Genetics of feed efficiency in dairy and beef cattle. D.P. Berry 1 * and J. J. Crowley 2

Transcription

1 1 GENETICS OF FEED EFFICIENCY 2 3 Genetics of feed efficiency in dairy and beef cattle 4 5 D.P. Berry 1 * and J. J. Crowley Animal and Grassland Research and Innovation Centre, Teagasc, Moorepark, Fermoy, Co. Cork, Ireland 2 Department of Agricultural, Food and Nutritional Science, University of Alberta, Edmonton, Alberta T6G 2P5, Canada *Corresponding Author: Donagh Berry, Animal and Grassland Research and Innovation Centre, Teagasc, Moorepark, Fermoy, Co. Cork, Ireland Tel: Fax: Donagh.berry@teagasc.ie

2 ABSTRACT Increasing food production for the growing human population off a constraining landbase will require greater efficiency of production. Feed efficiency in cattle can contribute to this, and breeding, which is cumulative and permanent, is one likely vehicle to achieving efficiency gains within animals. A meta-analysis of up to 39 scientific publications in growing cattle clearly showed that genetic variation in feed efficiency exists with a pooled heritability for residual feed intake (RFI) and feed conversion efficiency of 0.33 ± 0.01 (range of 0.07 to 0.62) and 0.23 ± 0.01 (range of 0.06 to 0.46), respectively. Heritability estimates for feed efficiency in cows was lower; a meta-analysis of up to 11 estimates revealed a heritability for gross feed efficiency and RFI of 0.06 ± and 0.04 ± 0.008, respectively. Meta-analysis of genetic correlations between feed intake, feed efficiency and other performance traits are presented and selection index theory is used to calculate the proportion of genetic variation in feed intake that can be explained by easy to measure, and often already collected data. A large proportion of the genetic variation in feed intake could be explained in both growing animals and lactating animals using up to 5 predictor traits including live-weight, growth rate, milk yield, body composition, and linear type traits reflecting body size and muscularity. Knowledge on genetic merit for feed intake can be used, along with estimates of genetic merit for energy sinks, to calculate genetic merit for feed efficiency. Therefore the marginal benefit of collecting actual feed intake data, using the genetic parameters used in this study, appears to be low. There is generally a lack of consistency across studies on regions of the bovine genome putatively associated with feed efficiency. There is now sufficient information available to develop a roadmap on how best to direct research to ensure long-term food security for a growing human population. Gaps in knowledge are identified here and possibilities to address these gaps are discussed. Key words: beef, dairy, cattle, efficiency, genetics - 2 -

3 INTRODUCTION The world human population is increasing and the demand for food in 2050 is expected to be approximately 70% greater than the demand in 2010 (FAO, 2009). Demand for meat and other livestock products is highly elastic to income (Delgado et al., 1999) and therefore as population affluence improves the demand for livestock products will increase further. Also, the global human population is aging and older people typically consume large quantities of protein than children (FAO, 2006b). The 70% increase in demand requires an annual increase in food production of 1.3% per annum. This required increase in food demand can only be met by increased efficiency of food production, both animal and crop derived. Although feed efficiency, as currently defined, is not synonymous with production efficiency, it undoubtedly has a major role to play in increasing production from an ever-decreasing food-producing land base. The global production of red meat is expected to increase from 229 million tonnes in 1999/2001 to 465 million tonnes in 2050 while milk production is expected to increase from 580 million tonnes globally to 1043 million tonnes over the same time period (Steinfield et al., 2006). This increased production, however, must be undertaken in an environmentally responsible and sustainable manner. There is considerable commentary nowadays on climate change and its implications, as well as possible mitigation strategies. Animal agriculture generates greenhouse gas emissions (GHG) as methane (CH 4 ) from enteric fermentation and manure, nitrous oxide (N 2 O) from the widespread use of nitrogenous fertilizers and animal manure, and carbon dioxide from the fossil fuels for energy usage plus land use change. Methane, however, is not only an environmental hazard but is also associated with a loss of carbon from the rumen and therefore an unproductive use of energy (Johnson and Johnson, 1995). There is wide variation in the documented calculations of animal agriculture contributions to GHG (Herrero et al., 2011). O Mara (2011) stated that animal agriculture is responsible for 8.0 to 10.8% of global GHG emissions based on calculations from the Intergovernmental Panel on Climate Change (IPCC) but if complete lifecycle analysis (i.e., accounting for the production of inputs to animal agriculture as well as change in land use such as deforestation) is undertaken this figure can be up to 18%. Cattle are the largest contributors to global GHG (O Mara, 2011). Thirty-seven industrialized countries plus the EU have signed up to the UN Kyoto Protocol and its target of reducing GHG between 2008 and 2012 by 5%, on average, from 1990 levels. The EU countries have further committed to reduce, from - 3 -

