Genetic background in partitioning of metabolizable energy efficiency in dairy cows

J. Dairy Sci. 101:1–11 https://doi.org/10.3168/jds.2017-13936 © American Dairy Science Association®, 2018.

Genetic background in partitioning of metabolizable energy efficiency in dairy cows T. Mehtiö,1 E. Negussie, P. Mäntysaari, E. A. Mäntysaari, and M. H. Lidauer Natural Resources Institute Finland (Luke), 31600 Jokioinen, Finland

ABSTRACT

indicating that some genetic variation may exist in the efficiency of using metabolizable energy for different pathways in dairy cows. Key words: feed efficiency, partitioning metabolizable energy, energy sink, heritability

The main objective of this study was to assess the genetic differences in metabolizable energy efficiency and efficiency in partitioning metabolizable energy in different pathways: maintenance, milk production, and growth in primiparous dairy cows. Repeatability models for residual energy intake (REI) and metabolizable energy intake (MEI) were compared and the genetic and permanent environmental variations in MEI were partitioned into its energy sinks using random regression models. We proposed 2 new feed efficiency traits: metabolizable energy efficiency (MEE), which is formed by modeling MEI fitting regressions on energy sinks [metabolic body weight (BW0.75), energy-corrected milk, body weight gain, and body weight loss] directly; and partial MEE (pMEE), where the model for MEE is extended with regressions on energy sinks nested within additive genetic and permanent environmental effects. The data used were collected from Luke’s experimental farms Rehtijärvi and Minkiö between 1998 and 2014. There were altogether 12,350 weekly MEI records on 495 primiparous Nordic Red dairy cows from wk 2 to 40 of lactation. Heritability estimates for REI and MEE were moderate, 0.33 and 0.26, respectively. The estimate of the residual variance was smaller for MEE than for REI, indicating that analyzing weekly MEI observations simultaneously with energy sinks is preferable. Model validation based on Akaike’s information criterion showed that pMEE models fitted the data even better and also resulted in smaller residual variance estimates. However, models that included random regression on BW0.75 converged slowly. The resulting genetic standard deviation estimate from the pMEE coefficient for milk production was 0.75 MJ of MEI/kg of energy-corrected milk. The derived partial heritabilities for energy efficiency in maintenance, milk production, and growth were 0.02, 0.06, and 0.04, respectively,

INTRODUCTION

Milk production, maintenance, and growth are the most important factors for energy use of lactating primiparous dairy cows. Maintenance requirements consist of the energy necessary to conduct voluntary body activity and to maintain the basal metabolism and body temperature. Traditionally, calorimetric chambers have been used to estimate partial efficiencies in converting energy intake to the energy sinks of the different physiological pathways. These partial efficiencies are defined as a ratio between the ME utilization for product and the energy requirement for production. So far no strong evidence has been found to assume genetic differences in the partial efficiencies, but it has been shown that high genetic merit cows are more efficient because they partition the available energy differently from low genetic merit cows (Veerkamp and Emmans, 1995; Agnew and Yan, 2000; Yan et al., 2006). Mäntysaari et al. (2012) also observed differences in the mobilization of body energy reserves between cows with different energy efficiency when efficiency was measured as energy conversion efficiency [ECM/metabolizable energy intake (MEI)] but not when measured as residual energy intake (REI). The most studied traits related to feed efficiency in dairy cows at the moment are DMI, REI, and residual feed intake (RFI), and different kinds of ratio traits, which are usually defined as the ratio of output over input or its inverse. Heritability estimates for DMI range from 0.27 to 0.63 in different studies (Spurlock et al., 2012; Berry et al., 2014; Liinamo et al., 2015). Residual traits such as REI and RFI can be calculated as the residual from a linear regression of energy or feed intake on various energy sinks, such as milk production, metabolic BW (for maintenance requirements), and BW change. Alternatively, these residual traits

Received October 4, 2017. Accepted January 8, 2018. 1 Corresponding author: [email protected]

1

2

MEHTIÖ ET AL.

can be calculated as the difference between actual MEI or DMI and MEI or DMI predicted from animal performance. In recent studies, the heritability estimates for RFI have varied from 0.01 to 0.32 (Pryce et al., 2014; Tempelman et al., 2015) and for REI from 0.07 to 0.09 (Liinamo et al., 2015; Hurley et al., 2018). However, some concerns in using RFI and REI have been addressed as well. The observations of these residual traits accumulate measurement errors that are associated with the component traits. Additionally, unknown genetic correlations may exist between residual traits and the regressors that are used to predict those (Pryce et al., 2014; Manzanilla-Pech et al., 2016). High phenotypic correlation between REI and energy balance was reported by Liinamo et al. (2015) and Hurley et al. (2018) and they argued that selecting for REI might also lead to greater negative energy balance. However, Hurley et al. (2018) concluded that REI could be used as a breeding objective for feed efficiency, but at the same time traits such as health and fertility need to be considered in the breeding program. Residual energy intake is describing the general efficiency of a cow in using ME. However, from a breeding point of view, it might be desirable to put more selection weight on efficiency with respect to a certain metabolic function. Therefore, models that are capable of partitioning a cow’s efficiency with respect to different pathways may be of interest. Such models would give more comprehensive information as to why some cows are more efficient than others. To establish the required efficiency trait analogous to REI, a model that directly includes regressions on energy sinks is fitted for MEI observations. This model is hereafter referred as metabolizable energy efficiency (MEE). In the usual derivation of REI, the MEI is first corrected with respect to the energy requirements for assumed needs, and the resulting REI is further analyzed by genetic models. Instead the effect of animal breeding values could be directly added into the model of MEI, and thereafter the breeding values should become estimated more accurately. Such a model can be extended with random regressions on energy sinks nested within the additive genetic effect, which would provide partial efficiencies for use of ME (pMEE), given genetic variation exists in the efficiency of using ME for different pathways. The objectives of this study were to assess the genetic variations in MEE and in its energy sink-specific components (pMEE) by fitting random regression models on weekly MEI measurements from Nordic Red dairy cattle, and further to compare the results with those from analyses of REI, which was used as a reference trait. Journal of Dairy Science Vol. 101 No. 5, 2018

