Methane prediction based on individual or groups of milk fatty acids for dairy cows fed rations with or without linseed

Abstract

Milk fatty acids (MFA) are a proxy for the prediction of CH4 emission from cows, and prediction differs with diet. Our objectives were (1) to compare the effect of diets on the relation between MFA profile and measured CH4 production, (2) to predict CH4 production based on 6 data sets differing in the number and type of MFA, and (3) to test whether additional inclusion of energy-corrected milk (ECM) yield or dry matter intake (DMI) as explanatory variables improves predictions. Twenty dairy cows were used. Four diets were used based on corn silage (CS) or grass silage (GS) without (L0) or with linseed (LS) supplementation. Ten cows were fed CS-L0 and CS-LS and the other 10 cows were fed GS-L0 and GS-LS in random order. In feeding wk 5 of each diet, CH4 production (L/d) was measured in respiration chambers for 48 h and milk was analyzed for MFA concentrations by gas chromatography. Specific CH4 prediction equations were obtained for L0-, LS-, GS-, and CS-based diets and for all 4 diets collectively and validated by an internal cross-validation. Models were developed containing either 43 identified MFA or a reduced set of 7 groups of biochemically related MFA plus C16:0 and C18:0. The CS and LS diets reduced CH4 production compared with GS and L0 diets, respectively. Methane yield (L/kg of DMI) reduction by LS was higher with CS than GS diets. The concentrations of C18:1 trans and n-3 MFA differed among GS and CS diets. The LS diets resulted in a higher proportion of unsaturated MFA at the expense of saturated MFA. When using the data set of 43 individual MFA to predict CH4 production (L/d), the cross-validation coefficient of determination (R²CV) ranged from 0.47 to 0.92. When using groups of MFA variables, the R²CV ranged from 0.31 to 0.84. The fit parameters of the latter models were improved by inclusion of ECM or DMI, but not when added to the data set of 43 MFA for all diets pooled. Models based on GS diets always had a lower prediction potential (R²CV = 0.31 to 0.71) compared with data from CS diets (R²CV = 0.56 to 0.92). Models based on LS diets produced lower prediction with data sets with reduced MFA variables (R²CV = 0.62 to 0.68) compared with L0 diets (R²CV = 0.67 to 0.80). The MFA C18:1 cis-9 and C24:0 and the monounsaturated FA occurred most often in models. In conclusion, models with a reduced number of MFA variables and ECM or DMI are suitable for CH4 prediction, and CH4 prediction equations based on diets containing linseed resulted in lower prediction accuracy.

Full text

1
ABSTRACT
Milk fatty acids (MFA) are a proxy for the prediction
of CH4 emission from cows, and prediction differs with
diet. Our objectives were (1) to compare the effect of
diets on the relation between MFA profile and mea-
sured CH4 production, (2) to predict CH4 production
based on 6 data sets differing in the number and type
of MFA, and (3) to test whether additional inclusion
of energy-corrected milk (ECM) yield or dry matter
intake (DMI) as explanatory variables improves predic-
tions. Twenty dairy cows were used. Four diets were
used based on corn silage (CS) or grass silage (GS)
without (L0) or with linseed (LS) supplementation.
Ten cows were fed CS-L0 and CS-LS and the other 10
cows were fed GS-L0 and GS-LS in random order. In
feeding wk 5 of each diet, CH4 production (L/d) was
measured in respiration chambers for 48 h and milk
was analyzed for MFA concentrations by gas chro-
matography. Specific CH4 prediction equations were
obtained for L0-, LS-, GS-, and CS-based diets and
for all 4 diets collectively and validated by an internal
cross-validation. Models were developed containing ei-
ther 43 identified MFA or a reduced set of 7 groups of
biochemically related MFA plus C16:0 and C18:0. The
CS and LS diets reduced CH4 production compared
with GS and L0 diets, respectively. Methane yield (L/
kg of DMI) reduction by LS was higher with CS than
GS diets. The concentrations of C18:1 trans and n-3
MFA differed among GS and CS diets. The LS diets
resulted in a higher proportion of unsaturated MFA at
the expense of saturated MFA. When using the data
set of 43 individual MFA to predict CH4 production
(L/d), the cross-validation coefficient of determination
(R2CV) ranged from 0.47 to 0.92. When using groups of
MFA variables, the R2CV ranged from 0.31 to 0.84. The
fit parameters of the latter models were improved by
inclusion of ECM or DMI, but not when added to the
data set of 43 MFA for all diets pooled. Models based
on GS diets always had a lower prediction potential
(R2CV = 0.31 to 0.71) compared with data from CS
diets (R2CV = 0.56 to 0.92). Models based on LS diets
produced lower prediction with data sets with reduced
MFA variables (R2CV = 0.62 to 0.68) compared with
L0 diets (R2CV = 0.67 to 0.80). The MFA C18:1 cis-9
and C24:0 and the monounsaturated FA occurred most
often in models. In conclusion, models with a reduced
number of MFA variables and ECM or DMI are suit-
able for CH4 prediction, and CH4 prediction equations
based on diets containing linseed resulted in lower pre-
diction accuracy.
