Definition of feed efficiency

A simple, oft-used definition of feed efficiency is based on FCE, a trait-based ratio calculated as energy-corrected milk (ECM) per kilogram DMI (Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016). The RFI definition — in its simplest form, the difference between actual intake and expected intake based on diet composition, BW change and milk energy production — is gaining favour. Unless live weight changes (ΔLW) are accounted for, this simple definition of RFI resembles that of energy balance, which may have confounding responses to fertility and health. Residual milk yield (the converse of RFI) has also been described as an efficiency trait (Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016). Each of these definitions has advantages and disadvantages. Because of time-related fluctuations in variables, such as BW, the error associated with any of them can be reduced by extending recording periods. In the following section, we will address these issues in more detail based on findings in the literature and a phenotypic example from Jersey cow data.

Residual feed intake

Efficiency trait estimation depends upon the recording of feed intake and some traits accounting for energy sinks. Commonly, RFI calculations are done by a two-step procedure (e.g. Berry and Crowley, Reference Berry and Crowley2013), where the first step is a linear regression of DMI encompassing the factors of metabolic BW (MBW), ECM and ΔLW to account for body tissue mobilisation (equation (1)). Other definitions have also included body condition score (BCS) changes, as well as combinations of ΔLW and BCS.

(1) $${\rm DMI}{\equals} \mu {\plus} \beta _{1} \,{\rm MBW}{\plus} \beta _{2} \,{\rm ECM}{\plus} \beta _{3} \,{\rm \Delta LW}{\plus}{\rm RFI}$$

The partial regressions in equation (1) are β ₁, β ₂ and β ₃, making RFI a residual from the regression model. In the second step (equation (2)), RFI is modelled against relevant fixed effects, such as feeding regime, herd, season and parity. The random part of the model includes animal effects at two levels if cows are recorded repeatedly, where genetic relationships (COW_A) form one level and ‘permanent animal effects’ (COW_PE) form another.

(2) $${\rm RFI}{\equals} \mathop{\sum}{{\rm Fixed}\,{\rm effects}{\plus}COW_{{PE}} {\plus}COW_{A} {\plus} {\varepsilon}} $$

As a residual of equation (1), RFI has a mean of zero. The random genetic effect of cows (COW_A) and the permanent animal effect (COW_PE), which includes learnt and non-genetic effects that persist in the life of the animal, come out as random solutions from equation (2). If this model is used at the phenotypic level only, the COW_A plus COW_PE effects are joined in a single COW effect. The two steps can be joined into a unified model, as shown in equation (3).

(3) $$\eqalignno {{\rm DMI}{\equals} &\mu {\plus} \beta _{1} \,{\rm MBW}{\plus} \beta _{2} \,{\rm ECM}{\plus} \beta _{3} \,{\rm \Delta LW}\cr &{\plus} \sum {\rm Fixed}\,\,{\rm effects}{\plus}COW_{{PE}} {\plus}COW_{A} {\plus} {\varepsilon}} $$

In equation (3), all effects are estimated simultaneously, reducing effect biases and residual variances (Tempelman et al., Reference Tempelman, Spurlock, Coffey, Veerkamp, Armentano, Weigel, de Haas, Staples, Connor, Lu and VandeHaar2015). If researchers employ serial (e.g. weekly) data recorded over a period of lactation (Berry et al., Reference Berry, Coffey, Pryce, de Haas, Løvendahl, Krattenmacher, Crowley, Wang, Spurlock, Weigel, Macdonald and Veerkamp2014; Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017), the linear regression coefficients in equation (1) change systematically with lactation stage, which impacts the estimated variance components, resulting, generally, in larger errors (Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017). Factors that change with lactation stage include, in addition to co-varying fixed regressions, a set of variance components, namely the genetic (COW_A), permanent animal (COW_PE) and residual components (Liinamo et al., Reference Liinamo, Mantysaari, Lidauer and Mantysaari2015; Tempelman et al., Reference Tempelman, Spurlock, Coffey, Veerkamp, Armentano, Weigel, de Haas, Staples, Connor, Lu and VandeHaar2015; Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016; Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017). Some studies have included records from multiple parities, assuming a repeatability model to be adequate (e.g. Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016), whereas others have focussed on primiparous cows (e.g. Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017). Although Holstein cows are common in Europe and used for most RFI studies, other breeds are also in commercial use, including Red Dairy cattle and Jersey cows. However, there are few reports on feed efficiency in these breeds (e.g. Liinamo et al., Reference Liinamo, Mantysaari, Lidauer and Mantysaari2015).

Residual milk yield

The RFI approach focusses on production cost, dividing it over a range of sinks, including milk production. Shifting focus from cost to income, DMI and ECM can exchange positions in the RFI model to obtain an estimated RMY (equation (4)) (Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016):

(4) $${\rm ECM}{\equals} \mu {\plus} \gamma _{1} \,{\rm MBW}{\plus} \gamma _{2} \,{\rm DMI}{\plus} \gamma _{3} \,{\rm \Delta LW}{\plus}{\rm RMY}$$

This RMY model can be extended in a manner similar to that used with the RFI model to include cow effects and fixed effects in one unified model.