4 level, emissions by 20% by 2020 (UNFCCC, 2011). Although most discussion is on the possible impact of food production on climate change, few have considered the converse which is the impact of possible climate change on food production. Climate change is expected to result in rising global temperature, changes in patterns of precipitation, and more extreme weather events. Therefore, the animal of the future, as well as being efficient, will have to be robust to these environments. Agricultural contributions to GHG and food security are, nonetheless, not mutually exclusive. The growing food demand from an ever decreasing land base can only be met by increased system efficiencies. Feed efficiency, both gross feed efficiency and net feed efficiency, as part of the entire production system, will contribute to achieving the goal. Moreover, the conflict between food for direct human consumption versus feed for the production of animal products for human food consumption is of increasing concern (Hume et al., 2011). Galloway et al. (2007) stated that globally the conversion rate of feed to meat was 20:1 in ruminants and 3.8:1 in non-ruminants; however after adjustment for feed not directly edible by humans (e.g., grass, crop residues) this ratio changed to 3:1 and 3.4:1, respectively. Therefore, animal efficiency in ruminants is particularly important. Grasslands cover over one-third of the ice-free land (Ellis and Ramankutty, 2008; Wang and Fang, 2009) which is approximately twice the arable cropland. Despite this, most research on feed efficiency is undertaken using high concentrate diets. This review summarises the literature on feed efficiency with particular reference to the genetics of feed efficiency in beef and dairy cattle. Both growing and lactating cattle are discussed. Gaps in knowledge as well as suggestions on how to address such gaps are also outlined. DEFINTIONS OF FEED EFFICIENCY Feed efficiency has generally been defined in either: 1) growing animals, or 2) lactating animals, with most studies undertaken on the former. In cattle, most research on feed efficiency in growing animals has been undertaken on beef animals (Herd and Bishop, 2000; Arthur et al., 2001a, 2001b; Schenkel et al., 2004a; Crowley et al., 2010), although not exclusively (Korver et al., 1991; Van Arendonk et al., 1991; Nieuwhof et al., 1992; Waghorn et al., 2012; Williams et al., 2012). Most research on feed efficiency in mature cows has mainly been undertaken in dairy animals (Nieuwhof et al., 1992; Veerkamp et al., 1995; Coleman et al., 2010; Prendiville et al., 2011; Vallimont et al

5 ) although feed efficiency in mature beef cows has also been undertaken (Fan et al., 1996b; Archer et al., 2002). In other species, most research on feed efficiency is undertaken in younger growing animals (Snowder and van Vleck, 2003; Gilbert et al., 2007) although research on older mature animals also exist (Gilbert et al., 2012). To the best of our knowledge all studies to-date on feed efficiency, irrespective of the definition of feed efficiency, the productive phase of the animal, or the breed of cattle, have focused solely on a given period of time of the animal s life and not the animal s life in its entirety. While the difficulties and expense associated with collecting a phenotype such as lifetime feed intake are acknowledged, this is a serious shortcoming of all such studies since: 1) any compensatory effects after the test period are not accounted for and possible influences before the test are also not fully accounted for, and 2) the impact of selection on feed efficiency on other long-term effects such as survival are generally not considered. Long-term controlled experiments (for review of both studies see Arthur et al., 2004) have however attempted to address this shortcoming. Growing animals Most research on feed efficiency has been undertaken on growing animals. Feed efficiency variables can be loosely defined as: 1) ratio traits, or 2) regression or residual traits. The time period of measurement required to obtain accurate estimates of feed efficiency have been discussed in detail elsewhere (Archer et al., 1997; Wang et al., 2006). Archer et al. (1997) concluded that a 70-day test period (following acclimatization), with animals weighed at least every 2 weeks, was sufficient for a performance test for feed efficiency; the British breed animals included in that study were, on average, 264 days of age when starting the performance test. Archer et al. (1997) also suggested that a shorter period of intake measurement of 35 days would result in little re-ranking of animals for genetic merit of feed efficiency but did result in considerable re-ranking of animals on phenotypic efficiency. Kearney et al. (2004) reported that using automatic weighing when animals were feeding could be used to reduce the duration of the test period to 56 days without affecting the precision of estimating average daily gain (ADG). It is prudent to mention the huge importance of accurate measures of the variables included in the definition of feed efficiency as discussed in detail by Robinson (2005)