Key words: dairy cow, methane emission prediction,
methane mitigation, methane proxy, milk fatty acids
INTRODUCTION
Methane is a greenhouse gas and a product of rumen
fermentation (Hristov et al., 2013). Methane produc-
tion (L/d) is a heritable trait, and genetic selection
for low-emitting cows is a promising mitigation option
(Pickering et al., 2015; Negussie et al., 2017). Meth-
ane emission quantification in respiration chambers
is considered as the gold standard but unsuitable for
large-scale individual animal measurements (Hammond
et al., 2016; Patra, 2016). One promising proxy for the
prediction of CH4 production is the concentration of
Methane prediction based on individual or groups of milk fatty
acids for dairy cows fed rations with or without linseed
Stefanie W. Engelke,1*UE]'Dú1 Michael Derno,1 Armin Tuchscherer,2 Klaus Wimmers,3 Michael Rychlik,4
Hermine Kienberger,5 Werner Berg,6 Björn Kuhla,1 and Cornelia C. Metges1,7*
1Institute of Nutritional Physiology “Oskar Kellner,” Leibniz Institute for Farm Animal Biology (FBN), Wilhelm-Stahl-Allee 2, 18196 Dummerstorf,
Germany
2Institute of Genetics and Biometry, Leibniz Institute for Farm Animal Biology (FBN), Wilhelm-Stahl-Allee 2, 18196 Dummerstorf, Germany
3Institute of Genome Biology, Leibniz Institute for Farm Animal Biology (FBN), Wilhelm-Stahl-Allee 2, 18196 Dummerstorf, Germany
4Analytical Food Chemistry, Technical University of Munich, Maximus-von-Imhof-Forum, 85354 Freising, Germany
5Bavarian Center for Biomolecular Mass Spectrometry, Gregor-Mendel-Strasse 4, 85354 Freising, Germany
6Department of Technology Assessment and Substance Cycles, Leibniz Institute for Agricultural Engineering and Bioeconomy (ATB),
Max-Eyth-Allee 100, 14469 Potsdam, Germany
7Nutritional Physiology and Animal Nutrition, Faculty of Agriculture and Environmental Sciences, University of Rostock, 18059 Rostock, Germany
J. Dairy Sci. 102:1–15
https://doi.org/10.3168/jds.2018-14911
© 2019, The Authors. Published by FASS Inc. and Elsevier Inc. on behalf of the American Dairy Science Association®.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Received April 11, 2018.
Accepted October 25, 2018.
*Corresponding author: metges@ fbn -dummerstorf .de

2ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
milk fatty acids (MFA; van Gastelen and Dijkstra,
2016; Negussie et al., 2017). Dietary composition affects
the relationship between MFA and CH4 emission pa-
rameters (Mohammed et al., 2011; Dijkstra et al., 2016;
Rico et al., 2016). Milk fatty acids synthesized de novo
are predominantly generated from rumen acetate and
BHB (Shingfield et al., 2013), which are derived from
fiber fermentation, and they are positively associated
with ruminal CH4 production (Shingfield et al., 2013;
Castro-Montoya et al., 2016a). Forage type in ruminant
diets can affect CH4 production as well as MFA pattern
(Beauchemin et al., 2008; Hart et al., 2015; van Gastel-
en et al., 2015). Likewise, a high starch content favors
propionate synthesis, which consumes fermentative
hydrogen and can thus reduce CH4 production (Hook
et al., 2010; Knapp et al., 2014). The MFA groups of
long-chain MUFA and PUFA and the individual C18:0
originate from plant fat-derived UFA (Ferlay et al.,
2013; Kliem and Shingfield, 2016; Meignan et al., 2017)
are either directly transferred into milk fat (e.g., C18:3
n-3) or influence rumen fermentation, thereby altering
the pattern and level of precursors for MFA synthesis
and affect the postabsorptive lipid metabolism (de novo
MFA synthesis, saturation or desaturation rates, or FA
elongation; Angulo et al., 2012; Lanier and Corl, 2015;
Meignan et al., 2017). In addition, dietary fat-associat-
ed decreases of fiber degradability and toxic effects of
PUFA on archaea reduce CH4 production (Maia et al.,
2007; Benchaar et al., 2015). Furthermore, fat supple-
ments can also affect DMI and, thus, net production of
short-chain MFA.
Several authors have used MFA concentrations to de-
velop CH4 prediction equations that differ in type and
numbers of explanatory MFA (e.g., van Lingen et al.,
2014; Castro-Montoya et al., 2016b; Rico et al., 2016).