MATERIALS AND METHODS Research Data

The data used in this study were collected from Luke’s experimental farms Rehtijärvi (tiestall) and Minkiö (loose housing) in Jokioinen between 1998 and 2014. The early data were from several consecutive and continuous feeding trials carried out between 1998 and 2008. Since 2009, the data collection was continued in Minkiö barn with automated feed intake, BW, and milk production data collection, with the main purpose of studying the animal variation in the components of feed efficiency. All cows in the data were fed grass silage and home blend concentrate mix. The proportion of concentrates in the diet depended on the experimental plan (1998–2008), stage of lactation, and digestibility of the grass silage (2009–2014). On average the proportion of concentrate in the diet of the cows in the data was 48.3%. To calculate energy and nutrient intake of the cows’ weekly representative subsamples of feeds were collected and combined to 4- to 8-wk samples for analyses based on the study. The silage samples were analyzed for pepsin-cellulase solubility and the solubility values were converted to digestible organic matter content in DM (D-values; Huhtanen et al., 2006). The ME content for grass silage was calculated as 0.016 × D-value (MAFF, 1975, 1984). The ME concentration of the concentrate was calculated from digestible nutrients (MAFF, 1975, 1984). The digestibility coefficients for the components of the concentrates were taken from the Finnish feed tables (Luke, 2015). The daily MEI was corrected by the total DMI and concentration of ME and protein in the diet according to the correction equation given by Luke (2015). More detailed explanation of data collection and feeding of the cows as well as feed sampling and analyses during years 1998 to 2008 are described in Mäntysaari et al. (2003, 2004, 2005, 2012) and during years 2009 to 2014 in Mäntysaari and Mäntysaari (2015). Data were from 495 primiparous Nordic Red dairy cows, of which 291 were from different feeding trials and 204 had been measured since changing to routine measuring of feed intake since 2009. The data from wk 1 were excluded due to the big variability in studied traits, which complicated genetic analyses. The analyzed data included 12,350 weekly observations from wk 2 to 40 of lactation. Feed intake was not recorded during the pasture period, which resulted in gaps in the feed intake for animals that were in lactation during summer months. Cows were mainly calving during fall and therefore most gaps were at the end of recording period (lactation wk 31 to 40). Milk yield was recorded

PARTITIONING OF METABOLIZABLE ENERGY EFFICIENCY

twice a week in the trials before 2009, and after 2009 daily. Before 2006 the BW of the cows were based on heart girth measurements. Heart girth measurements were conducted at intervals described by Mäntysaari et al. (2003, 2004, 2005). The heart girth was converted to BW using equations for Finnish Red dairy cows (Mäntysaari and Mäntysaari, 2008). During 2006 to 2009, the weighing of cows was done weekly during lactation wk 2 to 8 and monthly thereafter. From 2009 onward, the cows were weighted twice daily via automated scale when leaving the milking parlor after milking. To calculate BW and BW change for each animal, the weights measured were modeled by random regression fitting a second-order polynomial function plus a Wilmink term (Wilmink, 1987). The BW of cows was then predicted from the individual curves (Mäntysaari and Mäntysaari, 2015). Body weight gain values in lactation wk 2 and BW loss values after lactation wk 15 were rare and were thus set to zero. These were assumed to be unreliable predictions due to sparsely represented gain and loss values at these stages of lactation. The final data included MEI (MJ/d), ECM production (kg/d; calculated according to Sjaunja et al., 1990), metabolic body weight (BW0.75, kg), body weight loss (BWL, kg/d), body weight gain (BWG, kg/d), and REI (MJ of ME/d). The MEI was based on DMI of the feeds and their energy values. The energy values (MJ of ME/kg of DM) of the feeds were calculated according to Luke (2015). The daily MEI was corrected by the total DMI and concentration of ME and protein in the diet according to the correction equation given by Luke (2015). Residual energy intake was estimated by modeling MEI by a multiple linear first-order regression including ECM, BW0.75, and piecewise regressions of BWL and BWG. Weekly cow-wise means of residuals from the model were used as cows’ REI measures. All the 495 cows were from the Nordic Red dairy cattle breeding nucleus herd, and were well related to both paternal and maternal sides of their pedigrees. The pedigree used for the genetic analyses was traced back 4 generations from the animals with observations and included 2,409 informative animals. Modeling Metabolizable Energy Efficiency

Residual Energy Intake as a Reference Trait. Several publications have reported analyses of REI in dairy cows. Therefore, and because REI describes the same efficiency characteristics of a cow as MEE does, we chose REI as reference trait in this study. The applied repeatability animal model describes a REI observation of a cow l, recorded during the year × month period i, which addressed also the effect of feeding trials as the

3

trials in the older data set collected in Rehtijärvi were arranged in consecutive time periods, and in lactation week j as follows:

REI ijl = rymi + lw j + al + pel + εijl , [1]

where REIijl = REI (MJ of ME/d), rymi = the fixed effect of recording year-month i, lwj = the fixed effect of lactation week j, al = the random additive genetic effect for animal l [a ~N(0, Aσ2a), where A is the additive genetic relationship matrix among animals and σ2a is the additive genetic variance], pel = the random permanent environmental effect for animal l [pe ~N(0, Iσ2pe), where I is an identity matrix and σ2pe is the permanent environmental variance], and εijl = the random error term with variance σ2ε. Residual energy intake, which describes the efficiency of ME use, can be modeled alternatively by modeling MEI observations and including regressions on energy sinks (BW0.75, ECM, BWG, BWL) into the model, which will allow a better fit of the data. Including Energy Sinks into the Model for Metabolizable Energy Intake. In our data lactation class averages for REI deviated from the expectation of zero (Table 1), reflecting that on average the calculated energy requirements and actual MEI differ among different stages of lactation. Therefore, lactation week was included into the model for REI [1]. Likewise, it can be expected that regressions on energy sinks also differ across lactation, and if so, these regressions should be nested within lactation stages to avoid systematic bias in estimates for additive genetic effects of animals with lactation in progress. Therefore, a least square model was fitted to assess whether regression coefficients change significantly during lactation: MEI ijls = rymi + lw j + b1s BWjl0.75 + b2s ECM jl + b3s BWG jl + b4s BWLjl + εijls ,