6 Ratio traits. Feed conversion ratio (FCR) was traditionally the most commonly used measure of feed efficiency in beef cattle and is generally defined as dry matter intake divided by ADG, although the converse is sometimes used (Durunna et al., 2011a; Rolfe et al., 2011) and called feed conversion efficiency. Animals with a lower FCR are reported to be more efficient. Many studies have investigated the properties of FCR (Archer et al., 1997; Van der Westhuizen et al., 2004; Hoque et al., 2006; Crowley et al., 2010). Partial efficiency of growth (PEG; Kellner, 1909) is the efficiency of growth (i.e., the ratio of weight gain to feed) after accounting for energy requirements for maintenance and may be calculated as ADG divided by average feed intake less the feed intake required for maintenance. Maintenance requirements can be estimated using feed tables (e.g. NRC, 2001) and average body weight during the measurement period. However, neither FCR nor PEG assumes differences exist among animals in maintenance efficiency, which is not necessarily true (see Archer et al., 1999); maintenance efficiency may be defined in growing animals as the ratio of body weight to feed intake at zero body weight change (Archer et al., 1999). Although not a measure of feed efficiency per se, relative growth rate (RGR; Fitzhugh and Taylor, 1971), is nonetheless a measure of efficiency. Relative growth rate is defined as growth relative to instantaneous body size and is calculated as the logarithm of live weight at the end of the test period less the logarithm of live weight at the end of the test period divided by the days in test, all multiplied by 100. Also not a measure of feed efficiency per se, the Kleiber ratio (KR; Kleiber, 1961) is defined as ADG per unit metabolic live-weight. As ADG increases for the same metabolic live-weight, more growth is obtained without increasing maintenance energy cost. While both RGR and KR are not feed efficiency measures in their own right, they can be used as measures of feed efficiency in a scenario where all animals on test are fed the same restricted diet. None of these latter three efficiency traits are widely used. Residual traits. Net feed efficiency (NFE; Exton et al., 2000), now commonly known as residual feed intake (RFI), is increasing in popularity (Berry, 2009) as a measure of feed efficiency in growing animals. Residual feed intake may be defined as the difference between actual and predicted intake. Although Koch et al. (1963) is generally credited as the original proposer of RFI, Byerly (1941), in a technical article, actually previously - 6 -

7 described such a trait in laying hens. Residual feed intake, as proposed by Koch et al. (1963), is represented as the residuals from a regression model of intake (energy intake or dry matter intake) on the various energy sinks. The regression model may be developed using a least squares regression approach. Residual feed intake may also be generated using standard feed tables (e.g. NRC, 2001) or other information sources, as undertaken by Arthur et al. (2001a), to allocate the energy demand for each of the energy sinks and subtract the total from the energy intake; this measure of RFI is sometimes termed nutritional RFI. If calculated using a least squares approach, the average RFI of the study population will be zero in line with the mathematical properties of least squares regression; this may not be the case if RFI is derived using feed tables. Furthermore, correlations between RFI and the energy sinks may exist unless RFI is generated using least squares regression. Nevertheless, irrespective of the method used to derive RFI, animals with negative RFI are deemed to be more efficient during the test period; these animals eat less than expected. Traditionally the energy sinks used in the calculation of RFI in growing cattle were ADG and (metabolic) live-weight. Metabolic live weight is usually defined as live weight mid-way through the test-period to the power of 0.75 (Nkrumah et al., 2007a; Crowley et al., 2010; Duranna et al., 2011a), although a value of 0.73 has also been used (Arthur et al., 1996; Archer et al., 2002). Because, however, of the differences among animals in the composition of ADG and the differential in energy demands of both fat and protein gain (SCA, 1990), it is now recommended to include (ultrasound measures of) fat and protein (gain) as regressor variables in the multiple regression model when deriving RFI (Baker et al., 2006; Basarab et al., 2011). This is particularly important since all else being equal, animals depositing proportionally more protein than fat for the same ADG will, on average, be more efficient and if all animals are of similar age then this may result in long-term selection for later maturing animals which may have implications for the overall efficiency of the cow herd. Crowley et al. (2010) comparing breeds of animals managed similarly and of similar age, reported poorer RFI in the early maturing breeds (Angus and Hereford) compared to the later maturing breeds (Charolais and Limousin) with the Simmental intermediate; similar results were reported in performance tested bulls in Canada (Schenkel et al., 2004a). Therefore, all calculations of RFI should account for differences in the composition of growth and this can be best achieved through ultrasound measures of body fat. Ideally however, body - 7 -

8 composition gain rather than body composition at a single time-point should be used in the multiple regression. Including ultrasound fat depth in the multiple regression model for RFI, alongside ADG and metabolic live-weight, explains only an additional 0 to 7 % of the variation in DMI compared to a model that contains both ADG plus metabolic liveweight (Basarab et al., 2003; Baker et al., 2006; Crews et al., 2006; Basarab et al., 2011; Durunna et al., 2011a; Durunna et al., 2012). A similar conclusion is evident from the weak phenotypic correlations between different measures of animal fat and RFI calculated without body composition in the multiple regression model for RFI (Robinson and Oddy, 2004; Bouquet et al., 2010; Crowley et al., 2011b). However calculations based on the information provided by Hoque et al. (2006) showed that the proportion of phenotypic variance explained by ADG plus live-weight increased from 32% to 45% when backfat was included in the multiple regression model in Japanese Black cattle; Japanese Black cattle have been heavily selected for increased fat marbling. Other energy sinks that may also be included in the RFI models include activity which can be approximated by frequency of feeding (Basarab et al., 2011; Durunna et al., 2012); only 3 to 4% additional variation in DMI is explained with feeding activity over and above that already accounted for by ADG, metabolic liveweight, and ultrasound body fat (Basarab et al., 2011; Durunna et al., 2012). A similar conclusion (2.9%) was evident when the correlation matrix between a series of traits reported by Robinson and Oddy (2004) was used to quantify the increase in the variation in feed intake explained by both feeding time and number of eating sessions over and above that explained by ADG, metabolic live-weight, plus P8 rump fat estimated by ultrasound. Koch et al. (1963) also proposed residual body-weight gain (RG) as an alternative measure of identifying inter-animal variation in feed efficiency among growing animals. Using a similar principle to that for RFI, Koch et al. (1963) defined RG as the residuals from a regression model regressing ADG on both feed intake plus (metabolic) live-weight. Unlike RFI where negative values are deemed to be more efficient animals, more positive RG values (i.e., animals growing faster than expected) are deemed to be more efficient. Although RFI is generally the feed efficiency variable of choice in recent international literature in cattle, Koch et al. (1963) in the discussion on RFI and RG, actually favoured RG over RFI because it was considered a more - 8 -

