DOI: https://doi.org/10.21203/rs.3.rs-1397133/v1
This study aimed to estimate genetic parameters and the prediction ability and bias of the genomic values for residual feed intake (RFI) using genotypic and phenotypic records from 8,354 Nellore animals. The RFI was calculated using a regression equation for each test (RFItest) and for the whole population (RFIpop). An analysis of variance (ANOVA) using the same dataset and number of records was performed to verify the adequacy of the regression equations applied to estimate the RFItest and RFIpop. The (co)variance components were obtained using the single-step genomic best linear unbiased prediction (ssGBLUP) under single and two-trait animal model analyses. The ssGBLUP method was used for the genomic prediction and the cross-validation method as a validation strategy. The prediction ability and bias were estimated to compare the RFItest and RFIpop genomic breeding values (GEBVs). The RFIpop ANOVA showed a higher significance level (p < 0.0001) than did the RFItest for the fixed effects. The RFIpop displayed higher additive genetic variance estimated than the RFItest, although the RFIpop and RFItest displayed similar heritabilities. Overall, the RFItest showed higher residual correlations with growth, reproductive and carcass traits, while the RFIpop displayed higher genetic correlations with such traits. The GEBV for the RFItest was slightly biased than that obtained for the RFIpop. The method of calculating the RFI influences the decomposition and estimation of variance components, and genomic prediction. The genetic correlations between RFIpop and RFItest were high, implying that changes in sires ranking and genetic response are expected using one or the other. The application of RFIpop would be more appropriate for genetic evaluation purpose to adjust or correct for non-genetic effects and to decrease the prediction bias.
The beef cattle breeding programs have mainly focused on herd genetic improvement to increase growth, sexual precocity, reproductive, and carcass traits since these traits are directly associated with the output and meat production. Although the selection to improve such traits would increase the production efficiency by reducing the production cycle and the proportion of unproductive animals in the herd, this strategy disregards one of the main economic factors in beef cattle systems, the feeding costs (Savietto et al. 2014; Brunes et al. 2020). It is worth highlighting that the beef cattle profitability is highly associated with the feeding cost, reaching up to 75% of production costs in beef cattle operations (Basarab et al. 2003; Ceacero et al. 2016). Thus, other traits related to feed intake and feed efficiency need to be evaluated in beef cattle breeding programs (Savietto et al. 2014; Brunes et al. 2020).
The residual feed intake (RFI) is the most common trait applied in beef cattle to evaluate feed efficiency. Such trait allows the identification of animals that display less feed intake than that expected, and it is also independent of the live weight (LW) and average daily gain (ADG). The RFI is calculated through a regression equation involving the metabolic LW (MW0.75) and ADG (Santana et al. 2014; Ceacero et al. 2016; Brunes et al. 2020). The use of RFI allows the reduction of feed intake in beef cattle systems without compromising growth and reproductive performance, minimizing the feeding costs, and improving beef cattle profitability (Seabury et al. 2017; Elolimy et al. 2018). The heritability and additive genetic variation estimates for RFI in Nellore cattle pointed out that selection for this traits is feasible (Brunes et al. 2020, 2021; Ceacero et al. 2016; Santana et al. 2014).
The RFI is obtained as the difference between the observed and predicted dry matter intake (DMI), proposed as a tool to identify animals with differences in maintenance requirements (Koch et al. 1963). However, there are different methods to calculate the RFI in terms of variables considered to adjust the RFI, i.e., backfat thickness, contemporary groups (CG), and others. The difference in the RFI calculation approach is a source of RFI variation affecting the RFI animals’ ranking (Koch et al. 1963; Basarab et al. 2003; Hoque et al. 2009; Lancaster et al. 2009; Del Claro 2011). The RFI tests were commonly carried out in specified locations, such as in experimental stations or research centers. However, with the advent of electronic equipment to measure the DMI, there was an increase in the number of animals evaluated (Boaitey et al. 2017) as well as in the number of tests performed in commercial herds, increasing the variation among the tests due to genetic, environmental, and management factors.