[2] where MEIijls = metabolizable energy intake (MJ/d), rymi = fixed effect of recording year-month i, lwj = fixed effect of lactation week j, bcs = fixed regression coefficient bc nested within lactation class s, where c = 1, 2, 3, 4 is a regression coefficient for maintenance, milk production, BWG, and BWL, respectively, and where lactation classes s are 8 (8 classes: 2–5, 6–10, 11–15, 16–20, 21–25, 26–30, 31–35, and 36–40 wk of lactation), BWjl0.75 = metabolic body weight (kg) of cow l in lactation week j, ECMjl = ECM (kg/d) of cow l in lactation week j, BWGjl = BWG (kg/d) of cow l in

Journal of Dairy Science Vol. 101 No. 5, 2018

4

[4]

where MEIijls = the MEI (MJ/d), rymi = the fixed effect of recording year-month i, lwj = the fixed effect of lactation week j, bcs = the fixed regression coefficient bc nested within lactation class s, where c = 1, 2, 3, 4 is a Journal of Dairy Science Vol. 101 No. 5, 2018

26.23   22.13 23.36 24.69 26.65 27.44 26.78 27.94 26.54 −3.58   −16.47 −5.79 1.82 1.66 1.30 −0.02 −4.48 −7.80

SD

0.30   0.71 0.12 0.05           −0.07   −0.45 −0.05 −0.01          

Mean

0.23   0.16 0.20 0.17 0.15 0.16 0.21 0.25 0.29 0.27   0.07 0.19 0.22 0.24 0.28 0.35 0.46 0.51 8.22   7.59 7.28 7.39 7.64 7.84 7.84 7.71 8.30 119.11   116.51 116.27 117.07 118.18 119.32 120.98 123.28 125.71 4.32   4.89 4.33 4.16 4.05 4.10 4.05 3.92 3.92 28.46   27.99 29.57 29.53 29.01 28.31 27.94 27.50 26.50 31.86   25.18 26.56 29.43 31.70 33.57 31.72 32.20 31.66 208.69   183.28 205.49 214.92 214.80 214.26 215.21 214.07 210.91

SD

 

Mean

SD

 

Mean

SD

 

Mean

SD

 

BWL

SD

 

Mean

REI

12,350   1,639 1,988 1,798 1,686 1,541 1,383 1,184 1,131

+ pe4l BWLjl + εijls ,

Overall Lactclass  1  2  3  4  5  6  7  8

  + a4l BWLjl + pe0l + pe1l BW jl0.75 + pe2l ECM jl + pe3l BWG jl

Mean

  + b4s BWLjl + a0l + a1l BW jl0.75 + a2l ECM jl + a 3l BWG jl

Variable

MEI ijls = rymi + lw j + b1s BWjl0.75 + b2s ECM jl + b3s BWG jl

BWG

where MEIijls = the metabolizable energy intake (MJ/d), rymi = the fixed effect of recording year-month i, lwj = the fixed effect of lactation week j, bcs = the fixed regression coefficient bc nested within lactation class s, where c = 1, 2, 3, 4 is a regression coefficient for maintenance, milk production, BWG, and BWL, respectively, and where lactation classes s are 8 (8 classes: 2–5, 6–10, 11–15, 16–20, 21–25, 26–30, 31–35, 36–40 wk of lactation), BWjl0.75 = the metabolic BW (kg) of cow l in lactation week j, ECMjl = the ECM (kg/d) of cow l in lactation week j, BWGjl = the BWG (kg/d) of cow l in lactation week j, BWLjl = the BWL due to mobilization of tissue energy for milk production (kg/d) of cow l in lactation week j, al = the random additive genetic effect for animal l [a ~N(0, Aσ2a), where A is the additive genetic relationship matrix among animals and σ2a is the additive genetic variance], pel = the random permanent environmental effect for animal l [pe ~N(0, Iσ2pe), where I is an identity matrix and σ2pe is the permanent environmental variance], and εijls = random error term with variance σ2ε. Partial Metabolizable Energy Efficiency. Applying random regression models allows estimation of cow-specific energy sink efficiencies. Therefore, the model for MEE was extended with regressions on energy sinks nested within additive genetic and permanent environmental effects as follows:

BW0.75

[3]

ECM

+ b4s BWLjl + al + pel + εijls ,

MEI

MEI ijls = rymi + lw j + b1s BWjl0.75 + b2s ECM jl + b3s BWG jl

No. of observations

lactation week j, BWLjl = BWL due to mobilization of tissue energy for milk production (kg/d) of cow l in lactation week j, and εijls = random error term with variance σ2ε. Metabolizable Energy Efficiency. Regressions on energy sinks are included as fixed effects into the model for MEE. Otherwise the applied repeatability animal model had the same model effects as in [1]:

Table 1. Descriptive statistics of metabolizable energy intake (MEI, MJ/d), ECM (kg/d), metabolic body weight (BW0.75, kg), body weight gain (BWG, kg/d), body weight loss (BWL, kg/d) and residual energy intake (REI, ME MJ/d) across lactation and in 8 different lactation classes (Lactclass, 1 = 2–5, 2 = 6–10, 3 = 11–15, 4 = 16–20, 5 = 21–25, 6 = 26–30, 7 = 31–35, and 8 = 36–40 wk of lactation) of 495 animals’ weekly observations

MEHTIÖ ET AL.