9 accurate mathematical description of cause and effect. Of course, one could also define a trait residual maintenance which uses the same mathematical principles as the derivation of RFI and RG but regresses metabolic live-weight on the other energy sinks (e.g., ADG) as well as feed intake; the biological interpretation of residual maintenance is the equivalent kilograms of live-weight the intake of the animal can support after accounting for the average requirement of its ADG (and other energy sinks). Interestingly, there appears to be no recent studies that investigated whether non-linear associations between feed intake and either live-weight or ADG existed; most studies appear to just assume linearity which may be true in populations of animals performing similarly but Koch et al. (1963) reported non-linear associations in some populations. Biologically however, non-linear associations, especially in diverse populations and/or animals with inferior genetic merit for ADG may exist; for example ADG may eventually start to plateau as feed intake reaches relatively high levels. This is however more likely to be the case in older animals where an increased proportion of feed intake is used for maintenance rather than growth. Furthermore, in the equation to predict RFI, DMI is regressed on ADG and metabolic live-weight usually assumed to be the mid-test weight estimated from regressing live-weight on day of test (Nkrumah et al., 2007a; Montanholi et al., 2009; Durunna et al., 2011a). However, the regression model: FI = μ + b ADG + b actually represents FI MWT length 0.75 μ + b1 (WTd 1 WTd ) + b 2 (WTd ) + e + d= 0 d= 0 e length = Where FI is mean feed intake during the test period, μ is the intercept of the model (which may also include a contemporary group effect; Crowley et al., 2010), ADG is average daily gain, MWT is mid-test live-weight, WT d and WT d+1 is live-weight at day d and day d+1 of test, respectively, b 1 and b 2 are the regression coefficients on the first and second energy sink component, respectively, length is the length of the test and e is the residual of the model (i.e., RFI). Although very similar, MWT 0.75 is not mathematically equivalent to: 290 length (WTd d= 0 ) length

10 However, the latter is correct and was used by Crowley et al. (2010). Others studies assume metabolic live-weight to be the mean live-weight from the start to the end of the feeding test period raised to the power of 0.75 (Herd and Bishop, 2000; Hoque et al., 2007a), but this, is likely to suffer from residual noise associated with gut fill differences between the start and end of the test. Some studies do not clearly state how they calculated metabolic live-weight in the multiple regression model for RFI. Using phenotypic correlation matrices available from the literature the proportion of the variation in feed intake (Y) explained by differences in both ADG (X 1 ) and live-weight (X 2 ) was quantified as: R r + r 2r 2 Y,X1 Y,X2 Y,X1 Y,X2 Y,X,X = rx 1,X2 r r X1,X2 301 Where R 2 Y X 1, X, 2 is the proportion of variation in feed intake explained by differences in X 1 (i.e., ADG) and X 2 (i.e., live-weight), r *1, *2 is the correlation between trait *1 and *2. Between 38% and 67% of the variation in feed intake was explained by differences in (metabolic) live-weight and ADG combined in growing cattle (Herd and Bishop, 2000; Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2006; Crowley et al., 2010; Kelly et al., 2010). Using a similar approach between 16% and 54% of the variation in growth rate in cattle was explained by differences in feed intake and (metabolic) live-weight (Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2006; Crowley et al., 2010; Kelly et al., 2010). With the exception of the study by Nieuwhof et al. (1992) live-weight alone explained 38 to 58% (Herd and Bishop, 2000; Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2007a; Bouquet et al., 2010; Kelly et al., 2010) of the phenotypic variation in feed intake, while ADG alone explained 17 to 50% (Nieuwhof et al., 1992; Herd and Bishop, 2000; Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2007a; Kelly et al., 2010) of the phenotypic variation in feed intake; the respective values reported by Nieuwhof et al. (1992) were 1% and 8%. The phenotypic coefficient of variation for feed intake in cattle varies from 9% to 14% (Nieuwhof et al., 1992; Herd and Bishop, 2000;Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2007a; Bouquet et al., 2010; Crowley et al., 2010; Kelly et al., 2010) and for growth rate varies from 7% and 19% (Nieuwhof et al., 1992; Herd and Bishop, 2000; Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2007a; Crowley et al., 2010; Kelly et al., 2010), respectively. The coefficient of variation for RFI or RG when estimated using least squares regression is undefined because the mean

