Variance component estimation and genome-wide association of predicted methane production in crossbred beef steers

Abstract

Enteric methane is a potent greenhouse gas and represents an escape of energy from the ruminant digestive system. Additive genetic variation in methane production suggests that genetic selection offers an opportunity to diminish enteric methane emissions. Logistic and monetary difficulties in directly measuring methane emissions can make genetic evaluation on an indicator trait such as predicted methane production a more appealing option, and inclusion of genotyping data can result in greater genetic progress. Three predicted methane production traits were calculated for 830 crossbred steers fed in seven groups. The methane prediction equations used included mathematical models from Ellis et al. (2007), Mills et al. (2003), and IPCC (2019). Pearson correlations between the traits were all greater than 0.99, indicating that each prediction equation behaved similarly. Further, the Spearman correlations between the estimated breeding values for each trait were also 0.99, which suggests any of the predicted methane models could be used without substantially changing the ranking of the selection candidates. The heritabilities of Ellis, Mills, and IPCC predicted methane production were 0.60, 0.62, and 0.59, respectively. A genome-wide association study identified one single nucleotide polymorphism (SNP) that reached the threshold for significance for all of the traits on chromosome 7 related to oxidoreductase activity. Additionally, the SNP slightly below the significance threshold indicate genes related to collagen, intracellular microtubules, and DNA transcription may play a role in predicted methane production or its component traits.

Predicted methane traits are heritable and relatively simple to calculate, making them potential candidates for selecting animals for reduced methane production.

Introduction

Ruminant animals eructate enteric methane as byproduct of their digestive process. Methane is a greenhouse gas that has 84 times more global warming potential than carbon-dioxide over a 20-yr timeframe (Myhre et al., 2013). Further, enteric methane represents a loss of energy from the ruminant digestive system. Anywhere from 2% to 12% of gross energy (GE) consumed by cattle is emitted as methane (Johnson and Johnson, 1995).

Phenotypic and genetic variation in methane production between animals has been reported (Blaxter and Clapperton, 1965; Donoghue et al., 2016; Herd et al., 2016; Renand et al., 2019), which implies reduction of methane emissions through breeding is possible. Breeding for methane reduction offers several benefits that other methods such as feed additives (Beauchemin et al., 2020) do not. Namely, genetic improvements are permanent and cumulative if selection pressure on the trait is maintained. Further, it has the potential to reduce methane production in the cow calf sector, the sector that generates the greatest quantity of methane (Rotz et al., 2019). The drawback for selecting for animals solely based on methane production is methane production is genetically and phenotypically correlated to dry matter intake (DMI) and body weight (Herd et al., 2014). Therefore, selecting for reduced methane production would also reduce DMI and body weight unless multitrait selection was practiced to mitigate these impacts.

Several studies have been conducted to determine quantitative trait loci (QTL) associated with predicted methane production. Significant single nucleotide polymorphisms (SNP) were detected in two studies on predicted methane production in dairy cattle (de Haas et al., 2011; Pickering et al., 2015). However, there was no overlap in the SNP between the studies. Only one study (Uemoto et al., 2020a) has conducted a genome-wide association study (GWAS) on predicted methane production in beef cattle. No SNP in Uemoto et al. (2020a) passed the threshold for significance. Given the mixed results of previous studies, the loci related to predicted methane production remains unclear.

Selection of animals with low methane production is only possible with a large number of phenotyped animals. There are several techniques to phenotype animals for methane production; however, it can be costly and time-consuming to obtain direct methane production measurements, especially on the number of animals needed to perform genetic evaluation. Instead, it may be simpler and more cost-effective to select on an indicator trait such as predicted methane production, or to utilize these indicator traits to bolster genetic prediction while large numbers of animals with methane emissions data are phenotyped. Several equations which use variables that are simpler or more cost-effective to measure than direct methane production (e.g., DMI) have been formulated to estimate these values (as reviewed by Kebreab et al., 2016).

The objectives of this study were to calculate the heritabilities of three predicted methane prediction traits and identify QTL for each trait.

Materials and Methods

Study design and data collection

All animal procedures were approved by the Institutional Animal Care and Use Committee at Oklahoma State University (protocol AG 13-18) in accordance with Federation of Animal Science Societies (FASS, 2011) guidelines. Full details on study design can be found in Ahlberg et al. (2019). Briefly, daily DMI information was obtained on 830 crossbred steers from May 2014 to May 2018 at the Willard Sparks feedlot at Oklahoma State University. Steers were fed in seven groups with two feeding protocols. Three groups were fed using a slick bunk protocol (group 1, N = 119; group 2, N = 115; group 3, N = 120) and four groups had ad libitum access to feed (group 4, N = 105; group 5, N = 123; group 6, N = 126; group 7, N = 122). Within group, steers were blocked by weight (heavy and light) then randomly assigned to one of four pens. Each pen held approximately 30 animals. Each steer was implanted with Compudose (Elanco Animal Health, Greenfield, IN), an implant containing estradiol 17ß (E2 ß), per facility protocol.

