Genetics of feed intake traits in the Czech Large White pig population

Feed represents a substantial proportion of the variable costs of pig production. Feed efficiency is traditionally expressed as the feed conversion ratio (FCR) and more recently as residual feed intake (RFI). Although feed efficiency can be generally improved indirectly by selection for increased growth rate and decreased adipose tissue, a higher genetic response could be achieved through direct selection of feed intake traits. The aim of this study was to provide a pilot analysis of feed intake data of 281 Czech Large White boars. Data were recorded individually using the Feed Intake Recording Equipment in field performance testing from 2018 to 2020. The analysed feed intake traits were average daily feed intake (ADFI), FCR and RFI. RFI was calculated as the deviation of observed ADFI and average population ADFI predicted on the basis of the model, with mid-test metabolic weight and average daily gain as regressors. The heritability estimates were 0.35 and 0.34 for ADFI and FCR, respectively, and the estimate was slightly higher (0.43) for RFI. The genetic standard deviations ranged from 100 to 110 g of feed per day and 103 g of feed per kg of weight gain. The amounts of explained variability by environmental effects of jointly tested animals were from 0.20 to 0.46. The sufficient amount of genetic variability and moderate heritability estimates give the possibility for selection of feed intake traits, although a larger number of animals will be essential to estimate more precise breeding values.


Introduction
Feed represents a substantial proportion of the variable costs of pig production (Do et al., 2013). To increase profitability and produce saleable pork, producers need to follow technologies to improve feed efficiency. One of the most efficient ways is to include feed efficiency traits into pig genetic improvement programs. The economic weights of such traits are 16% to 24% of the overall economic importance of growth, reproduction, health and carcass traits in maternal and sire pig breeds (Krupa et al., 2020). In this context, decreasing the pressure of feed costs and reduction of environmental impact are the most significant challenges in the pig sector (Gillert et al., 2017).
The choice of feed intake traits to predict feed intake for production requirements differs between populations and studies. In pigs, the average daily feed intake (ADFI), feed conversion ratio (FCR) or residual feed intake (RFI) are commonly used. In general, FCR representing the ratio of feed intake (inputs) and weight gain (output) is expressed as the inverse trait of the feed intake efficiency during growth. Selection for reduced feed intake is not optimal because of the strong relationship between feed intake and body weight, body composition, and daily body weight gain (Do et al., 2013). Moreover, selection pressure oriented towards growth rate and fat content has an indirect impact on FCR (Gilbert et al., 2017).
Residual feed intake, also called net feed efficiency, was first introduced by Koch et al. (1963) and is defined as the difference between observed and expected feed intake for given production and maintenance needs. RFI does not suffer from undesirable properties that are joined to ratio traits such as FCR (Do et al., 2013). Selection for FCR reduces sow appetite and fatness which, together with increased prolificacy, has been seen as a hindrance to sow lifetime performance .
Selection for reduced RFI could improve the efficiency of energy use without reducing the feed intake capacity that is required for production (Kennedy et al., 1993).
The potential of RFI for the improvement of feed efficiency in pigs is evaluated in studies based on commercial populations or experimentally selected lines. The establishment of experimental lines is a common strategy to evaluate the direct and correlated responses to a criterion for selection and to study the impact of the selection on animal physiology (Gilbert et al., 2017). Recent studies showed that pigs with lower RFI had better feed conversion efficiency and meat quality (Gilbert et al., 2007, Gilbert et al. 2017. Pigs divergently selected for RFI consistently demonstrate differences in carcass composition and in feed intake (Patience et al., 2015). Low RFI pigs have less carcass fat, consume less feed and exhibit similar or slightly slower rates of gain compared with high RFI pigs (Cai et al., 2008). The influence of RFI selection on carcass composition and catabolic activity in the liver and muscle was reported by Naou et al. (2015). Gilbert et al. (2012) stated that metabolic changes observed during growth in response to selection might explain part of the better efficiency of the low RFI sows, decreasing basal metabolism and favouring rapid allocation of resources to lactation.
In the Czech Republic, breeding pigs are tested in field performance tests without information about feed intake. Now, data from the pilot project started in 2018 are available. Therefore, the aims of this study were to analyse phenotypic data, investigate different feed intake traits and estimate their characteristics in the Czech Large White population.

