Genomic selection for meat quality traits in Nelore cattle

Accepted Manuscript Genomic selection for meat quality traits in Nelore cattle

Ana Fabrícia Braga Magalhães, Flavio Schramm Schenkel, Diogo Anastácio Garcia, Daniel Gustavo Mansan Gordo, Rafael Lara Tonussi, Rafael Espigolan, Rafael Medeiros de Oliveira Silva, Camila Urbano Braz, Gerardo Alves Fernandes Júnior, Fernando Baldi, Roberto Carvalheiro, Arione Augusti Boligon, Henrique Nunes de Oliveira, Luis Arthur Loyola Chardulo, Lucia Galvão de Albuquerque PII: DOI: Reference:

S0309-1740(17)31003-3 doi:10.1016/j.meatsci.2018.09.010 MESC 7687

To appear in:

Meat Science

Received date: Revised date: Accepted date:

14 July 2017 11 September 2018 17 September 2018

Please cite this article as: Ana Fabrícia Braga Magalhães, Flavio Schramm Schenkel, Diogo Anastácio Garcia, Daniel Gustavo Mansan Gordo, Rafael Lara Tonussi, Rafael Espigolan, Rafael Medeiros de Oliveira Silva, Camila Urbano Braz, Gerardo Alves Fernandes Júnior, Fernando Baldi, Roberto Carvalheiro, Arione Augusti Boligon, Henrique Nunes de Oliveira, Luis Arthur Loyola Chardulo, Lucia Galvão de Albuquerque , Genomic selection for meat quality traits in Nelore cattle. Mesc (2018), doi:10.1016/ j.meatsci.2018.09.010

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Genomic selection for meat quality traits in Nelore cattle

Ana Fabrícia Braga Magalhãesa, Flavio Schramm Schenkelb, Diogo Anastácio Garciac, Daniel Gustavo Mansan Gordoa, Rafael Lara Tonussia, Rafael Espigolana, Rafael Medeiros de Oliveira Silvaa, Camila Urbano Braza, Gerardo Alves Fernandes Júniora, Fernando Baldia, Roberto

T

Carvalheiroa, Arione Augusti Boligond, Henrique Nunes de Oliveiraa Luis Arthur Loyola

Jaboticabal – SP, Brazil. c

Guelph – Canada.

b

CR

São Paulo State University (Unesp), School of Agricultural and Veterinarian Sciences, Centre for Genetic Improvement of Livestock, University of Guelph –

US

a

IP

Charduloe, Lucia Galvão de Albuquerquea

BRF Company, Curitiba – Paraná - Brazil. e

Federal University of Pelotas

São Paulo State University (Unesp), College of

AN

(UFPel), Pelotas – Rio Grande do Sul – Brazil.

d

M

Veterinary and Animal Science, Botucatu – SP, Brazil.

prediction

using

different

methods

for

meat quality traits in Nelore cattle.

PT

genomic

ED

Abstract. The objective of this study was to present heritability estimates and accuracy of

Approximately 5,000 animals with phenotypes and genotypes of 412,000 SNPs, were divided

CE

into two groups: (1) training population: animals born from 2008 to 2013 and (2) validation

AC

population: animals born in 2014. A single-trait animal model was used to estimate heritability and to adjust the phenotype. The methods of GBLUP, Improved Bayesian Lasso and BayesCπ were performed to estimate the SNP effects. Accuracy of genomic prediction was calculated using Pearson’s correlations between direct genomic values and adjusted phenotypes, divided by the square root of heritability of each trait (0.03 - 0.19). The accuracies varied from 0.23 to 0.73, with the lowest accuracies estimated for traits associated with fat content and the greatest

ACCEPTED MANUSCRIPT 2

accuracies observed for traits of meat color and tenderness. There were small differences in genomic prediction accuracy between methods. Keywords: fat deposition; genomics; meat composition; meat tenderness.

T

1. Introduction

IP

In Brazil, more than 80% of the bovine herd is of the Nelore breed and its crosses with other

CR

zebu breeds produced in pasture system with seasonal variations and short periods of feedlot. It is known that there is a negative relationship between meat tenderness and percentage of zebu

US

cattle, (Dransfield, 1994) hence we can expect some difficulties in obtaining products with

AN

superior quality in Brazil (Chardulo, Silveira, & Vianello, 2013). Even the processors observe wide variation in animal carcass standards with different size, muscularity, and backfat end

M

point. Therefore, mechanisms of classification and typification of carcasses are not always a

ED

good tool to predict meat quality traits, since hit rate not exceeding 80% of all carcasses

PT

evaluated (Chardulo, Silveira, & Vianello, 2013). Meat quality traits are directly related to the purchase decision of consumers, who are

CE

becoming more demanding and are looking for high-quality products (Scollan et al., 2006).

AC

Attributes most requested by consumers include tenderness (Williams, 2008), which is a determinant trait for meat quality and meat color which is a visually important and indicative of the freshness of meat at the time of purchase (Mancini & Hunt, 2005), and fat content, an important predictor of palatability, a determinant of carcass value (Ferraz & Felício, 2010) and associated with the sensory traits of meat such as juiciness and flavor (Scollan et al., 2006). Meat quality traits are difficult to improve by the traditional selection method since their measurement is expensive to measure and requires slaughter of the animals. Selection of sires for

ACCEPTED MANUSCRIPT 3

these measures requires the animals to be submitted to a progeny test, which is also expensive, and time consuming, increasing the generation interval and reducing annual genetic progress. In view of the difficulties in selecting these traits by the traditional method, genomic selection could be an alternative for the improvement of meat quality traits. Genomic selection

T

proposed by Meuwissen, Hayes, & Goddard (2001) is based on the use of information for a set

IP

of a molecular genetic marker, named SNP (single nucleotide polymorphism), distributed across

CR

the genome, assuming that quantitative trait loci (QTLs) are in linkage disequilibrium with these markers. Genomic selection can be implemented using multi-step, which requires firstly an

US

evaluation based on pedigree to create pseudo-phenotype and after a genomic evaluation is

AN

performed. The success of genomic selection depends on the accuracy of the direct genomic values (DGVs) of animals.