While on test, each group received a mixed ration of approximately 15% cracked corn, 51.36% wet corn sweet bran, 28.44% prairie hay, and 5.20% mineral supplement. The rations were analyzed by Dairy One, Inc. (Ithaca, New York) for percent dry matter and GE. The quality of ingredients differed slightly for each group which led to variations in the GE values. The GE of the diet was 18.99, 19.40, 18.26, 18.89, 18.82, 18.68, and 18.91 megajoule (MJ) per kilogram dry matter for groups 1 through 7, respectively. Similarly, the dry mater percentage was 74%, 74%, 73%, 73%, 70%, 74%, and 74%, for groups 1 through 7, respectively.

All animals were allowed a 21-d acclimation period followed by a 70-d feed intake trial period as outlined in the Beef Improvement Federation guidelines (BIF, 2016). Feed intake was measured using an Insentec system (Hokofarm Group, the Netherlands). The system consisted of six feed bunks and one water bunk placed under shade in each pen. Individual body weights were measured at the start of each trail, then every 14 d, and again at the end of the 70-d trial period.

Feed intake data were filtered using the procedures outlined in Allwardt et al. (2017). Briefly, the weight of the bunk before and after each animals’ assumed feeding period were filtered based on bunk volume and system settings to ensure accurate DMI averages. Body weight records were filtered for appropriateness by assuming that animals could gain or lose a maximum of 50 kg between weighing days.

Phenotypes

Average daily gain (ADG) was calculated by regressing time on the biweekly body weight measurements of each animal to account for differences in rumen fill. Midtest body weight was determined for each steer by multiplying each animal’s ADG by 35 then adding the animal’s intercept from the regression.

Exact breed composition of the animals was unknown; therefore, it was estimated using each animal’s genotypes and a framework developed by Chiang et al. (2010). The methodology described in Kuehn et al. (2011) which was employed here, uses breed specific allele frequencies from 16 breeds common across the United States, including taurine, indicine, and dairy breeds. Allele frequency estimates were found by recoding genotypes as the number of copies of the B allele (using the Illumina genotype calls) and then dividing by 2 (Kuehn et al., 2011). Breed composition was estimated using this model:

$${\displaystyle \begin{array}{l}{\displaystyle y=Xb+e}\end{array}}$$

where X is a 36,403 by 16 matrix of the frequencies of the B allele, b is the vector of regression co-efficients that represents the percentage of each breed in animal y, and e is the vector of random residuals. If the sum of the co-efficients was less than 1, the difference from 1 was assigned to an “other” category. If the sum of the co-efficients was greater than zero, the estimates were scaled as follows:

$${\displaystyle \begin{array}{l}{\displaystyle \frac{1}{\sum nonzerobreedcoefficients}\ast breedcoefficient}\end{array}}$$

All small negative co-efficients were assumed to be zero.

Methane emissions were predicted for each animal using three different methane prediction models. All models selected for use were chosen because: 1) all utilize DMI, a trait that was consistently measured across groups, 2) all have performed well when evaluated against observed methane datasets (Ellis et al. 2007, 2009; Kebreab et al., 2008), and 3) all are commonly used throughout methane prediction literature (de Haas et al. 2011; Pickering et al., 2015;, Hayes et al., 2016).

Methane prediction trait was first predicted using the model outlined in Ellis (Ellis et al., 2007) and detailed here:

$${\displaystyle \begin{array}{l}{\displaystyle EM{P}_i=\frac{[3.272+(0.736\ast DM{I}_i)]}{0.05565}}\end{array}}$$

where EMP_i is the average daily methane emission of animal i in grams from the Ellis et al. (2007) equation and DMI_i is average daily dry matter intake of animal i in kilograms. Ellis et al. (2007) used a dataset of observed methane measurements from 172 trials (83 beef and 89 dairy) to create 32 original methane prediction equations. Of those 32 equations, 14 were trained on the beef data, 8 were trained on the dairy data, and 10 were trained on the combined data. The authors then tested the 32 methane prediction equations against each other and several extant equations to determine which model was the most accurate for each dataset. This equation had the lowest error of all the equations when evaluating the combined dataset (Ellis et al., 2007). It was chosen for inclusion because the diet formulation of the steers in this study, a major factor in methane production, most closely aligned with the average diet composition in the combined dataset of Ellis et al. (2007).

Second, methane production was predicted with the first nonlinear equation from Mills et al. (2003) as follows:

$${\displaystyle \begin{array}{l}{\displaystyle MM{P}_i=\frac{56.27-\left(56.27\ast {\mathrm{e}}^{-0.028\ast DM{I}_i}\right)}{0.05565}}\end{array}}$$

where MMP_i is the average daily methane emission of animal i in grams from the Mills et al. (2003) equation and DMI_i is average daily dry matter intake of animal i in kilograms. Most methane equations are linear regressions (Ellis et al., 2007), and thus as the predictor variable (often DMI) continues to increase, the predicted methane production continues to increase. However, Mills et al. (2003) reasoned that a nonlinear, diminishing returns relationship between intake and methane production may be more biologically appropriate. This equation was chosen to be included in this study due to its nonlinear nature. It is worth noting that Mills et al. (2003) trained and evaluated this model on a dairy cattle dataset, and it was not tested in beef cattle data. By applying it to a beef dataset as in this study, it is assumed that there are no differences in rate of methane production between beef and dairy animals due to factors other than amount of feed intake.