11 of the residuals is zero. However when mean feed intake and mean ADG was used in the denominator for the calculation of the coefficient of variation for RFI and RG, respectively, excluding the study of Nieuwhof et al. (1992), the range in coefficient of variation for RFI was 0.04 to 0.08 while the range for RG, which was always larger than for RFI, varied from 0.10 to 0.14 (Arthur et al., 2001b; Robinson and Oddy, 2004; Hoque et al., 2007a; Bouquet et al., 2010; Crowley et al., 2010; Kelly et al., 2010); the corresponding coefficient of variation for RFI and RG in the study of Nieuwhof et al. (1992) was 0.11 and 0.06, respectively. Therefore, considerable phenotypic variation exists in both RFI and RG, but the variation in RG seems to be, on average, greater than the variation in RFI. The existence of heritable differences in both traits (discussed later) suggest that breeding for such traits in cattle will be fruitful once accurate estimates of genetic merit for these traits are available. There is nonetheless a perception that all of the variation in DMI not accounted for by the energy sinks is variation in feed efficiency and similarly variation in ADG not accounted for by differences in feed intake and (metabolic) live-weight. However, differences in contemporary groups and other systematic environmental effects (e.g., age of animal) also contribute to the observed variation. For example, Crowley et al. (2010) using a model to calculate RFI increased the proportion of variation in feed intake explained from 38% when the regression included only ADG and metabolic liveweight (calculated using the correlation matrix provided) to 72% when the regression model also included a contemporary group effect. Similarly the proportion of variation in ADG explained by differences in feed intake and metabolic live-weight increased from 16% to 51% when contemporary group was also included in the statistical model (Crowley et al., 2010). Crowley et al. (2010) did not include other systematic environmental effects in their multiple regression model for either trait but instead included them as fixed effects in the mixed models used to estimate variance components; these fixed effects accounted for more variance in the feed efficiency trait. Berry and Crowley (2012) proposed an amalgamation of both RFI and RG to generate an alternative feed efficiency trait which they called residual intake and gain (RIG). They cited the lack of a correlation between RFI and ADG as a possible reason for poor acceptance by industry of such a trait since individual animals growing slowly may rank highly on RFI, yet these animals may not be desirable. In addition, the lack of a correlation between RG and feed intake creates a similar situation. By combining both RFI and RG into RIG, the independence between RIG and metabolic live-weight was

12 maintained (i.e., mathematical properties of least squares regression approach they used) but RIG was negatively phenotypically correlated with feed intake and positively phenotypically correlated with ADG (Berry and Crowley, 2012). This therefore reduces (but does not eliminate) the likelihood of a slower growing animals ranking highly on RIG. Moreover, Berry and Crowley (2012) showed using a simple example that although animals ranking highly on RIG ate more per day than animals from the sample population ranking highly on RFI, the total quantity of feed ate during a hypothetical finishing period was lower in the superior RIG animals compared to the superior RFI or RG animals because of the dual objective of RIG to indentify, on average, faster growing animals that eat, on average, less per day. It should be noted that RIG as defined by Berry and Crowley (2012) is unit-less. Comparison of definitions of feed efficiency in growing animals. The main disadvantage of the vast majority of feed efficiency traits described (exceptions are kleiber ratio and relative growth rate which may actually be classified as comparative growth traits as opposed to feed efficiency traits per se) is the requirement for individual animal feed intake information. Accuracy of data recording for all traits is also important (Robinson, 2005) as is the case for all performance measures. The main advantage of most of the ratio traits is their ease of calculation (once the appropriate performance measures are available) and interpretation, as well as the ability to easily compare feed efficiency statistics across populations. The main disadvantages, however, of the ratio traits are: 1) an increase in the error variance as a proportion of the total variance in the statistical analysis can result, 2) strong correlations exist between the ratio trait and its component traits, and 3) no distinction is made between the energy used for separate functions. The expected responses to selection on ratio traits are difficult to ascertain (Gunsett, 1984) due to the poor statistical properties of ratio traits because of the antagonism between the desirable response in the numerator and the denominator and the unknown relative selection pressure on each. A disproportionate amount of selection pressure will be exerted on the trait in the ratio with the higher genetic variance (Sutherland, 1965); for FCR this is ADG since it generally has a higher coefficient of genetic variation. The contribution of ADG to genetic differences in FCR range from 28% to 74% (Robinson and Oddy, 2004; Hoque et al., 2006; Hoque et al., 2007a; Crowley et al., 2010) implying the FCR is strongly correlated with ADG. Therefore selection for FCR alone will result in faster