(5) $$\eqalignno {{\rm ECM}{\equals}&\mu {\plus} \gamma _{1} \,{\rm MBW}{\plus} \gamma _{2} \,{\rm DMI}{\plus} \gamma _{3} \,{\rm \Delta LW}\cr&{\plus}\sum {\rm Fixed}\,\,{\rm effects}{\plus}COW_{{PE}} {\plus}COW_{A} {\plus} {\varepsilon}$$

Note that, in (equation (5)), the partial regressions changed from β in the RFI model to γ in the RMY model, whereas the other terms remain similar. Consequently, efficient cows with negative RFI values would have positive RMY values. The interpretation of RMY becomes the deviation in milk yield at a constant feed intake, adjusted for MBW and ΔLW. Owing to the favourability of positive values, RMY is easier to comprehend than RFI.

Challenges associated with using residual feed intake and residual milk yield

The method of recording feed intake and feeding level may affect RFI and RMY results. Hurley et al. (Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016) used data from pasture-maintained Holstein cows and recorded intake by way of the indirect n-alkane method. Although this method is well documented, the grazing situation differs from that of housed total mixed ration-fed cows. International consortium findings indicate that environmental differences may be so substantial that genetic correlations with DMI deviate strongly from unity (Berry et al., Reference Berry, Coffey, Pryce, de Haas, Løvendahl, Krattenmacher, Crowley, Wang, Spurlock, Weigel, Macdonald and Veerkamp2014). The timeframe for estimation is also important. Recording windows may be shortened (to days or weeks) to enable more animals to be tested within a given facility each year. Thus, there is a need to select a timeframe likely to reflect the whole lactation-period or life-span efficiency, such as the more stable mid-lactation period (Hooven et al., Reference Hooven, Miller and Smith1972; VandeHaar et al., Reference VandeHaar, Armentano, Weigel, Spurlock, Tempelman and Veerkamp2016; Hardie et al., Reference Hardie, VandeHaar, Tempelman, Weigel, Armentano, Wiggans, Veerkamp, de Haas, Coffey, Connor, Hanigan, Staples, Wang, Dekkers and Spurlock2017).

Comparison of efficiency measures based on a study with Jersey cows

We have used recent data from Jersey cows kept at the Danish Cattle Research Centre (Foulum, Denmark) to compare efficiency measures. The calculations began with raw data, using weekly averages of traits (Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017), DMI, ECM, LW and MBW=LW^0.75. Daily ΔLW was considered more useful than LW and therefore calculated with smoothed data. Data recording began just after calving (5 days in milk (DIM)) and continued throughout lactation until drying off, or culling or 305 DIM. Cows in parity 1 and 2 with at least 32 weeks of data in the first 40 weeks post-parturition were included. All cows were fed ad libitum with partial mixed rations plus 3 kg of concentrates per day during milking with an automated milking system (see Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017 for recording procedures). Additional analyses included variables as lactation means: average DMI (AVG_DMI), average ECM (AVG_ECM) and average FCE (AVG_FCE), based on ECM/DMI ratios.

The specific models used were of the unified type, including an RFI model (6) and an RMY model (7), with each parity modelled separately.

(6) $$\eqalignno{ & {\rm DMI}_{{jkl}} {\equals} \mu {\plus}Y_{m} {\plus} S_{j} {\plus}\beta _{1} \,{\rm MBW}{\plus} \beta _{2} \,{\rm ECM}(S_{j} ){\plus} \beta _{3} \,\Delta {\rm LW}(S_{j} )\cr &{\plus} W_{l} {\plus}COW_{{kj}} {\plus}\mathop{\sum}\nolimits_{n{\equals}1}^3 {\left( \matrix{ \tau _{{1n}} \,\cos \left( {n\varphi } \right) \hfill \cr {\plus}\tau _{{2n}} \,\sin \left( {n\varphi } \right) \hfill \cr} \right){\plus} {\varepsilon}} $$

(7) $$\eqalignno{ & {\rm ECM}_{{jkl}} {\equals} \mu {\plus}Y_{m} {\plus}S_{j} {\plus}\gamma _{1} \,{\rm MBW}{\plus}\gamma _{2} \,{\rm DMI}(S_{j} ){\plus}\gamma _{3} \,{\rm \Delta LW}(S_{j} )\cr & {\plus}W_{l} {\plus}COW_{{kj}} {\plus}\mathop{\sum}\nolimits_{n{\equals}1}^3 {\left( \matrix{ \omega _{{1n}} \,\cos \left( {n\varphi } \right){\plus} \hfill \cr \omega _{{2n}} \sin \left( {n\varphi } \right) \hfill \cr} \right){\plus}{\varepsilon}{\rm }} $$

The effects are as in models (4) and (5), with added factors for recording year (Y _m ), seasons modelled as order-3 Fourier terms, where φ is the day of the year measured in radians, τ and ω are regression coefficients. Lactation curves were modelled by week of lactation (W _l ). Random cow effects (COW) were also nested within the segments to allow variance components to change during lactation. Random solutions for each cow within the segment were extracted and used to estimate correlations between and within traits. Estimates of fixed and random effects were obtained by the MIXED procedure in SAS software (SAS Institute Inc., Cary, North Carolina, USA).

An overview of the raw data is given in Table 1, with 269 first and 189second parity cow-lactations contributing to the data. Cows in second parity had a higher feed intake and larger yield, but gained less weight during the lactation period than first lactation cows.

Table 1 Feed efficiency data from first- and second-parity Jersey cows

DMI=dry matter intake; ECM=energy corrected milk yield; LW=live weight; ΔLW=daily change in live weight; MBW=metabolic BW; FCE=feed conversion efficiency.