Data
Feed Intake Recording Equipment (FIRE) electronic feeders were used to record the field performance test data of boars of the Czech Large White breed. Growth and feed intake data were collected on the farm from 2018 to 2020. Only one pig could take feed from each feeder at the same time. Each pen contained up to 12 boars fed by one feeder. Boars were fed ad libitum with a feed of the same composition for the whole test period. At the test end, all boars were measured with an ultrasound machine (Mindray LTD.) for lean meat content (LM) and backfat thickness (BF), both adjusted to a constant weight of 100 kg.
Individual feed intake and body weight were recorded when each pig visited the feeder. In total, there were 236,530 individual feeder visits of 294 boars, with age at the test start between 80 to 90 days and feeding for the next 50 to 70 days. These data were edited using the following rules: a) visits with feed intake less than 10 g per visit were omitted, b) outliers in body weight which substantially differed from the line estimated by the linear regression of the average daily body weight on age for each animal were marked as dubious and were omitted from analyses (4.14% of total visit records). The procedure for the regression coefficient estimation from only reliable data and the subsequent marking outliers was performed thirty times, c) only data up to the ultrasound measurement date were used, d) in the case of missing reliable body weight data on the current date, the weight was computed as the linear regression of weights on adjacent dates (3.35% of total daily weight records).
All boars had records regarding feeding at each day over the test period.
Next, growth and feed intake variables were computed for each pig: a) average daily gain (ADG) was calculated from FIRE data as the slope from the simple linear regression of the body weight at the start and end of the test (g/day), b) average daily feed intake (ADFI) was derived as the total feed intake (FI) over the test period divided by the number of days in the test (g/day), c) FCR was calculated as the ratio of FI and ADG (FI/ADG), d) the mean body weight (MBW) was derived as the average of the daily body weights from the FIRE data, e) metabolic body weight at mid-test (MWT) was calculated as the MBW raised to the power 0.75.
Only data from animals ranging between 1% and 99% quantiles for traits ADFI, FCR, and MWT were kept in the analysis. After editing, there were records of 281 boars fed in 35 pens (PEN) in the data set. The pedigree was traced back to the 1 st of January, 2000, and the file contained 1,422 animals in total. A basic statistical description of the data set is shown in Table 1.

Estimation of Residual Feed Intake (RFI)
RFI was computed as the difference between the observed ADFI and the predicted ADFI (ADFI p ). ADFI p was predicted on the basis of regression coefficients estimated by the General Linear Model implemented in statistical package SAS® (SAS Institute Inc., 2008). The model used for ADFI prediction (variables used are described above) was ADFI p = µ + β 1 * MWT + β 2 * ADG + e. Then, the residual feed intake values (RFI) were calculated as RFI = ADFI -(β 1 * MWT + β 2 * ADG).

Statistical analysis
Initially, the General Linear Model and Mixed Model procedures implemented in statistical package SAS® (SAS Institute Inc., 2008) were used to investigate the influence of various factors on analysed traits. Next, restricted maximum likelihood (REML) and optimisation using a quasi-Newton algorithm with analytical gradients (Neumaier and Groeneveld, 1998) as implemented in the VCE 6.0 program (Groeneveld et al., 2008) were used to estimate the variances in the univariate models as follows: ADFI ijk = YS i + P j + ADG ijk + MWT ijk + A ijk + e ijk , FCR ijk = YS i + P j + MWT ijk + A ijk + e ijk , RFI ijk = YS i + P j + A ijk + e ijk , where YS i is the fixed effect of year-season (defined on the basis of the quarter of the year of the feed test end), P j is the random effect of the given pen (animals fed by the feeder in whole test period together) with p ~ N(0,Iσ 2 p ), A ijk is a random animal genetic effect within YS i and P j with A l ~ N(0,Aσ 2 a ), where A is the relationship matrix, e ijk is the random residual with e ~ N(0,Iσ 2 e ), and ADG ijk and MWT ijk are linear regression covariates of average daily gain and mid-test metabolic weight.

