Section 20 – Methane Emission for Genetic Evaluation

Introduction

Increases in milk production through management and genetics have substantially improved feed efficiency and decreased costs per unit of product over recent decades. However, dairy systems are also associated with environmental costs (Baskaran et al., 2009), with methane (CH₄) emissions associated with rumen microbial fermentation being both an important contributor to global greenhouse gas (GHG) emissions, as well as an avoidable loss of energy that could otherwise be directed into milk production. The livestock sector is responsible for 14.5% of the global GHG (Gerber et al., 2013); dairy cattle account for 18.9% of these emissions, mainly in the form of enteric CH₄ emissions (van Middelaar et al., 2014). Methane is a greenhouse gas with a global warming potential 28 times that of CO₂ (Myhre et al., 2013). Methane from ruminant livestock is generated during microbial fermentation in the rumen and hindgut (enteric CH₄), and from decomposition of manure. Enteric CH₄ contributes 80% of CH₄ emissions by ruminants, and manure decomposition contributes 20%. Enteric CH₄ accounts for 17% of global CH₄ emissions and 3.3% of total global greenhouse gas emissions from human activities (Knapp et al., 2014). There is, therefore, a significant research interest to find ways to reduce enteric CH₄ emissions by ruminants. Ruminant animals have a digestive system to digest plant materials efficiently. Like most mammals, ruminants lack the cellulase enzyme required to break the beta-glucose linkages in cellulose, but they play host to diverse populations of rumen microbes that can digest cellulose and other plant constituents. When rumen bacteria, protozoa and fungi ferment carbohydrates and proteins of plant materials, they produce volatile fatty acids, principally acetate, propionate and butyrate. High fibre diets favour acetate synthesis. Synthesis of acetate and butyrate are accompanied by release of metabolic hydrogen, which, if allowed to accumulate in rumen fluid, has negative effects on microbial growth, and feed digestibility (Janssen, 2010). Rumen Archaea are microorganisms that combine metabolic hydrogen with CO₂ to produce CH₄ and water. Archaea play a vital role, therefore, in protecting the rumen from excess metabolic hydrogen, and the CH₄ they produce is an inevitable product of rumen fermentation. A number of CH₄ phenotypes have been defined (Hellwing et al., 2012); the most widely used is CH₄ production (MeP) in liters or grams per day. The CH₄ production trait is highly correlated with feed intake (Basarab et al., 2013; De Haas et al., 2017) and, thereby, with the ultimate breeding goal trait: milk production in dairy cattle. The economic value of daily dry matter intake and associated methane emissions in dairy cattle showed that increasing the feed performance estimated breeding value by one unit (i.e. 1 kg of more efficiently converted DMI during the cow’s first lactation) translates to a total lifetime saving of 3.23 kg in DMI and 0.055 kg in methane (Richardson et al., 2019). Feed Performance was defined as a 1 kg increase in more efficiently used feed in a first parity lactating cow. These results show not only the relation between DMI and CH₄ production, but also the economic relationship between these traits. Persistency of lactation was found to be positively associated with increased feed efficiency and decreased methane production and intensity. Feed efficiency was associated with lower methane intensity. Feed efficiency and methane emissions can be improved by selecting for dairy cattle that are smaller and have increased persistency of lactation. Efficiency and methane emissions can be further improved by improved management of body condition score and by extending lactations beyond the conventional 305-day length (Seymour, 2019). According to Ellis et al. (2007), DMI predicted MeP with an R² of 0.64, and ME intake (MJ/d) predicted MeP with an R² of 0.53 for dairy cattle. AlternativePhenotype definitions include CH₄ intensity (MeI), which is defined as liters or grams of CH₄ per kg of milk, and CH₄ yield (MeY), which is defined as liters or grams of CH₄ per kg of dry matter intake (DMI) (Moate et al., 2016). Residual CH₄ production (RMP) is calculated as observed minus predicted CH₄ production (Herd et al., 2014, Berry et al., 2015), with predicted values based on factors such as milk production, body weight and feed intake. At the moment, it is not obvious which of these phenotypes to use; but, it is important to monitor associations between the chosen CH₄ phenotype and the other important traits in the breeding goal (e.g. production, fertility, longevity) to avoid unfavorable consequences. Berry and Crowley (2012) describe advantages and limitations of ration traits. For example, because feed efficiency traits are a linear combination of other traits it is not recommended to include them in an overall total merit index, which is a clear limitation. For all applications it is necessary to measure the CH₄ emission of each animal individually. These guidelines are intended to make the right choices for this. Whilst diet changes and feed additives can be effective mitigation strategies for CH₄ emissions (Beauchemin et al., 2009; Martin et al., 2010; Hristov et al., 2013), their effects depend on the continued use of a particular diet or additive and there have been issues with the rumen microbiomes adapting to additives. Rumen bacterial communities are highly dynamic after a diet switch and did not stabilize within 5 wk of cows grazing pasture (Bainbridge et al., 2016). In contrast, breeding for reduced CH₄ emissions should result in a permanent and cumulative reduction of emissions (Wall et al., 2010). Several studies have shown that CH₄ emissions by ruminants have a genetic component, with heritability in the range 0.20 – 0.30 (de Haas et al., 2011; Donoghue et al., 2013; Pinares-Patiño et al., 2013, Kandel et al., 2014A, B; Lassen and Lovendahl, 2016; López-Paredes et al. 2020). Breeding for reduced CH₄ emissions, alone or together with other mitigation strategies, could therefore be effective in reducing the environmental impact of cattle farming and, possibly, also in increasing feed efficiency. Such a breeding scheme would require, as a fundamental starting point, accurate measures of individual CH₄ emissions on a large scale. Several techniques have been developed for the measurement of CH₄ emissions from ruminants, with varying degrees of accuracy (see reviews by Cassandro et al., 2013 and Hammond et al., 2016A), but routine individual measurements on a large scale (a requisite for genetic selection) have proven to be difficult to obtain and expensive to measure (Pickering et al., 2015; Negussie et al., 2016). Therefore, identifying proxies (i.e. indicators or indirect traits) that are correlated to CH₄ emissions, but which are easy and relatively lowcost to record on a large scale, would be a welcome alternative. Proxies might be less accurate but could be measured repeatedly to reduce random noise and in much larger populations. These guidelines are highly indebted to Garnsworthy et al. (2019). In this paper the methods to measure CH₄ are compared with special emphasis to the genetic evaluation of dairy cattle.

Disclaimer

The fact that specific device manufacturers are mentioned in these guidelines is in no way an endorsement of the devices or their accuracy by ICAR.

Scope

A variety of technologies are being developed and employed to measure CH₄ emissions of individual dairy cattle under various environmental conditions, as is evidenced by frequent reviews (Storm et al., 2012; Cassandro et al., 2013; Hammond et al., 2016A; de Haas et al., 2017). The first objective of the current guidelines is to review and compare the suitability of methods for large-scale measurements of CH₄ output of individual animals, which may be combined with other databases for genetic evaluations. Comparisons include assessing the accuracy, precision and correlation between methods. Combining datasets from different countries and research centres could be a successful strategy for making genetic progress in this difficult to measure trait if the methods are correlated (de Haas et al., 2017). Accuracy and precision of methods are important. Data from different sources need to be appropriately weighted or adjusted when combined, so any methods can be combined if they are suitably correlated with the ‘true’ value. The second objective of the current guidelines, therefore, is to examine correlations among results obtained by different methods, ultimately leading to an estimate of confidence limits for selecting individual animals that are high or low emitters (see also Garnsworthy et al., 2019).

Methane determining factors

Diet and rumen microbiota

Table 1 contains a list of dietary or microbiota factors that determine CH₄ production.