M

The aim of this paper is to estimate the heritability using phenotypic data for the following

ED

meat quality traits: tenderness, marbling, lipid percentage and meat color. As well as present the accuracy of genomic prediction using three methods (GBLUP, Improved Bayesian Lasso and

CE

PT

BayesCπ) in Nelore cattle.

2. Material and methods

AC

2.1 Description of the data

The data were obtained on male animals of the Nelore breed, belonging to 13 farms distributed across different regions of Brazil. These farms participating in four breeding programs (DeltaGen, Paint, Cia do Melhoramento and Qualitas). The animals were feedlot finished for approximately 90 days and slaughtered in commercial slaughterhouse at a mean age of 701 ± 88 days and hot weight carcass of 280.5 ± 28.53 kg, under the approval of ethics

ACCEPTED MANUSCRIPT 4

committee of the São Paulo State University (UNESP), School of Agricultural and Veterinarian Sciences, Jaboticabal – SP, Brazil (Nº 18.340/16). Carcasses were chilled for 48 h and longissimus thoracis muscle samples were removed between the 12th and 13th rib of the left half-carcasses. Laboratory analysis of the following meat

T

traits was performed: tenderness, lipid percentage, marbling, and meat color. Meat tenderness

IP

was analyzed in 2.54-cm thick samples of the longissimus thoracis muscle with bone using the

CR

procedure standardized and proposed by Wheeler, Koomaraie, & Shackelford (1995), grilled to an internal temperature of 71ºC. Shear force was measured with a Warner-Bratzler Shear Force

US

machine (GR-Manufacturing, Kansas, USA). Shearing was performed on core samples 1.27 cm

AN

in diameter removed longitudinally to the muscle fibers. The results were reported as Newton (N) and eight measurements were performed per sample to increase accuracy of the results.

M

Total lipids were quantified using the method described by Bligh & Dyer (1959), which

ED

determines the lipid percentage in the samples. The marbling score of the sample was determined by visual grading (USDA, 1989), which ranges from 1 to 10, where 1 = practically devoid; 2 =

PT

traces; 3 = slight; 4 = small; 5 = modest; 6 = moderate; 7 = slightly abundant; 8 = moderately

CE

abundant; 9 = abundant, and 10 = very abundant. Meat color (L*, lightness; a*, redness; b*, yellowness) was measured using CIELab system

AC

of the Chroma Meter CR-400 (light source A, observer angle 10°, aperture size 5 cm, Konica Minolta Sensing, Inc., Tokyo, Japan) as described by Borges et al. (2014). The colors were obtained after a 30 min bloom time under low temperature (~5°C).

2.2 Estimation of heritability

ACCEPTED MANUSCRIPT 5

Pedigree file contained a total of 2,688,124 animals, from 9,811 sires and 915,371 dams, distributed in 12 generations. Phenotypic data, structured in half sib, with 5,084 measures of tenderness, 3,871 of lipid percentage and 5,089 of marbling and color (L*, a* and b*) were used in this study. The lipid

T

percentage did not follow a normal distribution and was therefore submitted to square root

IP

arcsine transformation.

CR

Animals were arranged in 167, 158 and 173 contemporary group (CG) for tenderness, lipid percentage and marbling, a*, b* and L* colors, respectively, based on year of birth (from 2008 to

US

2014), farm (n=13) and management group at yearling (n=69). Measures that were 3.5 standard

AN

deviations below or above the mean of the CG were excluded from the data analysis and CG containing fewer than three animals were also excluded.

M

A single-trait animal model was used to estimate variance components for the meat quality

ED

traits using an animal model under Bayesian inference, including the fixed effects of contemporary group, age at slaughter and time between slaughter and physicochemical analysis

𝑌 = 𝑋𝛽 + 𝑍𝑢 + 𝑒,

CE

PT

as covariable, and random animal and residual effects. The model can be represented as follows:

In which Y is the vector of phenotype, β is the vector of fixed effects, and u is the vector

AC

of direct additive genetic effect. X and Z are incidence matrices for effects contained in β and u, respectively, e is random residual effects vector. The analysis consisted of a single chain of 1,000,000 cycles with burn-in de 150,000, taking sample every 10 iterations. So, 85,000 samples were used to obtain the parameters. These analyses were performed using GIBBS2F90 and POSTGIBBSF90 (Misztal et al., 2002). Chain convergence was assessed by visual examination.

ACCEPTED MANUSCRIPT 6

The phenotypes were adjusted (Yc) for the same fixed effects described in the previous model, using PREDICTF90 (Misztal et al., 2002): 𝒀𝒄 = 𝒀 − 𝑿𝜷

T

2.3 Analysis of genomic data

IP

A number of 1,695 animals were genotyped using the Illumina Bovine HD Beadchip

CR

(Illumina, San Diego, CA, USA) that contains 777,962 SNPs. And 3,195 animals were genotyped using the GeneSeek® (Genomic Profiler Indicus HD - GGP75Ki NEOGEN) which

US

contains 74,677 SNPs. Animals genotyped using lower density panel were imputed to the HD

AN

panel using FImpute v.22 (Sargolzaei, Chesnais, & Schenkel, 2014) with imputation accuracy expected to be higher than 0.98 (Carvalheiro et al., 2014). This approach uses both family and

M

population information to perform imputation.