For example, van Lingen et al. (2014) developed CH4
predictions based on 21 individual and 3 groups of MFA
concentrations, whereas Rico et al. (2016) included 83
individual MFA and sums of different MFA propor-
tions. van Gastelen and Dijkstra (2016) pointed out
that only C17:1 cis-9 and C18:1 cis-11 appeared in 2
or more of the prediction equations published, indicat-
ing that explanatory MFA are extremely diverse, which
likely is a consequence of differences in diet composition
(Dijkstra et al., 2011; Mohammed et al., 2011; Rico et
al., 2016). We were interested to know how prediction
equations change when the MFA data changes in the
number of variables and whether individual MFA or
groups of related MFA are used.
It has been also discussed whether the combination
of MFA concentrations with additional explanatory
parameters would improve prediction (Castro-Montoya
et al., 2016b; Negussie et al., 2017; van Gastelen et al.,
2017). The DMI of cows is known as the main determi-
nant for CH4 production (Knapp et al., 2014), and DMI
is associated with milk yield (Hristov et al., 2013). Con-
sequently, milk yield could be a potential alternative
for DMI as a variable in prediction equations (Hristov
et al., 2013; Negussie et al., 2017). We have reported
that a prediction equation based on MFA, determined
by mid-infrared spectroscopy, which contained DMI as
an additional explanatory variable, showed a similar
coefficient of determination when DMI was replaced by
ECM (Engelke et al., 2018).
Thus, our hypothesis was that CH4 prediction equa-
tions based on MFA depend on the number of MFA
or MFA group variables, and that inclusion of DMI
or ECM together with MFA improves prediction equa-
tions. Hence, the objectives of this study were (1) to
determine the effect of dietary composition on the
relationship between MFA concentrations and CH4
production, with special emphasis on the effect of lin-
seed supplementation; (2) to compare CH4 prediction
models based on differently sized and composed MFA
data sets; and (3) to test whether the inclusion of DMI
or ECM in addition to MFA concentrations as variable
improves prediction.
MATERIALS AND METHODS
Animals, Experimental Design, and Diets
The procedures performed in our study were in
agreement with the German Animal protection law
and approved by the relevant authority (Landesamt für
Landwirtschaft, Lebensmittelsicherheit und Fischere-
iwesen Mecklenburg-Vorpommern, Germany; permis-
sion no. 7221.3–1-014/14). Twenty lactating German
Holstein cows (106 ± 28 DIM, 29.5 ± 7.7 kg of ECM/d,
580 ± 57 kg of BW; mean ± SD), of which 15 cows
were in second and 5 cows were in third lactation, were
purchased from a dairy farm located in the region of
Mecklenburg-Western Pomerania, Germany. All cows
were sired by 1 bull (Omega 802670; Rinderzucht
Mecklenburg-Vorpommern GmbH, Woldegk, Ger-
many). Cows were kept in tiestalls, had free access to
water, and were offered a TMR for ad libitum intake.
Milking occurred at 0630 and 1630 h. To produce a
wide range of CH4 production values, we used 2 basal
TMR, which were composed close to what is commonly
used in German dairy farming. The major forage com-
ponent was either corn silage (CS) or grass silage (GS),
supplemented with (LS) or without linseed (L0; Table
1). Diets contained grass silage and corn silage at DM
levels of 130 and 450 g/kg (CS) or 360 and 190 g/kg
(GS), respectively. The LS diets contained 60 g of fat/

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 3
kg of DM, whereas the L0 diets had 30 g of fat/kg of
DM. The diets had the same target energy level of 7 MJ
of NEL/kg of DM, which was achieved by replacing the
linseed product with an isoenergetic amount of starch
in the TMR. Forage-to-concentrate ratio differed from
66:34 (GS-L0), 73:27 (GS-LS), 65:35 (CS-L0), to 68:32
(CS-LS). We randomly selected 5 of the 20 cows to be
fed the CS-L0 diet in the first 5-wk period (period A)
and the CS-LS diet in the second 5-wk period (period