The RFI estimation is performed within the feed efficiency tests by considering only the animals of each test at a time, and this effect is included in the regression equation such as MW0.75 and ADG (BIF 2021). Subsequently, the feed efficiency test is included in the animal model as a fixed effect to estimate the genetic parameters and breeding values for the RFI (Gomes et al. 2013; Lin et al. 2013; Santana et al. 2014; Brunes et al. 2020). Thus, when the RFI is calculated within the test, the RFI records are pre-adjusted for the fixed effects before the estimated breeding value (EBV) prediction using the animal model, and the non-genetics and random animal effects are not estimated simultaneously. There is no consensus on the most appropriate method to calculate the RFI for genetic evaluations of feed efficiency. Some authors carried out a unique regression equation for each test and others as a general regression equation considering the whole population (Brunes et al. 2021; de Moraes et al. 2017; Del Claro 2011; Herd and Bishop 2000; Hoque and Oikawa 2004). So far, there is no recommendation on the literature or by any regulatory institution addressing this issue. The objective of this study is to estimate genetic parameters for the RFItest and RFIpop, and their association with growth, reproductive, and carcass-related traits. Further, this study aimed to evaluate the prediction ability for RFI estimated within each test (RFItest) and by considering the whole population (RFIpop) in Nellore beef cattle.
DNA samples were obtained from hair follicles, and the animals were genotyped using the CLARIFIDE® Nellore 3.1 low-density panel containing approximately 29,000 SNP markers. Quality control was applied for minor allele frequency (MAF), call rate and p-value for Hardy-Weinberg equilibrium test less than 0.02; 0.95 and 0.15, respectively. Samples with call rate lower than 0.95 were also excluded from the analysis. This process was performed with the PREGSF90 (Misztal et al. 2019), resulting in a dataset with 16,851 SNPs and 8,354 animals.
Records from 8,354 purebred Nellore cattle (Bos Taurus indicus) with genotypic and phenotypic records for RFI were used. Phenotypic data were obtained from a compilation of 125 commercial feed efficiency tests carried out between 2011 and 2019. The dataset was provided by the Nellore Brazil Breeding Program, coordinated by the National Association of Breeders and Researchers (ANCP, Ribeirão Preto, Brazil). The relationship matrix was built based on pedigree records from 58,374 animals provided by the ANCP. The animals composing the dataset had an average inbreeding of 1.72%, and the proportion of inbred individuals was 64.2% over the total population. These parameters were estimated using the INBUPGF90 program (Misztal et al. 2019).
In each of the 125 tests, the animals were grouped in pens (CG) based on the farm, sex, management group, and year and season of birth (maximum range of 90 days from the date of birth). Records outside the range of 3.5 standard deviation from the mean of the CG were removed. Animals of CG with fewer than four individuals were also removed. Descriptive statistics of the studied traits are provided in Table 1. The animal’s LW was recorded at the beginning, end, and every 14 days of the feed efficiency test. Each pen was equipped with an individual feed intake system (Integrado® Ltda., Contagem, Minas Gerais, Brazil; GrowSafe System® Ltda., Calgary, Canada). The experimental period lasted 70 days, with 14 days of adaptation to the feedlot diet and environment conditions. Daily feed intake data were excluded when the feed intake system showed any malfunctioning. The diet composition was modified over the tests, but it was equivalent in the content of total crude protein (13%) and total digestible nutrients (64%) on a DM basis. The experimental diet was fed in excess to achieve on average 5–10% of refusal daily. All animals had ad libitum access to feed and water throughout the experimental period.