Table 1. Methane determining factors related to diet and rumen microbiota.
Factors	Reference
The main determinants of daily methane production are dry matter intake and diet composition: the more feed consumed, and/or the greater the fibre content of the diet, the more methane is produced per day. However, per unit of DMI, and per unit of fat+protein yield the grass diet produced less enteric CH4 per cow than the TMR diet. Nutritional approaches for methane mitigation include reducing the forage to concentrate ratio of diets, increasing dietary oil content, and dietary inclusion of rumen modifiers and methane inhibitors.	Beauchemin et al., 2009; Cottle et al., 2011; Knapp et al., 2014; O’Neill et al., 2011; Sauvant et al., 2011
Methane output per kg of product is affected mainly by cow milk yield or growth rate, and by herd-level factors, such as fertility, disease incidence and replacement rate.	Garnsworthy, 2004
Methane output varies considerably between individual animals. For animals fed the same feed, the between-animal coefficient of variation (CV) in methane was 8.1%.	Blaxter and Clapperton, 1965
The amount of digestible nutrients consumed especially of the carbohydrate fraction (starch, sugar, N-free residuals) is reliable to estimate CH4 release with high precision. Furthermore, diets rich in fat reduced CH4 formation in the rumen.	Jentsch et al., 2007
DMI was also the most important determining factor, but there were different regression lines for maize silage and dried grass as the main roughage component respectively: $CH4\ (g)=93+16.8\times DMI(kg)$ $CH4\ (g)=81+14.0\times DMI(kg)$ Methane release was particularly dependent on the intake of crude fiber (CF) and ether extract (EE): $CH4\ (g)=63+80xCF\ (kg)+11xNFE\ (kg)+19xCP(kg)-195xEE\ (kg)$	Kirchgessner et al., 1991
Methane linearly increased with NDF intake $CH4\ (L)=59.4\times NDF[kg]+64.6$ for cows together with their calves independent of the breed.	Estermann et al., 2002
Enteric CH4 could be predicted with the equation: $CH4\ (g/d)=84+47\times cellulose(kg/d)+32\times starch(kg/d)+62\times sugars\ (kg/d)$	Hindrichsen et al., 2005
The higher the percentage concentrate the lower Ym.	Zeitz et al., 2012
Additives can sometimes have a methane reducing effect: higher dosages mitigate methane more. Saponins mitigate methanogenesis by reducing the number of protozoa, whereas condensed tannins act both by reducing the number of protozoa and by a direct toxic effect on methanogens.	Beauchemin et al., 2008; Jayanegara et al., 2012; Zmora et al., 2012; Cieslak et al., 2013; Guyader et al., 2014
Plant essential oils have been shown as promising feed additives to mitigate CH4 and ammonia emission, but results were inconsistent.	Cobellis et al., 2016; Moate et al., 2011
Nitrate and sulphate addition decreased the enteric methane emissions negatively affecting diet digestibility and milk production. The effects of the salts are additive.	van Zijderveld et al., 2010; van Zijderveld et al., 2011
The methanogenesis in the rumen of calves is associated with the development of the ruminal protozoa population. The absence of protozoa in the rumen reduced both the CH4 production and the digestibility of carbohydrates.	Schönhusen et al., 2003
Implementing good grazing management reduced gross energy intake loss as CH4 by 14%.	Wims et al., 2010

Host genetics, physiology and environment

A low-moderate proportion of variation in CH₄ emissions among ruminants is under genetic control. Heritability coefficients of MeY and RMPR were h²=0.22 and 0.19 respectively in a population of 1,043 Angus growing steers and heifers measured during 2 days in RC (Donoghue et al., 2016). The heritability coefficient of MeY was h²=0.13 in a population of1,225 dual-purpose growing sheep measured during 2 days in RC (Pinares-Patino et al., 2013). Table 2 contains information of heritability of traits related to CH₄ production.

Table 2. Heritability information of methane-related traits and measurements.
Factors	Reference
List with several h2	Pickering et al., 2015
List with several h2	MPWG White paper Dec 18
Methane emissions from individual cows during milking varied between individuals with the same milk yield and fed the same diet. Between-cow variation in MERm is greater than within-cow variation and ranking of cows for CH4 emissions is consistent across time. Variation related to body weight, milk yield, parity, and week of lactation/days in milk. The monitored variation might offer opportunities for genetic selection.	Garnsworthy et al., 2011A; Garnsworthy et al., 2011B
Mechanistic modelling approach: potential for dietary intervention as a means of substantially reducing CH4 emissions without adverse effects on dietary energy supply.	Mills et al., 2001
The CH4-to-CO2 ratio measured using the non-invasive portable air sampler and analyzer unit based on Fourier transform infrared (FTIR) detection method is an asset of the individual cow and may be useful in both management and genetic evaluations.	Lassen et al., 2012
The estimated heritability for CH4 g/day and CH4 g/kg of FPCM were lower than common production traits but would still be useful in breeding programs.	Kandel et al., 2013
Genetic correlation between CH4 intensity and milk yield (MY) was - 0.67 and with milk protein yield (PY) was -0.46 in Holstein cows.	Kandel et al., 2014A, B
Milk production and CH4 emissions of dairy cows seemed to be influenced by the temperature humidity index.	Vanrobays et al., 2013A
Estimate the heritability of the estimated methane emissions from 485 Polish Holstein-Friesian dairy cows at 2 commercial farms using FTIR spectroscopy during milking in an automated milking system by implementing the random regression method. The heritability level fluctuated over the course of lactation, starting at 0.23 (SE 0.12) and then increasing to its maximum value of 0.3 (SE 0.08) at 212 DIM and ending at the level of 0.27 ± 0.12. Average heritability was 0.27 ± 0.09.	Pszczola et al., 2017
CH4 measured with a portable air-sampler FTIR detection method on 3,121 Holstein dairy cows from 20 herds using automatic milking systems. The heritability of CH4_MILK was 0.21 ± 0.06. It was concluded that a high genetic potential for milk production will also mean a high genetic potential for CH4 production. The results suggested that CH4 emission is partly under genetic control, that it is possible to decrease CH4 emission from dairy cattle through selection, and that selection for higher milk yield will lead to higher genetic merit for CH4 emission/cow per day.	Lassen and Løvendahl, 2016
CH4 production was measured of 184 Holstein-Friesian cows in. the milking robot with a in total 2,456 observations for CH4 production. Heritability for CH4 production ranged from 0.12 ± 0.16 to 0.45 ± 0.11, and genetic correlations with MY ranged from 0.49 ± 0.12 to 0.54 ± 0.26. The positive genetic correlation between CH4 production and milk yield indicates that care needs to be taken when genetically selecting for lower CH4 production, to avoid a decrease in MY at the animal level. However, this study shows that CH4 production is moderately heritable and therefore progress through genetic selection is possible.	Breider et al., 2019
CH4 concentration was measured with NDIR, and CH4 production was estimated from CH4 concentration and body weight. Heritability for CH4 concentration was 0.11 ± 0.03 and for CH4 production 0.12 ± 0.04. Positive genetic correlation was observed with MY (0.17-0.21), PY (0.22-0.31) and FY (0.27-0.29). Other type traits showed positive correlation with methane production (chest width=0.26, angularity =0.19, stature = 0.43 and capacity = 0.31) possibly associated to higher milk feed intake from these animals. Rumination time was negatively correlated to CH4 production (-0.24) and CH4 concentration (-0.43). However, larger CH4 production and CH4 concentration was associated with shorter days open.	López-Paredes et al. (2020)
Genetic parameters of CH4 emissions predicted from milk fatty acid profile (FA) and those of their predictors in 1,091 Brown Swiss cows reared on 85 farms showed that enteric CH4 emissions of dairy cows can be estimated on the basis of milk fatty acid profile. Additive genetic variation of CH4 traits was shown which could be exploited in breeding programmes.	Bittante and Cecchinato, 2020
A total of 670 test day records were recorded on lactating Holstein Friesian cows reared in 10 commercial dairy herds. Predicted methane production (PMP) was estimated to be 15.33±1.52 MJ/d in dairy cows with 23.53±6.81 kg/d of milk yeild (MY) and 3.57±0.68% of fat content (FC). Heritability of MY was 0.09 with a posterior probability for values of h2 greater than 0.10 of 44%. Estimates of heritability for FC and protein content (PC) were 0.17 and 0.34, respectively, with a posterior probability for values of h2 greater than 0.10 of 77% and 99%. For somatic cell score (SCS), heritability was 0.13 with a posterior probability for values of h2 greater than 0.10 of 67%. Heritability for the trait PMP was moderate to low (0.12); however, posterior probability for values of h2 greater than 0.10 was 60%. Medians of the posterior distributions of genetic correlations between PMP and milk production traits were: 0.92, 0.67, 0.14, and 0.14 between PMP and MY, PMP and FC, PMP and PC, and PMP and SCS, respectively. Reduction of PMP seems to be viable through selection strategies without affecting udder health and PC.	Cassandro et al., 2010
GWAS to study the genetic architecture of CH4 production and detected genomic regions affecting CH4 production. Detected regions explained only a small proportion of the heritable variance. Potential QTL regions affecting CH4 production were located within QTLs related to feed efficiency, milk-related traits, body size and health status. Five candidate genes were found: CYP51A1 on BTA 4, PPP1R16B on BTA 13, and NTHL1, TSC2, and PKD1 on BTA 25. These candidate genes were involved in a number of metabolic processes that are possibly related to CH4 production. One of the most promising candidate genes (PKD1) was related to the development of the digestive tract. The results indicate that CH4 production is a highly polygenic trait.	Pszczola et al., 2018
A 1000-cow study across European countries revealed that the ruminant microbiomes can be controlled by the host animal. A 39- member subset of the core microbiome formed hubs in co-occurrence networks linking microbiome structure to host genetics and phenotype (CH4 emissions, rumen and blood metabolites, and milk production efficiency).	Wallace et al., 2019

