract. SLWGT is slaughter weight. HCW is hot carcass weight. MTW is the weight of the muscles in the carcass. ATW is the weight of the adipose tissue in the carcass.

Bivariate random regression model between environments. The analysis of LWGT by the

random regression model resulted in the following genetic variance and covariance matrices:

405.6 164.6 161.3 164.6 179.4 97.45 161.3 97.45 324.5 −     =   −    11 G 76.95 32.20 18.68 12.37 28.26 64.88 47.62 34.54 26.21 −     = _ − _     12 G 283.9 169.5 14.05 169.5 123.5 3.057 14.05 3.057 61.28 −     = _ − _ − −    22 G

The estimates of the genetic parameters for LWGT at each age are shown by graphs due to the dimension of the matrices. All the estimated genetic parameters were considered more reliable after 300 days because of the higher number of observations available. Figure 3 summarizes the phenotypic and genetic variances of LWGT within each fattening system and the genetic covariance between LWGT expressed in INT and LWGT expressed in PAST. Both the genetic and phenotypic variances tend to increase with the age in both systems, with a more steady increase in PAST and an overall higher increase in INT. While the genetic covariance is quite constant.

Figure 4 summarizes the estimates at each age of heritability in INT, of heritability in PAST, of the genetic correlation between LWGT expressed in INT and LWGT expressed in PAST

(rg) and of the correlation between the EBVs of all animals estimated from the observations

collected within INT and the EBVs of all animals estimated from the observations collected

within PAST (rbv). The trajectories of the heritabilities agree with the trend of their genetic

and phenotypic variances shown in figure 3: from 300 days on the trajectory of the genetic and phenotypic variances in PAST are quite linear and the trajectory of heritability in PAST is constant on the value of 0.5. While the trajectory of the heritability in INT oscillates around the same value between the range of 0.3 and 0.7, following the curves observed in the graph of its variances. Being higher than heritability in PAST until 450 and lower after that day. Figure 5 and 6 show the genetic correlation of LWGT between ages estimated in INT and in PAST, respectively. In INT, the graph of genetic correlation across ages displays a highly correlated surface between 300 and 450 days of age which is wider than the corresponding surface in PAST. While the highly correlated surface from 450 days on is wider in PAST than in INT. These results are in accordance with the trajectories of the sire’s EBVs estimated within each fattening system (which are shown in figure 7 and 8, respectively). In INT, most of the EBVs trajectories are parabolic with downward concavity, some of them have upward concavity and a few linear trajectories occur as well. The range of the EBV estimates span from -40 to 40, and most of the curves reach their maximum (or minimum) between 350 and 500 days. In PAST, most of the EBVs trajectories in PAST are parabolic with downward concavity, some have upward concavity and some are linear. However, the concavity of the

parabolas are not very pronounced. Between 350 and 450 days, most of the parabolas reach their maximum or minimum, within a range of 20 and -20. After that day, they diverge constantly. The EBVs trajectories show more variation in INT than they do in PAST and their concavity is much more pronounced in INT than in PAST. These graphs together with figure 9, showing the EBV of each sire for the whole growth trajectory estimated in INT and in PAST, proof that the effect of the fattening system on the EBV is not the same for all the sires but it depends on the genotype of the sire itself. Indeed, the difference between the EBV of each sire estimated in INT and its EBV estimated in PAST at each age (figure 10) are not the same for all the sires. This graph is in accordance with the plot of rg and rbv along the ages

shown in figure 4. Indeed, the differences between the EBVs estimated in each system are smaller at the ages featuring higher correlations and vice versa.

The graphs showing the dependency between EBV and fattening system together with the plot of the overall EBV of each sire in each fattening system and the graph showing that the genetic correlations between LWGT expressed in INT and LWGT expressed in PAST at the same age are always lower than 0.8, proof that GEI affects LWGT significantly during the whole growth trajectory. These results are in accordance with previous investigations on GEI affecting growth traits in the tropics (Menendez and Mandonnet, 2006; Pegolo et Al., 2008).