B), whereas we randomly selected 5 other cows to be
fed the CS-LS diet during the first 5-wk period and
CS-L0 diet during the second 5-wk period. Another 5
cows were fed GS-L0 diet first and GS-LS diet second,
whereas the remaining 5 cows were given the opposite
treatments. The transition from the standard diet to an
experimental diet or from one experimental diet to the
other was made over 5 d in experimental week (EW) 0
and 6, respectively, in a step-wise fashion by replacing
20, 40, 60, 80, and 100% of DM of the standard diet
by the experimental diet. The study was conducted in
5 experimental blocks. Each block ran over a total of
12 EW and included 4 cows, all 4 diets, and 2 peri-
ods (A and B). That means 2 cows per each diet were
measured within each block. Cows assigned to the CS
Table 1. Ingredients and chemical composition of the experimental TMR consisting of basal rations based
largely on grass silage (GS) or corn silage (CS) without (L0) and with (LS) linseed supplementation (means
± SD; n = 5)
Item
GS
CS
L0 LS L0 LS
Ingredient (g/kg of DM)
Grass silage 343 ± 31.0 380 ± 15.0 122 ± 21.5 142 ± 34.3
Corn silage 181 ± 33.2 195 ± 37.9 445 ± 46.7 452 ± 44.0
Straw, barley 52.1 ± 7.0 60.8 ± 23.2 39.8 ± 7.2 34.0 ± 10.3
Grass hay 80.8 ± 12.0 93.5 ± 22.0 43.2 ± 3.1 43.5 ± 3.0
Corn, ground — — 92.7 ± 9.2 25.4 ± 27.2
Soy extract meal 25.4 ± 26.8 40.1 ± 19.2 114 ± 14.2 87.1 ± 9.2
Barley, ground 99.5 ± 17.9 — — —
Wheat, ground 101 ± 9.8 — — —
Linseed product1— 143 ± 11.4 — 137 ± 9.9
Concentrate2109 ± 19.7 78.4 ± 13.0 129 ± 7.3 65.0 ± 32.6
Mineral/vitamin mix39.1 ± 0.5 9.5 ± 0.6 10.1 ± 0.6 10.1 ± 0.6
Calcium carbonate4— — 4.1 ± 0.3 4.1 ± 0.3
Grass to corn silage ratio 1.90 1.95 0.27 0.31
Forage: concentrate ratio 66:34 73:27 65:35 68:32
DM (g/kg of FM5) 468 ± 33.0 452 ± 45.0 446 ± 32.0 439 ± 33.0
Nutrients (g/kg of DM)
Crude ash 69.9 ± 2.8 75.8 ± 2.5 66.1 ± 3.0 65.7 ± 4.2
CP 161 ± 9.0 173 ± 14.4 169 ± 11.7 163 ± 11.1
Crude fiber 166 ± 7.9 186 ± 7.1 155 ± 3.3 161 ± 4.5
Crude fat 28.1 ± 1.7 58.3 ± 7.8 29.8 ± 2.4 56.4 ± 6.5
Sugar 51.3 ± 15.7 50.1 ± 16.2 31.4 ± 7.9 30.3 ± 11.1
Starch 218 ± 12.0 110 ± 18.7 261 ± 25.4 216 ± 34.5
aNDF6371 ± 18.1 410 ± 12.9 330 ± 10.9 352 ± 9.2
ADF 198 ± 9.8 224 ± 7.9 186 ± 5.4 194 ± 5.0
NEL (MJ/kg of DM) 6.9 ± 0.2 6.9 ± 0.1 7.0 ± 0.1 7.1 ± 0.1
1Omegalin 60 (Spezialfutter Neuruppin GmbH und Co.KG, Neuruppin, Germany; per kg feed; 88% DM): 60%
extruded linseed Tradilin and 40% bran. Composition: 19.5% CP, 25% crude fat, 8.5% crude fiber; fatty acids
(% of fat): 6% palmitic acid (C16:0), 18.4% oleic acid (C18:1 cis-9), 18.5% linoleic acid (C8:2 cis-9,cis-12), 55%
linolenic acid (C18:3 cis-9,cis-12,cis-15), and 12.8 MJ of NEL/kg of DM.
2Concentrate MF 2000 (Vollkraft Mischfutterwerke GmbH, Güstrow, Germany; per kg feed; 88% DM): 33%
extracted soy meal, 20% corn, 17% wheat gluten, 8% extracted rapeseed meal, 5% sugar beet pulp, 2% sodium
hydrogen carbonate, 1.3% calcium carbonate, 0.2% sodium chloride. Composition: 24% CP, 3.3% crude fat,
6.2% crude fiber, 8.4% crude ash, 0.7% calcium, 0.5% phosphorus, 0.65% sodium, and 7.1 MJ of NEL/kg of
DM.
3Rinderstolz 9522 Salvana (Tierernährung, Kl.-O. Sparrieshoop, Germany; per kg feed; 88% DM): 39.3% cal-
cium carbonate, 21.7% monocalcium phosphate, 21% sodium chloride, 11.9% magnesium oxide, 2% sugar beet
molasses. Composition: 92% crude ash, 20% calcium, 8% sodium, 6% magnesium, 5% phosphorus, 1,000,000
IU of vitamin A, 200,000 IU of vitamin D3, and 4500 mg of vitamin E.
4Kreidekalk (Spezialfutter Neuruppin GmbH und Co.KG): calcium carbonate; 37% calcium.
5Fresh matter.
6aNDF = (amylase) neutral detergent fiber.

4ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
and GS diets did not differ in their mean baseline CH4
production (506 and 477 L/d, respectively; P = 0.441).
To keep feed composition as constant as possible
over time, experimental diets were mixed once a week,
conserved with 1% granulated propionic acid (BERGO
TMR-stabil G, Bergophor Futtermittelfabrik Dr.