Results and discussion
The analysis of factors affecting feed intake traits (ADFI and FCR phenotypes) are summarised in Table 2. Values in the table represent the decrease in R 2 when the given factor is removed from the model (R 2 of the full model -R 2 of reduced model). In our data set, the environmental conditions (PEN, LIT, YS), MWT, production and body composition (ADG, BF) explained 91.39% (ADFI) and 86.01% (FCR) of phenotypic variance. Saintilan et al. (2013), using similar models, published from 70.5% to 79.9% of explained ADFI phenotypic variance by models in four studied populations. Similarly, Rauw et al. (2006) published 61% to 81% of explained phenotypic variance by models with factors MWT, BF, age and body weight gain in different test periods defined on the basis of age in the Duroc population. The higher values in the current study could be explained by smaller numbers of animals.
The highest amount of phenotypic variance was explained by all environmental factors (27.91 and 44.40% for ADFI and FCR variance, respectively). The exclusion of environmental factors particularly from the full model showed almost zero influence of PEN and YS and a significant influence only of the common litter environment (P <0.001). The next most important factors were ADG and MWT with 3.72 -6.80% and 2.12 -3.14% of explained phenotypic variability, respectively. Both these factors were significant, with P < 0.001 for both analysed traits. BF explained less than 1% of the variability and was not significant for any trait. Saintilan et al. (2013) found the highest influence of environmental factors on ADFI (24 -28%) and ADG (29 -47%), followed by LM (5 -16%), keeping the BF and dressing percentage below 1%. The MWT in their work removed only up to 1.5% of phenotypic variability due to managing feeding performance testing based on the animals' weights. Similarly, Do et al. (2013) reported that models with or without MWT did not lead to significantly different results. Performance testing in the Czech Republic is arranged in herds, and the test start/end are managed on the basis of the animals' age. Therefore, there are not constant weights at the beginning and end of the test, and therefore the MWT for animals differs. In station trials, which are managed on a weight basis, MWT is relatively constant and could be omitted from the models.  There are several different ways to calculate RFI in pigs and other species.  summarised models for RFI prediction in beef cattle populations and recommended an equation with covariates ADG and MWT. This equation seems to be the best choice for the studied data set, due to the insignificant influence of BF on ADFI. The prediction of ADFI was based only on linear regression of covariates (SAS GLM procedure) and did not take into account the other environmental factors. It could be argued that a relatively low number of boars and tested pens in the study disrupt assumptions about the average influence of other factors and resulted in overestimation or underestimation of regressors. On the other hand, RFI values computed via regression coefficients estimated with the model shown in Table 2 resulted in highly correlated (92.0%) RFI values (not shown).
Heritability, additive genetic variances and the fractions of phenotypic variances explained by pen effects are shown in Table 3. For all studied feed intake traits (ADFI, FCR, RFI), heritabilities were moderate with the highest value estimated for RFI (0.43) and values of 0.35 for ADFI and 0.34 for FCR. Similar results were found in Yorkshire pigs and other populations. Heritabilities ranged from 0.30 to 0.45, 0.41 to 0.66 and 0.10 to 0.39 for FCR, ADFI and RFI, respectively (Kadarmideen et al., 2004, Cai et al., 2008, Do et al., 2013, Jiao et al., 2014, Gilbert et al., 2017. Although heritability estimates in our study are significant, levels of estimated standard errors were higher than in other studies. The values of the standard errors are about twice those in other previously cited studies (taking into account the level of heritability estimate). Although the pen variance in the phenotypic analysis seemed to be not significant, the inclusion of additive animal factor into the animal model changed the previously reported ratio of explained variance of LIT and PEN. In animal models, there were found estimations of common litter environment close to zero and not significant for any of the traits. On the other hand, the estimations of the environment of jointly tested animals (PEN) were high and significant (0.20, 0.21 and 0.46 for ADFI, RFI and FCR, respectively). Saintilan et al. (2013) reported the ratio of the variance of the common litter environment on the phenotypic variance as 0.0 to 0.13 in populations of three breeds, and some of these variance estimations were significant. PEN, for Czech data, described from 20% to 47% of phenotypic variability. These values are much higher than values reported by Do et al. (2013), from 2% to 5%, although phenotype variabilities of feed efficiency traits in that study are similar. This high ratio of phenotypic variability explained by the PEN factor is probably due to little experience with feeders in the herd (the data are from the pilot project) and the smaller amount of data in the current study.

Conclusions
In the current study, variance components for feed intake traits ADFI, FCR and RFI were estimated using univariate animal models. Investigated traits were moderately heritable. The heritability estimates were 0.35 and 0.34 for ADFI and FCR, respectively, and slightly higher for RFI (0.43). The genetic standard deviations from 100 to 110 g of feed per day and 103 g of feed per kg of weight gain were found. A high amount of phenotypic variation was explained by pen effect, probably due to an unstable herd environment, which could be resolved in the future with an increased number of animals tested and higher accommodation and standardisation of the feed Intake Recording Equipment during the performance testing.