Data in the table reflect weekly records of DMI, ECM, LW, ΔLW, MBW and FCE (FCE=ECM/DMI).

The partial regressions of DMI on ECM and ΔLW reflect the amount of DMI per kilogram ECM or per kilogram ΔLW, respectively. The coefficients changed over the trajectory of lactation in similar ways in first and second parity cows (Figure 1a and b). Similar partial regression changes were described by Li et al. (Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017) in first parity Holstein cows. The changes in regression coefficients were gradual over the lactation period for regression of DMI on ECM (Figure 1b). The regressions of DMI on ΔLW showed a peak in early lactation and another peak 6 to 8 months into lactation (Figure 1b). There is no obvious, simple reason for these unexpected peaks, which need to be studied further, especially given that they were seen in both in first and second parity cows. Regressions on MBW (and thereby also on LW) were constant over the lactation period (0.0974±0.0055 kg DMI/MBW×day) and therefore not shown in figures. Changes in regression coefficients over the lactation period may reflect interactions with mobilisation of body energy (body composition) that were ignored in the models, or may reflect digestibility changes that were also not accounted for in the models.

Figure 1 Estimated regression coefficients for dry matter intake (DMI) v. energy-corrected milk (ECM) (a) and DMI v. live weight changes (ΔLW) (b) across ten 4-week lactation periods in first-parity (_1) and second-parity (_2) Jersey cows.

Partial regressions of ECM on DMI and ΔLW during lactation (Figure 2a and b) differed clearly from those of the DMI regressions presented in Figure 1. The regression coefficients cannot be compared directly because they have different units, and the pattern of development over lactation differs for the covariates DMI and ΔLW. Although they are supposed to be mirror images, they are not exactly so, possibly due to correlations between these covariates. The regression of ECM on MBW was small and not significant.

Figure 2 Estimated regression coefficients for energy-corrected milk (ECM) v. dry matter intake (DMI) (a) and ECM v. live weight changes (ΔLW) (b) ratios across ten 4-week lactation periods in first-parity (_1) and second-parity (_2) Jersey cows.

Trait consistency over weeks in lactation was examined vis-à-vis correlations between random cow solutions from each of 10 4-week segments in first parity cows (Figure 3), and for each segment with an overall estimate per lactation (AVG). For RFI, adjacent segments correlated strongly, whereas distant segment correlations approached zero, especially between early and late lactation stages (Figure 3). However, from the fourth 4-week lactation segment onwards, correlation coefficients surpassed 0.40, and correlations with average lactation were high by the third segment.

Figure 3 Rank correlations between random animal solutions for residual feed intake (below diagonal) and residual milk yield (above diagonal) for ten 4 weeks of lactation segments in first-parity Jersey cows. Correlations between RFI and RMY in the same segment are in the diagonal. Correlations to complete lactation solutions are indicated as ‘1–40’. Colour intensity reflects correlation strength. Correlations above 0.12 or below −0.12 were significant (P<0.05).

Rank correlations between RMY solutions at different lactation stages were, in general, somewhat higher than those for RFI, albeit with similar reductions for correlations between distant segments (Figure 3). A similar pattern of results with very similar values was obtained for second parity cows (data not shown). These results indicated that treating RFI and RMY as trait characteristics based only on a single 4-week segment for each cow does not yield results that are highly consistent across short-term lactation periods, though the results correlate well with complete lactation values. These findings support previous findings in Holstein cows (Li et al., Reference Li, Berglund, Fikse, Lassen, Lidauer, Mäntysaari and Løvendahl2017), indicating that if RFI or RMY is evaluated in only part of the lactation period, it is best done in the more stable part of lactation (months 3~6) when segments correspond most closely with the entire lactation period. However, limiting subjects to the stable lactation phase may reduce contemporary group sizes due to cows being in different stages of lactation at the time of recording.

Although RFI and RMY are both used as indices of efficiency, there are divergences in the animal rankings they yield (Figure 3). All rank correlations were in the range of −0.25 to 0.25, with positive and negative correlation values being observed mainly in early and late lactation phases, respectively. Theoretically, we would expect negative correlations between RFI and RMY values. However, in second parity cows, we tended to see RFI–RMY correlation values close to zero (data not shown). These findings contrast with prior results seen in grazing Holstein cows (Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016). Notwithstanding, the lack of a strong correlation indicates that the two measures are distinct indicators of efficiency.

Summary data covering entire lactation periods may be more relevant than data limited to particular phases within the lactation period. Therefore, additional analyses were carried out with models (6) and (7), where the cow effect was assumed to be constant across the entire lactation period, and the results were aligned with the simple means of DMI, ECM and FCE (Table 2).

Table 2 Correlations among cow random solution parameters in first- and second-parity (italics) Jersey cows

DMI=dry matter intake; ECM=energy-corrected milk; FCE=feed conversion efficiency; ΔLW=live weight change; MBW=metabolic BW; RFI=residual feed intake; RMY=residual milk yield.

Analysed parameters include RFI, RMY and lactation means for DMI, ECM, FCE (FCE=ECM/DMI), ΔLW and MBW.

In the diagonal are correlations between first and second parity for the same trait (bold).

Correlations with absolute value above 0.13 were significant.