Methane measurements methods

Several factors influence the choice of measurement method such as cost, level of accuracy, precision, scope of application, and scale, which vary across disciplines (Cassandro et al., 2013; Hammond et al., 2016A; Garnsworthy et al., 2019). For instance, genetic selection programs require CH₄ measurements on thousands of related individuals under the environmental conditions in which the animals are expected to perform (Falconer and Mackay, 1996). This can be challenging because dairy cattle perform in a wide range of conditions (e.g. grazing vs indoor housing). There are a number of different measurement methods currently being employed, each with advantages and disadvantages in terms of the factors listed above. The currently accepted and widely used measurement methods are listed and described below. The main features of methods for measuring CH₄ output by individual animals are summarised in Table 3. Values for each feature are based on experience of experts in METHAGENE WG2 who have used the methods. All values are relative, and somewhat subjective, because absolute values will depend on installation and implementation of each method at different research centres. It should be noted that the measuring methods can be divided in two major sections: methods that measure the concentration and flux of CH₄ (e.g. the respiration chamber), and methods that measure the flux of CH₄ through the device (e.g. GreenFeed). This affects the useability of the methods for answering research questions – please see also the recommendations at the end of these guidelines.

• Respiration chamber
• Portable Accumulation Chambers
• SF6
• Breath sampling during milk and feeding
• Sniffer SOP
• Greenfeed
• Greenfeed SOP
• Laser Methane Detector

Table 3. Summary of the main features of methods for measuring CH4 output by individual animals.
Method	Purchase cost	Running costs	Labour	Repeatability	Behaviour alteration	Throughput
Respiration chamber	High	High	High	High	High	Low
SF6 technique	Medium	High	High	Medium	Medium	Medium
Breath sampling during milking and feeding	Low	Low	Low	Medium	None	High
GreenFeed	Medium	Medium	Medium	Medium	Medium	Medium
Laser methane detector	Low	Low	High	Low	Low-Medium	Medium

Discussion of methods

SF6 vs Respiration Chamber

For large-scale evaluation of CH₄ emissions by individual animals, the SF6 technique is more useful than respiration chambers. Animal behaviour and intake might be affected by wearing the apparatus, and by daily handling to exchange canisters, but the technique is considerably less intrusive than respiration chambers because cows remain in the herd. Labour and monetary costs for changing canisters each day and for lab analysis are high. Throughput is limited by the number of sets of apparatus available, handling facilities, labour, and the capacity of the lab for gas analysis. Animals need to be measured for 5 to 7 days, and it is recommended that group size should be less than 15 animals (Berndt et al., 2014), so maximum throughput would be about 750 animals per year. The method may be better suited for in housed conditions because of the labour and the potential movement restriction of the animals due to wearing the apparatus.

Breath sampling during milking and feeding vs Respiration Chamber

For large-scale evaluation of CH₄ emissions by individual animals, breath-sampling methods have significant advantages compared with other methods. Breath-sampling methods are non-invasive because, once installed, animals are unaware of the equipment and are in their normal environment. Animals follow their normal routine, which includes milking and feeding, so no training of animals, handling, or change of diet is required. Equipment is relatively cheap, although more expensive gas analysers are available, and running costs are negligible.

The compromise for non-invasiveness of breath-sampling is that concentrations of gases in the sampled air are influenced by cow head position relative to the sampling tube (Huhtanen et al., 2015). The use of head position sensors and data filtering algorithms can remove the effects when the cow’s head is completely out of the feed bin (Difford et al., 2016), but not within the feed bin. Consequently, sniffer measurements are more variable than flux methods, with factors like variable air flow in the barn increasing measurement error (imprecision), and head position, a highly repeatable character, inflating between-cow variability.

Using CO₂ as a tracer gas partly addresses the issue but, because CO₂ arises from metabolism as well as rumen fermentation, variability of CO₂ emissions has to be considered. A further consideration is diurnal variation in breath concentrations of CH₄ and CO₂ because animals are spot-sampled at different times of day and night. Diurnal variation can be accounted for either by fitting a model derived from the whole group of animals, or by including time of measurement in the statistical model (Lassen et al., 2012).

The number of observations per analyser is limited only by number of cows assigned to one automatic milking station or concentrate feeding station and length of time equipment is installed. Typically, each analyser will record 40 to 70 animals 2 to 7 times per day for 7 to 10 days, although the number of sampling stations per analyser can be increased by using an automatic switching system (Pszczola et al., 2017). Throughput per analyser is likely to be 2,000 to 3,000 animals per year.

NDIR vs LMD

Both methods are low invasive. LMD needs larger labor force, wheras NDIR can be used during milking and feeding. According to Rey at al. (2019), the repeatability of the CH₄ concentration was greater for NDIR (0.42) than for LMD (0.23). Correlation between methods was moderately high and positive for CH₄ concentration (0.73 and 0.74,respectively) and number of peaks (0.72 and 0.72, respectively), and the repeated measures correlation and the individual-level correlation were high (0.98 and 0.94, respectively). A high coefficient of individual agreement for the CH₄ concentration (0.83) and the number of peaks (0.77) were observed between methods. The study suggests that methane concentration measurements obtained from NDIR and LMD cannot be used interchangeably. But the use of both methods could be considered for genetic selection purposes or for mitigation strategies only if sources of disagreement, which result in different between subject and within-subject variabilities, are identified and corrected for.