Finally, methane production was predicted using the IPCC Tier 2 (IPCC, 2019) equations as follows:

$${\displaystyle \begin{array}{l}{\displaystyle IM{P}_{ij}=\frac{G{E}_j\ast \left(\frac{6.3}{100}\right)\ast DM{I}_i}{0.05565}}\end{array}}$$

where IMP_ij is the average daily methane emission in grams of animal i consuming ration j as calculated by the IPCC (2019) equation, GE is the GE of ration j in MJ per kilogram, DMI_i is the average daily dry matter intake of animal i. The International Panel on Climate Change (IPCC) is the intergovernmental body tasked with assessing the science concerning anthropogenic climate change, of which ruminant emissions are a part. The IPCC has created three models used for estimating enteric methane production. Tier 2 (IPCC, 2019) was chosen for this study as it is more accurate than Tier 1 while Tier 3 required inputs, such as feed degradation characteristics, not recorded during the trial.

Genotypes

On a day when weights were collected during the feeding period, two blood samples were drawn from each animal. Each sample was 8.5 milliliters of blood collected in a vacutainer tube containing 1.5 mL of the anticoagulant citrate dextrose. Samples of DNA were extracted using a phenol:chloroform:isoamyl alcohol extraction and ethanol precipitation. The DNA samples were sent to GeneSeek (Lincoln, NE) for genotyping on the GeneSeek Genomic Profiler genotyping array (GGP 150K). Thresholds for quality control were set so that SNP with minor allele frequency less than 0.05 and SNP and animals with call rates less than 0.90 were removed from the analysis. After quality control, 802 animals and 123,940 SNP were used in the analysis.

Statistical analysis

For statistical analysis, data from all groups were included. The minimum, mean, maximum, and standard deviation for each trait, as well as Pearson and Spearman correlations between the traits were calculated with the “stats” package in R (R Core Team, 2020). The “stats” package in R (R Core Team, 2020) was also used to determine differences (P < 0.05) between the trait means with a two-tailed t-test.

Multivariate models of the three traits (or any combinations thereof) would not reach convergence because of the extremely high correlations between them. Therefore, genetic correlations could not be calculated and variance components and heritabilities for each trait were instead estimated using an average information restricted maximum likelihood (AIREML) algorithm in the BLUPF90 software package (Misztal et al., 2014) using the following univariate animal model:

$${\displaystyle \begin{array}{l}{\displaystyle y_i=X_i{b}_i+Z_i{u}_i+e_i}\end{array}}$$

where y is a vector of phenotypes for predicted methane trait i, b is a vector of fixed effects, group, midtest weight, management style, and breed percentages, for trait i, X is an incidence matrix relating phenotypes to the fixed effects and covariates in b for trait i, u is a vector of additive direct genetic effects for trait i, Z is an incidence matrix relating phenotypes to the additive direct genetic effects in u for trait i, and e is a vector of random residuals for trait i. Body weight and DMI have a strong, positive correlation (Martin et al., 1955) and all the predicted methane equations chosen for this study are functions of feed intake (DMI and/or GE intake);.thus, to account for differences in predicted methane production caused by differences in body size, midtest weight was added as a covariate to the model.

Genetic analyses were conducted using genomic best linear unbiased prediction (GBLUP) methodology where all relationships were defined solely using genomic data (Aguilar et al., 2010; Christensen and Lund, 2010; VanRaden, 2008). The univariate animal model above was used to calculate each animal’s estimated breeding value (EBV) and standard error of prediction for each trait utilizing the BLUPF90 suite of programs (Misztal et al., 2014). The standard error of prediction for each animal and each trait was squared to calculate the prediction error variance. The prediction error variance for each EBV was then used to estimate the accuracy of the EBV in R (R Core Team, 2020). The following accuracy equation defined in the Beef Improvement Federation Guidelines (BIF, 2020) was used:

$${\displaystyle \begin{array}{l}{\displaystyle Accurac{y}_{BIF}=1-\left(\sqrt{\frac{predictionerrorvariance}{additivegeneticvariance}}\right)}\end{array}}$$

Further, classical animal breeder accuracy was also calculated using the following conversion equation from the Beef Improvement Federation Guidelines (BIF, 2020):

$${\displaystyle \begin{array}{l}{\displaystyle r_{EBV,BV}=\sqrt{1-{(1-Accurac{y}_{BIF})}^2}}\end{array}}$$

where r_EBV,BV is the correlation between the EBV and the true breeding value. Pearson and Spearman correlations between the EBV for each trait were also calculated using the “stats” package in R (R Core Team, 2020).

A GWAS was conducted for each trait using the postGSf90 function in the BLUPF90 suite of programs (Misztal et al., 2014). Marker effects were calculated from EBV using the methodology described by Wang et al. (2012). Markers were considered statistically significant at the Bonferroni corrected threshold of 7.3 on the −log₁₀P-value scale. A post-GWAS analysis was conducted on the top 20 QTL regions for each trait. The SNP were sorted by significance for each trait, and QTL regions were formed as follows: the SNP with the highest −log₁₀P-value for each trait became the first QTL region, the SNP next closest in significance was used to either form a new QTL region or combined with an existing QTL if it was within 250 kb of the SNP already in the QTL region. This process was repeated until 20 unique regions were obtained for each trait. Thus, each QTL region is at least 500 kb in size to account for moderate linkage disequilibrium (McKay et al. 2007), and regions with multiple SNP are slightly larger (range 500 to 637 kb) The JBrowse genome browsing system (Buels et al., 2016) was used to identify genes in these regions based on the ARS 1.2 cattle genome assembly and annotations (Rosen et al., 2020). The The UniProt Consortium (2021) was used to investigate the function of these genes. Candidate genes were then selected from all genes within the QTL regions based on their function and possible relationship to predicted methane production. In addition, the cattle QTL database (Hu et al., 2019) was used to determine if QTL in this study overlapped previously reported QTL in the scientific literature.