5

PARTITIONING OF METABOLIZABLE ENERGY EFFICIENCY

regression coefficient for maintenance, milk production, BWG, and BWL, respectively, and where lactation classes s are 8 (8 classes: 2–5, 6–10, 11–15, 16–20, 21–25, 26–30, 31–35, 36–40 wk of lactation), acl = the random additive genetic regression coefficient c = 0 (intercept), 1 (maintenance), 2 (milk production), 3 (BWG), 4 (BWL) for animal l, pecl = the random permanent environmental regression coefficient c = 0 (intercept), 1 (maintenance), 2 (milk production), 3 (BWG), 4 (BWL) for animal l, BWjl0.75 = the metabolic BW (kg) of cow l in lactation week j, ECMjl = the ECM (kg/d) of cow l in lactation week j, BWGjl = the BWG (kg/d) of cow l in lactation week j, BWLjl = the BWL due to mobilization of tissue energy for milk production (kg/d)   of cow l in lactation week j, BW jl0.75 , ECM jl , and BWG jl are the same covariables as given above but standardx − µx ized by x = , where x is the original covariable, σx and µx and σx are the mean and standard deviation of covariable x, respectively, as given in Table 1, and εijls = the random error term with variance σ2ε. The covariance matrices were var(a) = G5x5⊗A and var(pe) = P5x5⊗I, where G was the covariance matrix for the random additive genetic effects, A was the additive genetic relationship matrix, P was the covariance matrix for the random permanent environmental effects, and I was an identity matrix. As the model [4] is very complex, 3 simpler random regression models were tested. The submodels of model [4] differed in the number of random regression effects included into the permanent environment and additive genetic animal effects, but otherwise the models were the same as [4]. The alternative models included the following random regressions on intercept and ECM (pMEE1); intercept, BW0.75, ECM, and BWG (pMEE2); and intercept, BW0.75, ECM, BWG, and BWL (pMEE3). Hence, model pMEE3 was equal to [4]. Estimation of Genetic Parameters. Variance components were estimated by REML method applying either average information (AI-REML) or for the most complex random regression models, expectation maximization (EM-REML) implemented in the DMU software package (Madsen and Jensen, 2013). The fit of the models were assessed using the Akaike’s information criterion (AIC) and the size of residual variances. Heritabilities from repeatability models for both REI and MEE were calculated as the ratio of genetic variance (σ2a) to total phenotypic variance (σ2pe). In contrast, for the random regression models different sets of heritabilities can be derived. Overall heritability was ˆ and P ˆ calculated by summing the (co)variances of G with respect to the energy sinks. For BWL, standard-

izing was not possible, and thus the overall data average of −0.07 is used. However, standardizing other covariables for random effects, as given in [4], allowed calculating partial heritabilities for energy sinks, assuming that a cow’s genetic regression coefficient deviates on average 1 genetic standard deviation from the mean. By that, genetic and permanent environment variances for overall heritability and partial heritability calculations for the most complex model (pMEE3) were obtained by applying the summation matrix 1  1  0 ϕ(pMEE 3) =  0  0  0



1 0 1 0 0 0

1 0 0 1 0 0

1 0   0 0  0 0  . 0 0   1 0   0 −0.07 

Additive genetic and permanent environment variances from random regression model i were calculated as 2 ˆ is the = ϕ(i ) Pˆ(i )ϕ(′i ), where G σa2 = ϕ(i )Gˆ(i )ϕ(′i ) and σpe (i ) (i )

(i )

covariance matrix for the random additive genetic efˆ is the covariance matrix for the random fects and P (i ) permanent environmental effects. RESULTS

Descriptive statistics of the data are presented in Table 1. Metabolizable energy intake increased from 183.28 to 214.92 MJ/d from lactation wk 2 to 15 and then stayed somewhat constant until a slight decline toward the end of the recording period. Energy-corrected milk production increased from 27.99 to 29.57 kg/d in lactation classes 1 and 2 and was followed by a decrease from 29.53 to 26.50 kg/d. Metabolic BW increased from 116.51 to 125.71 kg, and BWG increased from 0.07 to 0.51 kg/d during the studied lactation period. Body weight loss was highest in early lactation (−0.45 kg/d) and then dropped rapidly to −0.05 in the second and to −0.01 in the third lactation class. Residual energy intake was on average negative in early and late lactation and positive only from lactation wk 11 to 25, varying from −16.47 to 1.82. Regression Coefficients for Energy Sinks

Table 2 presents the least squares solutions from fitting the model [2] on MEI. The results suggest significant differences between energy sink regression coefficients at different lactation stages. The overall estimate

Journal of Dairy Science Vol. 101 No. 5, 2018

6

MEHTIÖ ET AL.

Table 2. Estimates and SE for energy sink regression coefficients from the least squares model across lactation and for 8 lactation classes Energy sink Lactation class

Maintenance1

Overall 1 2 3 4 5 6 7 8

0.81 1.08 1.09 0.82 0.91 0.78 0.66 0.58 0.48

± ± ± ± ± ± ± ± ±

Milk production2

0.03 0.08 0.07 0.08 0.08 0.08 0.08 0.09 0.09

2.67 1.27 1.85 2.76 2.93 3.31 3.33 3.70 3.67

± ± ± ± ± ± ± ± ±

0.06 0.14 0.14 0.15 0.16 0.16 0.16 0.18 0.19

BW gain 13.80 6.35 10.14 22.68 20.55 23.01 14.65 9.58 14.24

± ± ± ± ± ± ± ± ±

1.29 3.87 3.02 3.76 4.34 4.18 3.32 3.02 2.79

BW loss 8.05 5.52 11.73 14.95

± ± ± ±          

0.94 0.98 4.81 10.87

1

Metabolic BW. ECM.

2

for ME used for maintenance was 0.81 MJ × BW0.75 (kg), and estimates decreased from 1.09 to 0.48 MJ as lactation progressed. The overall estimate for milk production was 2.67 MJ × ECM (kg/d), and estimates increased from 1.27 to 3.70 MJ as lactation progressed. For BWG and loss, the overall estimates were 13.80 and 8.05 MJ/kg, respectively. Estimates for BWG varied between 6.35 and 23.01 MJ/kg during lactation. On the other hand, for BWL the estimates increased from 5.52 to 14.95 MJ/kg from the beginning of lactation to wk 14. Thus, the estimates for milk production, BW gain, and BW loss were lower and the estimates for maintenance were on average higher than the values from the official Finnish feeding recommendations (Luke, 2015). Genetic Parameters for Residual Energy Intake and Metabolizable Energy Efficiency

Estimates of variance components from the repeatability models for REI and MEE resulted in smaller genetic and residual variances for MEE compared with REI, but the variance for permanent environment was slightly higher for MEE (Table 3). The phenotypic variance for MEE was 7% lower compared with the phenotypic variance of REI, and the estimated heritability for MEE was 0.26, whereas it was 0.33 for REI. The estimated repeatability was 0.54 and 0.53 for MEE and REI, respectively. Genetic Parameters for Partial Metabolizable Energy Efficiency