Trait |
N |
Mean ± SD |
Median |
Min |
Max |
CV (%) |
---|---|---|---|---|---|---|
RFItest (kg/day) |
11,256 |
0.001 ± 0.64 |
0.01 |
-3.69 |
3.71 |
- |
RFIpop (kg/day) |
11,256 |
0.003 ± 1.15 |
-0.02 |
-4.78 |
3.95 |
- |
BW (kg) |
690,025 |
34.07 ± 4.61 |
34.00 |
21.00 |
50.00 |
13.53 |
W240 (kg) |
612,946 |
190.74 ± 31.24 |
190.00 |
101.00 |
300.00 |
16.38 |
W450 (kg) |
364,992 |
292.88 ± 52.16 |
287.00 |
201.00 |
476.00 |
17.81 |
MCW (kg) |
119,333 |
467.50 ± 70.66 |
460.00 |
251.00 |
715.00 |
15.11 |
SC365 (cm) |
28,102 |
21.59 ± 2.27 |
21.20 |
18.10 |
29.70 |
10.51 |
AFC (months) |
426,534 |
34.57 ± 4.63 |
35.00 |
21.00 |
44.00 |
13.40 |
ACP (kg/calf/year) |
179,470 |
147.60 ± 32.41 |
147.00 |
61.00 |
261.00 |
21.96 |
PP30 (%) |
123,262 |
1.27 ± 0.45 |
1.00 |
1.00 |
2.00 |
35.01 |
STAY (%) |
250,938 |
1.27 ± 0.44 |
1.00 |
1.00 |
2.00 |
34.94 |
REA (cm2) |
140,049 |
55.34 ± 11.52 |
54.39 |
20.39 |
96.13 |
20.81 |
BF (mm) |
180,924 |
3.08 ± 1.67 |
2.67 |
1.01 |
22.09 |
54.34 |
RF (mm) |
181,062 |
4.16 ± 2.04 |
3.62 |
1.01 |
11.56 |
49.05 |
MAR (%) |
38,801 |
2.41 ± 0.87 |
2.34 |
0.17 |
5.50 |
36.10 |
FRAME |
13,397 |
5.55 ± 2.00 |
5.62 |
-1.36 |
12.55 |
36.08 |
DMI (kg/day) |
10,308 |
8.11 ± 1.62 |
8.00 |
3.34 |
12.00 |
19.94 |
Abbreviations: Min, minimum; Max, maximum; CV, coefficient of variation; RFItest, residual feed intake estimated in each test; RFIpop, residual feed intake estimated considering the whole population; BW, birth weight; W240, weight at 240 days of age; W450, weight at 450 days of age; MCW, mature cow live weight; SC365, scrotal circumference at 365 days of age; AFC, age at first calving; ACP, accumulated productivity; PP30, early heifer pregnancy; STAY, stayability; REA, rib eye area; BF, backfat thickness; RF, rump fat thickness; MAR, marbling; FRAME, frame score; DMI, dry matter intake |
The ADG, MW0.75, and DMI were calculated for each animal within each feed efficiency test group, as previously described by Brunes et al. (2020). The ADG was estimated as the slope of the linear regression between LW and days of the RFI test. The MW was calculated as the LW midpoint and the product of the ADG multiplied by half of the days on the feed efficiency test. The residual feed intake (Koch et al. 1963) was calculated using a regression for each test (RFItest, n = 125) or considering the whole population (RFIpop). The RFItest and RFIpop were assumed to represent the residuals from a multiple regression model regressing DMI on ADG and MW0.75 in the following model:
where \({y}_{i}\) is individual DMI of the ith animal; \({\beta }_{o}\) is the intercept; \({\beta }_{1}\)and \({\beta }_{2}\) are the partial regression coefficient of \(ADG\) and \({MV}_{i}^{0.75}\), respectively; and \({\epsilon }_{ij}\)is the residual error (i.e., RFI).
An analysis of variance using the same dataset and number of records was performed to verify the adequacy of the regression equations applied to estimate the RFItest and RFIpop. This analysis was performed using the Proc mixed procedure of SAS (version 9.3) that contemplated the sire random effect, the fixed effects of CG and animal age at the beginning of test as covariable (linear effect). Also, the RFIpop and RFItest traits were tested for normality using Shapiro–Wilk statistics and for constant across years using a linear model with the interactions of MW0.75, ADG, and DMI. The RFItest and RFIpop displayed data with normal distribution.