Berger GmbH & Co.KG, Kulmbach, Germany), and
vacuum-packaged in thirty 40-kg plastic bags (NeuRo
Planen GmbH, Neuendorf, Germany). Cows were fed
ad libitum from these bags twice daily at 0730 and
1730 h. Feed intake was recorded daily. Three cows
had to be removed in the GS-LS period because of
illness or feed refusal. The diet composition was cal-
culated in accordance to recommendations of the Ger-
man Society of Nutritional Physiology (Gesellschaft für
Ernährungsphysiologie, 2001).
Feed Sampling and Analyses
Feed samples were collected during TMR mixing
and stored at −20°C. At the end of each experimental
block, feed samples were pooled to determine DM. The
DM content was determined by drying at 60°C for 24
h and then at 103°C for 4 h (Naumann et al., 1976).
Analyses of nutrient composition in feedstuffs (Table
1) were performed according to the Weender standard
analysis (Naumann et al., 1976), with modifications by
Van Soest et al. (1991), by the accredited feed labora-
tory of Landwirtschaftliche Untersuchungs-und Forsc-
hungsanstalt der LMS Agrarberatung GmbH (LUFA,
Rostock, Germany). The energy content (NEL) was cal-
culated according to the German Society of Nutritional
Physiology (Gesellschaft für Ernährungsphysiologie,
2001), except for the linseed product Omegalin 60 (Spe-
zialfutter Neuruppin GmbH und Co. KG, Neuruppin,
Germany) and concentrate MF 2000 (Vollkraft Mis-
chfutterwerke GmbH, Güstrow, Germany), which were
analyzed according to German feed regulations (Ger-
man Federal Law Gazette, 1981) §13 by LUFA. Nutri-
ent intake of cows was calculated from the DMI and
analyzed nutrient contents in feedstuffs (Supplemental
Table S1; https: / / doi .org/ 10 .3168/ jds .201814911).
Methane Measurements
Methane production (L/d) was recorded in EW 5
and 11 for 2 subsequent 24-h periods, each using 4
open-circuit respiration chambers as described by
Derno et al. (2009) and Bielak et al. (2016). Cows were
fed with respective TMR diets at 0730 and 1730 h and
feed intake was recorded continuously by feed troughs
placed on balances and summarized over 24 h. Milking
occurred at 0630 and 1630 h. Drinking water was freely
available. The concentrations of CH4 were measured at
6-min intervals throughout 23.9 h; the data were nor-
malized to 24 h. The temperature and relative humidity
in the chambers were 15°C and 65%, respectively. The
mean recovery rate of the chambers was 99.9 ± 0.96%.
Methane production parameters were calculated as
CH4 production (L/d), CH4 yield defined as liters of
CH4 per kilogram of DMI, and CH4 intensity defined
as liters of CH4 per kilogram of ECM (Table 2), where
ECM (kg/d) = [1.05 + 0.38 × milk fat (%) + 0.21 ×
milk protein (%)]/3.28 × milk yield (kg/d) (Spiekers et
al., 2009).
Milk Sampling and Analyses
During the CH4 measurements, aliquots of milk
from the evening of the first and the morning of the
second 24-h period were pooled proportionally ac-
cording to milk yield at each milking (0.25% of milk
yield). Proximate milk composition was analyzed by
the state control association Mecklenburg-Western
Pomerania (Landeskontrollverband für Leistungs-und
Qualitätsprüfung Mecklenburg-Vorpommern e.V.,
Table 2. Performance and methane emission parameters of cows fed basal rations based on grass silage (GS) or corn silage (CS) with or without
linseed supplementation (LS or L0, respectively)
Item
GS1
CS1
SE
P-value2
L0 LS L0 LS B L B × L
DMI (kg/d) 15.94 14.80 18.91 17.76 0.985 0.024 0.061 0.990
ECM (kg/d) 22.61 22.01 29.04 28.28 1.359 0.002 0.334 0.910
Milk fat (%) 4.48 4.50 4.08 3.83 0.202 0.035 0.364 0.292
CH4 emission parameters
CH4 production (L/d) 506.8 453.0 587.7 500.6 30.40 0.089 0.002 0.373
CH4 intensity (L/kg of ECM) 23.3 21.3 20.2 17.7 1.27 0.050 0.006 0.750
CH4 yield (L/kg of DMI) 32.7 31.9 31.1 28.2 1.84 0.289 0.011 0.138
1LSM.
2ANOVA F-test for the effects of basal ration (B), linseed supplementation (L), or their interaction (B × L).

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 5
Güstrow, Germany) for milk fat and protein content us-
ing mid-infrared spectroscopy (MilkoScan FT6000 and
MilkoScan FT+, Foss, Hillerød, Denmark). Aliquots of
the pooled milk samples were stored at −20°C until
MFA analysis using GC (Firl et al., 2014) performed
by the Bavarian Biomolecular Mass Spectrometry
Center (Freising, Germany). Milk lipid was extracted
by a 1:1 chloroform: methanol mixture and esterified
with trimethylsulfonium hydroxide to form FAME.