Results and Discussion

Summary statistics

Summary statistics for each methane production trait are listed in Table 1. All traits had significantly different means (P < 0.001). Ellis et al. (2007, 2009), and van Lingen et al. (2019) have all reported differences in predicted methane estimates when applying multiple prediction models to the same dataset, which is consistent with our finding. The standard deviation for EMP was much lower than that of the other two traits (19.7 vs. 31.3 and 32.0 for MMP and IMP, respectively). This is because the prediction models were formulated in such way that the change in grams of methane resulting from a one-unit change in DMI for the equation from Ellis et al. (2007) was less than the rate of change from the other equations until the equation from Mills et al. (2003) reaches 28 kg of DMI. In other words, the difference in methane production between an animal with a DMI of 10 kg and an animal with a DMI of 11 kg is not as great for the Ellis et al. (2007) equation as it was for the other equations. This formulation restricted the predicted methane production of the animals in this dataset to a narrower margin, and consequently a smaller standard deviation, than the other two equations. As previously stated, the equation from Mills et al. (2003) was chosen for inclusion because a nonlinear association between methane production and intake was thought to be more biologically appropriate as it would be unlikely for methane production to continue to increase at the same rate as DMI increases. In short, the Mills et al. (2003) equation was intended to limit the maximum methane predicted. Therefore, it is interesting to note that MMP had the greatest mean and maximum of all the traits (a mean of 261.8 g methane/d compared to 200.9 and 229.0 g methane/d for EMP and IMP, respectively). The relative rate of increase in predicted methane from one additional unit of DMI at the levels of DMI seen in this data was much greater for the Mills et al. (2003) equation compared to the equations detailed by Ellis et al. (2007) and IPCC (2019). Further, the Mills et al. (2003) equation was trained on feed intake observations in mature dairy cattle. The co-efficients detailed in the model may have been accurate for the high DMI observed in dairy cattle, but the model may not perform as well when using data from smaller beef animals, such as the animals in this study, which likely have lower intake. This relatively poor performance was demonstrated by the Mills et al. (2003) model’s high root mean square prediction error when attempting to predict the observed methane production of a beef dataset (Ellis et al., 2007).

Table 1.

Summary statistics for average daily methane production (g methane/d) for each methane prediction equation

Predicted methane trait	Number of animals	Mean	Minimum	Maximum	Standard deviation
EMP 1	730	200.9a	138.2	272.7	19.7
MMP 2	730	261.8b	156.4	368.2	31.3
IMP 3	730	229.0c	131.8	344.5	32.0

¹EMP is methane predicted using an equation adapted from Ellis et al. (2007).

²MMP is methane predicted using an equation adapted from Mills et al. (2003).

³IMP is methane predicted using an equation adapted from IPCC (2019).

^{a ,}

^{b ,}

^cMeans with different superscripts are statistically different (P < 0.05).

The Pearson phenotypic correlations between all three traits were above 0.99, likely because each trait was derived from feed intake (Table 2). The high Pearson correlations indicate the same trends among the values themselves, even though the actual predicted methane measurements differed. The Spearman correlation for EMP and MMP was 1, indicating these traits ranked animals identically. The Spearman correlations between EMP and IMP and between IMP and MMP were also very high at 0.99. High Spearman coefficients indicate the animals’ phenotypes ranked similarly.

Table 2.

Phenotypic correlations (Pearson above the diagonal, Spearman below the diagonal) between methane prediction traits

	EMP1	MMP2	IMP3
EMP1		0.99*	0.99*
MMP2	1*		0.99*
IMP3	0.99*	0.99*

¹EMP is methane predicted using an equation adapted from Ellis et al. (2007).

²MMP is methane predicted using an equation adapted from Mills et al. (2003).

³IMP is methane predicted using an equation adapted from IPCC (2019).

^* P < 0.01.