The (co)variance components were obtained by single-step genomic best linear unbiased prediction (ssGBLUP) (Aguilar et al. 2010) under a single and two-trait animal model analyses using the BLUPF90 program (Misztal et al. 2019). The genomic matrix (\(G\)) was built as VanRaden (2008). The model included the direct additive genetic and residual effects, in addition to the CG as a fixed effect and the linear effect of the animal’ age as a covariate. The model used for the analysis was:
where \(y\) is the vector of the phenotypes; \(X\) is the incidence matrix associating \(\beta\) with \(y\); \(\beta\) is the vector of fixed effects including the CG and the animal’s age as covariate; \({Z}_{1}\) is the incidence matrix associating \(a\) with \(y\); \(a\) is the vector of random direct additive genetic effects; and \(e\)is the vector of random residual effects. It was assumed that \(\text{E}\left[\mathbf{y}\right]=Xβ; with the direct additive genetic and residual effects assumed to be normally distributed with mean zero and Var(a)=H ⨂ Sa and \(\text{V}\text{a}\text{r}\left(\mathbf{e}\right)=\mathbf{I} ⨂{\mathbf{S}}_{\text{e}}\), in which \({\mathbf{S}}_{\text{a}}\),and \({\mathbf{S}}_{e}\) are the additive genetic and residual covariance matrix, respectively. \(\mathbf{I}\)is an identity matrix of appropriate order. Descriptive analysis of the fixed effects solution obtained in the variance components estimation were carried out to demonstrate the amplitude of these results (Table 2).
Trait |
Mean ± SD |
Median |
Min |
Max |
CV (%) |
---|---|---|---|---|---|
RFItest (kg/day) |
0.35 ± 0.48 |
0.33 |
-2.17 |
1.82 |
136 |
RFIpop (kg/day) |
0.89 ± 1.14 |
0.85 |
-2.83 |
3.83 |
127 |
Abbreviations: Min, minimum; Max, maximum; CV, coefficient of variation |
A total of 1,000 individuals out of 4,000 with RFI phenotype and genotype information and without progenies records were randomly selected to perform the genomic prediction using the BLUPF90 (Misztal et al. 2019). The ssGBLUP method was used for genomic prediction, in which phenotypes, pedigrees, and genotypes are blended in a single analysis (Aguilar et al. 2010; Christensen and Lund 2010). The ssGBLUP uses the H matrix that combines the marker-based (G) and pedigree-based relationship matrix (A) in the classical animal model (Aguilar et al. 2010):
where \({A}_{22}\) is a subset of the additive relationship matrix for the genotyped animals.
The validation strategy was the cross-validation, considering the k-fold technique and repeated ten times. At each repetition, the phenotype information was removed from the 1,000 individuals to predict the genomic breeding value (GEBV), being subsequently correlated with the adjusted phenotype (Y*). The phenotype adjustment was performed using the PREDICTF90 software (BLUPF90 family) (Misztal et al. 2019), using the complete dataset containing all phenotypic and genotypic information available in both validation and training subsets. The validation analyses was implemented in the R program (2021). The phenotype prediction ability was obtained by dividing the correlation between Y* and GEBV by the square root of the heritability of the trait (Pryce et al. 2012). The regression between the adjusted-phenotype and the GEBVs was used to express the GEBV dispersion towards them (inflation and deflation in relation to the GEBV).
The analysis of variance (ANOVA) for the RFIpop showed statistical significance (p < 0.0001) for the fixed effects (CG and age) than did the RFItest ANOVA results (Table 3).
Traits |
P-value |
|
---|---|---|
Age |
Contemporary group |
|
RFItest |
0.0388 |
0.0074 |
RFIpop |
< .0001 |
< .0001 |
The RFIpop displayed higher additive genetic variance estimate than the RFItest, although the RFIpop and RFItest displayed similar heritability estimates (Table 4). Moderate heritability estimates were observed for almost all the evaluated traits, with values ranging from 0.20 to 0.36. Age at first calving and accumulated productivity displayed the lowest heritabilities, with values close to 0.12. Low heritability estimate were observed for BF (0.19).