Greenfeed

A limitation of the GreenFeed system is that animals require training to use the system, although animals which have been trained to use the system will readily use it again (Velazco et al., 2014). However, some animals will not use the system or will use it infrequently, and frequency of visits is affected by diet (Hammond et al., 2016B). This can be a challenge when screening commercial herds for CH₄ emission under genetic evaluation. On the other hand, animals seem to get used to the equipment rapidly, and the sound produced by the system is remembered by the animals easily (personal information Dr. Finocchiaro). Alternatively, as practised in Canada, the unit is moved to individual animals in a tie-stall setting multiple times a day (personal information Prof C.F. Baes). Thus, action of individual animals is not needed. The manufacturer recommends 15 to 25 animals per GreenFeed unit, and recordings are made typically for 7 days. If all animals visit the unit adequately, throughput per unit is likely to be 750 to 1,250 animals per year. Sebek et al. (2019A, B) and Bannink et al. (2018) showed the usefulness of the GreenFeed method in an on farm setting.

Laser Methane Detector

The LMD can be used in the animal’s normal environment, although for consistency restraint is required during measurement. Because the LMD measures CH₄ in the plume originating from the animal’s nostrils, results can be affected by factors such as: distance from the animal; pointing angle; animal’s head orientation and head movement; air movement and temperature in the barn; adjacent animals; and operator variation (Sorg et al., 2017). Operator variation is likely to be one of the biggest factors because the operator controls distance and pointing angle, and is responsible for ensuring the laser remains on target. The structure of the barn and the resulting ventilation conditions and wind speed at the location of the measurement are also considerable sources of variation in recorded CH₄. Assuming operator fatigue does not limit measurements, each LMD could record up to 10 animals per hour. If each animal is recorded 3 times (on 3 consecutive days, for example, as in Mühlbach et al. (2018)), throughput is likely to be up to 1000 animals per year.

Comparison of methods to measure methane

Correlations among methods

Correlations among methods Table 4 shows correlations between the respiratory chamber method as the gold standard to measure CH₄ emission from cows and other methods from Garnsworthy et al. (2019).

Table 4. Correlations between CH4 measuring methods. Data were taken from Garnsworthy et al. (2019).
Method	Correlation	S.E.
Respiratory chamber - SF6	0.87	- 0.08
Respiratory chamber - Greenfeed	0.81	- 0.1
Respiratory chamber - NDIR	- 0.07	0.88
Respiratory chamber - NDIR peak	0.72	- 0.11
Respiratory chamber - PAIR	- 0.08	0.7
SF6 - Greenfeed	0.4	- 0.8
LMD - Greenfeed	0.77	- 0.23
NDIR - Greenfeed	0.64	- 0.18
NDIR - LMD	0.6	- 0.11
FTIR - LMD	0.57	- 0.25
NDIR - NDIR peaks	0.58	- 0.15
FTIR - NDIR	0.97	- 0.02

In method comparison studies, simultaneous repeated measures per cow with two or more methods are required in order to assess systematic differences between methods (means) and random differences (precision) and correlation between methods free of residual error. Furthermore, adequately short time differences between repeated measures per subject are needed to ensure the underlying biology of the cow has not changed. Not all methods can be recorded simultaneously and CH₄ emission of cows’ changes both within day and over the lactation period. In such instances either cross-over designs or matched pair repeated measures designs are needed. Members of METHAGENE WG2 provided data from studies in which two or more methods had been used to measure CH₄ output (g/day) by individual dairy cows. Methods were applied to each cow either concurrently or consecutively within a short timeframe.

Seven main methods were represented: respiration chambers; SF6; GreenFeed; LMD; and three breath-sampling systems based on different gas analysers. Gas analysers incorporated different technologies to measure CH₄, which were NDIR (e.g. Guardian Plus, Edinburgh Instruments, Edinburgh, UK), FTIR (e.g. Gasmet 4030, Gasmet Technologies Oy, Helsinki, Finland), or PAIR (e.g. F10, Gasera Ltd, Turku, Finland). In the contributing studies, NDIR and FTIR were used in automatic milking stations, and PAIR was used in concentrate feeding stations. One NDIR study and all FTIR and PAIR studies used CO₂ as a tracer gas, with daily CO₂ output calculated either from milk yield, live weight and days pregnant or from metabolisable energy intake. Two NDIR studies were based on CH₄ concentration in eructation peaks rather than mean CH₄ concentration, so were treated as separate methods. By separating NDIR studies, a total of 8 distinct methods were available giving a matrix of 28 potential combinations for comparisons. Data were available for 13 method combinations (Garnsworthy et al., 2019).

Method comparisons were conducted using bivariate models (repeatability animal models) to obtain correlations between ‘true values’, also known as repeated measures correlations or individual level correlations (Bakdash and Marusich, 2017). Variance components including between cow variation and within cow variation (precision) and means (accuracy) were used in the calculation of between cow coefficient of variation (CV, %) and total CV and repeatability. Where single measurements were available for each method Pearson’s correlation was reported and where repeated measures per subject were available repeated measures correlation was reported.

Respiration chambers were the most precise method, as can be seen by the smaller between cow CV% and total CV compared to alternative methods, and respiration chambers are by definition the most accurate. All methods tested showed high correlations with respiration chambers but none of the correlations exceeded 0.90. This is in part due to the increased imprecision of alternative methods, as even the most accurate and precise method will compare poorly to a less precise method. These correlations are also likely to be underestimated because none of the methods could be recorded simultaneously with respiration chambers and had to be recorded in cross over designs. Consequently, the true value for each cow may have changed due to changes in the underlying biology of the cow over time between measurements. Comparisons among alternative methods generally had lower correlations than comparisons with respiration chambers, despite having relatively higher numbers of animals and in most cases simultaneous or near simultaneous repeated measures per cow per method due to the increased variability and imprecision of alternative methods as is seen by the increased CVs or due to the possibility that different aspects of CH₄ emission are captured using different methods.

For the methods with repeated measures per cow the two mass flux methods, SF6 and GreenFeed, had the highest repeated measures correlations (0.87 ± 0.08 and 0.81 ± 0.10) which outperformed the concentration based NDIR method using CO₂ tracer gas. Of the two concentration methods evaluated against respiration chambers using single measurements, NDIR Peaks had a higher correlation (0.89 ± 0.07) than the PAIR CO₂ tracer gas (0.80 ± 0.10). The study of Hristov et al. (2016) comparing SF6 and GreenFeed reported a low Pearson correlation of 0.40, despite having a large number of animals with repeated measures per method, the authors appear not to have estimated a repeated measures correlation, which could be larger. Estimating a repeated measures correlation between these two mass flux methods is a priority as it would clarify the inexplicable disagreement between two methods which both correlate highly with the gold standard method. With the exception of the aforementioned study, the imprecision was low in the mass flux measure comparisons as compared to the concentration-based methods.

Two of the sniffer methods evaluated, FTIR CO₂t1 and NDIR CO₂t1, correlated close to unity (0.97), most likely due to the shared prediction equation for CO₂ tracer gas. Nevertheless, all correlations derived from actual data were positive. This suggests that combination of datasets obtained with different methods is a realistic proposition for genetic studies. Calculation of adjustment or weighting factors for bias, accuracy and precision would improve the value of combined datasets.

Pro’s and con’s of devices

Daily methane emission measures

Due to the large diurnal variation in enteric CH₄ emission in relation with feeding pattern (Grainger et al., 2007; Jonker et al. 2014), the highest accuracy of daily CH₄ production rate (DMPR) will be obtained with methods that encompass the whole day emissions. Two methods are available: Respiration Chambers (RC) and SF6 methods.

Alternative methods are based on short-term measures of CH₄ production rate: Portable Accumulation Chambers (PAC) for sheep and GreenFeed Emission Monitoring (GEM) systems for cattle and sheep (Hegarty, 2013).

DMPR with Respiration Chamber (RC)

It should be noted that CH₄ emissions recorded in RC also include gases from flatulence in addition to eructed and expired CH₄. Compared with mouth exhaled CH₄, CH₄ from flatulence is generally considered as limited.