ED

The quality control was performed by trait, considering the following criteria for exclusion of SNPs: non-autosomal SNPs, SNPs at the same position, a minor allele frequency ≤ 0.05, a p-

PT

value for Hardy-Weinberg equilibrium ≤ 10-6 , a GC score ≤ 0.70, and call rate ≤ 0.98. During

CE

quality control of the samples, SNPs with a call rate ≤ 0.90 were excluded. The numbers of

Table 1.

AC

genotyped animals and SNPs, by trait, which remained after quality control were summarized in

2.4 Predictions of genomic breeding values The animals with known phenotypes and genotypes were used to divide the population into two groups: training population (1), which included the animals born from 2008 to 2013 (90% of the total population) and where SNP effects were estimated; validation population (2), with

ACCEPTED MANUSCRIPT 7

animals born in 2014 (10% of the total population), where DGVs were predicted for each animal based on the SNP effects estimated in the training population (Table 1). This dataset was structured in half-sib family, with average of the maximum genomic relationship between an individual in the validation and individuals in the training of about 0.25.

T

The model used for the prediction genomic in matrix notation was:

IP

𝑌𝑐 = 1𝜇 + 𝑋𝑔 + 𝑒,

CR

where Yc is the vector of adjusted phenotypes for fixed effects, 1 is the vector of 1s; μ is the

US

overall mean; X is the matrix (n x p) that consists of p SNPs for n animals; g is the random vector of SNP effects, and e is the random residual vector, which was assumed normally distributed

AN

with mean zero and variance of Iσ2 e, where I is an identity matrix and σ2 e is the residual variance. The genotypes were coded as 0, 1, 2 and 5 for genotypes AA, Aa (or aA), aa and missing

M

genotypes, respectively.

ED

The GS3 software developed by Legarra, Ricard, & Filangi (2016) was used for the analyses. A chain of 300,000 iterations, with a burn-in period of 50,000 iterations and taking sample every

PT

10 iterations was used for Gibbs sampler algorithm. A complete description of methods used in

CE

this study for SNP effects follows below: GBLUP: This method assumes a normal distribution with the same priori variance for all

AC

markers. Var(g) = Gσ2 g, where σ2 g is the variance due to the markers and G is the genomic relationship matrix created according to VanRaden (2008):

𝑚

𝐺 = [(𝑀 − 𝑃)(𝑀 − 𝑃)′ ]/[2 ∑

𝑗=1

𝑃𝑗 (1 − 𝑃𝑗 )]

ACCEPTED MANUSCRIPT 8

where M is a matrix of marker alleles with n lines (n = total number of genotyped animals), m is total number of markers) and P is a matrix containing 2 times the observed frequency of the second allele 𝑃𝑗 Elements of M are 0 or 2 for both homozygous and for the heterozygous. Improved Bayesian LASSO (iBLASSO): This method was proposed by Tibshirani

T

(1996) and modified by Legarra, Robert Granie, Croiseau, & Fritz (2011). The prior for SNP

IP

effects (g) can be written as: 𝜆

CR

𝑔 |𝜆 ~ ∏𝑗 2 𝑒𝑥𝑝(−𝜆|𝑔𝑗| ) 𝑎𝑛𝑑 𝑒| 𝜎𝑒2 ~ 𝑀𝑉𝑁 (0, 𝐼𝜎𝑒2)

US

The prior distribution of the marker effect (g= {gj}) is exponential. λ is the "sharpness" parameter which determines the shape of distribution of the SNP effect. The prior for λ was

AN

vague, being uniform between 0 and 1,000,000. e is a random vector of residual effect and the

M

prior distribution for the residual variance (σe²) was an inverted chi-squared distribution. BayesCπ: For this method a mixed distribution for effect marker was assumed according to

ED

Habier, Fernando, Kizilkaya, & Garrick (2011), in which an additional variable is included (δ i),

PT

informing whether the marker exerts an effect (1) or not (0). The priori distribution of δ is binomial, with parameters n and π, where n is SNPs number and π is the fraction not included in

CE

the model. The parameter π was fixed to 0.99 (Legarra, A., Ricard, A., & Filangi, 2016), i.e. a

AC

prior beta distribution with parameters α = 10d8 and β= 10d10. In this method, a scaled inverse chi-squared priori distribution for the variance of SNP effects (σ²g) and for the residual variance (σ²e).

The DGVs were predicted from the estimated SNP effects using the following formula:

𝑛

𝐷𝐺𝑉 = ∑ 𝑋𝑖 𝑔𝑖 𝑖

ACCEPTED MANUSCRIPT 9

where n is the number of SNPs, Xi is the incidence matrix of SNPs for all individuals, and gi is the vector of SNP effects for each marker.

2.5 Comparison of methods

T

The following criteria were used to compare the methods considering animals of the

IP

validation population:

CR

1) Accuracy of prediction methods was calculated using Pearson’s correlation between

US

DGV and Yc. The correlation was divided by the square root of heritability (Acc) as described by Legarra, Robert-Granié, Manfredi, & Elsen (2008).

AN

2) Bias in the prediction of DGVs was estimated by the regression coefficient of Yc on predicted DGVs (b). This criterion measures the degree of inflation or deflation of the

M

genomic prediction in relation to the Yc. A regression coefficient > 1 indicates that the

ED

model is underestimating DGVs and a value < 1 indicates overestimation of DGVs.

PT

Estimates of regression close to 1 indicate that the predictor is in the same scale as the

CE

Yc.

3. Results and discussion

AC

3.1 Estimation of heritability Heritability estimates for tenderness, lipids, marbling, and meat color (a*, b* and L*), using whole population are summarized in Table 2. These estimates were moderate for traits associated with fat (marbling and lipids) and L* color, ranging from 0.13 to 0.19, while the traits of tenderness and parameters of a* and b* colors presented lower heritability estimates (0.03 to 0.09) than other traits.