As expected, milk yield and feed intake correlated strongly in both first and second parity cows, such that higher yield and higher feed intake followed each other. The FCE ratio (ECM/DMI) correlated robustly with milk yield but not DMI, reflecting the fact that more efficient cows had higher milk yields, but not higher feed intake. Residual feed intake correlated positively with feed intake, but not with milk yield, MBW or ΔLW, suggesting that efficient (i.e. low RFI) cows eat less. In contrast, efficient RMY cows have higher milk yields and, by definition, at similar feed intake levels, higher MBW and ΔLW values, suggesting that efficient RMY cows (i.e. cows with a high RMY) produce more milk. Residual feed intake and RMY tended to correlate negatively with each other, as found by Hurley et al. (Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016), demonstrating that the choice of efficiency index can have a pronounced impact on data interpretation and the ranking of animals. It is noteworthy that ECM/DMI ratio data, a traditional measure of FCE, correlated substantially better with RMY data than with RFI data. Furthermore, FCE-associated traits, with the exception of ΔLW, were found to have moderate to high repeatability, across first and second parity cows.

The strengths of our estimated RFI–DMI correlations were similar to or higher than those reported previously for lactating Holstein cows (Connor, Reference Connor2015; Hurley et al., Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016). Our RMY–ECM correlations were strong in both first and second parity cows (Table 2), largely in agreement with the findings of Hurley et al. (Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016). Furthermore, we found that RMY correlated positively with FCE, extending the findings of Hurley et al. (Reference Hurley, López-Villalobos, McParland, Kennedy, Lewis, O’Donovan, Burke and Berry2016) to Jersey cows housed indoors and sustained on total mixed ration feed.

It is tempting to suggest that efficiency can be seen in two dimensions, namely an input and an output. If RMY is used for selection, cows with a higher yield at a fixed feed intake will be found to be the more efficient ones. If RFI is used, cows who eat less while producing a given fixed yield will be found to be the more efficient ones. In any case, the key to identifying the most efficient cows is to have comprehensive records of both milk yield and feed intake for large groups of cows, preferably over most or all lactation stages.

The presently observed close correlation between RFI and energy balance, representative of efficient cows having a more negative energy balance, might impact fertility trait selection negatively. This aspect needs to be investigated further using empirical data before solid recommendations can be made as to the implementation of an efficiency trait-based breeding programme.

Rumen microbiome and rumen fermentation

Between-animal variation in CH₄ emissions can be attributed to a variety of factors, including individual differences in rumen microbiome, digestive functions and metabolic regulation, and the interactions between these factors. Rumen microbiome effects on CH₄ emissions have been studied intensively, but the results remain inconclusive. Firkins and Yu (Reference Firkins and Yu2006) concluded that the abundance of methanogens might not be a reliable indicator of ruminant CH₄ emissions. Methanogen densities did not differ between rumen samples from high and low emitters, but archaeal community structure did differ across rumen samples associated with different CH₄ yields (Kittelmann et al., Reference Kittelmann, Pinares-Patiño, Seedorf, Kirk, Ganesh, McEwan and Janssen2014). Part of the inconsistency between CH₄ emissions and archaea abundance could be related to differences in the expression of genes involved in the methanogenic pathway. Shi et al. (Reference Shi, Moon, Leahy, Kang, Froula, Kittelmann, Fan, Deutsch, Gagic, Seedorf, Kelly, Atua, Sang, Soni, Li, Pinares-Patiño, McEwan, Janssen, Chen, Visel, Wang, Attwood and Rubin2014) found that although the abundance of methanogens and methanogenesis pathway genes were similar between low and high CH₄ emitters, the transcription of these genes was substantially higher in high emitters.

If rumen microbiomes play a major role in explaining inter-animal variations in CH₄ emissions, then such differences should be reflected in rumen fermentation patterns because molar proportions of rumen volatile fatty acids (VFAs) reflect the balance between H₂ production (acetate and butyrate) and H₂ utilisation (propionate). In a recent meta-analysis of rumen fermentation data from dairy cow-digestion studies, Cabezas-Garcia et al. (Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017) found a between-cow CV in CH₄ yield per mole VFA (calculated according to Wolin, Reference Wolin1960) of only 0.01, with low repeatability (t=0.11). Such results suggest that rumen microbiome differences are not strongly related to the end-products of fermentation and thus CH₄ emission. Pinares-Patiño et al. (Reference Pinares-Patiño, Ulyatt, Lassey, Barry and Holmes2003) found a CV for CH₄ yield of 0.098 in a study of 10 sheep with low variability in molar proportions of acetate, propionate and butyrate (0.011, 0.047 and 0.036, respectively). Small between-animal variation in rumen fermentation patterns relative to CH₄ emission variation is inconsistent with the rumen microbiome being a major factor in determining between-animal differences in CH₄ emission.

It is reasonable to expect that VFA concentrations may be related to animal factors such as rumen volume, passage rate and absorption rate, whereas VFA proportions are more likely to be influenced by the rumen microbiome than animal factors. Cabezas-Garcia et al. (Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017) found that between-cow variability and repeatability of ruminal VFA concentrations was much greater than the corresponding parameters for molar proportions of VFA, suggesting that diet and animal factors have a greater impact on CH₄ emissions than the rumen microbiome. Notwithstanding, it remains a concern that currently available microbiome characterisation methods are not adequate for determining what traits are predictive of CH₄ emission.