Feed intake in the RC may not be representative of the normal animal feed intake (Bickell et al., 2014; Llonch et al., 2016; Troy et al., 2016). As a consequence, the DMPR measured could be biased. Animals are usually not fed ad libitum when recorded in RC. It is therefore recommended to compare animal or diet effects on Methane Yield (MY) calculated as the ratio of the observed DMPR/DMI during the RC recording in order to take into account possible differences among animals in DMI bias. Animal effects can also be compared on the Residual Methane Production Rate (RMPR) the difference between the observed DMPR and the expected DMPR obtained by regression of observed DMPR on DMI recorded during RC test. Residual traits, however, require a large number of recorded animals for valid adjustment.

Repeatability coefficients between measures taken on consecutive days are very high, rep=0.85 [0.75 to 0.94] for MeY and RMPR of cattle and sheep (Grainger et al., 2007; Donoghue et al., 2016; Pinares-Patino et al., 2013). It has been concluded that 1-day measurement duration could be recommended as it will have a limited impact, less than 5%, on the efficiency of selection of MeY as compared to a selection on a 2-day measurement duration.

When repeated measures of CH₄ emission of sheep are taken few days to two weeks apart the repeatability coefficients of MeY and RMPR drops to rep=0.36 [0.26 to 0.41] on average (Pinares-Patino et al., 2013; Robinson et al., 2014a). Interestingly, repeatability maintains at a moderate level, rep=0.27 [0.23 to 0.53], when animals were measured several months or even years apart. Similar results were found in Angus cattle, rep=0.20, between MeY and RMPR measures taken more than 60 days apart (Donoghue et al., 2016).

Conclusions and reccomendations

All these results show that animal effects exist on daily CH₄ emissions and animal differences are partially under genetic determinism. This trait, as any other physiology trait, is subject to number of environmental effects and to evolution with time. Ranking animals on their CH₄ emission requires standardization of the testing environment. Although highly precise, a single measure recorded in RC is not sufficient for characterizing an animals emission aptitude. In order to characterize a long term phenotype it is therefore recommended to record several 1-day measures, each a few weeks apart, instead of one single 2-day measure, keeping the testing environment as constant as possible.

DMPR with GEM

At each visit CH₄ and CO₂ fluxes are measured and animal emission rates are obtained by averaging the short-term flux measures recorded during the testing period. In a review of published results (Dorich et al., 2015; Hammond et al., 2015; Velazco et al., 2016) Hammond et al. (2016A) concluded that the GEM system provides similar DMPR values as the RC or SF6 methods. Similar accuracy was found by Arbre et al. (2016) for CH₄ yield measured with GEM as compared with RC and SF6 measures.

The spot measures are highly variable since they include, in addition to the animal and environment effects, an important within-animal and within-day variance. The latter is considered as an error term. Consequently, the precision of the animal estimates increase with the number of spot measures averaged per animal. From the results reported by Renand and Maupetit (2016) with 124 beef heifers controlled indoors, it can be shown that the coefficient of variation of that error term (CV_e) decreases exponentially with the number of spot measures: 13.7%, 10.8%, 7.9% and 4.9% with 5, 10, 25 and 100 measures respectively. Results reported by Arbre et al. (2016) with 7 lactating dairy cows controlled indoors, also show that CV_e decreases from 12.8% to 11.4%, 9.5% and 6.8% when the number of measures increases from 5 to 10, 25 and 100. With dairy cows at pasture, Waghorn et al. (2016) showed that the coefficient of variation among 36 dairy cows at pasture was half (6.6 and 7.5%) when CH₄ production rate was averaged over 16 days with approximately 18 to 26 measures per cow, as compared with 4 day averages with 4 to 6 measures per cow (13.0 and 17.2%). These authors concluded that at least 16 days are required to give confident estimates.

With 45 to 50 spot measures recorded during 2 weeks Arbre et al. (2016) and Renand and Maupetit (2016) obtained repeatabilityof 0.78 and 0.73 for DMPR estimates of 7 dairy cows and 124 beef heifers, respectively. A similar repeatability coefficient (0.74) was obtained by Huhtanen et al. (2015) with 25 dairy cows recorded during 3 weeks, with 20 to 30 samples per cow. Interestingly, these latter authors fitted gas concentration, airflow and head position measurement equipments into two automatic milking systems that were used to measure CH₄ emission of 59 dairy cows during two periods of 10 days. After filtering data for acceptable head-position, the repeatability of DMPR was 0.75.

Considering the need to average enough spot measures and the advantage of measuring DMPR over long periods to take into account the emission variability with time, the GEM system should be run over several weeks. Averaging 40 to 50 spot measures per animal should provide a precise measure of the animal DMPR. The minimum duration of CH₄ recording will depend on the number of spot measures actually recorded per day.

The GEM system relies on animals that voluntarily visit the GEM unit when attracted with pellets dispensed by a feeder at a controlled rate. The visitation frequency appears to be highly variable among different studies reported up to now. While some experiments report a very high frequency of cattle visiting the GEM units (up to 96%), the proportion of not visiting animals may be very high in other studies (up to 60%) (Dorich et al., 2015; Hammond et al, 2015A, Arbre et al., 2016; Renand and Maupetit, 2016; Velazco et al., 2016; Waghorn et al., 2016). The reason why some animals may not visit the unit is not obvious. That problem of no or low visiting frequency may jeopardize the precise ranking of animals on their DMRP. Training them is an important requisite for the success of DMPR recording with the GEM system (see recommendations on the C-Lock website). Palatability of the pellets used to attract the cattle should be high compared with the diet they receive in the trough or the grass they are grazing.

In addition to the effect on precision, the low visiting frequency may have an impact on accuracy if associated in some animals with specific time of visiting. Enteric CH₄ emissions have a diurnal variation with a minimum at the end of night, before the first feeding, and a steady increase after each feeding. A weak diurnal pattern in CH₄ emission was detected by Velazco et al. (2016) using GEM systems. Renand et al. (2013) observed significant differences between visit hours (CV=10%). If some animals visit the GEM at specific hours of the day, the rough average of spot measures will be biased. In order to get rid of this time effect on the DMPR measure, Dorich et al. (2015) and Hristov et al. (2016) came up with a protocol where the GEM units were moved sequentially from one cow to the next one over several days, so that all the cows were equally measured during different hours of the day. That protocol is possible only with tie stall cattle and is obviously not applicable for measuring large number of animals. However, with animals controlled in their production environment, the bias generated by potential specific visiting patterns can actually be removed if the measuring hour is taken into account in the linear model when estimating the animal effect.

As voluntary visiting of the GEM system may be a limiting factor under some conditions, measures of DMPR can be designed when animals are drinking or eating, i.e. several times per day. Velazco et al. (2016) showed that a GEM water unit prototype designed and built by C-Lock Inc., displayed different eructation patterns as compared with a plain GEM unit. They concluded that further development appears necessary before any application. Troy et al. (2016) tested a CH₄ hood (MH) system placed above an automated feeding bin. That system includes an air extraction fan for each hood with continuously recorded airflow. Methane concentration was measured using 4 infrared analyzers, one for 8 hoods. In this system one CH₄ concentration value was recorded every 6 min. With 9 to 12 feeding events per day on average and feeding visits averaging 8 min, there were between 12 to 16 CH4 concentration values recorded and CH₄ production rates calculated per day. The measurements were recorded during 46 days and ranking of animals in relation with the test duration was studied. However no repeatability coefficient was given for comparison with other methods. That system was compared with respiratory chambers results in two experiments with 82 and 80 steers fed different diet-treatment combinations. Over the whole experimental design, a good concordance was found between MH and RC results as a consequence that both methods detected similar effects for the diet-treatment effects. However no correlation was given between both methods within diet-treatment samples that are the essential information needed to evaluate the ability of this new method to predict individual DMPR.