ACCEPTED MANUSCRIPT 10

Heritablity estimates of tenderness trait related in the present study was lower than estimates found by Akanno et al. (2014) and Magnabosco et al. (2016). However, our estimate are similar to Zwambag et al. (2013) and De Castro et al. (2014). This difference was expected, since it is know that heritability is a populational parameter, being difficult to make comparations between

T

studies, for example Johnston et al. (2003) estimated heritabiliy for tenderness in temperate and

IP

tropically adapted beef breeds and found estimates of 0.11 in temperate breeds and 0.42 in

CR

tropical adapted breeds, proving that estimates of the heritability for tenderness vary a lot and probably, the reason of this can be related with nature of tenderness, which is sensitive to

US

environmental factors such as, difficulty in maintaining environmental control in pre and post-

AN

slaughter, sample preparation, cooking methods and number of samples analyzed (Burrow, 2001).

M

The estimate of the heritability of marbling trait was moderate (0.19) and consistent with

ED

previous findings (Akanno et al, 2014; Bolormaa et al., 2013), while the other trait associated to fat, lipid percentage, was also moderately heritable (0.13). However, few genetic studies were

PT

published using lipid percentage as fat indicator, probably because, being a chemical analysis

CE

method, it is more difficult to perform. As these traits are genetically highly correlated (Hocquette, Renand, & Levéziel, 2006), studies with genetic parameter estimates may use

AC

marbling score as indicative of intramuscular fat. The estimates of heritability for a* and b* colors were low, but L* color was moderately heritable. The results of estimates for meat color in the literature reinforce the same findings here, i.e. L* color is more heritable than a* and b* colors (Johnston et al., 2003; King et al., 2010; Wolcott et al., 2009). Probably, a* and b* colors were more affect by environmental variations

ACCEPTED MANUSCRIPT 11

than L* color. These variations may be related with antemortem and postmortem management, such as age of animal, nutritional regime, stress, packaging methods, time and temperature at storage, chilling rate, exposure to oxygen, microbial load, postmortem age and others factor (Faustman & Cassens, 1990; Murray, 1989).

T

The estimates of heritability presented in this study were based on phenotypic recordings and

IP

exhibited estimates from low to moderate between traits, indicating that meat quality traits can

CR

be strongly affected by environmental factor. This can lead to the unsuccessful traditional selection, since these traits were not highly heritable, in addition they are only available after

US

slaughtering of the animal. However, this should not be a discouraging factor for selection of

AN

animals for meat quality traits, since advances in molecular genetics has allowed that information of many thousands of SNP markers distributed by entire genome of the animal (Meuwissen,

M

Hayes, & Goddard, 2016) be used to select the animals, commonly denominated to as genomic

ED

selection (Meuwissen, Hayes, & Goddard, 2001).

PT

3.2 Accuracy of genomic selection for meat quality traits

CE

The genomic prediction accuracy for traits of tenderness, lipids, marbling and meat color (a*, b* and L*) using GBLUP, iBLASSO and BayesCπ as prediction methods are summarized in

AC

Table 3. The accuracies varied from 0.23 to 0.73, between traits, with lowest accuracy estimated for traits associated with fat (percentage of lipids and marbling) and greater accuracy observed for meat color and tenderness. Several factors can be causing the difference in the accuracies of genomic selection between traits studied here, since this accuracy are dependent on the effective population size, number of animals in the training population, marker density, statistical method, heritability and genetic architecture of the trait ( Daetwyler, Pong-Wong, Villanueva, &

ACCEPTED MANUSCRIPT 12

Woolliams, 2010; De Los Campos, Hickey, Pong-Wong, Daetwyler, & Calus, 2013; Goddard, 2009). The accuracies estimated in this study for tenderness were close to those reported by Akanno et al. (2014); Magnabosco et al. (2016) and Miller, Lu, Vander Voort & Mandell, (2014).

T

Probably, one reason for this is higher quantity of animals used in this study, when compared

IP

with previous studies. In addition, these authors used different approaches, such as Miller, Lu,

CR

Vander Voort, & Mandell (2014), which evaluated different validation populations and estimated the correlation between DGV and Yc for tenderness (from 0.10 to 0.50), using GBLUP method,

US

in Canadian multi-breed population, and related that the estimate of accuracy was higher when

AN

genomic relationship between validation and training populations increased. Comparing the results of both studies, our maximum genomic relationship was of 0.25 and accuracy was of 0.57

M

using GBLUP method, while Miller, Lu, Vander Voort, & Mandell (2014) for the same genomic

ED

relationship found accuracy of 0.15. When the accuracy was 0.50, the average of the highest genomic relationship between an individual in the validation and individuals in the training

PT

population was 0.36. The accuracy of genomic selection observed in this study for tenderness is

CE

a good example of how the genomic selection can contribute to gain genetic by providing a prediction with moderate accuracy to select a young animal.

AC

The prediction accuracy achieved in this study for traits indicative of intramuscular fat in meat, marbling and lipid percentage, were different, with accuracy found for marbling higher than lipid percentage (Table 3). Comparing with other traits of this study, marbling and lipid percentage obtained low accuracy. One of the main reasons cited in the literature for low accuracy is the low heritability of trait studied (Daetwyler, Pong-Wong, Villanueva, & Woolliams, 2010; De Los Campos, Hickey, Pong-Wong, Daetwyler, & Calus, 2013), but in this

ACCEPTED MANUSCRIPT 13

study there was no relation of accuracy with heritability of trait, for instance, the lipid percentage obtained an estimate of moderate heritability (Table 2) and greater than most traits in the present study. The accuracy for marbling found here was close to the findings by Bolormaa et al. (2013) when evaluated genomic prediction of several traits in multibreed population. By contrast, our

T

accuracy was lower than found by Boddhireddy et al. (2014) in Angus cattle. Probably, this

IP

difference in results can be attributing to smaller size of reference population in this study when

CR

compared with Boddhireddy et al. (2014), which used about 8,000 animals.