13 growing animals which can be associated with larger mature size (and thus greater maintenance requirement). However, inclusion of mature live-weight as a goal trait in a breeding objective can be used to negate against unwanted responses to selection in mature cow live-weight. Residual feed intake and RG, in contrast, if calculated using least squares regression are independent of their regressors owing to the mathematical properties of least squares regression. In general, RFI and RG are generated at the phenotypic level and are thus phenotypically independent of the regressors. However, this does not necessarily imply genetic independence (Kennedy et al., 1993) unless the heritability estimates of the feed intake and the production traits are identical and the residual covariance between the traits is equal to the genetic covariance. In general, as the genetic correlation between the regressor traits and feed intake becomes more positive, the genetic correlation between RFI and the regressor traits also becomes more positive (i.e., unfavourable) unless the residual correlation between feed intake and the regressor traits is also strongly positive. Negative genetic correlations between RFI and the regressor traits are evident when the genetic correlation between feed intake and the regressor traits is weak and the residual correlation is strongly positive. Studies have reported non-zero genetic correlations between phenotypically derived RFI and either ADG and live-weight (Bouquet et al., 2010), although most of the correlations have large associated standard errors. Although most estimates of genetic correlations between RFI derived from phenotypic regression and the regressor traits are not significantly different from zero (Herd and Bishop, 2000; Arthur et al., 2001b; Barwick et al., 2009; Crowley et al., 2010), genetic regression is also possible (Kennedy et al., 1993; Crews, 2005) to ensure zero genetic covariances between RFI and the regressors. Estimation of RFI using genetic (co)variances will avoid any possibility that an identified genetic variance in RFI is not a consequence of the genetic correlations between feed intake and the energy sinks (Veerkamp, 2002). RFI and RG-type traits can also be derived using restricted selection indexes (Eisen, 1977). A common misconception already alluded to, which is relevant to all measures of feed efficiency, but is particularly cited for RFI is that the unexplained feed intake or residual is assumed to represent true feed efficiency. In fact, this residual component can be due in part to random noise, such as measurement and prediction error, or due to inaccurate recording, feed losses, or bias in the regression coefficients for the respective regressors (Robinson, 2005). Explaining a large proportion of the variation in RFI with

14 different animal measures (Basarab et al., 2011; Robinson and Oddy, 2004; Kelly et al., 2010) has proved difficult substantiating the fact that true residual noise is contributing to differences in RFI. Furthermore, where genetic independence between RFI and the regressor traits does not exist a proportion of the heritability of RFI may simply be an artefact of picking up the genetic correlations (Veerkamp, 2002); the contribution to the heritability of RFI will be dictated by the strength of the genetic correlation between RFI and the regressor traits and can be derived using selection index methodology. To illustrate this we simulated daily dry matter intake of the 2,605 performance tested analysed by Crowley et al. (2010) from a normal distribution N(10.73,1.52) with a phenotypic correlation structure with the observed ADG and live-weight as reported by Crowley et al. (2010). Residual feed intake was calculated and its variance components estimated exactly as defined by Crowley et al. (2010). The heritability for RFI was 0.06 ±0.03. Therefore the observed heritability of RFI (and similar traits) must be interpreted with caution. One of the main disadvantages of the residual traits, compared to the aforementioned ratio traits, is the greater difficulty associated with calculating these traits. Once the data are available, ratio traits can easily be calculated but residual traits must be calculated by individuals familiar with least squares regression unless feed tables are used to determine the regression coefficients on the energy sinks. If calculated using least squares approaches, residual traits cannot be calculated for individual animals and needs several animals to provide accurate comparison. Wulfhorst et al. (2010) in their social assessment of producers perception of RFI in the US, concluded that the RFI concept is complex and not readily understood when first encountered, even for trained scientists. Furthermore, if RFI is generated using least squares regression, comparing populations without access to the original data is not possible; this can however be overcome by predicting feed intake using regression coefficients from feed tables and using these to derive RFI or by publishing the regression coefficients from the least squares analysis. Therefore, to facilitate comparison of populations, all studies should report regression coefficients from their least squares prediction models used to generate RFI and ideally the units of measurement of the traits should be standardised. Otherwise the phenotypic and genetic, where available, (co)variance components between feed intake and the energy sinks should be presented which will facilitate the calculation of different definitions of feed efficiency and estimation of associated information (e.g., regression coefficients in the multiple