Conclusions and reccomendations

With only a single gas analyzer for 8 feed bins, the time when useful CH₄ concentration is recorded is certainly too short for including several eructation peaks. Fitting one gas analyzer per feed bin will combine advantages of the measurement time during visits of the GEM system with the visiting frequency allowed by the MH system.

MPR with PAC

The delay between the measurement and the last feeding has to be carefully monitored and taken into account when calculating animal emission values. As individual DMI is difficult to record, direct measurement of CH₄ yield (MY=MPR/DMI) turns out to be impossible. Although not representative of a whole day production rate, that method can be used to characterize individual CH₄ emission rates if standardized protocols are applied. It was first validated with 40 ewes measured 1 hour in PAC after three 22-hour measures in RC: a correlation of 0.71 was found between the two measures of CH₄ production rate over 1 or 22 hours (Goopy et al., 2011). The 1-hour CH₄ production measure in PAC has a moderate repeatability of rep=0.50 [0.37 to 0.60] when taken few days to seven weeks apart (Robinson et al., 2015; Goopy et al., 2016). Heritability coefficient of this 1-hour CH₄ production measure is estimated to h²=0.12 in a population of 2,279 sheep (Robinson et al., 2014b) with a repeatability coefficient rep=0.25.

Conclusions and reccomendations

The authors recommend using the mean of 3 PAC measurements in order to get accurate phenotype estimates.

Proxies

Introduction

Large-scale measurements of enteric CH₄ emissions from dairy cows are needed for effective monitoring of strategies to reduce the carbon footprint of milk production, as well as for incorporation of CH₄ emissions into breeding programs. However, measurements on a sufficiently large scale are difficult and expensive. Proxies for CH₄ emissions can provide an alternative, but each approach has limitations. Negussie et al. (2019) recently showed the potential of proxies proxies that are easy to record in the farm. These proxies can be gathered in most farms and are a realistic threshold accuracy that can be obtained without more fancy proxies. Several techniques have been developed for the measurement of CH₄ emissions from ruminants, with varying degrees of accuracy (see reviews by Cassandro et al., 2013 and Hammond et al., 2016A), but routine individual measurements on a large scale (a requisite for genetic selection) have proven to be difficult and expensive (Pickering et al., 2015; Negussie et al., 2016). Therefore, identifying proxies (i.e. indicators or indirect traits) that are correlated to CH₄ emission, but which are easy and relatively low-cost to record on a large scale, is a much needed alternative. Proxies might be less accurate, but could be measured repeatedly to reduce random noise. The (potential) proxies range from simple and low-cost measurements such as body weight, to high-throughput milk MIR, to more demanding measures like rumen morphology, rumen metabolites or microbiome profiling.

Combining proxies that are easy to measure and cheap to record could provide predictions of CH₄ emissions that are sufficiently accurate for selection and management of cows with low CH₄ emissions.

Available Proxies

A large array of CH₄ proxies differing widely in accuracy and applicability under different conditions have been reported. The ideal proxy would be highly phenotypically and genetically correlated with CH₄ emissions and could easily, and potentially repeatedly, be measured on a large scale. A systematic summary and assessment of existing knowledge is needed for the identification of robust and accurate CH₄ proxies for future use. Table 5 summarizes proxies for CH₄ production, and Table 6 summarizes results from combining proxies to improve predictability of proxies for CH₄ prediction.

Table 5. Available methane proxies include: (1) feed intake and feeding behaviour, (2) rumen function, metabolites and microbiome, (3) milk production and composition, (4) hind-gut and faeces, and (5) measurements at the level of the whole animal. It is evident that no single proxy offers a good solution in terms of all of these attributes, though the low cost and high throughput make milk MIR a good candidate for further work on refining methods, improving calibrations and exploring combinations with other proxies.
Proxy	Description / conclusion	Reference
(1) Feed intake and feeding behavior
Dry Matter Intake (DMI)	DMI predict MeP with R2= 0.06-0.64, and ME intake predict MeP with R2= 0.53-0,55	Ellis et al. (2007); Mills et al. (2003); Negussie et al. (2019)
Gross Energy intake (GE)	Predict MeP with RMSPE= 3.01.	Moraes et al. (2014)
Feeding behavior	Magnitude and direction of relation to MeP varies across studies	Nkrumah et al. (2006); Jonker et al., 2014
Rumination time	High rumination relates to more milk, consume more concentrate and produce more CH4, lower RMP and MeI	Watt et al. (2015); López- Paredes et al. (2020)
Rumen microbiome	The metagenome can predict DMI, and classify high vs low intakes	Delgado et al. (2019)
(2) Rumen function, metabolites and microbiome
Dietary antimethanogenic compounds	Inhibitors of the enzyme methyl coenzyme-M reductase: bromochoromethane; chloroform; 3- nitrooxypropanol (not always)	Denman et al., 2007; Knight et al., 2011; Haisan et al., 2014; Romero-Perez et al., 2014
Dietary antimicrobial compounds	Induce reductions in both MeP and methanogens numbers: nitrates, anacardic acid (cashew nut shell liquid), monensin, isobutyrate	Iwamoto et al., 2002; Kubo et al., 1993; van Zijderveld et al., 2010; Veneman et al., 2015; Shinkai et al., 2012; Wang et al., 2015
Rumen microbiome profile	High Fibrobacteres, Quinella ovalis and Veillonellaceae and low Ruminococcaceae, Lachnospiraceae and Clostridiales associate with low- CH4 phenotypes and high propionate Protozoa concentration	Kittelmann et al., 2014; Wallace et al., 2014; Sun et al., 2015; Guyader et al., 2014
Rumen microbiome profile	Predict MeP with R2 up to 0.55	Ross et al. 2013a; Ross et al. (2013b)
Microbial genes	20 (out of 3970 identified) related to CH4 emissions	Roehe et al. (2016)
Rumen volume (Xray Computed Tomography) and retention time	Low-MeY sheep had smaller rumens. Faster passage= less time to ferment substrate - explained 28% of variation in MeP	Pinares Patiño et al., 2003; Goopy et al., 2014; Okine et al. (1989)
Blood triiodothyronine concentration	Reduced MeY	Barnett et al. (2012)
Acetate to propionate ratio in ruminal fluid	Positively associated with CH4 emissions, but not confirmed in all studies, sometimes opposite relation	Mohammed et al., 2011; Fievez et al., 2012; Chung et al., 2011; Van Zijderveld et al., 2010
(3) Milk production and composition
Modelling approach	Doubling milk production only adds 5 kg to the MeP and so greatly reduces MeY	Kirchgessner et al. (1995); Hristov et al. (2014)
Milk fat content	key explanatory variable for predicting CH4: A moderate negative genetic correlation with infrared predicted MeI: correlations MeP = 0,08 and MeI = - 0.13	Moraes et al. (2014); Kandel et al., 2014A, B; Vanlierde et al. (2015)
Milk fat content	A positive relationship between VFA proportions and methanogenesis is expected as a consequence of the common biochemical pathways; Dietary unsaturated fatty acids are negatively associated with CH4 emissions	Vlaeminck et al., 2006; Van Lingen et al., 2014
Milk protein yield	Correlation with Mel = - 0.47 or -0.09, MeP = 0.53	Kandel et al. (2014); Vanlierde et al. (2015)
Lactose	Variable correlations: MeP = 0,33; MeI = - 0.21; R = 0.19 for CH4 emission	Miettinen and Huhtanen (1996); Dehareng et al. (2012)
Somatic cell score	Genetic correlation with infrared predicted MeI: R = 0.07	Kandel et al. (2014A, B)
Prediction equations Milk FA and CH4 emissions, including from MIR data	R² ranged between 47 and 95%; relationships between the individual milk FA and MeP differed considerably and the correlations between CH4 and milk FA vary throughout the lactation	Chilliard et al. (2009); Delfosse et al. (2010); Castro-Montoya et al. (2011); Dijkstra et al. (2011); Kandel et al. (2013); Mohammed et al. (2011); Van Lingen et al. (2014); Williams et al. (2014); Dijkstra et al. (2016); Rico et al. (2016); Van Gastelen and Dijkstra (2016); Vanrobays et al. (2016); Bougoin et al., (2019)
(4) Hind-gut and feces
Whole tract digestibility (potential as supporting factors in the prediction of enteric CH4 emissions)	Main effects relate to rumen (see above), but energy digestibility as a supporting factor to GE intake improved the accuracy of CH4 prediction, despite the fact that there was no direct linear relationship between energy digestibility and MeY and in % of GE intake	Yan et al., 2009 C
Ratio of acetic and butyric acid divided by propionic acid	Methane yield positive relation	Moss et al., 2000
(5) Whole animal measurements
Body weight and conformation	Prediction models; primary predictor for enteric MeP	Moraes et al. (2014); Holter and Young, 1992; Yan et al., 2009
Body weight	Relationship with MeI: r = 0.44; relationship between body weight and rumen capacity	Antunes-Fernandes et al. (2016); Demment and Van Soest, 1985
Body weight	Key explanatory variable for enteric MeP	No reference available
Conformation traits: affects enteric MeP	Indicators for rumen volume (via feed intake and rumen passage rates); BCS	Agnew and Yan, 2000
Lactation stage	Complementary proxy	Vanlierde et al. (2015)