Figure 3. Plot of the variances obtained for live weight at each age. Auge VarP and Auge VarG are the phenotypic and genetic variances in

intensive fattening system, respectively. PAST VarP and PAST VarG are the phenotypic and genetic variances in pasture, respectively. CovG is the genetic covariance between live weight expressed in intensive fattening system and liveweight expressed in pasture.

Figure 4. Plot of the genetic parameters obtained for live weight at each age. Auge h2 and PAST h2 are the heritabilities obtained in intensive

fattening system and pasture, respectively. CorG is the genetic correlation between liveweight expressed in intensive fattening system and pasture. CorEBV is the correlation of the breeding of all the animals estimated at each age in intensive fattening system with their breeding values estimated at the same age in pasture.

Figure 5. Contour plot of the genetic correlation between live weight expressed at each age in intensive fattening system. aage1 and aage2 are

Figure 6. Contour plot of the genetic correlation between live weight expressed at each age in pasture. page1 and page2 are the ages. pgcor is the

Figure 9. Plot of the EBV for live weight of each sire for the whole growth trajectory in each system. INT and PAST are the EBVs of each sire

Figure 10 Plot of the difference between the EBV for live weight of each sire at each age estimated in intensive fattening system and its EBV for

CONCLUSIONS

Significant genotype by environment interaction affects yearling weight, average daily gain, slaughter weight, hot carcass weight, muscular tissue weight and adipose tissue weight expressed across intensive fattening system and pasture. Genotype by environment interaction was not significant for empty digestive tract weight. These results indicate that genotype by environment interaction is not negligible in the design of a breeding plan for improving growth and meat production traits of Creole cattle in Guadaloupe. However, the fattening systems had no significant effect on the genetic correlations’ estimates between yearling weight, average daily gain, slaughter weight, hot carcass weight, muscular tissue weight and adipose tissue weight. Hence the correlated responses to selection for these traits in each fattening environment won’t differ between systems. This results suggests the hypothesis that the efficiency of the digestive system of Creole cattle is such that it can extract sufficient resources to allocate both to production traits and its own energy deposits simultaneously. Further, the higher genetic correlations of the empty digestive tract weight with the other traits and the higher average weight of the empty digestive tract observed in pasture suggest a relationship between the digestive efficiency and the weight of the digestive tract.

AKNOWLEDGEMENTS

I’m grateful to Alberto Menendez Buscadera for his help with the random regression analysis, to Jean-Luc Goudrine for having written the code we used implemented in Matlab and to Michel Naves for its supervision.

REFERENCES

Barré N. (1997) Les tiques des ruminants dans les Petites Antilles: biologie, importance economique, principes de lutte. INRA Prod. Anim.10, 111-119

Barwick S.A., Wolcott M.L., Johnston D.J., Burrow H.M. and Sullivan M.T. (2009) Genetics of steer daily and residual feed intake in two tropical beef genotypes, and relationships among intake, body composition, growth and other post-weaning measures. Anim. Prod. Sci.49, 351-366.

Berbigier P. and Sophie S.A. (1986) Performances de croissance et d’abbatage de taurillons limousins x creoles et Créoles élevés au soleil et à l’ombre en guadeloupe (antilles françaises) rev. elev. méd. vét pays trop. 39, 81-88.

Bru J.C., Berbigier P., and Sophie S.A. (1987) Estimation of Sweat Rate and Thermal Tolerance of Pure Creole and of Limousin x Creole Crossbred Growing Bulls in Guadeloupe (French West Indies). Int. J. Biometeor. 31, 77-84.