Variance components and heritabilities

The genetic variance, residual variance, and heritability estimates for each trait are reported in Table 3. The heritabilities ranged from 0.59 to 0.62. In comparison, Brito et al. (2018) performed a meta-analysis of eight studies with heritability estimates for predicted methane production. From the available literature, the authors estimated the heritability of predicted methane in cattle to be 0.26 ± 0.02. However, the majority of the eight predicted methane studies were conducted in dairy cattle. Of all the literature compiled by Brito et al. (2018), only one study reported the heritability of predicted methane in beef cattle, Sobrinho et al. (2015). The authors of Sobrinho et al. (2015) used three different equations to estimate predicted methane for 955 Nellore cattle. None of the equations overlapped between and this study and Sobrinho et al. (2015); however, body weight was also fitted as a covariate in the genetic analysis for both. The heritability of predicted methane was 0.32 ± 0.07, for all three prediction models used. Another study that estimated the heritability of predicted methane traits in beef cattle was Uetmoto et al. (2020a). The authors of Uemoto et al. (2020a) utilized two methane prediction equations. One was developed by Uemoto et al. (2020b) for cattle on high concentrate diets and the other from Shibata et al. (1993) which has been adopted for national greenhouse gas evaluations in Japan. Uemoto et al. (2020a) calculated heritabilities of 0.54 ± 0.05 of 0.56 ± 0.05 for predicted methane emission of Japanese Black steers using the equations of Uemoto et al. (2020b) and Shibata et al. (1993), respectively. The heritabilities reported for predicted methane traits in this study are greater than the heritabilities reported for predicted methane traits in Brito et al. (2018) or Sobrinho et al. (2015), possibly due to the small sample size or population dynamics. Nonetheless, the heritabilities presented here, when standard errors are taken into account, are on par with those of Uemoto et al. (2020a).

Table 3.

Variance components (standard errors) for each predicted methane trait

Predicted methane trait	Genetic variance	Residual variance	Phenotypic variance	Heritability
EMP1	125.3 (29.9)	86.1 (24.5)	211.4	0.59 (0.13)
MMP2	333.0 (77.3)	207.4 (62.8)	540.4	0.62 (0.13)
IMP3	328.1 (78.1)	224.8 (64.0)	552.9	0.59 (0.13)

¹EMP is methane predicted using an equation adapted from Ellis et al. (2007).

²MMP is methane predicted using an equation adapted from Mills et al. (2003).

³IMP is methane predicted using an equation adapted from IPCC (2019).

Alternatively, the traits presented in this study can be viewed as functions of average daily DMI in a population of growing animals. From that perspective, heritabilities of 0.59 ± 0.13, 0.62 ± 0.13, and 0.59 ± 0.13 are within literature estimates. For example, Ahlberg et al. (2019), reported the heritability of DMI to be 0.67 (± 0.04). This is to be expected since the animals in Ahlberg et al. (2019) had substantial overlap with the animals analyzed in this study (578 animals in common with an additional 252 animals in this study). Further, Koch et al. (1963) and Archer et al. (1997) estimated the heritability of DMI for cohorts of Angus, Hereford, and Shorthorn growing animals to be 0.64 (± 0.12) and 0.62 (± 0.12), respectively. More recently, Freetly et al. (2020) reported a heritability of 0.82 (± 0.12) for average daily DMI in growing heifers. While these examples are provided to show that the heritability of DMI in certain populations can be relatively high, it is worth noting that a meta-analysis done by Diaz et al. (2014), which included 128 heritability estimates for DMI, reported a pooled heritability of 0.39 for the trait.

Estimated breeding values

EBVs were generated for each steer and each trait (Table 4). As expected, the mean EBV for each trait was approximately zero. Generally, these estimates followed the same pattern as the summary statistics for the phenotypic traits. Ellis methane production had the smallest range of values and the correspondingly lowest standard deviation. Curiously, while both MMP and IMP had a greater range than EMP, IMP had a greater standard deviation, whereas the standard deviations for EMP and MMP were relatively similar. The difference may lie in the fact that IMP has a slightly larger phenotypic range than MMP or perhaps it is because the IPCC et al. (2019) methane prediction equation was different for each group due to the inclusion of the group-specific GE for the ration. This latter hypothesis would be supported by the relatively large residual variance of IMP as seen in Table 3.

Table 4.

Summary statistics and mean accuracies of EBVs for average daily methane production (g methane/d) for each methane estimation equation

	Number of animals	Mean	Minimum	Maximum	Standard deviation	BIF accuracy	Classical accuracy
EMP1	810	0.0004	−25.6	23.6	8.61	0.33	0.74
MMP2	810	0.0006	−43.7	36.0	13.7	0.34	0.75
IMP3	810	0.0006	−42.8	37.5	13.9	0.33	0.74

¹EMP is methane predicted using an equation adapted from Ellis et al. (2007).

²MMP is methane predicted using an equation adapted from Mills et al. (2003).

³IMP is methane predicted using an equation adapted from IPCC (2019).

Like the phenotypes, the EBV all share a very high Spearman correlation of 0.99. This supports the premise that any of the predicted methane traits described in this study could be utilized for selection and the animals would rank essentially the same.

When discussing the accuracy, it is important to remember the accuracy is not how accurate the models used were in predicting actual methane production of the animals. Instead, accuracy is defined here as the correlation between the EBV and the true breeding value of any given individual. The BIF accuracies for each trait were lower than the conventional accuracies because the BIF Guidelines (2020) utilize a more conservative estimate. With the more stringent calculation, the BIF accuracies for the EMP, MMP, and IMP were moderate. In addition, the conventional breeder’s accuracies were moderately high. The high accuracies of these traits are to be expected due to the correspondingly high heritability estimates. The higher the heritability, the more an individual’s own phenotype can be relied on as an indicator for that individual’s true breeding value.