15 regression model). Also the biological sensibility of the regression coefficients (i.e., comparison to feed tables) should be used, along with the associated standard errors, to test the validity of the multiple regression equation and ensure no multi-collinearity among the independent variables exists, which could become more of an issue as the number of energy sinks in the multiple regression model increase. This is particularly so in lactating animals discussed later. Lactating animals Defining feed efficiency in lactating animals is more complicated than defining feed efficiency in growing animals during the linear phase of growth because lactating cows undergo lactation cycles characterised by rapid catabolism of body reserves immediately post calving, followed by anabolism of body reserves until next calving (Berry et al., 2006b; Drennan and Berry, 2006; Roche et al., 2007a). Any proposed measure of feed efficiency in lactating animals must take cognizance of the contribution of mobilization of body reserves to the energy supply of the animal (Berry, 2009). Failure to properly account for changes in body-weight or body condition score is mathematically equivalent to energy balance (Veerkamp, 2002), a term commonly used in lactating dairy cattle (Berry et al., 2006b) which is known to be correlated with fertility and health (Beam and Butler, 1999; Collard et al., 2000). Ratio trait measures of feed efficiency used in lactating animals include milk production (either milk yield, milk composition corrected milk yield or milk solids yield) per unit intake this trait is commonly referred to as feed conversion efficiency (Nieuwhof et al., 1992; Coleman et al., 2010; Prendiville et al., 2011; Vallimont et al., 2011), milk production per kg body weight (Coleman et al., 2010; Prendiville et al., 2011), and intake per kg body weight (Coleman et al., 2010; Prendiville et al., 2011). Although not a feed efficiency trait per se the weight of calf weaned relative to weight of cow is a commonly used measure of efficiency in beef production systems. Coleman et al. (2010) in their derivation of feed efficiency in lactating Irish Holstein-Friesian dairy cows applied similar mathematical procedures in defining RFI to those previously used in growing animals but differences in body tissue mobilization were accounted for (as far as was possible). Coleman et al. (2010) defined RFI as the residuals from a least squares regression model that regressed daily dry matter intake on daily milk yield, fat yield, protein yield, lactose yield, metabolic live-weight, body condition score and change in body weight; they also included year as a fixed effect in

16 the multiple regression model to remove temporal effects. Because the energy generated from a 1-kg loss in BW is less than the energy required for a 1-kg gain in BW (O Mara, 2000), Coleman et al. (2010) applied piece-wise regression to body weight change in the RFI model to account for this differential. Others (Nieuwhof et al., 1992; Veerkamp et al., 1995; Lopez-Villalobos et al., 2008; Prendiville et al., 2011) in contrast, did not use piece-wise regression; Vallimont et al. (2011) fitted body condition score change rather than body-weight change in their multiple regression model definition of RFI but also did not use piecewise regression to model body condition score change. Vallimont et al. (2011) in their model to define RFI also included an interaction between body weight and body condition score change as well as days in milk; multi-collinearity between variables should be monitored in other populations when such a large number of correlated variables are included in a multiple regression model. Veerkamp et al. (1995) in one of their definitions of RFI also included a two-way interaction between body condition score and live-weight as well as between body condition score and liveweight change. Gilbert et al. (2012) in lactating sows defined RFI as the residuals from a multiple regression model regressing average daily feed intake from farrowing to weaning on metabolic live-weight, change in backfat thickness averaged across lactation, change in body-weight averaged across lactation, number of piglets at weaning, and change in the weight of the litter between birth and 21-days of lactation; the latter two components were to capture the energy required for milk production and could be a useful approach in the calculation of RFI in beef cattle where milk yield is not available. Irrespective of the species, when defining feed efficiency in lactating animals it is important to measure live-weight and fatness regularly to be able to appropriately account for body tissue mobilization. For example, two lactating animals starting and ending a test period with the same body weight and fatness level may actually have considerably different body tissue mobilization profiles and this has implications for their energy use kinetics. Analogous to the similarities in the definition of both RFI and RG in growing animals, Coleman et al. (2010) proposed an alternative definition of feed efficiency in lactating animals which they termed residual solids production (RSP). Instead of this trait representing the residuals of a multiple regression model regressing dry matter intake on energy sinks, RSP was represented by the residuals from a least squares model regressing milk solids production on the remaining energy sinks plus dry matter intake. Coleman et al. (2010) did not include days of gestation in their model as an energy sink

17 because of a lack of an association; this was primarily because records late in gestation, where the dependent variable may be affected by the developing foetus, were not included in their analysis. Residual solids production may be defined as the actual milk solids produced relative to expected solids production based on the individual animal s feed intake and other energy sinks (e.g., maintenance, growth) or energy sources (e.g., body tissue mobilization). In contrast to RFI, positive RSP values are indicative of more feed efficient (but not like RFI necessarily efficient animals discussed later). Comparison of definitions of feed efficiency in lactating animals. Other than the aforementioned disadvantages of ratio traits (e.g., difficult to predict response to selection and strong correlations with component traits) the ratio traits in lactating cows take no cognizance of body tissue mobilization which will influence such definitions of feed efficiency if undertaken over a relatively short period (e.g., stages of lactation). McCarthy et al. (2007a) and Roche et al. (2006) in Irish and New Zealand dairy cattle, respectively reported greater BCS loss in Holstein-Friesian dairy cattle originating from North American ancestry and of greater genetic merit for milk production. Using FCE in early lactation as a measure of feed efficiency would clearly identify such a strain of animal as the most efficient, yet these animals, because of the greater loss of BCS in early lactation and its associated deleterious consequences on fertility (Roche et al., 2007b) and health (Berry et al., 2007b), would have dramatically inferior fertility (Horan et al., 2004) and survival (McCarthy et al., 2007b) and would therefore be less profitable (McCarthy et al., 2007c) and result in a less efficient production system. RFI and RSP in lactating animals, although measuring feed efficiency per se, does not accurately reflect production efficiency. This is because the models used to calculate both residual traits do not account for the partitioning of energy into the individual components, some of which are more economically important (e.g., milk fat and protein yield) than others (metabolic live-weight). This will be discussed in detail in a subsequent section. Gaps in knowledge on definitions of feed efficiency Undoubtedly one of the greatest gaps in knowledge hindering the widespread use of feed intake or feed efficiency measures in both animal breeding and animal management is the availability of routine access to large quantities of feed intake information on individual animals. However, several methods to predict feed intake are