Proxies discussion

The greatest limitation of proxies today is the lack of robustness in their general applicability. Future efforts should therefore be directed towards developing combinations of proxies that are robust and applicable across diverse production systems and environments. Here we present the present status of the knowledge of proxies and their predictive value for CH₄ emission. Proxies related to body weight or milk yield and composition are relatively simple, low-cost, high throughput, and are easy to implement in practice. In particular, DMI and milk MIR, along with covariates such as lactation stage, are a promising option for prediction of CH₄ emission in dairy cows. No single proxy was found to accurately predict CH₄, whilst combinations of two or more proxies are likely to be a better solution. Combining proxies can increase the accuracy of predictions by up to 15 - 35%, mainly because different proxies describe independent sources of variation in CH₄ and one proxy can correct for shortcomings in the other(s). One plausible strategy could be to increase animal productive efficiency whilst reducing CH₄ emissions per animal. This could be achieved by reducing MeY and/or decreasing DMI provided that there is no concomitant reduction in productivity or increase in feed consumption (Pickering et al., 2015).

Combining diet-based measurements with other proxies for methane emissions

Feed intake appears a reasonably adequate predictor of MeP: generally, heavier animals have higher maintenance requirements, so eat more and produce more CH₄. However, a substantial level of variation is left unaccounted for. This suggests that more detailed information on dietary composition is needed. This is also important when one wants to account for MeP on diets of similar DMI but of different nutrient profiles.

The prediction accuracy of MeP strongly depends on the accuracy of quantifying the VFA produced in the rumen (Alemu et al., 2011). The type of VFA formed during rumen fermentation depends on the type of substrate fermented (Bannink et al., 2011), such as the dietary content of neutral detergent fiber and starch. The type of substrate fermented thus appears a useful factor for predicting MeP (Ellis et al., 2007), indicating that including a description of variation in dietary quality caused by nutritional factors results in improved prediction accuracy of CH₄ emission (Ellis et al., 2010; Moraes et al., 2014).

Rumen

When feed intake is kept constant, a higher rumen capacity results in a lower passage rate (Demment and Van Soest, 1985), resulting in a higher MeP (Moraes et al., 2014). Proxies based on rumen samples are generally poor to moderately accurate predictors of CH₄, and are costly and difficult to measure routinely on-farm. VFA are a proxy for rumen CH₄ emissions. Using rumen fermentation data obtained from in vitro gas production, Moss et al. (2000) reported a negative linear relationship between CH₄ production and the ratio of (acetic + butyric acid)/propionic acid. However, by combining different information sources, either related to feed intake or to the impact of feed intake on the VFA composition, a better proxy with an improved accuracy can be achieved. This way, the prediction equation for CH₄ production can be optimized (higher accuracy).

The relationship between rumen methanogen abundance and methanogenesis is less clear when changes in enteric CH₄ emissions are modulated by diet or are a consequence of selecting phenotypes related to feed efficiency or MeY. Whereas in some reports there was a significant positive relationship (Aguinaga Casanas et al., 2015; Arndt et al., 2015; Sun et al., 2015; Wallace et al., 2015), in many others the concentration of methanogens was unrelated to methanogenesis (Morgavi et al., 2012; Kittelmann et al., 2014; Shi et al., 2014; Bouchard et al., 2015). Bouchard et al. (2015) even reported a reduction in methanogens withoutsignificant decrease in MeP for steers fed sainfoin silage. Sheep selected for high or low MeY showed no differences in methanogen abundance, though there was a strong correlation with expression of archaeal genes involved in methanogenesis (Shi et al., 2014).

Hindgut and Feces: whole tract digestibility variables cannot serve as primary predictors for enteric MeP in cattle or sheep, but might be used as supporting factors to improve the accuracy of prediction of CH₄ output.

Protozoa and other rumen microbes

Protozoa are net producers of H₂ and their absence from the rumen is associated with an average reduction in enteric MeP of approximately 11% (Hegarty, 1999; Morgavi et al., 2010; Newbold et al., 2015). Using a database of 28 experiments and 91 dietary treatments, Guyader et al. (2014) showed a significant decrease of 8.14 g CH₄/kg DMI for each log unit reduction in rumen protozoal abundance. About 21% of experiments within this dataset reported CH₄ changes unrelated to protozoal abundance, highlighting the multifactorial nature of methanogenesis. Roehe et al. (2016) observed that the ranking of sire groups for CH₄ emissions measured with respiration chambers was the same as that for ranking on archaea/bacteria ratio, providing further evidence that host control of archaeal abundance contributes to genetic variation in CH₄ emissions - at least in some circumstances. Across a wide geographical range, the methanogenic archaea were shown to be highly conserved across the world (Henderson et al., 2015). This universality and limited diversity could make it possible to mitigate CH₄ emissions by developing strategies that target the few dominant methanogens. However, one clear limitation of metagenomic predictions compared to genomic predictions was that the microbiome of the host is variable - that is, it may change in response to diet or other environmental factors over time, whereas the hosts DNA remains constant.

Rumen microbial genes

These included genes involved in the first and last steps of methanogenesis: formylmethanofuran dehydrogenase subunit B (fmdB) and methyl-coenzyme M reductase alpha subunit (mcrA), which were 170 times more abundant in high CH₄ emitting cattle. Whilst gene-centric metagenomics is not low-cost or high-throughput, these results point to potential future proxy approaches using low-cost gene chips.

The difference in gene expression activity as opposed to abundance was also reported by others (Popova et al., 2011). However, there are also studies in which there was no relationship with gene expression (Aguinaga Casanas et al., 2015). There are some methodological and experimental differences that might explain some of the apparent contradictions, such as the type of gene target and primers used for nucleic acid amplification. Effects are seen most clearly when the difference in MeP between groups of animals is large (e.g. Wallace et al. (2015) used treatments that generated a 1.9-fold difference CH₄ emissions).

Proxies based on measurements in milk

Milk yield alone does not provide a good prediction of MeP by dairy cows. Yan et al. (2010) indicated that CH₄ as a proportion of GE intake or milk energy output was negatively related to milk production. It is less clear if MeY can be predicted from milk yield when making comparisons across studies.