Animal factors

Cabezas-Garcia et al. (Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017) found that the between-cow variability and repeatability of the NDF pool (g/kg BW) of rumen and the indigestible-NDF passage rate were much greater than the variability and repeatability of variables describing rumen fermentation patterns. Smuts et al. (Reference Smuts, Meissner and Cronjé1995) reported that rumen digesta retention time data were repeatable (t=0.45). Levels of CH₄ produced appear to be strongly inversely related to the rate of digesta passage from the rumen (Pinares-Patiño et al., Reference Pinares-Patiño, Ulyatt, Lassey, Barry and Holmes2003 and Reference Pinares-Patiño, Ebrahimi, McEwan, Clark and Luo2011; Goopy et al., Reference Goopy, Donaldson, Hegarty, Vercoe, Haynes, Barnett and Oddy2014). Pinares-Patiño et al. (Reference Pinares-Patiño, Ulyatt, Lassey, Barry and Holmes2003) obtained a particulate passage rate-to-CH₄ yield correlation coefficient of −0.75, a level of CH₄-relatedness similar to that seen with feeding level. It is well known that increased feeding reduces digesta retention time and CH₄ yield (Johnson and Johnson, Reference Johnson and Johnson1995). Analyses of Yan et al. (Reference Yan, Agnew, Gordon and Porter2000) and of Ramin and Huhtanen (Reference Ramin and Huhtanen2013) showed that CH₄ yield decreases about 10% per multiple of maintenance increase in feeding level. Increasing feeding raises the digesta passage rate, which reduces diet digestibility and increases the efficiency of microbial cell synthesis partitioning of fermented carbon from gasses and VFA to microbial cells (Russell et al., Reference Russell, O’Connor, Fox, Van Soest and Sniffen1992); microbial cells are more reduced than fermented carbohydrates (Czerkawski, Reference Czerkawski1986; Van Soest, Reference Van Soest1994).

Animals as hosts for the microbiome

Modern ruminants and the diverse ecosystems of rumen microorganisms within them have been coevolving for some 50 million years. The rumen provides an anaerobic environment hospitable to microorganisms, which ferment and digest plant material into nutrients that are absorbed by the host. This commensal relationship has been highly successful in allowing ruminants and rumen microorganisms to thrive in harsh and widely distributed environments on high cellulose plant diets, which are indigestible to most other domains of life. This co-dependence is such that the ruminant host cannot survive without the microorganisms and some rumen microorganisms are not found outside of ruminants (e.g. rumen protozoa) or are highly differentiated from related microbes (e.g. methanogenic archaea) (Knapp et al., Reference Knapp, Laur, Vadas, Weiss and Tricarico2014).

The rumen microbial ecosystem, henceforth referred to as the rumen microbiome, contains many domains of life, most of which have not been cultured or annotated, but can be described by culture-independent molecular techniques, such as ribotyping, metagenomics or meta-transcriptomics (Denman and McSweeney, Reference Denman and McSweeney2014). The relative microbial domain biomass percentages have been estimated as follows: protozoa, 40% to 50%; bacteria, 26% to 42%; fungi, 1% to 20%; and archaea, 0.3% to 4.0% (Wallace et al., Reference Wallace, Rooke, Duthie, Hyslop, Ross, McKain, De Souza, Snelling, Waterhouse and Roehe2014; Tapio et al., Reference Tapio, Snelling, Strozzi and Wallace2017). From the limited annotations and cultured representatives described, it has been determined that plant structural carbohydrates are degraded to 5- and 6-carbon sugars and then further converted into VFAs for host energy metabolism, with CO₂ and H₂ by-products being absorbed primarily by bacteria, protozoa and fungi (Moss et al., Reference Moss, Jouany and Newbold2000). Fermentation efficiency is limited to a narrow range of H₂ partial pressure and pH, such that a drop in pH inhibits fermentation (McAllister and Newbold, Reference McAllister and Newbold2008). Methanogenic archaea are thought to play an important role in modulating the partial pressure of H₂ and regulating pH in the rumen by converting H₂ and CO₂ into CH₄. This interspecific hydrogen transfer is seen as a link between CH₄ production and the efficiency of fibre digestion.

Digestibility

Herd et al. (Reference Herd, Oddy and Richardson2004) suggested that diet digestibility may underlie, at least in part, RFI variation. Herd and Arthur’s (Reference Herd and Arthur2014) review paper also provided evidence showing that digestibility is negatively associated with RFI (low residual feed intake=high efficiency), indicating that more efficient animals have greater diet digestibility. To have a major contribution to variation in feed efficiency, digestibility should vary between animals and be repeatable. Based on carefully conducted digestibility experiments, Van Soest (Reference Van Soest1994) concluded that between-animal variation is about 20 g/kg with respect to starch digestibility. Part of this variation appears to be random when measurements are based on one replicate per animal. In a meta-analysis by Cabezas-Garcia et al. (Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017), between-cow variation in starch digestibility was small (starch digestibility=10 g/kg; CV=0.013). These data originated from changeover studies in which digestibility was determined mostly by the total faecal collection method. Mehtiö et al. (Reference Mehtiö, Rinne, Nyholm, Mäntysaari, Sairanen, Mäntysaari, Pitkänen and Lidauer2016) reported a between-cow organic matter digestibility (OMD) variation of 12.3 g/kg, determined using acid-insoluble ash as an internal marker. Huhtanen et al. (Reference Huhtanen, Ramin and Cabezas-Garcia2016) obtained lower variability (SD=8.8 g/kg; CV=0.012). Berry et al. (Reference Berry, Horan, O’Donovan, Buckley, Kennedy, McEvoy and Dillon2007) reported a greater variability for grazing cows when digestibility was determined using alkanes as a marker, probably because they used only one replicate per animal, and the alkane technique is less precise than intake recorded using weigh bins. Repeatability of digestibility was moderate (0.28 and 0.37) when digestibility was determined with acid-insoluble ash as an internal marker (Huhtanen et al., Reference Huhtanen, Cabezas-Garcia, Utsumi and Zimmerman2015) or a total faecal collection method (Cabezas-Garcia et al., Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017), respectively. These studies indicate that between-cow variation in digestibility alone is too small to explain the observed variation in feed efficiency.