The accuracies observed for meat color were moderate, ranging from 0.40 to 0.73. Genomic

US

selection studies with meat color in bovines are restricted, however we found some papers

AN

published with other species. For instance, Brito et al. (2017) performed genomic selection in about 10,000 sheep for meat color measure with 24, 48, 96 and 168 h postmortem and found

M

accuracy, using GBLUP method, lower than results of genomic accuracy of this study (0.08 –

ED

0.35, 0.24 – 0.32 and 0.31 – 0.33 for a* color, b* color and L* color, respectively). While Baby et al. (2014) using the same CIE L*, a* and b* system in swine related DGV accuracies only for

PT

the training samples, which makes it difficult the comparison with present study, since for the

CE

training samples, the DGV prediction is based on both genotypes and phenotypes, consequently the accuracy is higher than for the testing samples. However, the same authors used another

AC

measurement based on color score and found accuracy ranged from 0.30 to 0.38 in testing samples, considering different methods of genomic prediction. The most papers found in the literature for genomic selection of meat quality traits in beef cattle are limited to study tenderness and fat content, while results of genomic selection for trait of meat color are scarce in the literature.

Probably, this is because tenderness and marbling are priorities for the beef industry.

ACCEPTED MANUSCRIPT 14

However, our results approached the genetic importance of meat color, evidencing estimates of heritability from low to moderate and genomic prediction accuracy moderate for these traits. The moderate genomic prediction accuracies for meat quality traits reported in the study suggest the feasibility of implementation of genomic selection, i.e. these traits should be

IP

T

incorporate in genetic evaluation scheme of genetic improvement program in Nelore cattle.

CR

3.3 Predictive ability of the methods

Three different prediction methods were used in this study, GBLUP, iBLASSO and BayesCπ

US

(Table 3). The prior distributions for the effects of SNPs is the major difference between

AN

methods, i.e., GBLUP assumes a normal distribution, iBLASSO assumes an exponential distribution for SNP effect and BayesCπ assumes a mixed distribution, where only a portion of

M

the SNPs have an effect. We observed that there was no difference between methods for traits of

ED

lipids, marbling and a* color, while there was small variation in the prediction accuracy for traits of tenderness (0.03), b* color (0.04) and L* color (0.05), i.e. prediction accuracy was greater for

PT

Bayesian methods than GBLUP method, but there was no difference between the Bayesian

CE

methods.

Early publications with evaluation of genomic prediction methods were mainly based on

AC

simulated data and found difference between genomic prediction methods, confirming that the accuracy is greater for Bayesian model (Daetwyler, Pong-Wong, Villanueva, & Woolliams, 2010; Clark, Hickey, & Van Der Werf, 2011), but in recent years genomic selection using real data have showed that there are no difference between methods of genomic selection (Colombani et al., 2013; Fernandes Júnior et al., 2016).

However, this discordance between studies was

expected, since the accuracy of methods does not depend only on theoretical properties and type

ACCEPTED MANUSCRIPT 15

of data used,

but also on the different situations in which selection was applied (Zhang, Zhang,

& Ding, 2011), i.e., differences in breeds, populations, number of animals in the training population and, mainly, in the traits studied. Other factor that can affect the accuracy of genomic prediction methods is the genetic architecture of trait of interest, i.e. the accuracy is higher for

T

the methods that best fitted the genetic architecture of a trait (Lund, Sahana, Koning, Su, &

IP

Carlborg, 2009). The genetic architecture of a trait can be mostly inferred based on heritability

CR

and number of QTLs affecting the trait (Zhang et al., 2011). However, in this study the heritability estimates did not affect the accuracy of genomic prediction. The number of QTLs

US

affecting the trait is associated with predictive ability of methods, since the accuracy of GBLUP

AN

is independent of the number of QTLs (NQT L) that affect the trait, while the accuracy of Bayesian methods will be greater than of GBLUP when N QT L is smaller than the number of chromosome

M

segments (Daetwyler et al., 2010). This approach about genetic architecture was observed by

ED

Daetwyler, Pong-Wong, Villanueva, & Woolliams (2010) using simulated data, but in studies with real data the genetic architecture is less well known and grouping traits based on their

PT

architecture is not straightforward (De Los Campos, Hickey, Pong-Wong, Daetwyler, & Calus,

CE

2013).

In beef cattle, it is difficult to find examples of traits where major genes explain a high

AC

proportion of genetic variation. In this study, traits of tenderness, b* and L* colors showed an accuracy higher for iBLASSO and BayesCπ methods than GBLUP method, but in this case, genome wide association studies with meat color to prove the theory that the better performance of the Bayesian methods can be associated with major effect gene are limited. However, genome wide association studies for tenderness in Nelore cattle found the Calpastatin gene explained

ACCEPTED MANUSCRIPT 16

0.46% of additive genetic variance in window of 150 SNPs (Magalhães et al., 2016). Probably, this is one reason for the better performance of Bayesian methods for tenderness in this study. The estimates of the regression coefficients (b) showed that the predictor can be deflated for traits of tenderness, lipid percentage and marbling, i.e. the values of DGVs can be

T

underestimated, while b for other traits can be inflated and overestimating the DGVs (Table 3).

IP

When the estimates of the regression coefficients were deflated, the Bayesian methods obtained

CR

b values more distant of 1 than GBLUP method, which agree with results from Neves et al. (2014) and Silva et al. (2016). In opposite, when regression slopes inflated, the Bayesian

US

methods were close to 1.

AN

Inflation and deflation are more serious when genotyped and non-genotyped animals are compared. When the prediction is inflated, the selection of genotyped animals will be favored in

M

the same way, as a deflated prediction will benefit non-genotyped animals. In the present study,

ED

all candidates for selection were genotyped and animals with and without genotypes were, therefore, not compared. There are currently no animals with phenotypes for these traits in

PT

Brazil; hence, in practice, only genotyped animals will be used. When the predictions are

CE

deflated, the differences between progeny of selected animals will be greater than those predicted by DGVs and the opposite will occur when the prediction is inflated.