A GC-flame ionization detector instrument (Hewlett
Packard 6890, Palo Alto, CA) with a 7683 autosampler
was equipped with a 100 m × 0.25 mm column (CP
7420; 0.25 μm film thickness; Agilent Technologies,
Böblingen, Germany; Firl et al., 2014). Identifications
of peaks were made by comparison with known FAME
standards. The GC analyses resulted in the separation
and quantification (% of total lipids) of 46 individual
MFA. Because not all individual MFA were detected
in all milk samples, we excluded those which occurred
in less than 85% of the milk samples (C18:1 trans-10,
C18:2 trans-10,cis-12, and C21:0), which resulted in
a total of 43 individual MFA (Table 3). Because SFA
have been shown to be positively associated and UFA,
MUFA, PUFA, C18:1 cis and trans isomers, and n-3
MFA have been shown to be negatively associated with
CH4 emission (Chilliard et al., 2009; Castro-Montoya
et al., 2016a; van Gastelen and Dijkstra, 2016), we
were interested to know the predictive power of equa-
tions containing groups of biochemically related MFA
instead of individual MFA. Thus, MFA concentrations
were summed as groups of SFA, UFA, MUFA, PUFA,
C18:1 cis and trans isomers, and n-3 MFA (Supplemen-
tal Table S2; https: / / doi .org/ 10 .3168/ jds .2018 -14911)
according to our previous study in which MFA were
predicted by infrared spectroscopy (Engelke et al.,
2018).
Calculations and Statistical Analysis
ANOVA. The dependent variables CH4 production,
DMI, and ECM yield were measured on 2 consecutive
days in EW 5 and 11 and were averaged per day in
each EW. Data were analyzed with repeated measures
ANOVA using PROC MIXED (SAS/STAT 9.3; SAS
Institute Inc., Cary, NC). The model contained the
fixed effects of basal ration (CS, GS), linseed supple-
mentation (LS, L0), the interaction effect between basal
ration and linseed supplementation, as well as effects of
experimental blocks (1–5), periods (A or B), and the
order (LS in period A or B first). The covariance struc-
ture was set to be compound symmetry. Effects were
considered significant at P < 0.05 and least squares
means were compared using the Tukey test with the
SLICE statement for performing a partitioned analysis
of the least squares means for the interaction. Data are
presented as least squares means ± standard error if
not given otherwise.
Regression Models and Validation. Correlations
were calculated between CH4 production (L/d) values
and concentrations of individual MFA and MFA groups
using PROC CORR of SAS. Regression equations for
the dependent variable CH4 production (L/d) were
constructed from MFA data of each diet separately
using PROC REG of SAS with the STEPWISE vari-
able selection method. Regression models for MFA
data from combined data categorized by basal diets or
linseed supplementation, as well as for data from all 4
diets collectively, were also estimated. Six data sets of
variables were used for the stepwise variable selection
to enter in the regression equations. Data sets 1 to 3
included 43 different individual MFA, whereas data
sets 4 to 6 were generated according to the MFA group-
ing of Engelke et al. (2018) and contained 7 groups
of MFA (SFA, UFA, MUFA, and PUFA, C18:1 trans,
C18:1 cis, and n-3 MFA) as summed concentrations
plus the individual MFA C16:0 and C18:0 as key MFA
in milk fat metabolism (Supplemental Table S2; https:
/ / doi .org/ 10 .3168/ jds .2018 -14911). Data set 1 included
the 43 individual MFA only and data set 2 included
43 individual MFA and ECM data, whereas data set 3
included 43 individual MFA and DMI data. Data set
4 included the reduced data set of 7 groups of MFA
and the 2 individual MFA only. Data set 5 included
the same reduced data set of MFA variables as well as
ECM, whereas data set 6 included the same reduced
data set of MFA as well as DMI.
To have the most relevant explanatory variables in the
regression equations and to avoid over-fitted regression
models, we preselected candidate explanatory variables
through their correlation coefficients with the depen-
dent variable (i.e., CH4 production). For this purpose,
only variables that had a significant correlation (P <
0.05) with CH4 production were allowed to potentially
enter in the models. Thereafter, the preselected vari-
ables were allowed to enter and remain in the regression
equations with a significance threshold of P ≤ 0.05 (i.e.,
entry and stay levels of the stepwise variable selection
method). Subsequently, explanatory variables in the
final regression equations derived from data set 1 and 4
were made available to potentially enter in the regres-
sion models for data sets 2 and 5 together with ECM
or 3 and 6 together with DMI, respectively. In this way,
ECM or DMI was used to additionally improve regres-
sion models derived from data sets 1 and 4 or could
replace 1 or more of the explanatory variables in the
final regression models. In the final regression models

6ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
of each data set, multicollinearity was assessed through
variation inflation factor, and explanatory variables
were allowed to have a variation inflation factor <10,
as suggested by Kaps and Lamberson (2004).