Several alternative random regression models were tested, but only model pMEE1 showed good convergence characteristics (10 AI-REML rounds to reach convergence). When, in addition to intercept and ECM, BW0.75 was included in the model, the heritability esJournal of Dairy Science Vol. 101 No. 5, 2018

timates became unexpectedly high and convergence of the model was only achieved by applying EM-REML and performing more than 1,000 EM steps. However, when BWG and BWL were also included in the random part of the model, results improved and became more logical (Table 3, pMEE2 and pMEE3), but still EM-REML iterations instead of AI-REML iterations were needed to achieve convergence. Compiled variance component estimates and the derived overall heritability estimate (0.21–0.23) for random regression models were in line with the estimates from the repeatability model for MEE. Most of the variance was explained by the intercept, for which heritability estimates ranged from 0.13 to 0.23 depending on the applied random regression model. The heritability estimates for partial ME efficiency for milk production changed between 0.04 and 0.06. The heritability estimates for partial ME efficiency with respect to maintenance and growth were 0.02 and 0.04, respectively. The variance explained by BWL was practically nonexistent. Because of this, model pMEE3 was not studied any further. Model validation based on AIC and residual variances showed that even if poorer convergence characteristics, more complex models pMEE2 and pMEE3 had the best fit for the data. Genetic and permanent environmental variances for regression coefficients from models pMEE1 and pMEE2 are presented in Table 4. The estimated genetic variances were highest for the intercepts in both models, whereas estimated genetic variances were clearly smaller for the linear regression coefficients BW0.75, ECM, and BWG. Variance estimates obtained by pMEE1 or pMEE2 for the same genetic (intercept and ECM) and permanent environmental (intercept and ECM) effects differed considerably. The genetic variances, estimated with model pMEE2 for the linear regression coefficients on BW0.75 and BWG, were smaller than their

REI is a repeatability model for residual energy intake, MEE is a repeatability model for metabolizable energy efficiency and pMEE1 to pMEE3 are random regression models for partitioning metabolizable energy efficiency. Random regression models included random regression effects for intercept (int), metabolic body weight (BW0.75), ECM, body weight gain (BWG), and body weight loss (BWL). σ2a = additive genetic variance; σ2pe = permanent environmental variance; σ2ε = random error variance.

1

81,333.1 80,369.5 79,793.6 78,126.9 78,089.6         0.00       0.04 0.04     0.04 0.06 0.05       0.02 0.02     0.23 0.14 0.13 0.33 0.26 0.23 0.22 0.21 268.57 246.68 223.81 167.37 165.50 115.57 148.81 244.05 705.16 685.55 Intercept Intercept Intercept, ECM Intercept, BW0.75, ECM, BWG Intercept, BW0.75, ECM, BWG, BWL           REI MEE pMEE1 pMEE2 pMEE3

187.99 137.23 141.91 249.53 225.61

AIC η2BWL η2BWG η2ECM η2BW0.75 η2int h2 σ2ε σ2pe σ2a Additive genetic animal effect   Model

Table 3. Variance component estimates, h2, partial heritabilities (η2) for random regression effects of energy sink specific efficiencies and Akaike’s information criterion (AIC) given by models1

PARTITIONING OF METABOLIZABLE ENERGY EFFICIENCY

7

associated standard errors, which indicate large uncertainty associated with estimates for energy efficiency for maintenance and growth. However, analyses with model pMEE1 resulted in combined variance component estimates, which resemble with the estimates from MEE (Table 3). Based on pMEE1, the genetic standard deviation estimate for the intercept was 10.8 MJ of MEI/d and the genetic standard deviation estimate for the regression coefficient for milk production was 0.75 MJ of MEI/kg of ECM. The genetic standard deviation estimates for regressions on maintenance and growth, obtained with model pMEE2, were 0.47 MJ of MEI/ kg of BW0.75 and 18.0 MJ of MEI/kg, respectively, but associated standard errors were large. Genetic correlations between regression coefficients were positive, except between energy efficiency for maintenance and for growth, which was practically zero (Table 5). The genetic correlation between energy efficiency in milk production and maintenance was 0.44, and the genetic correlation between energy efficiency in milk production and growth was 0.54. However, the standard errors for genetic correlations were high, which again is an indication of uncertainty associated with the genetic covariance estimates. Mostly permanent environmental correlations were weakly negative except that the correlation between energy efficiency in growth and maintenance was −0.71, and the correlation between energy efficiency in growth and intercept was 0.24. Phenotypes of the high and low genetic merit cows were compared with assess the response of selection if using EBV for partial energy efficiencies. Means and standard deviations of the phenotypic data in high and low genetic merit groups are presented in Table 6. Comparing cows based on the EBV for the intercept showed that best cows had 11% lower MEI intake, produce 3.7% less milk, had 1.8% lower BW and had 17.4 MJ lower REI compared with average cows. Cows with high genetic merit based on EBV for energy efficiency in milk production had 3.5% lower MEI, produced 3.7% less milk, had the same BW, and had 3.0 MJ lower REI compared with average cows. Cows with high genetic merit based on EBV for energy efficiency in growth had 6.2% lower MEI, produced 2.0% less milk, had the same BW, and had 11.0 MJ lower REI compared with average cows. In contrast, when comparing cows based on the EBV for energy efficiency in maintenance, results looked illogical. In this case, cows with high genetic merit performed similar to average cows by having the same MEI, producing 1.3% less milk, and having 0.8% lower BW and 1.0 MJ higher REI, whereas low genetic merit cows were more efficient by having 5.1% lower MEI, producing 2.3% less milk, and having 2.2% lower BW and 6.4 MJ lower REI. Journal of Dairy Science Vol. 101 No. 5, 2018

8

MEHTIÖ ET AL.

Table 4. Genetic (σ2a) and permanent environmental (σ2pe) variance estimates with SE for regression coefficients from 2 random regression models including random regression on intercept and ECM (pMEE1); intercept, metabolic body weight (BW0.75), ECM, and body weight gain (BWG; pMEE2) Regression coefficient Intercept BW0.75 ECM BWG

σ2a pMEE1 116.74 ± 48.52   10.60 ± 8.91  

σ2pe pMEE2 78.18 15.21 12.21 17.07

± ± ± ±

54.37 71.68 7.63 27.83

DISCUSSION

 

pMEE1

       

174.46 ± 40.78   48.90 ± 9.58  

pMEE2 320.47 627.92 25.18 219.35

± ± ± ±

58.63 90.95 7.65 33.11

mation, regression coefficients have modeled well the average MEI intake of the cows across lactation stages.