AC

In addition, regarding computing time, Bayesian methods were less efficient than GBLUP method, since GBLUP requires only the inversion of the genomic relationship matrix, which takes less 1 hour, while Bayesian methods took about 10 days to run our data set.

4. Conclusions The estimates of heritability for meat quality traits were from low to moderate.

ACCEPTED MANUSCRIPT 17

There was small difference of prediction accuracy between methods for traits of tenderness, b* and L* colors, while for other traits there was no difference. The meat quality traits should be incorporate in genetic evaluation scheme of genetic improvement program, since the accuracies reported in this study support the feasibility of

IP

T

implementation of genomic selection in Nelore cattle.

CR

Acknowledgements

US

The authors would like to thank Fapesp (Fundação de Amparo à Pesquisa do Estado de São Paulo) for the first author’s scholarship (Nº 2012/21969-7) and for funding this study (Nº

ED

M

AN

2009/16118-5).

5. References

PT

Akanno, E. C. , Plastow, G., Woodward, B. W., Bauck, S., Okut, H., Wu. X.-L., ...Basarab, J. A.

CE

(2014). Reliability of molecular breeding values for Warner-Bratzler shear force and carcass traits of beef cattle - an independent validation study. Journal of Animal Science, 92(7),

AC

2896–2904.

Baby, S., Hyeong, K. E., Lee, Y. M., Jung, J. H., Oh, D. Y., Nam, K. C., … Kim, J. J. (2014). Evaluation of genome based estimated breeding values for meat quality in a Berkshire population using high density single nucleotide polymorphism chips. Asian-Australasian Journal of Animal Sciences, 27(11), 1540–1547. https://doi.org/10.5713/ajas.2014.14371 Bligh, E.G., & Dyer, W. J. (1959). A rapid method of total lipid extraction and purification.

ACCEPTED MANUSCRIPT 18

Canadian Journal of Biochemistry and Physiology, 37(8), 911–917. Boddhireddy, P., Kelly, M. J., Northcutt, S., Prayaga, K. C., Rumph, J., & DeNise, S. (2014). Genomic predictions in Angus cattle: Comparisons of sample size, response variables, and clustering methods for cross-validation. Journal of Animal Science, 92(2), 485–497.

IP

T

https://doi.org/10.2527/jas.2013-6757

CR

Bolormaa, S., Pryce, J. E., Kemper, K., Savin, K., Hayes, B. J., Barendse, W., … Bunch, R. J. (2013). Accuracy of prediction of genomic breeding values for residual feed intake and

US

carcass and meat quality traits in, and composite beef cattle. Journal of Animal Science,

AN

91(7), 3088–3104. https://doi.org/10.2527/jas.2012-5827

Borges, B. O., Curi, R. A., Baldi, F., Feitosa, F. L. B., De Andrade, W. B. F., De Albuquerque,

M

L. G., … Chardulo, L. A. L. (2014). Polymorphisms in candidate genes and their

ED

association with carcass traits and meat quality in Nellore cattle. Pesquisa Agropecuaria

PT

Brasileira, 49(5), 364–371. https://doi.org/10.1590/S0100-204X2014000500006 Brito, L. F., Clarke, S. M., McEwan, J. C., Miller, S. P., Pickering, N. K., Bain, W. E., …

CE

Schenkel, F. S. (2017). Prediction of genomic breeding values for growth, carcass and meat

AC

quality traits in a multi-breed sheep population using a HD SNP chip. BMC Genetics, 18(7). https://doi.org/10.1186/s12863-017-0476-8 Burrow, H. (2001). Breed and crossbreeding effects on marbling. Marbling Symposium, 1–14. Carvalheiro, R., Boison, S. A., Neves, H. H. R., Sargolzaei, M., Schenkel, F. S., Utsunomiya, Y. T., … Garcia, J. F. (2014). Accuracy of genotype imputation in Nelore cattle. Genetics Selection Evolution, 46(69), 1–11. https://doi.org/10.1186/s12711-014-0069-1

ACCEPTED MANUSCRIPT 19

Chardulo, L. A. L., Silveira, A. C., & Vianello, F. (2013). Analytical Aspects for Tropical Meat Quality. In Springer-Verlag Wien (Ed.), Food Quality, Safety and Technology (1st ed., pp. 53–62). Viena - Austria. Clark, S. A., Hickey, J. M., & Van Der Werf, J. H. J. (2011). Different models of genetic

IP

T

variation and their effect on genomic evaluation. Genetics Selection Evolution, 43(18), 9.

CR

https://doi.org/10.1186/1297-9686-43-18

Colombani, C., Legarra, a, Fritz, S., Guillaume, F., Croiseau, P., Ducrocq, V., & Robert-Granié,

US

C. (2013). Application of Bayesian least absolute shrinkage and selection operator (LASSO) and BayesCπ methods for genomic selection in French Holstein and Montbéliarde

AN

breeds. Journal of Dairy Science, 96(1), 575–91. https://doi.org/10.3168/jds.2011-5225

M

Daetwyler, H. D., Pong-Wong, R., Villanueva, B., & Woolliams, J. A. (2010). The impact of

ED

genetic architecture on genome-wide evaluation methods. Genetics, 185(3), 1021–1031.

PT

https://doi.org/10.1534/genetics.110.116855 De Castro, L. M., Magnabosco, C. U., Sainz, R. D., De Faria, C. U., & Lopes, F. B. (2014).