The RFItest showed the highest residual correlations within the evaluated traits while the RFIpop displayed the highest genetic correlations (Table 5). Although the genetic and phenotypic correlation estimates were mostly low, W450 and DMI displayed moderate to high correlations. The genetic and phenotypic correlation estimates between the RFItest and RFIpop were high, with values of 0.72 and 0.71, respectively.
Traits |
RFItest |
RFIpop |
||
---|---|---|---|---|
rg |
rp |
rg |
rp |
|
RFItest |
- |
- |
0.72 ± 0.03 |
0.71 ± 0.02 |
BW |
0.21 ± 0.01 |
0.14 ± 0.03 |
0.27 ± 0.18 |
0.11 ± 0.05 |
W240 |
0.16 ± 0.02 |
0.04 ± 0.05 |
0.20 ± 0.02 |
0.08 ± 0.07 |
W450 |
0.31 ± 0.03 |
0.16 ± 0.06 |
0.32 ± 0.04 |
0.14 ± 0.08 |
MCW |
0.04 ± 0.06 |
0.10 ± 0.04 |
0.09 ± 0.03 |
0.07 ± 0.04 |
SC365 |
0.22 ± 0.01 |
-0.24 ± 0.01 |
0.26 ± 0.01 |
-0.21 ± 0.01 |
AFC |
0.24 ± 0.03 |
0.25 ± 0.11 |
0.28 ± 0.04 |
0.22 ± 0.02 |
ACP |
0.02 ± 0.03 |
0.12 ± 0.02 |
0.10 ± 0.00 |
0.07 ± 0.01 |
PP30 |
0.24 ± 0.05 |
0.41 ± 0.01 |
0.28 ± 0.03 |
0.28 ± 0.03 |
STAY |
0.12 ± 0.03 |
0.19 ± 0.04 |
0.17 ± 0.03 |
0.16 ± 0.04 |
REA |
0.19 ± 0.15 |
-0.22 ± 0.11 |
0.21 ± 0.17 |
-0.16 ± 0.11 |
BF |
0.11 ± 0.13 |
0.14 ± 0.13 |
0.17 ± 0.16 |
0.14 ± 0.14 |
RF |
0.21 ± 0.14 |
0.21 ± 0.14 |
0.28 ± 0.15 |
0.18 ± 0.15 |
MAR |
0.07 ± 0.13 |
0.22 ± 0.14 |
0.12 ± 0.10 |
0.18 ± 0.11 |
FRAME |
0.16 ± 0.03 |
-0.09 ± 0.10 |
0.17 ± 0.02 |
-0.07 ± 0.09 |
DMI |
0.65 ± 0.08 |
0.57 ± 0.02 |
0.71 ± 0.09 |
0.55 ± 0.02 |
Abbreviations: BW, birth weight; W240, weight at 240 days of age; W450, weight at 450 days of age; MCW, mature cow live weight; SC365, scrotal circumference at 365 days of age; AFC, age at first calving; ACP, accumulated productivity; PP30, early heifer pregnancy; STAY, stayability; REA, rib eye area; BF, backfat thickness; RF, rump fat thickness; MAR, marbling; FRAME, frame score; DMI, dry matter intake. |
The prediction ability for the RFIpop and RFItest were similar, with values close to 0.30. Further, the GEBVp for the RFItest was more inflated (0.57) than the GEBVp obtained for RFIpop (Table 6).
Model |
Prediction ability |
Bias |
||||||
---|---|---|---|---|---|---|---|---|
Mean |
SD |
Min |
Max |
Mean |
SD |
Min |
Max |
|
RFItest |
0.29 |
0.04 |
0.23 |
0.35 |
0.57 |
0.07 |
0.47 |
0.68 |
RFIpop |
0.30 |
0.04 |
0.25 |
0.38 |
0.61 |
0.07 |
0.49 |
0.74 |
Abbreviations: SD, standard deviation; Min, Minimum; Max, Maximum. |
The significance (p < 0.0001) of the fixed effects for the RFIpop implies that the environmental effects and animal age were better adjusted for the RFIpop than did for the RFItest. Thus, it is expected that non-genetic differences affecting RFI were better corrected or less biased for RFIpop. It should be highlighted that different regression equations to estimate the RFI was applied for the each individual test (RFItest) whereas a unique regression equation was employed for the whole population (RFIpop). Additionally, the random sire variance effect was higher for the RFIpop (0.017 ± 0.02) than for the RFItest (0.014 ± 0.01), pointing out that differences in the RFI due to the sire effect were more noticeable for such population.