Digestibility was positively associated with ECM production in Huhtanen et al.’s (Reference Huhtanen, Cabezas-Garcia, Utsumi and Zimmerman2015) data set when the effects of DMI, LW and DIM were used as covariates in a mixed model analysis with random effects of experiment, diet within experiment and period within experiment, as follows:

$${\rm ECM}\,\left( {{\rm kg\,/\,day}} \right){\rm {\equals}8}{\rm .82{\plus}1}{\rm .69{\times}DMI }{\minus}{\rm 0}{\rm .0148{\times}LW }{\minus}{\rm 0}{\rm .0289{\times}DIM {\plus} 0}{\rm .0249{\times}OMD}$$

When these factors are held constant, a 10 g/kg increase in OMD increases ECM production by 0.25 kg/day. The difference in digestibility corresponds to an increase in metabolisable energy (ME) intake of ~3 MJ (DMI 21 kg/day organic matter, 0.95 dry matter (DM); 1 kg digestible organic matter=16 MJ of ME). The marginal response to additional ME owing to improved digestion was 0.083 kg ECM/MJ of ME. This value is slightly lower than the typical marginal efficiency of ME utilisation for milk production (0.10 kg ECM/MJ of ME). This slightly lower value could might be associated with a greater marginal CH₄ production and increased digesta retention time in the rumen. The relationship between CH₄ yield and digestibility suggests that some 25% to 30% of incremental digestible energy can be lost as CH₄ in response to a slower passage rate.

Overall, the potential to improve feed efficiency by selecting cows for low CH₄ emission or high digestibility at a constant feed intake level appears to be limited, especially given the positive relationship between CH₄ yield and diet digestibility. Considering that differences in digesta passage rate affect both CH₄ yield and digestibility, and that digestion rate is primarily a feed characteristic, rather than an animal trait, it seems unlikely that one can increase ME supply by selecting for improved digestibility while reducing CH₄ emissions. The potential to improve digestibility is limited mainly to the cell wall fraction of feed because true digestibility of neutral detergent soluble materials is almost complete (Weisbjerg et al., Reference Weisbjerg, Hvelplund and Søegaard2004). Because the digesta passage rate is the main (perhaps only) animal factor contributing to cell wall digestibility, the potential for improving feed efficiency through genetic selection on digestibilty in high yielding dairy cows could be limited. Slowed passage will increase ruminal digesta pool size (rumen fill) and, consequently, limit intake potential.

Diet digestibility has been shown to be lower in low CH₄ emitters than in high CH₄ emitters in sheep (Pinares-Patiño et al., Reference Pinares-Patiño, Ebrahimi, McEwan, Clark and Luo2011), although one study found only a non-significant trend (Goopy et al., Reference Goopy, Donaldson, Hegarty, Vercoe, Haynes, Barnett and Oddy2014). In a modelling study, Huhtanen et al. (Reference Huhtanen, Ramin and Cabezas-Garcia2016) observed a negative relationship of digesta passage rate with CH₄ yield and digestibility, with a positive relationship between CH₄ yield and digestibility. The model simulations were based on observed variability in digesta passage rate determined by the rumen evacuation technique. The aforementioned model simulations and studies with sheep (Pinares-Patiño et al., Reference Pinares-Patiño, Ulyatt, Lassey, Barry and Holmes2003; Goopy et al., Reference Goopy, Donaldson, Hegarty, Vercoe, Haynes, Barnett and Oddy2014) indicated that CH₄ yield was positively associated with rumen digesta pool size. On average, a decrease of 1.0 g/kg DM in CH₄ yield was associated with a 10 g/kg reduction in diet digestibility (Table 3). Assuming a gross energy concentration of 18.5 MJ/kg DM, a reduction of the CH₄ yield by 1.0 g/kg DM would be expected to be associated with a decrease ME intake by 0.185 MJ (dietary gross energy concentration 18.5 MJ/kg DM), an effect three times greater than that associated with reducing CH₄ (0.055 MJ; 1 g CH₄=0.055 MJ).

Table 3 Diet digestibility and methane (CH₄) yield (g/kg dry matter intake (DMI)) in high emitter (HE) and low emitter (LE) groups, and difference (Δ) in digestibility between HE and LE groups

Positive correlations (0.66 and 0.74) have been reported between CH₄ yield and cellulose digestibility (Pinares-Patiño et al., Reference Pinares-Patiño, Ulyatt, Lassey, Barry and Holmes2003) and between CH₄ yield and NDF digestibility (Pinares-Patiño and Clark, Reference Pinares-Patiño and Clark2010). Strong positive relationships between CH₄ yield expressed as a proportion of gross energy intake and energy digestibility could be derived from individual cow data published by Schiemann et al. (Reference Schiemann, Jentsch, Hoffmann and Wittenburg1970a and Reference Schiemann, Jentsch and Wittenburg1970b; Figure 4), indicating that the relationship between CH₄ yield and digestibility is rather similar when differences in CH₄ result from differences between cows at the same intake level or differences in feeding level, with differences in passage rate being the most likely causative factor. Hence, selection of low-emitting animals may reduce NDF digestibility, a more important characteristic of ruminants in human food production over monogastrics.