De Almeida Teixeira R. and De Albuquerque L.G., De Alencar M.M. and Dias L.T. (2006) interação genótipo-ambiente em cruzamentos de bovinos de corte. R. Bras. Zootec. 35, 1677-1683

De Jong J., van Noordwijk A.J. (1992) Acquisition and allocation of resources: genetic (co)variances, selection, and life histories. Am. Nat. 139, 749-770

Falconer D. S. (1952) The Problem of Environment and Selection Am. Nat. 86, 293-298

Gilmour A.R., Thompson R., Cullis B.R. (1995) Average Information REML: An Efficient Algorithm for Variance Parameter Estimation in Linear Mixed Models. Biometrics 51, 1440-1450

Hill W.G. and Thompson R. Probabilities of Non-Positive Definite between-Group or Genetic Covariance Matrices.Biometrics 34, 429-439

Hofmann R.R. (1998) Evolutionary Steps of Ecophysiological Adaptation and Diversification of Ruminants: A Comparative View of Their Digestive System. Oecologia 78, 443-457

Kay R. (1989) adaptation of the ruminant digestive tract to the diet. Acta Vet Scand Suppl. 86, 196-203

Kirkpatrick M. and Heckman N.J. (1989) A quantitative genetic model for growth, shape, reaction norms, and other infinite-dimensional characters Math. Biol. 27, 429-450 Koots K.R. and Gibson J.P. (1996) Realized Sampling Variances of Estimates of Genetic

Parameters and the Difference Between Genetic and Phenotypic CorrelationsGenetics

143, 1409-1416.

Liu B.H., Knapp S.J., Birkes D. (1997) Sampling distributions, biases, variances, and confidence intervals for genetic correlations. Theor App Genet 94, 8-19

Lynch M. and Walsch B. (1998) Genetics and analysis of quantitative traits, 1st edn., Sinauer Associates, Sunderland, 980pp.

Mahieu M. and Naves M. (2008) Incidence of Toxocara vitulorum in Creole calves of Guadeloupe Trop Anim Health Prod 40, 243–248

Marshall D.M. (1994) Breed differences and genetic parameters for body composition traits in beef cattle J Anim Sci 72, 2745-2755.

Mello de Alencar M., dos Santos Mascioli A., Ribeiro de Freitas A. (2005) Evidências de Interação Genótipo x Ambiente sobre Características de Crescimento em Bovinos de Corte R. Bras. Zootec. 34, 489-495

Mourão G.B., Ferraz J.B.S., Eler J.P., Balieiro J.C.C., Bueno R.S., Mattos E.C. and Figueiredo L.G.G. (2007) Genetic parameters for growth traits of a Brazilian Bos taurus x Bos indicus beef composite. Genetics and Molecular Research 6, 1190-1200

Mrode R.A. (2005) Linear models for the prediction of aimal breeding values 2nd _{edn., CABI}

Publishing, Cambridge, 334pp.

Naves M. (2003) Caracterisation et gestion d’une population bovine locale de la zone

tropicale : le bovin creole de la Guadeloupe. PhD thesis Istitute national agronomique

Paris-Grignon.

Ott R.L. and Longnecker M. (2001) An introduction to statistical methods and data analysis, 5th edn., Duxbury, Pacific Grove, 1152pp.

Reverter A. and Kaiser C.J. (1997) The role of different pedigree structures on the sampling variance of heritability estimates J Anim Sci. 75, 2355-2361.

Robertson A. (1959) The Sampling Variance of the Genetic Correlation Coefficient Biometrics 15, 469-485

Swallow W.H. and Monahan J.F. (1984) Monte Carlo Comparison of ANOVA, MIVQUE, REML, and ML Estimators of Variance Components Technometrics 26, 47-57

Tamassia Pégolo N., Nunes de Oliveira H., Galvão Albuquerque L., Framartino Bezerra L. A., Barbosa Lôbo R. (2008) Normas de reação adaptativas para peso em diferentes

idades em bovinos Nelore 45a _{reuniao anual da sociedade brasileira de zootecnia,}

Lavras, Brazil, 22-25 july, 2008

Wang C.S., Yandell B.S. and Rutledge J.J. (1992) The dilemma of negative analysis of variance estimators of intraclass correlation Theor Appl Genet 85, 79-88.