CE

Quantitative genetic analysis for meat tenderness trait in polled nellore cattle. Revista

AC

Ciencia Agronomica, 45(2), 393–402. https://doi.org/10.1590/S1806-66902014000200022 De los Campos, G., Hickey, J. M., Pong-Wong, R., Daetwyler, H. D., & Calus, M. P. L. (2013). Whole-genome regression and prediction methods applied to plant and animal breeding. Genetics, 193(2), 327–345. https://doi.org/10.1534/genetics.112.143313 Dransfield, E. (1994). Optimisation of tenderisation, ageing and tenderness. Meat Science, 36(1– 2), 105–121. https://doi.org/10.1016/0309-1740(94)90037-X

ACCEPTED MANUSCRIPT 20

Faustman, C., & Cassens, R.G. (1990). The biochemical basis for discolration in fresh meat: A Review. Journal of Muscle Foods, 1(3), 217–243. https://doi.org/https://doi.org/10.1111/j.1745-4573.1990.tb00366.x Fernandes Júnior, G. A., Rosa, G. J. M., Valente, B. D., Carvalheiro, R., Baldi, F., Garcia, D. A.,

IP

T

… De Albuquerque, L. G. (2016). Genomic prediction of breeding values for carcass traits in Nellore cattle. Genetics Selection Evolution, 48(1). https://doi.org/10.1186/s12711-016-

CR

0188-y

US

Ferraz, J. B. S., & Felício, P. E. de. (2010). Production systems - An example from Brazil. Meat

AN

Science, 84(2), 238–243. https://doi.org/10.1016/j.meatsci.2009.06.006 Habier, D., Fernando, R. L., Kizilkaya, K., & Garrick, D. J. (2011). Extension of the bayesian

M

alphabet for genomic selection. BMC Bioinformatics, 12(186), 12.

ED

https://doi.org/10.1186/1471-2105-12-186

PT

Hocquette, J., Renand, G., & Levéziel, H. (2006). The potential benefits of genetics and genomics to improve beef quality-a review. Animal Science Papers …, 24(3), 173–189.

CE

Retrieved from http://www.ighz.edu.pl/files/objects/2142/64/strona173-190.pdf

AC

Johnston, D. J., Reverter, A., Ferguson, D. M., Thompson, J. M., & Burrow, H. M. (2003). Genetic and phenotypic characterisation of animal, carcass, and meat quality traits from temperate and tropically adapted beef breeds. 3. Meat quality traits. Australian Journal of Agricultural Research, 54(2), 135–147. https://doi.org/10.1071/AR02086 King, D. A., Shackelford, S. D., Kuehn, L. A., Kemp, C. M., Rodriguez, A. B., Thallman, R. M., & Wheeler, T. L. (2010). Contribution of genetic influences to animal-to-animal variation in myoglobin content and beef lean color stability. Journal of Animal Science, 88(3), 1160–

ACCEPTED MANUSCRIPT 21

1167. https://doi.org/10.2527/jas.2009-2544 Legarra, A., Ricard, A., & Filangi, O. (2016). GS3. France. Retrieved from http://genoweb.toulouse.inra.fr/~alegarra/

T

Legarra, A., Robert Granie, C., Croiseau, P., Guillaume, F., & Fritz, S. (2011). Improved Lasso

IP

for genomic selection. Genetics Research Cambridge, 93(1), 77–87.

CR

https://doi.org/10.1017/S0016672310000534

Legarra, A., Robert-Granié, C., Manfredi, E., & Elsen, J. M. (2008). Performance of genomic

US

selection in mice. Genetics, 180(1), 611–618. https://doi.org/10.1534/genetics.108.088575

AN

Lund, M.S., Sahana, G., Koning, D., Su, G., & Carlborg, O. (2009). Comparison of analyses of the QTLMAS XIV common dataset. I: genomic selection. BMC Proceedings, 5(Suppl 3).

M

https://doi.org/10.1186/1753-6561-5-S3-S1

ED

Magalhães, A. F. B., de Camargo, G. M. F., Fernandes, G. A., Gordo, D. G. M., Tonussi, R. L.,

PT

Costa, R. B., … de Albuquerque, L. G. (2016). Genome-Wide Association Study of Meat Quality Traits in Nellore Cattle. Plos One, 11(6), e0157845.

CE

https://doi.org/10.1371/journal.pone.0157845

AC

Magnabosco, C. U., Lopes, F. B., Fragoso, R. R., Eifert, E. C., Valente, B. D., Rosa, G. J. M., & Sainz, R. D. (2016). Accuracy of genomic breeding values for meat tenderness in Polled ellore cattle. Journal of Animal Science, 94. https://doi.org/10.2527/jas2016-0279 Mancini, R. A., & Hunt, M. C. (2005). Current research in meat color. Meat Science, 71(1), 100– 121. https://doi.org/10.1016/j.meatsci.2005.03.003 Meuwissen, T. H. E. , Hayes, B. J., & Goddard, M. E. (2001). Prediction of total genetic value

ACCEPTED MANUSCRIPT 22

using genome-wide dense markers maps. Genetics, 157, 1819–1829. https://doi.org/11290733 Meuwissen, T., Hayes, B., & Goddard, M. (2016). Genomic selection: A paradigm shift in

T

animal breeding. Animal Frontiers, 6(1), 6. https://doi.org/10.2527/af.2016-0002

IP

Miller, S., Lu, D., Vander Voort, G., & Mandell, I. (2014). Genomic Prediction of Beef

CR

Tenderness in Canadian Beef Cattle. In 10th World Congress of Genetics Applied to Livestock Production Genomic. Vancouver: Proceedings, 10th World Congress of Genetics

AN

http://www.gsejournal.org/content/48/1/7

US

Applied to Livestock Production Genomic. Retrieved from

Misztal, I., Tsuruta, S., Strabel, T., Auvray, B., Druet, T., & Lee, D. H. (2002). BLUPF90 and