As cows were used rotationally with 2 different di-
ets (L0 and LS) at 2 different periods, the regression
equations based on data specific to basal ration and
all diets included pooled data of periods (e.g., EW 5
and 11). This was possible because period did not af-
fect CH4 production and MFA profiles. The regression
equations based on data specific to linseed supplemen-
tation were not concerned with pooled data due to the
experimental setup. The performance of the developed
CH4 prediction equations was assessed by an internal
cross-validation with the existing data set. A cross-
validation was performed by leaving 1 animal out at a
time and performing a calibration with the remaining
animals per model (each diet separately, basal diets,
linseed supplementation or not, all diets collectively;
Moraes et al., 2014). The root mean square error of
Table 3 Individual milk fatty acid (MFA) composition (% of total lipids) of cows fed basal rations based on grass silage (GS) or corn silage (CS)
with (LS) or without (L0) linseed supplementation
MFA (% of total lipids)
GS1
CS1
SE
P-value2
L0 LS L0 LS B L B × L
C4:0 3.81 3.86 3.78 4.14 0.173 0.467 0.123 0.248
C6:0 2.46 2.11 2.50 2.22 0.094 0.472 0.001 0.577
C8:0 1.45 1.08 1.17 1.48 0.056 0.368 0.001 0.410
C10:0 3.25 2.04 3.34 2.32 0.122 0.165 0.001 0.265
C10:1 0.41 0.25 0.39 0.24 0.019 0.681 0.001 0.734
C11:0 0.13 0.03 0.10 0.05 0.016 0.869 0.001 0.125
C12:0 3.92 2.20 3.97 2.57 0.147 0.184 0.001 0.135
C12:1 0.12 0.06 0.11 0.06 0.008 0.885 0.001 0.651
C13:0 0.16 0.06 0.15 0.09 0.015 0.728 0.001 0.069
C14:0 11.13 8.79 11.55 9.45 0.239 0.038 0.001 0.510
C14:0 iso 0.10 0.09 0.12 0.09 0.013 0.398 0.019 0.128
C14:1 cis-9 1.36 0.91 1.32 0.91 0.068 0.791 0.001 0.637
C15:0 1.37 0.80 1.25 0.89 0.095 0.851 0.001 0.142
C15:0 iso 0.19 0.17 0.24 0.18 0.015 0.082 0.012 0.165
C15:0 anteiso 0.42 0.38 0.44 0.44 0.036 0.269 0.473 0.518
C16:0 34.40 24.05 33.07 22.93 1.007 0.312 0.001 0.860
C16:0 iso 0.25 0.21 0.28 0.24 0.030 0.430 0.031 0.936
C16:1 cis-9 2.06 1.39 1.89 1.11 0.137 0.120 0.001 0.593
C16:1 trans-9 0.03 0.03 0.03 0.04 0.004 0.090 0.058 0.558
C17:0 0.58a0.45c 0.56a0.51b0.020 0.306 0.001 0.015
C17:0 iso 0.28 0.26 0.30 0.27 0.021 0.420 0.132 0.747
C17:0 anteiso 0.45 0.36 0.40 0.47 0.028 0.363 0.001 0.547
C17:1 cis-9 0.26 0.17 0.23 0.16 0.017 0.244 0.001 0.373
C18:0 7.50 12.81 8.52 14.22 0.495 0.022 0.001 0.632
C18:0 iso 0.05 0.03 0.05 0.04 0.004 0.127 0.010 0.338
C18:1 cis-9 17.33b26.20a* 17.21b23.91a* 0.889 0.306 0.001 0.013
C18:1 cis-11 0.61 0.58 0.48 0.62 0.095 0.585 0.588 0.377
C18:1 cis-12 0.23 0.63 0.29 0.74 0.034 0.017 0.001 0.455
C18:1 trans-9 0.37 1.06 0.57 1.29 0.125 0.014 0.001 0.936
C18:1 trans-11 0.97 2.80 1.05 3.02 0.162 0.283 0.001 0.640
C18:2 cis-9,cis-12 2.15 2.92 2.24 1.87 0.576 0.396 0.666 0.219
C18:2 cis-9,trans-11 0.52 1.21 0.52 1.18 0.073 0.828 0.001 0.836
C18: 3n -6 0.03 0.03 0.03 0.03 0.002 0.655 0.309 0.843
C18: 3n -3 0.62 1.15 0.38 1.12 0.068 0.021 0.001 0.120
C20:0 0.11 0.14 0.13 0.15 0.007 0.126 0.001 0.429
C20:1 cis-11 0.08 0.11 0.05 0.06 0.023 0.117 0.175 0.474
C20: 2n -6 0.02 0.02 0.03 0.02 0.006 0.944 0.652 0.764
C20: 3n -6 0.07b0.04c 0.10a0.05c0.006 0.016 0.001 0.003
C20: 4n -6 0.07 0.07 0.07 0.06 0.008 0.822 0.385 0.416
C20: 5n -3 0.05 0.07 0.04 0.05 0.012 0.082 0.055 0.819
C22:0 0.03 0.02 0.02 0.02 0.003 0.100 0.495 0.784
C22: 5n -3 0.08a0.06b 0.07a0.08ab 0.007 0.642 0.168 0.043
C24:0 0.03 0.03 0.03 0.03 0.003 0.251 0.916 0.077
a–cValues within a row with differing superscripts denote B × L interactions (P < 0.05).