Regression Coefficients for Energy Sinks

Modeling of MEE requires the inclusion of regression coefficients for energy sinks into a model. The results from fitting models that included regression coefficients on energy sinks showed that estimates of regression coefficients varied across lactation stages. An upward trend in estimates for energy use for milk production and a downward trend in estimates for energy use for maintenance may indicate co-linearity between covariables in the data. Therefore, the resulting estimates for a particular energy sink at a particular lactation stage may not be comparable with the estimates of the requirement norms from nutritional studies. For instance, the estimates of the first 4 lactation stages for milk production were smaller than the energy content in milk (3.14 MJ/kg; Sjaunja et al., 1990). Moreover, it is not unlikely that modeling energy use at early stages of lactation is suboptimal. For instance, a drop in BW0.75 during first lactation weeks, due to negative energy status of the cow, may not necessarily cause an equivalent drop in energy requirements for maintenance, which may yield inflated regression coefficients for maintenance at the beginning of lactation. Thus, although each particular regression coefficient estimated in this study for the various energy sinks were somewhat different than those values from Finnish feed table recommendations (BW0.75: 0.515 MJ/kg of BW0.75; ECM production: 5.15 MJ/kg of ECM; BWG: 34.0 MJ/kg; BWL: 28.0 MJ/kg; Luke, 2015); in sum-

Genetic Parameters for Residual Energy Intake and Metabolizable Energy Efficiency

When fitting a repeatability animal model, our estimate of heritability for REI was 0.33. This heritability estimate was slightly higher than most estimates reported in the literature (Liinamo et al., 2015; Tempelman et al., 2015; Hurley et al., 2018). Liinamo et al. (2015) and Tempelman et al. (2015) have shown that heritability for REI and RFI changes during lactation. Liinamo et al. (2015) reported heritability estimates of 0.20 in the beginning (wk 2–5) and 0.20–0.38 at the end of a 30-wk lactation period, but observed low heritability estimates (<0.10) between 10 and 23 wk of lactation, which could indicate that REI in different lactation stages is a different trait. On the other hand, the lactation-wise heritability of REI over the 2- to 30-wk period was 0.09 (Liinamo et al., 2015), and Tempelman et al. (2015) reported heritability estimates ranging from 0.06 to 0.39 during the different DIM. Fitting regressions for energy sinks simultaneously with all other model effects (model MEE) yielded not only a smaller heritability (0.26) and residual variance than REI, but also a slight increase in the repeatability, indicating that the MEE model should have a better predictability compared with the model for REI. This is in line with Tempelman et al. (2015) who also observed heterogeneity in partial regression coefficients

Table 5. Genetic (upper triangle) and permanent environmental (lower triangle) correlations with SE between genetic values for partial efficiencies for intercept [pMEE (int)], maintenance [pMEE (BW0.75)], milk production [pMEE (ECM)], and growth [pMEE (BWG)] based on the pMEE2 model Item pMEE pMEE pMEE pMEE

(int) (BW0.75) (ECM) (BWG)

pMEE (int)

pMEE (BW0.75)

pMEE (ECM)

pMEE (BWG)

  −0.040 ± 0.11 −0.084 ± 0.17 0.243 ± 0.11*

0.182 ± 1.26   −0.207 ± 0.147 −0.712 ± 0.06**

0.523 ± 0.48 0.441 ± 1.75   0.096 ± 0.15

0.753 ± 0.70 −0.006 ± 2.20 0.526 ± 0.71  

*Estimate deviates more than 1.960 × SE from 0 (P < 0.05). **Estimate deviates more than 2.576 × SE from 0 (P < 0.01). Journal of Dairy Science Vol. 101 No. 5, 2018

9

PARTITIONING OF METABOLIZABLE ENERGY EFFICIENCY

Table 6. Mean of the EBV, metabolizable energy intake (MEI), ECM, metabolic body weight (BW0.75), and residual energy intake (REI) for the 100 most (lowest EBV) and 100 least (highest EBV) efficient cows given by ranking based on partial efficiency for intercept [pMEE (int)], maintenance [pMEE (BW0.75)], milk production [pMEE (ECM)], and growth [pMEE (BWG)] when breeding values were estimated with model pMEE2 MEI

EBV Mean

SD

 

Mean

SD

 

Mean

Lowest pMEE (int) Highest pMEE (int) Lowest pMEE (BW0.75) Highest pMEE (BW0.75) Lowest pMEE (ECM) Highest pMEE (ECM) Lowest pMEE (BWG) Highest pMEE (BWG)

11.6 −10.8 1.8 −1.6 0.8 −0.6 36.1 −36.5

5.7 8.4 1.1 1.1 0.3 0.4 23.5 26.5

               

222.3 186.1 198.1 208.5 214.0 201.4 214.0 195.8

21.3 22.5 26.1 23.6 21.1 28.1 24.1 27.9

               

28.8 27.4 27.8 28.1 29.1 27.4 28.9 27.9

when modeling RFI on an international consortium of feed efficiency data. If such heterogeneities are ignored in modeling RFI or REI, it may lead to biased inferences about the traits in breeding programs. This could be especially critical if REI observations are not recorded during the same lactation periods for all cows because mean and standard deviation of REI observations change across lactation as shown in this study. Hence, modeling MEE allows more accurate modeling of the data and thereby reduces the residual variance but also increases robustness against an inflation of additive genetic variance. Genetic Parameters for Partial Metabolizable Energy Efficiency