M

related programs (BGF90). In Proceedings of the 7th World Congress on Genetics Applied

ED

to Livestock Production (Vol. 28, pp. 21–22). https://doi.org/9782738010520

PT

Murray, A. C. (1989). Factors Affecting Beef Color At Time of Grading. Canadian Journal of Animal Science, 69(2), 347–355. https://doi.org/10.4141/cjas89-039

CE

Sargolzaei, M., Chesnais, J. P., & Schenkel, F. S. (2014). A new approach for efficient genotype

AC

imputation using information from relatives. BMC Genomics, 15(478). https://doi.org/10.1186/1471-2164-15-478 Scollan, N., Hocquette, J. F., Nuernberg, K., Dannenberger, D., Richardson, I., & Moloney, A. (2006). Innovations in beef production systems that enhance the nutritional and health value of beef lipids and their relationship with meat quality. Meat Science, 74(1), 17–33. https://doi.org/10.1016/j.meatsci.2006.05.002

ACCEPTED MANUSCRIPT 23

Silva, R. M. O., Fragomeni, B. O., Lourenco, D. A. L., Magalhães, A. F. B., Irano, N., Carvalheiro, R., … Albuquerque, L. G. (2016). Accuracies of genomic prediction of feed efficiency traits using different prediction and validation methods in an experimental Nelore cattle population. Journal of Animal Science, 94(9), 3613–3623.

IP

T

https://doi.org/10.2527/jas2016-0401 Tibshirani, R. (1996). Regression Shrinkage and Selection via the Lasso. Journal of Thw Royal

CR

Statistical Society, 58(1), 267–288.

US

USDA. (1989). Official USDA marbling photographs. Washington, DC. U.S.A: National

AN

Cattlemen’s Beef Association. U.S. Depth. of Agriculture.

VanRaden, P. M. (2008). Efficient Methods to Compute Genomic Predictions. Journal of Dairy

M

Science, 91(11), 4414–4423. https://doi.org/10.3168/jds.2007-0980

ED

Wheeler, T. L., Koomaraie, M., Shackelford, S. D. (1995). Standardized Warner-Bratzler shear

PT

force procedures for meat tenderness measurement. (Clay Cente). Williams, J. L. (2008). Genetic Control of Meat Quality Traits. In F. Toldrá (Ed.), Meat

CE

Biotechnology (pp. 21–61). New York: Springer Science & Business Media.

AC

https://doi.org/10.1007/978-0-387-79382-5_2 Wolcott, M. L., Johnston, D. J., Barwick, S. A., Iker, C. L., Thompson, J. M., & Burrow, H. M. (2009). Genetics of meat quality and carcass traits and the impact of tenderstretching in two tropical beef genotypes. Animal Production Science, 49(6), 383–398. https://doi.org/10.1071/EA08275 Zhang, Z., Zhang, Q., & Ding, X. D. (2011). Advances in genomic selection in domestic

ACCEPTED MANUSCRIPT 24

animals. Chinese Science Bulletin, 56(25), 2655–2663. https://doi.org/10.1007/s11434-0114632-7 Zwambag, A., Kelly, M., Schenkel, F., Mandell, I., Wilton, J., & Miller, S. (2013). Heritability of beef tenderness at different aging times and across breed comparisons. Canadian Journal

CR

IP

T

of Animal Science, 93(3), 307–312. https://doi.org/10.4141/cjas2012-100

Lipids

AnimalsA

4,743

3,576

TrainB

4,206

3,165

TestC

537

411

SNPsD

412,719

a* color

b* color

L* color

4,721

4,733

4,731

4,751

4,186

4,196

4,194

4,211

535

537

537

540

412,700

412,715

412,685

412,735

M

413,213

: number of animals with genotypes and phenotypes; B: number of animals in train population;

PT

A

Marbling

AN

Tenderness

ED

US

Table 1. Number of animals and SNPs used for genomic selection in meat quality traits

C

AC

CE

: number of animals in test population;

D

: number of SNPs after of quality control.

ACCEPTED MANUSCRIPT

IP

T

25

CR

Table 2. Descriptive statistics and heritability estimates (h2 ) of meat quality traits in Nelore

Tenderness (N)

5,062

Lipids

3,812

Marbling

5,039

a* color

5,052

ED 5,046

PT

b* color L* color

5,071

Mean±SDB

h2 ±SDC

62.37±18.83

0.09±0.01

0.09±0.02

0.13±0.04

2.82±0.47

0.19±0.04

13.73±4.33

0.03±0.01

9.35±3.44

0.07±0.01

31.32±6.24

0.17±0.10

CE

: number of observations; B: Mean and standard deviation; C: heritability and standard deviation.

AC

A

AN

N

M

Traits

US

cattle

ACCEPTED MANUSCRIPT

T

26

IP

Table 3. Genomic prediction accuracy and bias of different methods for meat quality traits in

CR

Nelore cattle ACCGBLUP A ACCiBLASSOB

ACCBayesCπ C

bGBLUP D

biBLASSE

bBayesCF

Tenderness

0.57

0.60

0.60

1.48

1.60

1.54

Lipids

0.23

0.23

0.23

1.78

1.75

1.75

Marbling

0.32

0.32

0.32

1.33

2.18

2.08

a* color

0.40

0.40

0.40

0.53

0.61

1.75

b* color

0.49

0.53

0.53

0.59

0.85

0.96

L* color

0.68

0.73

0.63

1.16

1.23

AN M

0.73

:Genomic prediction accuracy for GBLUP method; B:Genomic prediction accuracy for iBLASSO method; C:Genomic prediction accuracy for BayesCπ method; D:Genomic prediction bias for GBLUP method; E:Genomic prediction bias for iBLASSO method; F:Genomic prediction bias for BayesCπ method.

AC

CE

PT

ED

A

US

Traits