Genome wide association study

As shown in Figure 1, one SNP exceeded the significance threshold for all of the traits. This SNP was rs43076526 on Bos taurus autosome (BTA) 7. To our knowledge, the traits presented here are the first associations tied to that SNP. To further explore functionality, QTL regions that approached significance for each trait were reported for validation in future studies and evaluated to determine whether these associations have obvious ties to the phenotypes studied. A total of 30 SNP nearest to significance were organized into 22 distinct QTL regions (Table 5). Of those 30 SNP, 23 (76.7%) were shared by all traits, 2 (6.7%) were shared by EMP and MMP, 3 (10%) were near significance only for MMP, and 2 (6.7%) were near significance only for IMP. Supplementary Table S1 details each region, the SNP in that region, the candidate genes, and their functions. To our knowledge, none of the SNP identified for this study have been previously identified as significant in a GWAS of either predicted or observed methane production in either beef or dairy cattle. This may be due to a variety of reasons including because there are few of these studies in the literature, it is a fairly novel trait and expensive to collect, and many are in populations quite different from the animals in this study and/or in other countries. However, many of these regions have been identified as associated with the component traits of predicted methane production. Specifically, the data and methodology described here were also used in a GWAS focusing on DMI (E. Dressler, Kansas State University, Manhattan, KS, unpublished data). That analysis yielded five significant SNP, including the SNP that reached significance in this study. The remaining four significant SNP of the DMI GWAS, however, were not identified in any of the remaining QTL regions described here.

Figure 1.

Manhattan plot showing results of genome-wide association mapping for methane production predicted by (A) Ellis et al., 2007 (B) Mills et al., 2003, and (C) IPCC, 2019. The genome-wide significance threshold of 7.3 is depicted by the horizontal red line.

Table 5.

QTL regions and SNP most closely associated with predicted methane production traits

Region	QTL range (Chromosome: base pair)	Largest −log10P-value in QTL region for Ellis methane production	Largest −log10P-value in QTL region for Mills methane production	Largest −log10P-value in QTL region for IPCC methane production	SNP	SNP location (Chromosome: base pair)
1	3:85840597 - 86340597	4.09	3.79	4.16	rs110220315	3:86090597
2	3:87885131 - 88414394	3.79	3.7	3.82	rs110058749	3:88164394
3	4:104329265 - 104829265	3.72	3.53	NA	rs134296722	4:104579265
4	5:1620477 - 2120477	3.91	3.7	3.92	rs42740586	5:1870477
5	5: 21225995 - 21766232	4.31	4.01	4.3	rs110309656 rs109244569	5:21475995 5:21516232
6	7:60245101 - 60745101	3.82	3.65	3.85	rs137645685	5:60495101
7	7:21941462 - 22578341	4.98	4.78	5	rs43508672 rs43508667 rs43508661 rs43141114 rs43509246 rs43508669 rs29023390	7:22191462 7:22197623 7:22204260 7:22268814 7:22328341 7:22195580 7:22202959
8	7:27002564 - 27502564	7.76	7.57	7.78	rs43605790	7:27252564
9	9:89557465 - 90057465	3.8	3.75	3.73	rs41621748	9:89807465
10	12:9790464 - 10290464	4.01	4.01	4.06	rs134083327	12:10040464
11	12:45993629 - 46493629	3.97	3.71	3.98	rs110121749	12:46243629
12	12:54039535 - 54539535	4.03	3.88	4.06	rs134046542	12:54289535
13	14:28148738 - 28648738	NA	NA	3.73	rs133977719	14:28398738
14	16:62894583 - 63394583	3.72	3.63	3.76	rs42691659	16:63144583
15	17:31870758 - 32370758	4.1	3.87	4.13	rs110766243	17:32120758
16	19:36174186 - 36674186	4.27	4.15	4.27	rs134127572	19:36424186
17	19:59132179 - 59632179	NA	NA	3.75	rs110820800	19:59382179
18	20:18222746 - 18722746	3.89	3.67	3.89	rs109366906	20:18472746
19	20:20244060 -20744060	3.74	3.66	NA	rs110957960	20:20494060
20	20:39252283 - 39772823	4.33	4.14	4.35	rs133044483 rs136158794	23:39502283 23:39522823
21	24:22174171 - 22674171	3.8	3.55	3.77	rs135137105	24:22424171
22	24:53794457 - 54294457	4.12	3.86	4.05	rs110389869	24:54044457

Candidate genes

The candidate SNP for the significant SNP (QTL region 8) was identified as ALDH7A1. The protein produced by ALDH7A1 is involved in synthesizing glycine betaine from choline, thereby protecting the cell from oxidative stress (The UniProt Consortium, 2021). Interestingly, ALDH7A1 was not only identified as being associated with feed efficiency by de Oliveira et al. (2014) but was also associated with yearling weight in Canchim cattle by Buzanskas et al. (2014). Further, Grubbs et al. (2013) discovered that swine with increased residual feed intake (RFI) have lower oxidative stress than their high RFI contemporaries. Therefore, animals with an ALDH7A1allele that provides less oxidative stress protection would consume more (independent of body weight) and would have greater predicted methane production.