The largest additive genetic variance estimated for the RFIpop implies a greater chance of identifying genetically superior animals, being expected higher response to selection for such population. Hoque and Oikawa (2004) comparing the RFI obtained for each feed efficiency testing stage with that obtained using genetic and multiple regression (ordinary RFI) in Wagyu cattle, suggested that the genetic and ordinary RFI would contribute more for genetic selection than the RFI obtained for each testing stage, as it displayed superior additive variance estimates.
When the RFI was calculated within the population (RFIpop), the prediction of the genetic value is conditioned to the solution of the CG effect (fixed effects) and random effects. On the other hand, this condition is not accomplished for the RFItest, influencing the partition of the variance components, and consequently, a fraction of the genetic variance is absorbed by the pre-adjusted fixed effects. This pattern can be supported by the calculation leading to an increase in the additive variance of the RFIpop. The heritability estimates obtained for RFIpop and RFItest (Table 4) were within the interval described in the literature for beef cattle, with values ranging from 0.13 to 0.28 (Grion et al. 2014; Oliveira et al. 2014; Olivieri et al. 2016; Silva et al. 2016).
Moderate heritability estimates were observed for DMI, which is in accordance with Oliveira et al. (2014) (0.29) and de Moraes et al. (2017) (0.25 to 0.36) in Nellore cattle. Heritabilities for growth, reproductive, longevity, and carcass traits reported in this study were consistent with the literature, ranging from low to moderate (Yokoo et al. 2010; Zuin et al. 2012; Santana et al. 2014; Bonin et al. 2015; Tonussi et al. 2015; Grossi et al. 2016; Lopes et al. 2016; Pires et al. 2017; Kluskaa et al. 2018; Gordo et al. 2018; Bonamy et al. 2019; Sainz et al. 2020). As expected, the lowest heritability estimates were obtained for sexual precocity and maternal heritability related traits, such as AFC and ACP. The heritability estimates for BW, W240, W45, MCW, PP30, STAY, REA, BF, MAR, RF, SC365, and FRAME pointed out that selection for these traits is feasible when phenotypic records are available.
Higher genetic correlations between the RFIpop and some traits (BW, STAY, BF, RF, MAR, and DMI) were obtained when compared with those for the RFItest. Differences in (co)variance decomposition between the RFIpop and RFItest might explain these results. Hypothetically, biased variance components were obtained for the RFItest since the RFI was pre-adjusted for the feed efficiency test. Therefore, it is possible to infer when the RFItest estimation is performed within the feed efficiency tests, and the differences between the CG may not be properly corrected or adjusted. Hence, calculated RFI within the test is adjusted for the CG without considering the random effects. In contrast, for the RFIpop calculate considering all feed efficiency tests, the CG and other fixed effect were estimated simultaneously with random animal effect. It is important to emphasize that in the mixed and animal model, the solution of the fixed effects relies on the random effects and vice versa (Perri and Iemma 1999).
Despite the low genetic correlation estimates between RFItest and RFIpop with the other evaluated traits (BW, W240, MCW, SC365, AFC, ACP, PP30, STAY, REA, BR, RF, MAR, and FRAME), the estimates obtained with the DMI were moderate. Such a pattern may be due to the fact that the RFI is estimated as a function of the DMI. The correlation estimates obtained were in agreement with those presented in the literature (0.33 to 0.95) (Grion et al. 2014; Santana et al. 2014; Ceacero et al. 2016; de Moraes et al. 2017; Polizel et al. 2018). Low correlation estimates between RFI and growth, reproduction, longevity, and carcass traits were also reported in Nellore cattle (Grion et al. 2014; Santana et al. 2014; Ceacero et al. 2016; de Moraes et al. 2017; Ferreira Júnior et al. 2018; Bonamy et al. 2019; Moraes et al. 2019; Brunes et al. 2021).