Figure 4 Relationship between methane (CH₄) yield and gross energy (GE) digestibility in dairy cows fed at maintenance and lactation production levels. Adapted from Schiemann et al. (Reference Schiemann, Jentsch, Hoffmann and Wittenburg1970a and Reference Schiemann, Jentsch and Wittenburg1970b). DM=dry matter.

Energy supply is affected by methane emission

Johnson and Johnson’s (Reference Johnson and Johnson1995) finding that energy losses in CH₄ ranged from 2% to 12% of gross energy intake is often interpreted as indicating a great potential for improving efficiency of feed utilisation. However, this range represents extreme conditions, rather than variation among animals fed the same diet at the same level of intake. The lowest values were obtained from animals consuming a high concentrate ad libitum diet (>90% on a DM basis) and feedlot cattle, whereas the highest values were related to highly digestible diets fed at maintenance levels. For dairy cows eating 20 kg of DM/day (gross energy,18.5 MJ/kg DM), each cow loses about 24 MJ of energy as CH₄ each day (6.5% of gross energy intake) that otherwise could supply ME. Assuming that between-cow CV in CH₄ yield is 10% at the same feeding level, 1 SD unit reduction in CH₄ represents 2.4 MJ of ME. Using a value of 0.62 for the efficiency of ME utilisation for milk production, incremental ME supply from reduced CH₄ would support 0.48 kg/day greater ECM yield, assuming that all incremental ME is partitioned towards the mammary glands and that diet digestibility is unchanged. Although this calculation assumes that there is no relationship between CH₄ yield and digestibility, all experimental and modelling evidence indicates otherwise. It could be argued that it is possible to select animals that have low CH₄ emissions and high digestibility. However, this is not very likely since digestibility is a function of digestion (k _d ) and passage (k _p ) rates: digestibility=k _d /(k _d +k _p ). Both CH₄ yield and digestibility are reduced when k _p is increased, unless k _d is increased enough to compensate for the change in k _p . However, k _d is a feed-specific characteristic and therefore unlikely to be influenced by animal characteristics. The only feasible animal-related change that can positively influence digestibility would be increasing the buffering of rumen contents to maintain more optimal conditions for cell wall digestion. However, Cabezas-Garcia et al. (Reference Cabezas-Garcia, Krizsan, Shingfield and Huhtanen2017) found that there was no relationship between rumen pH and diet digestibility when the variations due to the diet and period effects were removed. Rumen pH was found to be a moderately repeatable trait (t=0.46).

Linking the rumen microbiome to feed efficiency and methane emission

A critical challenge in rumen microbiology has been to link mircobiome composition, in terms of the functional roles of individual species, with cow phenotypes (Bickhart and Weimer, Reference Bickhart and Weimer2017). In the case of feed efficiency, small studies have been inconsistent in identifying operational taxonomic units (OTUs) in dairy cattle (Jami and Mizrahi, Reference Jami and Mizrahi2012; Jewell et al., Reference Jewell, McCormick, Odt, Weimer and Suen2015) and sheep (Shi et al., Reference Shi, Moon, Leahy, Kang, Froula, Kittelmann, Fan, Deutsch, Gagic, Seedorf, Kelly, Atua, Sang, Soni, Li, Pinares-Patiño, McEwan, Janssen, Chen, Visel, Wang, Attwood and Rubin2014). Grouping of OTUs or meta-transcriptomics at the species/genus level, or higher levels, has been variable across studies, often with opposing (positive v. negative) correlation values (Jami and Mizrahi, Reference Jami and Mizrahi2012; Shi et al., Reference Shi, Moon, Leahy, Kang, Froula, Kittelmann, Fan, Deutsch, Gagic, Seedorf, Kelly, Atua, Sang, Soni, Li, Pinares-Patiño, McEwan, Janssen, Chen, Visel, Wang, Attwood and Rubin2014). A notable example of this challenge can be seen with respect to OTUs belonging to the genus Prevotella, a predominant rumen bacterial genera with a large variety of metabolic functions but few cultured representatives (Stevenson and Weimer, Reference Stevenson and Weimer2007). In Holstein cattle divergent for RFI as well as Holstein and Swedish red cattle divergent for CH₄ emission, a number of significant associations of Prevotella OTUs have been found across both sets of divergent groups (Jewell et al., Reference Jewell, McCormick, Odt, Weimer and Suen2015). Prevotella species need to be better characterised before it can be concluded that this genus underlies high and low divergent groups for CH₄ production and RFI.

The known methanogenic capabilities of Archaea spurred research into Archaea associations with CH₄ production (Wallace et al., Reference Wallace, Rooke, Duthie, Hyslop, Ross, McKain, De Souza, Snelling, Waterhouse and Roehe2014). It has been postulated that Archaea biomass yield should be proportional to CH₄ production because Archaea rely almost solely on methanogenesis for ATP synthesis (Wallace et al., Reference Wallace, Rooke, Duthie, Hyslop, Ross, McKain, De Souza, Snelling, Waterhouse and Roehe2014). However, reviews of recent studies have suggested that production of hydrogen and other substrates of methanogenesis drive CH₄ production (Tapio et al., Reference Tapio, Snelling, Strozzi and Wallace2017; Wallace et al., Reference Klop, Dijkstra, Dieho, Hendriks and Bannink2017). Thus, further research into other groups of rumen microbes, such as bacteria, protozoa and anaerobic fungi, is required to better characterise how the rumen microbiome relates to CH₄ production.