Milk MIR spectroscopy is relatively inexpensive, rapid and already routinely used technology in milk recording systems to predict fat, protein, lactose and urea contents in dairy milk to assist farm management decisions and breeding. It can be used as a promising strategy to exploit the link between enteric CH₄ emission from ruminants and microbial digestion in the rumen by assessing the signature of digestion in milk composition. Milk MIR data can be obtained through regular milk recording schemes, as well as, on a herd level, through analysis used for milk payment systems. Diverse milk phenotypes can be obtained by MIR spectrometry – including detailed milk composition (e.g. FA as reported by Soyeurt et al., 2011), technological properties of milk, and cow physiological status (De Marchi et al., 2014; Gengler et al., 2016). Several of these novel traits (i.e. FA composition) have been identified as potential indicators of CH₄ emission. Therefore, using MIR to predict MeP (Dehareng et al. 2012; Vanlierde et al. 2013, 2015; Van Gastelen and Dijkstra, 2016) is also a logical extension of its use to quantify the major milk components (i.e. fat, protein, casein, lactose, and urea) and minor components (e.g. FA). Dehareng et al. (2012) assessed the feasibility to predict individual MeP from dairy cows using milk MIR spectra. Their initial results suggest that this approach could be useful to predict MeP at the farm or regional scale, as well as to identify low-CH₄ emitting cows. According to Van Gastelen and Dijkstra (2016), MIR spectroscopy has the disadvantage that it has a moderate predictive power for CH₄ emission, both direct and indirect (i.e. via milk FA), and that it lacks the ability to predict important milk FA for CH₄ prediction. They concluded that it may not be sufficient to predict MeP based on MIR alone. It is, however, possible to improve the accuracy of prediction through the combination of MIR with some animal characteristics such as lactation stage (Vanlierde et al., 2015). The advantage of this latter development is that this type of prediction can be done on a very large scale inside a routine milk recording system (Vanlierde et al., 2015).

Proxy: future developments and perspectives

There is currently limited consensus on which phenotype to use to lower the carbon footprint of milk production through genetic selection. This could be MeP, MeI or MeY. The direct goal would be CH₄ production; the relationship with milk production and/or feed intake could be accounted for by including these in the final selection index or scheme. However, one might argue that it would be more effective/accurate to directly use milk production- or feed intakecorrected CH₄ (e.g. CH₄ intensity or yield) as breeding goal.

The analysis of proxies in terms of their attributes shows that proxies that are based on samples from the rumen or related to rumen sources are poor to moderately accurate predictors of CH₄. In addition, these proxies are too costly and difficult for routine on-farm implementation. On the other hand, proxies related to BW, milk yield and composition (e.g. milk FA) are moderately to highly accurate predictors of CH₄ and relatively simple, low-cost and easier to implement in practice (Cassandro et al.,2010; Cassandro, 2013). Particularly, milk MIR and the prediction of CH₄ based on milk MIR along with other covariates such as lactation stage is a promising alternative: that is accurate, cheaper and easy to be implemented in routine milk analysis at no extra cost.

Therefore, in the future advances in infrared, photoacoustic and related technologies will push the boundaries, particularly in focusing on developments of fast and portable technologies. Such developments will lead to better and promising proxies for CH₄ that will enable a sizable throughput of CH₄ phenotypes in dairy cows. Antunes-Fernandes et al. (2016) already presented the use of metabolomics on milk to better understand the biological pathways involved in CH₄ production in dairy cattle. The techniques used in that study are not suitable for large scale measurements, but rapid developments in omics may offer tests and assay methodologies on blood, urine or milk samples that will provide an additional tool for developing new / additional proxies for CH₄ emissions in dairy cattle.

Merging and sharing data in genetic evaluations

Early 2016 an attempt to make cross country evaluations of CH₄ emissions from Holstein dairy cattle was initiated. The work was based on data from NL, DK, AUS, UK and IR. In total, 12,820 weekly CH₄ emission records from 2,857 cows were available. Although different equipment was used across countries to measure CH₄ emissions, the research aimed to define similar CH₄ output phenotypes in each country. The analysed CH₄ traits, that are available in each country, are (1) CH₄ production in g/d, and (2) CH₄ intensity in g/d per kg fat protein corrected milk (FPCM). In addition to these CH₄ traits, CH₄ concentration (in ppm) was available in Denmark, the Netherlands and UK, and the ratio between CH₄ and CO₂ concentration was available in Denmark and the Netherlands.

Bivariate analyses were carried out to estimate genetic correlations between countries, using an animal linear mixed model for all traits. Both univariate and bivariate analyses were repeated with the GRM as well. With all weekly records, standardizing the trait in the full dataset increased the heritability for CH₄ production from 0.03 to 0.06. The heritability for CH₄ intensity was slightly higher. The highest heritability with the full dataset is estimated for the standardized CH₄ concentration (0.19). Correlations estimated among CH₄ traits estimated with either the pedigree or the GRM were in same direction and of similar magnitude. The genetic correlations show that when CH₄ production increased, the CH₄ concentration and the ratio between CH₄ and CO₂ increased as well.

The approach is novel, and no other attempt has been performed to make genetic analysis of CH₄ traits across countries. The analysis can be repeated in future studies where more data hopefully will be available, and more effort can be made into improving both the fixed and random part of the model.

Recommendations

The most important question: what method to use if you need to measure CH₄? The answer may be: it depends on what you like to do. In the Table 6 we summarize some experimental conditions and designs, and make recommendations.

Table 6. Recommendations for measuring methane in diverse experimental conditions and designs.
Experimental condition and design	Methane measurement method recommendation
Need to measure absolute methane values – animal numbers and location not important	Respiration chamber; SF6; GreenFeed
Need to rank animals from low to high methane emission	Sniffer method
Need to measure methane on farm	Sniffer method; GreenFeed; PAC
Low budget measurements needed	Proxy; Proxies measurement
High animal numbers required	Sniffer method; Proxies measurement; LMD

Conclusions

Measuring CH₄ emission on large numbers of cows is a challenge. The high costs and low throughput of RC restrict their use to research studies measuring CH₄ emissions on small numbers of individual animals. Respiration chambers remain the gold standard method, but benchmarking alternative methods against RC is challenging because simultaneous replicate measures per cow are not feasible. Methods like SF6 and GreenFeed require lower capital investment and running costs than RC, and have higher throughput and potential for use in extensive and grazing situations, but costs are still prohibitive for recording large numbers of animals. Methods based on concentration are less precise and accurate than flux methods, but they are viable for large scale measurement, which is a prerequisite of genetic evaluations. Further development is needed to increase accuracy and precision of concentration methods. Several reviews of methods for measuring CH₄ have made qualitative judgements based on individual comparison studies without expanding scope to genetic evaluations and considering repeated measure correlations between methods as proxies for genetic correlations. Results confirm that there is sufficient correlation between methods for all to be combined for international genetic studies and provide a much needed framework for comparing genetic correlations between methods should these be made available. Proxies have the potential to be used as predictors of CH₄ production and emission. Although proxies are less accurate than direct CH₄ measurements they can be easier, cheaper, and at high throughput, and may be therefore the best method in practical situations, especially proxies related to milk measurements. Therefore, these proxies at the population level, can provide useful information at genetic improvement that can be used to reduce emissions following 3 ways: (1) intensification of animal production; (2) improving of system efficiency and (3) the direct reduction of GHG emissions by breeding for reduced predicting animals that are high or low GHG emitters.

Summary of Changes

Date of change	Nature of change
March 2020	Draft from Feed & Gas WG put into standard template for ICAR Guidelines. Separate out EDGP database to become a standalone appendix.
April 2020	Edits and acknowledgements added by Feed & Gas WG.
May 2020	Approved by ICAR Board on 26th May subject to addition of disclaimer. Disclaimer added as new chapter 2 - the fact specific device manufacturers are mentioned in these guidelines is in no way an endrosement of the devices or their accuracy by ICAR.
December 2023	Creation of Methane Emission for Genetic Evaluation Wiki Page.