High genetic (0.72) and phenotypic (0.71) correlation estimates were obtained between RFItest with RFIpop. It has been widely accepted that just traits with a correlation over than 0.80 can be assumed to be genetically the same trait, what doesn't apply here (Robertson 1959). Despite the high genetic and phenotypic association, these values were less than unity, implying that there are some differences between these traits (Table 5). According to Herd and Bishop (2000), differences in the RFI estimation can be attributed to the information used in the estimation. These results pointed out that differences in the genetic progress are expected using RFItest or RFIpop as selection criteria to improve RFI, and changes in sires ranking are also expected using one or another.
The highest inflation of the GEBVp for the RFItest might be attributed to the differences between the CG, which are not properly corrected or adjusted when the RFI calculation is performed within the feed efficiency tests. Differences due to the management, environment, and other non-genetic components would be better corrected for the RFIpop since the adjustment for the fixed effects was performed with the mixed model. As a result, the fixed effect solutions and genetic predictions were less biased in the RFIpop calculation. The implication is that genetic selection based on RFItest might not lead to the identification of the best feed efficient animals since overestimated GEBV’s are expected for them.
Despite the differences in (co)variance decomposition between RFIpop and RFItest, the prediction ability for them were similar. Thus, there is non-advantage or differences in terms of prediction ability between RFIpop and RFItest. The prediction ability observed in this study for the RFIpop and RFItest were similar to those reported in the literature for Nellore cattle (Silva et al. 2016; Brunes et al. 2020). However, the genomic selection main goal is the phenotype approximate prediction, and in this case, the use of RFIpop is the strategy that would guarantee lower inflation. To best our knowledge, no study has been conducted evaluating the GEBVs obtained for the RFI considering each and all feed efficiency tests to corroborate these results.
From a nutritional assessment point of view, it is not recommended to combine tests or exclude the feed efficiency test effect in the RFI regression estimation. The RFI calculation within the feed efficiency test groups may be more suitable for datasets in which the differences between the metabolizable energy concentrations and diet composition or environmental effects are higher (Arthur et al. 2001; Lancaster et al. 2009). For genetic evaluation purpose, to obtain the RFI calculated within the test, initially a pre-adjustment is performed for the feed efficiency test effect, and it influences the variance component estimation and the genetic value solutions. Thus, considering the standardization of environmental and management conditions, the use of the RFIpop is supported, mainly for genetic evaluation purpose.
In conclusion, the RFI estimation influences the decomposition of the variance components and, consequently, the estimation of variance components and genomic prediction. The genetic correlation between RFIpop and RFItest was moderate, implying that both traits have different genetic background and differences in sires ranking and genetic response are expected between both traits. The use of a unique regression equation with all evaluated animals to estimate the RFI (RFIpop) would be more appropriate to adjust or correct non-genetic effects and to decrease the genomic prediction inflation.
Acknowledgements
The authors thanks to the National Association of Breeders and Researchers (ANCP) for providing dataset and genealogical data.
Author Contribution
All authors contributed to the study conception and design. Material preparation and data collection were performed by Fernando Baldi, Carina Ubirajara de Faria and Raysildo Barbosa Lobo. The analysis was performed by Fernando Baldi and Ludmilla Costa Brunes. The first draft of the manuscript was written by Fernando Baldi and Ludmilla Costa Brunes, and all authors commented on previous versions of the manuscript. All authors read and approved the manuscript before submission.
Funding
The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.
Declarations
Ethics approval
The collection of phenotypic information is not categorized as an experiment as the interventions are related to farming practices according to the law N° 11.794 (8 October 2008; subsection VII of § 1o of clause 225 of Brazilian Federal Constitution), which lays down procedures for the scientific use of animals. Hence, this study was not submitted to an ethics committee, considering that a data set from a commercial production system was used.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing Interests
The authors declare no competing interests.