Potential diet–microbe interactions

The rumen can be seen as a flow-through batch fermenter wherein the substrate and flow rate impact CH₄ emissions and nutrient availability to the host through effects on the rumen microbiome. An extreme illustration of this notion can be seen in in vitro batch culture rumen fermentation studies, in which rumen fluid samples are pooled (to reduce host effects on microbes) and direct effects of additives on CH₄ production and efficiency of VFA production are assessed (Yáñez-Ruiz et al., Reference Huhtanen, Ramin and Cabezas-Garcia2016). Such in vitro studies are often followed up with in vivo studies to validate the observed effects of screened compounds. This approach has resulted in numerous dietary strategies for CH₄ mitigation based on substrate rumen microbiome interactions; this line of research is well reviewed and will not be discussed here (Hristov et al., Reference Hristov, Oh, Firkins, Dijkstra, Kebreab, Waghorn, Makkar, Adesogan, Yang, Lee, Gerber, Henderson and Tricarico2013; Knapp et al., Reference Knapp, Laur, Vadas, Weiss and Tricarico2014; Patra, Reference Patra2016). In general, these methods inhibit protozoa and, thus, associated abundant archaea serve as electron or hydrogen sinks, or are toxic to methanogens or inhibit methanogenesis directly (Patra, Reference Patra2016). These strategies have aimed mostly to reduce CH₄ emission, but in some instances have also aimed to improve feed efficiency (Hristov et al., Reference Hristov, Oh, Firkins, Dijkstra, Kebreab, Waghorn, Makkar, Adesogan, Yang, Lee, Gerber, Henderson and Tricarico2013).

Interestingly, many of these strategies appear to be temporary and not cumulative, or to have negative influences on milk production (de Haas et al., Reference Knapp, Laur, Vadas, Weiss and Tricarico2014). Microbial adaption is the primary cause of effect waning (Wallace et al., Reference Wallace, McEwan, McIntosh, Teferedegne and Newbold2002). However, these limitations can, in principle, be overcome by rotating and combining complementary additives (Klop et al., Reference Klop, Dijkstra, Dieho, Hendriks and Bannink2017).

There are exceptions, including nitrate and the patented compound 3-nitrooxypropanol, a methyl coenzyme M reductase inhibitor, which decreases CH₄ emissions by up to 30% in lactating cows without compromising feed intake or milk production (Hristov et al., Reference Hristov, Oh, Giallongo, Frederick, Harper, Weeks, Branco, Moate, Deighton, Williams, Kindermann and Duval2015). Its persistence is likely related to its narrow specificity, which reduces the risk of target microbes acquiring resistance. Thus, diet formulation, digestion rate and substrate passage rate have direct effects on rumen microbiome composition, feed efficiency and CH₄ emissions. Rumen microbes may simply be indicators of digestibility and substrate availability. Notwithstanding, the rumen microbiome can adapt to some feed additives, demonstrating that substrate–microbe interactions are not unidirectional.

Potential host–rumen interactions

Increasing evidence suggests that microbes affect the development, health and metabolism of their hosts, and that the host itself affects its own rumen microbiome. Weimer et al. (Reference Weimer, Mantovani and Man2010) exchanged ~95% of the rumen contents between cows and found that the rumen microbiome reverted almost completely back to pre-exchange characteristics, suggesting some level of host specificity of the rumen microbiome. Similar archaea:bacteria ratio differences were found among the progeny of eight animals from different breeds, suggesting that this ratio may be heritable (Roehe et al., Reference Roehe, Dewhurst, Duthie, Rooke, McKain, Ross, Hyslop, Waterhouse, Freeman, Watson and Wallace2016), in agreement with a large study of 750 Danish Holsteins in which 5%~10% of rumen 16s rRNA OTU counts were found to have significant heritability (Difford et al., Reference Difford, Lassen and Løvendahl2016b). Such findings suggest that it may be possible to breed for the relative abundance of certain rumen microbes associated with favourable phenotypes. However, while ruminants’ genomes are inherited, they are not born with a functional rumen microbiome but rather acquire it from their mothers and other animals in the environment (Jewell et al., Reference Jewell, McCormick, Odt, Weimer and Suen2015). Large cohort studies are needed to characterise genetic influences on host–microbe interactions, the results of which would also be useful for characterising and annotating rumen microbial species.

Holobionts – a holistic view of the host and its microbiome

Complex digestive system phenotypes, such as CH₄ emission or feed efficiency, are an expression of animal genetic, microbial community and environmental factors (Bickhart and Weimer, Reference Bickhart and Weimer2017). Holobiont theory holds that where the holobiont is the collective of host and associated microbiomes, it is the holobiont that is subjected to joint selection for the best performance in an environment (Rosenberg and Zilber-Rosenberg, Reference Rosenberg and Zilber-Rosenberg2011; Bordenstein and Theis, Reference Bordenstein and Theis2015). Consequently, breeding or dietary management directed at traits like feed efficiency and CH₄ emission without consideration of microbial factors are likely to be suboptimal. Alternatively, incorporating knowledge of the rumen microbiome in the selection for animals with higher than typical feed efficiency on particular diets may be key to future improvements in ruminant production.