With the partial MEE models, we obtained heritability estimates that are in line with the estimates from the repeatability model analysis of MEE. Also including into the MEE model partial regression on ECM (pMEE1) or partial regressions on BW0.75, ECM, and BWG (pMEE2) was feasible and resulted in partial heritabilities between 0.04 and 0.06 for energy efficiency in milk production. Applying standardization of the covariables for estimating the random partial regression coefficients by pMEE models was found to be crucial for modeling the partial efficiencies for the different pathways. When including a partial regression on BW0.75, the model did not converge. One explanation for this is that, when the animal intercepts (both the overall breeding value and permanent environment) are part of model, the information for the animal-wise BW0.75 related maintenance cost comes from the different within-animal BW levels. Although variation in BW exists across animals, the variation within animals is limited, and thus the model might not be able to describe this. However, including partial regression on ECM and on BWG (pMEE2), or for all energy sinks (pMEE3) made the models converge, but convergence

BW0.75

ECM

Groups of 100 cows with lowest or highest efficiency based on the partial breeding value

REI

SD  

Mean

SD  

Mean

SD

3.5 3.3 3.5 4.1 3.4 3.2 4.0 3.4

119.3 117.7 117.3 118.7 118.1 118.1 120.0 119.4

7.3 7.7 5.7 6.4 7.4 7.4 7.2 8.1

8.34 −21.0 −10.0 −2.6 1.4 −6.6 0.0 −14.6

16.7 18.3 21.1 20.8 15.5 22.4 20.7 21.8

               

               

was slow. Nevertheless, AIC and residual variances were lower for these models, indicating a better fit. Overall, we were able to model partial energy efficiencies for milk production and growth, but we encountered difficulties in modeling energy efficiency in maintenance. This is evidenced by the unexpected high estimate for genetic variance and the higher standard errors associated with it (Table 4), as well as the illogical relationship between EBV for efficiency in maintenance and phenotypic observations. One possible reason may be the size and structure of the data used. Cows in these data were all primiparous and high genetic merit cows of the Nordic Red dairy cattle breeding nucleus. Cows are above average in feed efficiency compared with the whole Nordic Red dairy cattle population, and therefore the variability in the efficiency component traits and MEI between cows might be smaller. Further, the data used for these analyses were recorded over a long time period, resulting in many calving years with low numbers of cows per calving year. The effect of genetic selection for milk yield is widely described in the literature. One important aspect of this is that genetic selection for milk yield has affected physiological pathways that are controlling feed efficiency. Among others, Veerkamp et al. (2003) reviewed several studies to assess the effects of genetic merit for milk yield on hormone and metabolite profiles in lactating dairy cows. They concluded that selection for higher yield has changed the energy partitioning in lactating dairy cows, most likely due to genetic effects on the somatotropic axis, including growth hormone and IGF-I. This genetically induced energy partitioning forces a high genetic merit cow to partition more energy to milk, even if the maintenance requirements are not fulfilled. Similarly, Agnew and Yan (2000) presented that high genetic merit cows have the ability to partition more energy into milk than low genetic merit cows, and Yan et al. (2006) showed differences between dairy cattle breeds in partitioning energy between milk Journal of Dairy Science Vol. 101 No. 5, 2018

10

MEHTIÖ ET AL.

and body tissue. Earlier in their review, Veerkamp and Emmans (1995) also indicated that high and low genetic merit cows partition the available energy differently, but that there was no strong evidence to ascertain the existence of genetic differences in the partial efficiencies. The results in this study indicate genetic differences in the partial efficiencies and not only in allocation of energy resources in dairy cows. Despite the difficulties we encountered with modeling efficiency of the maintenance pathway, we found promising results for the milk production pathway. The estimated genetic standard deviation of 0.75 MJ of ME/kg of ECM in the efficiency of utilization energy for milk production indicates genetic variation, which could be improved by genetic selection. Our variance component estimates indicated a positive genetic correlation between energy efficiencies of the different energy sinks, which indicates that an efficient cow is efficient in utilization of ME in all pathways. Phenotypes of high and low genetic merit groups showed that cows with high genetic merit on energy efficiency for milk production eat less but produce slightly more milk per unit of MEI than low genetic merit cows, which could indicate that highly feed efficient cows compensate for the lower MEI by higher efficiency of converting ME into milk energy or by better digestibility. Genetic variation in cow-specific digestibility was found to be small (Berry et al., 2007); however, it can contribute to the positive genetic correlations between partial efficiencies found here. This is because cows with higher ability to digest feed will make more MEI available for all pathways than cows with lower ability to digest feed. Selecting for feed efficiency based on the models studied here should have a small effect on BW given the genetic correlation with BW is close to zero. In this study, we found that the genetically best cows based on the EBV for the intercept had on average lowest REI but had also 1.8% lower BW, which indicates that there might be a positive genetic correlation between REI and BW. CONCLUSIONS

In this study, MEI was modeled by including regressions on energy sinks, resulting in an efficiency trait that is analogous to RFI. Metabolizable energy efficiency improved modeling of the data and resulted in better properties for prediction of breeding values compared with RFI. Moreover, the model for ME efficiency can be expanded to include random regressions on different energy pathways to model genetic variation in efficiency with respect to specific energy pathways. In this study, we found evidence for genetic variation among cows in how efficiently cows are using ME for Journal of Dairy Science Vol. 101 No. 5, 2018

specific energy pathways and also that there is a positive genetic correlation among these partial efficiencies. By fitting random regression models, we were able to model cow-specific partial efficiency breeding values for milk production and growth. However, modeling partial efficiency breeding values for maintenance was difficult and will require additional research. ACKNOWLEDGMENTS