Outside of the significant SNP region, three closely related themes were identified in the candidate gene functions. The first was associated with collagen. QTL regions 7 and 16 both have candidate genes which relate to collagen, P4HA2 and COL1A1, respectively (UniProt Consortium, 2021). This is of particular interest because QTL region 7 contained the largest number of SNP (7) of any of the QTL regions and regions 7 and 16 were both identified for all three methane production traits. Second, two of the 22 QTL regions had candidate genes with functions related to cell adhesion. Cell adhesion is the process of cells binding either to other cells or the extracellular matrix (ECM). Finally, three candidate genes are involved with microtubule and cytoskeleton organization. Thus far the focus has been on genes linked to the ECM either by encoding a constituent like collagen (COL1A1), encoding a posttranslational catalyst for collagen formation (P4HA2), or being involved in ion binding enabling cellular adhesion to other cells and to the ECM.

The cytoskeleton is a key feature of cell adhesion (Hynes and Zhao, 2000). If one assumes that predicted methane traits are similar to DMI (which is reasonable given that DMI is the most influential component trait for the models we included) then having several candidate genes related to the ECM is reasonable for multiple possible reasons. First, the strength of the ECM has been found to vary between cattle differing feed efficiencies (Chen et al., 2012). Secondly, the primary component of the ECM is collagen. Collagen turnover rate has been shown to increase growth, and likely DMI, of beef cattle (Wu et al., 1981). Although the specific candidate genes of this study are novel regarding predicted methane production, genes involved in the ECM and cell adhesion have been associated with feed efficiency, body weight, and growth in the literature. In fact, Chen et al. (2011) identified several genes related the ECM, including COL1A1 which was also identified in this study, are up-regulated in low-RFI cattle. The authors of Chen et al. (2011) suggested cattle with higher ECM activity have lower feed intake (and therefore would be expected to have a lower predicted methane) while maintaining the same level of growth as their contemporaries. The identification of COL1A1 as a gene important for feed efficiency was further supported by Kern et al. (2016). The authors of that paper found COL1A1 was down-regulated in the ruminal epithelial cells of high-gain beef animals. de Oliveria et al. (2014) noted that ADAM12, which regulates the ECM, was significantly associated with partial efficiency of growth in Nellore cattle. When examining differential expression of genes in the rumen epithelium of cattle with differing RFI, Kong et al. (2016) found that genes related to intercellular adhesion were enriched in low RFI animals. The authors suggested this may be due to increased tissue morphogenesis. In addition, Seabury et al. (2017) identified two candidate genes that encode extracellular proteins that were significantly associated with midtest metabolic weight in Hereford and SimAngus cattle.

Like the way the ECM provides structure between cells, the cytoskeleton provides intracellular structure and one of the key components of the cytoskeleton is microtubules. Both HOOK1 (QTL region 1) and DAB1 (QTL region 2) are linked to microtubule binding, while KIF13A (QTL region 20) depends on microtubules for intracellular transport. Microtubules provide structure to the cytoskeleton and differences in genes controlling their arrangement may indicate differences in cell energy efficiency. In Kong et al. (2016), the authors posit the up-regulation of genes related to cytoskeletal organization allows for greater dynamic remodeling within the epithelial cells of cattle with low RFI, which may create gaps between cells resulting in easier absorption of nutrients.

Two candidate genes are similar to the candidate gene of the most significant SNP in that they also influence intake independent of body weight. Both DACH1 and DTNA (QTL regions 11 and 21, respectively) have been linked to animals that either eat or gain differently than their contemporaries. DACH1 has been associated with RFI, residual intake gain, and efficiency of intake by Serão et al. (2013). In the same study, DTNA was associated with residual ADG. Intriguingly, DACH1 and DTNA were both identified as a ­breed-­dependent mode of action. DACH1 having a larger effect in more Simmental influenced animals, while DTNA greater impact on purebred Angus, both breeds being heavily represented in this dataset. Additionally, DTNA was also associated with feed efficiency by Chen et al. (2011) and Manca et al. (2021). Given these results, it is possible that differences in the DACH1 and DTNA genes can lead to differences in feed intake and thus differences in predicted methane production.

Other candidate genes in this study have broader functions that make their ties to predicted methane harder to elucidate. For example, the candidate gene for QTL region 12 (RBM26) is involved in RNA binding and mRNA processing as is YTHDF3 in QTL region 13. The binding, processing, and regulation of RNA is a necessary step to encode every protein in any cell. Another example is the group of genes related to the regulation of the binding of RNA polymerase II for DNA transcription (ELK3 and ZNF648 in QTL regions 6 and 14, respectively), which must take place to encode every protein in the cell. It should be noted, however, that ELK3 has been identified as a possible regulator for genes involved in FCR and feed efficiency ratio traits in Nelore cattle in de Lima et al. (2020).

QTL region overlap with previous literature

While no SNP in any of the QTL regions had been previously identified for traits that reasonably relate to predicted methane production; almost every QTL region overlapped a QTL that may be tied to predicted methane production (full details in Table S1). It is worth noting, however, that several of these overlapping QTL had extremely wide ranges across the genome because they were derived from linkage analyses. Therefore, any inference based on overlap of QTL regions should take QTL window length into consideration and more heavily favor smaller windows.

As previously established, predicted methane traits are based on DMI, thus DMI and predicted methane are intimately tied together. Therefore, it is surprising that none of the 22 QTL regions in this study overlapped QTL previously associated with DMI.