18 currently under research, some of which show promise. McParland et al. (2011; 2012) proposed the use of the spectra generated from mid-infrared (MIR) spectroscopy analysis of individual milk samples as predictors of energy intake (as well as energy balance) in lactating dairy cows. McParland et al. (2011; 2012) reported correlations between actual and MIR predicted energy intake in Holstein-Friesian dairy cows across two contrasting production systems of up to 0.75 when assessed using external validation. Similar accuracy of predicting daily dry matter intake was observed from near-infrared spectroscopy analysis of faecal samples in growing Angus bulls (Huntington et al., 2011). Moreover, methods such as infrared thermography (Montanholi et al., 2009), although at an early stage of research, may have some potential in explaining at least some of the variation in RFI among animals. The correlations between the average temperature of the feet of cattle estimated from a thermal image and RFI was 0.36 to 0.43; it has to be remembered that a likely substantial proportion of the variation in RFI is due to measurement error (Robinson, 2005) and therefore unity correlations are not expected. Also, the utility of group feed intakes in genetic evaluations for feed efficiency (i.e., FCR) is under investigation (Tedeschi et al., 2006; Cooper et al. 2010). Moderate phenotypic correlations have been reported between feeding behaviour and dry matter intake, RFI and FCR (Robinson and Oddy, 2004; Nkrumah et al., 2007c). Using the phenotypic correlation matrices provided in their study, 6% (Robinson and Oddy, 2004) to 25% (Nkrumah et al., 2007a) of the phenotypic variance in RFI in growing animals could be explained by daily feeding duration and feeding frequency; 12 to 14% of the phenotypic variance in dry matter intake could be explained by these two feeding behaviour traits. Automatic measurement of these two traits is relatively simple by the use of sensor technology with a sensor on the animal and a receiver at the feed bunk. The contribution of such feeding behavioural measurements to the genetic prediction of feed intake and efficiency is considerably greater and will be discussed later. Basarab et al. (2011) noted that many of the studies on feed efficiency in males occur around the period when bulls are reaching sexual maturity. Not taking cognizance of this may result in the selection of bulls that have not yet reached sexual maturity (i.e., later maturing bulls) because of the increased activity associated with young bulls following puberty and its implications on energy usage and thus feed efficiency. This substantiates further the hypothesis that selection for RFI alone may delay maturity

19 which may also result in larger mature size. However, such antagonistic responses to selection can be negated through the use of selection indexes that include traits like age at first calving and/or mature cow size with the appropriate weighting. Age at first calving and mature weight are both moderately heritable (Berry and Evans, 2012; Crowley et al. 2011a; McHugh et al., 2011) and information for age at first calving, at least, is generally available nationally, implying that it can be included in breeding objectives to counteract any unfavourable responses to selection. Nonetheless, Basarab et al. (2011) reported no association between RFI in heifers and age at puberty, weight at puberty, or the rate at which puberty was reached although the low RFI animals reached puberty at a numerically older age (353 in low RFI animals compared to 347 days in high RFI animals); the difference, however, was significant when ultrasound fat was included in the multiple regression model for the derivation of RFI corroborating the negative correlation (-0.16; P=0.06) between RFI and age at puberty in heifers reported by Shaffer et al. (2011). Also lacking is information on the repeatability and genetic correlations between feed intake and feed efficiency between nulliparae and lactating animals, or in other words the extent of genotype by environment (i.e., physiological state) interactions. Albeit from a very limited sized, yet well designed experiment, Nieuwhof et al. (1992) reported that selection for RFI in growing animals was genetically correlated (r=0.58) with RFI in lactating mature animals in early lactation; however no standard errors of the genetic correlations were provided. Moreover, although RFI was not calculated on growing bulls in the same study, Nieuwhof et al. (1992) reported a weak genetic correlation between gross feed efficiency in growing bulls and both growing heifers (0.21) and first lactation animals (0.40); bulls were fed a high energy density diet while the heifers were fed roughage. Therefore it was not possible to elucidate whether the genotype by environment (GxE) interaction was due to different diets, gender, or physiological state. Gilbert et al. (2012) reported a genetic correlation of 0.29 ±0.23 between RFI in growing pigs and lactating sows. Accurate knowledge of the genetic correlations between growing and lactating animals has implications for breeding programs, especially in dairy cattle, since selection of animals on feed efficiency during the growing phase may impact the selection intensity on the breeding goal usually derived for lactating animals, thereby possibly reducing overall genetic gain and thus profitability which is contrary to the purpose of selection on feed efficiency in the first place