The authors thank biometrician Timo Pitkänen from Natural Resources Institute Finland (Luke) for valuable comments and acknowledge the financial support provided for this study by the Finnish Ministry of Agriculture (DNRO:​1844/​312/​2012; Helsinki, Finland); Valio Ltd. (Helsinki, Finland), Faba co-op (Vantaa, Finland), VikingGenetics (Hollola, Finland), The Finnish Cattle Breeding Foundation (Hamina, Finland), and RAISIOagro Ltd. (Raisio, Finland). REFERENCES Agnew, R. E., and T. Yan. 2000. Impact of recent research on energy feeding systems for dairy cattle. Livest. Prod. Sci. 66:197–215. Berry, D. P., M. P. Coffey, J. E. Pryce, Y. de Haas, P. Løvendahl, N. Krattenmacher, J. J. Crowley, Z. Wang, D. Spurlock, K. Weigel, K. Macdonald, and R. F. Veerkamp. 2014. International genetic evaluations for feed intake in dairy cattle through the collation of data from multiple sources. J. Dairy Sci. 97:3894–3905. Berry, D. P., B. Horan, M. O’Donovan, F. Buckley, E. Kennedy, M. McEvoy, and P. Dillon. 2007. Genetics of grass dry matter intake, energy balance, and digestibility in grazing Irish dairy cows. J. Dairy Sci. 90:4835–4845. Huhtanen, P., J. Nousiainen, and M. Rinne. 2006. Recent developments in forage evaluation with special reference to practical applications. Agric. Food Sci. 15:293–323. Hurley, A. M., N. López-Villalobos, S. McParland, E. Lewis, E. Kennedy, M. O’Donovan, J. L. Burke, and D. P. Berry. 2018. Characteristics of feed efficiency within and across lactation in dairy cows and the effect of genetic selection. J. Dairy Sci. 101:1267–1280. Liinamo, A.-E., P. Mäntysaari, M. H. Lidauer, and E. A. Mäntysaari. 2015. Genetic parameters for residual energy intake and energy conversion efficiency in Nordic Red dairy cattle. Acta Agric. Scand. Anim. Sci. 65:63–72. Luke. 2015. Feed tables and feeding recommendations. Accessed Jul. 21, 2017. https://​portal​.mtt​.fi/​portal/​page/​portal/​Rehutaulukot/​ feed​_tables​_english/​nutrient​_requirements/​Ruminants/​Energy​ _dairy​_cows. Madsen, P., and J. Jensen. 2013. DMU A package for analysing multivariate mixed models. Version 6, release 5.2. Center for Quantitative Genetics and Genomics, Department of Molecular Biology and Genetics. University of Aarhus Research Centre Foulum, Tjele, Denmark. MAFF. 1975. Energy allowances and feeding systems for ruminants. In Ministry of Agriculture, Fisheries and Food Technical Bulletin, No. 33. Ministry of Agriculture, Fisheries and Food, London, UK. MAFF. 1984. Energy allowances and feeding systems for ruminants. In ADAS Reference Book, No. 433. Ministry of Agriculture, Fisheries and Food, London, UK. Mäntysaari, P., P. Huhtanen, J. Nousiainen, and M. Virkki. 2004. The effect of concentrate crude protein content and feeding strategy of total mixed ration on performance of primiparous dairy cows. Livest. Prod. Sci. 85:223–233.

PARTITIONING OF METABOLIZABLE ENERGY EFFICIENCY

Mäntysaari, P., P. Huhtanen, J. Nousiainen, and M. Virkki. 2005. The effect of protein-feeding strategy during lactation on performance of primiparous dairy cows fed total mixed ration. Livest. Prod. Sci. 94:189–198. Mäntysaari, P., A.-E. Liinamo, and E. A. Mäntysaari. 2012. Energy efficiency and its relationship with milk, body, and intake traits and energy status among primiparous Nordic Red dairy cattle. J. Dairy Sci. 95:3200–3211. Mäntysaari, P., and E. A. Mäntysaari. 2008. Relationship of body measurements and body condition score to body weight in modern Finnish Ayrshire cows. Acta Agric. Scand. Anim. Sci. 58:170–178. Mäntysaari, P., and E. A. Mäntysaari. 2015. Modeling of daily body weights and body weight changes of Nordic Red cows. J. Dairy Sci. 98:6992–7002. Mäntysaari, P., J. Nousiainen, and P. Huhtanen. 2003. The effect of constant or variable forage to concentrate ratio in total mixed ration on performance of primiparous dairy cows. Livest. Prod. Sci. 82:27–37. Manzanilla-Pech, C. I. V., R. F. Veerkamp, R. J. Tempelman, M. L. van Pelt, K. A. Weigel, M. VandeHaar, T. J. Lawlor, D. M. Spurlock, L. E. Armentano, C. R. Staples, M. Hanigan, and Y. De Haas. 2016. Genetic parameters between feed-intake-related traits and conformation in 2 separate dairy populations – the Netherlands and United States. J. Dairy Sci. 99:443–457. Pryce, J. E., W. J. Wales, Y. de Haas, R. F. Veerkamp, and B. J. Hayes. 2014. Genomic selection for feed efficiency in dairy cattle. Animal 8:1–10. Sjaunja, L. O., L. Baevre, L. Junkkarinen, J. Pedersen, and J. Setälä. 1990. A Nordic proposal for an energy corrected milk (ECM) for-

11

mula. Pages 156–157 in Performance Recording of Animals: State of the Art, 1990. EAAP Publication 50. P. Gaillon and Y. Chabert, ed. Centre for Agricultural Publishing and Documentation, Wageningen, the Netherlands. Spurlock, D. M., J. C. M. Dekkers, R. Fernando, D. A. Koltes, and A. Wolc. 2012. Genetic parameters for energy balance, feed efficiency, and related traits in Holstein cattle. J. Dairy Sci. 95:5393–5402. Tempelman, R. J., D. M. Spurlock, M. Coffey, R. F. Veerkamp, L. E. Armentano, K. A. Weigel, Y. de Haas, C. R. Staples, E. E. Connor, Y. Lu, and M. J. VandeHaar. 2015. Heterogeneity in genetic and nongenetic variation and energy sink relationships for residual feed intake across research stations and countries. J. Dairy Sci. 98:2013–2026. Veerkamp, R. F., B. Beerda, and T. van der Lende. 2003. Effects of genetic selection for milk yield on energy balance, levels of hormones, and metabolites in lactating cattle, and possible links to reduced fertility. Livest. Prod. Sci. 83:257–275. Veerkamp, R. F., and G. C. Emmans. 1995. Sources of genetic variation in energetic efficiency of dairy cows. Livest. Prod. Sci. 44:87– 97. Wilmink, J. B. M. 1987. Adjustment of test-day milk, fat and protein yield for age, season and stage of lactation. Livest. Prod. Sci. 16:335–348. Yan, T., C. S. Mayne, T. W. J. Keady, and R. E. Agnew. 2006. Effects of dairy cow genotype with two planes of nutrition on energy partitioning between milk and body tissue. J. Dairy Sci. 89:1031–1042.

Journal of Dairy Science Vol. 101 No. 5, 2018