While previously identified QTL associated with DMI were surprisingly underrepresented, overlap of QTL associated with traits highly correlated with DMI were not. Of the 22 QTL regions in this study, 17 of them overlapped with QTL previously associated with some measurement of body weight; either birth weight, weaning weight, yearling weight, finished weight, or combinations thereof. Additionally, ADG also had multiple overlapping QTL (5). All the QTL regions in this study that overlapped with an ADG QTL also overlapped with a body weight QTL. The repeated occurrence of these traits is not unexpected for multiple reasons. First, body weight and ADG are both relatively simple traits to measure, making them easy traits upon which to perform GWAS. Second, ADG and body weight are highly polygenic traits (Seabury et al., 2017) that have been included in numerous GWAS studies. The large number of documented QTL connected to ADG, and body weight likely stems from these factors, which, in turn, increases the likelihood of the QTL in our study overlapping with a previously established QTL for either trait. However, that does not diminish the fact ADG, and body weight are both moderately to highly correlated with DMI and observed methane (Donoghue et al., 2016; Renand et al., 2019) and that these results provide evidence that some of the effects identified are relevant to understanding the biology of predicted methane production.

Feed conversion ratio (FCR), defined as the ratio of feed intake to body weight gain, and RFI are two measures of feed efficiency for which previously reported QTL overlapped QTL identified in this study (2 and 6, respectively). Both traits require observed or estimated DMI for their formulation and are correlated to DMI (Herd et al., 2016). Only one study (Herd et al., 2014) has reported a correlation between FCR and observed methane production (−0.31), though the two traits involved in the correlation were FCR when animals were in a feedlot being fed with a high concentrate diet and methane production when the animals were housed in a respiration chamber being fed with roughages. Studies of the relationship between observed methane production and RFI are contradictory, and no consensus has emerged. These studies have reported various results including animals with lower RFI also have lower methane production (Nkrumah et al., 2006), that there is a negative correlation between observed methane and RFI (Herd et al., 2016), and that there is no relationship between the traits whatsoever (Jones et al., 2011). Our results would indicate there is some relationship between feed efficiency and predicted methane production, though the directionality of the relationship remains ambiguous.

Interestingly, previously identified QTL for carcass traits often overlapped with QTL from this study. Longissimus muscle area (LMA) QTL overlapped 6 of our QTL, 12th rib backfat (BF) had 5, and marbling score (MS) had 12 overlapping QTL. The presence of overlapping QTL was consistent with the positive correlations reported in the literature between DMI and carcass quality traits (Nkrumah et al., 2007), though the number of associations was unexpected. Upon further investigation, the vast majority of the previously identified carcass quality QTL came from two studies: McClure et al. (2010) and Leal-Gutiérrez et al. (2020). Both papers included mostly Angus and Angus-influenced animals, which reflects the genetic background of many of the animals in this study (Ahlberg, 2018). The strong correlation between observed methane and DMI and ADG, thus making them more likely to have overlapping QTL with predicted methane, does not apply to all carcass quality traits. Herd et al. (2016) reported moderate correlations between observed methane and LMA and between observed methane and ultrasound intramuscular fat (a proxy for marbling). The authors did not find a correlation different from zero between observed ­methane and BF, although the correlation between DMI and BF has been well-established (Nkrumah et al, 2007).

There were also overlaps for meat quality trait like tenderness score (TS) and shear force (SF). Of these traits, TS QTL each overlapped with QTL regions 4, 6, 7, 11, and 19 while SF QTL overlapped with QTL regions 11, 15, and 20. Of particular note are the two QTL regions (11 and 18) which overlap with previously identified QTL for connective tissue amount. The reason these QTL were of particular interest was because all three traits are related to collagen and the ECM, which ties back into the functions of several candidate genes. Li et al. (2022) utilized a meta-analysis to establish an inverse relationship between the volume of ECM components, especially collagen, and various measures of meat tenderness including SF and TS. To our knowledge, no relationship between TS, SF, or connective tissue amount and either DMI or observed methane has been quantified in the literature. Nonetheless, the intersection between predicted methane, the ECM, and various tenderness characteristics suggests a connection that should be studied further.

Conclusion

All three predicted methane traits, as determined by equations from Ellis et al. (2007), Mills et al. (2003), and IPCC (2019), produced similar rankings of EBVs for methane production. All predicted methane traits had heritabilities ranging from 0.59 to 0.62, which demonstrates that genetic selection on predicted methane is possible. One of the 123,940 SNP investigated reached the threshold for significance for all traits. In addition, the 20 genomic regions closest to that threshold for each trait were almost identical for all three traits. Both individual candidate genes and several functional groups appeared related to predicted methane production through its relationships with DMI and growth, but no SNP were identified with obvious ties to methane production independent from DMI.

Glossary

Abbreviations:

ADG

average daily gain

BTA

Bos taurus autosome

DMI

dry matter intake

EBV

estimated breeding value

ECM

extracellular matrix

EMP

Ellis methane prediction

FCR

feed conversion ratio

GE

gross energy

GWAS

genome wide association study

IMP

IPCC methane prediction

MJ

megajoules

MMP

Mills methane prediction

QTL

quantitative trait loci

SF

shear force

SNP

single nucleotide polymorphism

TS

tenderness score

Conflict of interest statement

The authors declare no real or perceived conflicts of interest.