1LSM.
2ANOVA F-test for the effects of basal ration (B), linseed supplementation (L), or their interaction (B × L).
*P < 0.10.

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 7
cross-validation and the cross-validation coefficient of
determination (R2
CV) were estimated (Moraes et al.,
2014). In addition, we report adjusted coefficient of
determination (R2
Adj) to account for the number of
variables in the final models. We further evaluated the
models using the concordance correlation coefficient
(CCC) as reported earlier (Dijkstra et al., 2016; van
Gastelen et al., 2017). Due to the low number of obser-
vations (n = 7–10) and variables with multicollinearity,
we refrained from reporting predictions for individual
diets.
RESULTS
Animal Performance and Methane Production
Cows fed the CS rations had a higher DMI (P =
0.024) and ECM yield (P = 0.002) than those fed GS
rations (Table 2). Accordingly, CP, crude fat, starch,
and NEL intakes (P < 0.05) were higher with CS than
with GS rations (Supplemental Table S1; https: / / doi
.org/ 10 .3168/ jds .2018 -14911). The intake of sugar was
higher (P = 0.005) with GS rations but fiber intake was
similar with both basal diets. The percentage of milk
fat was higher (P = 0.035) with GS compared with
CS feeding (Table 2). Linseed supplementation had no
effect on ECM and milk fat (P > 0.3), but we noted
a tendency toward a lower DMI (P = 0.061) with LS
compared with L0 diets (Table 2). Feeding LS rations
resulted in twice the crude fat intake (P = 0.001) as
with L0 rations (Supplemental Table S1).
Cows fed the CS basal diets tended to have higher
CH4 production compared with those cows fed the GS
rations (P = 0.089; Table 2). The CH4 yield ranged
from a maximum 50 L/kg of DMI in cows fed the GS-
L0 ration to a minimum of 21 L/kg of DMI with the
CS-LS diet (Supplemental Table S3; https: / / doi .org/
10 .3168/ jds .2018 -14911). Basal diet had no effect on
CH4 yield (P = 0.289), but CH4 intensity was lower
(P = 0.050) by about 15% with CS than with GS ra-
tions. Furthermore, LS decreased the level of all CH4
emission parameters as compared with cows fed L0.
The reduction of enteric CH4 emission levels due to LS
amounted to 10 and 13% for CH4 intensity and produc-
tion, respectively (P < 0.01). Methane yield (L/kg of
DMI) was reduced (P = 0.011) by approximately 6%
when diets with LS were fed.
MFA Composition and Correlations
with Methane Production
With CS compared with GS diets, C14:0, C18:0,
C18:1 cis-12, C18:1 trans-9, C20:3 n-6, and the sum of
C18:1 trans MFA concentrations were higher whereas
C18:3 n-3 and the sum of n-3 MFA were lower (Tables
3 and 4; P < 0.04). The LS diets resulted in lower
concentrations of individual even-chain SFA from C6:0
to C16:0, the group of SFA (P < 0.001), the off-chain
SFA from C11:0 to C17:0 (P < 0.001), and the indi-
vidual MUFA C10:1, C12:1, C14:1 cis-9, C16:1 cis-9,
C17:1 cis-9, as well as C18:0 iso and C20:3 n-6 (P <
0.05; Table 3). Most of C18 PUFA and C20:0 MFA
Table 4. Milk fatty acid (MFA) groups (% of total lipids) as sums of concentrations of individual MFA of cows fed basal rations based on grass
silage (GS) or corn silage (CS), with (LS) or without (L0) linseed supplementation
MFA (% of total lipids)
GS1
CS1
SE
P-value2
L0 LS L0 LS B L B × L
∑ SFA372.06 60.32 72.32 62.41 1.453 0.457 0.001 0.369
∑ UFA427.43 39.09 27.10 36.60 1.440 0.377 0.001 0.279
∑ MUFA 23.81 33.88 23.63 32.15 1.044 0.470 0.001 0.153
∑ PUFA 3.62 5.55 3.47 4.45 0.652 0.332 0.011 0.357
∑ C18:1 cis518.17 27.32 17.99 25.26 0.948 0.362 0.001 0.053
∑ C18:1 trans61.33 3.72 1.62 4.31 0.193 0.014 0.001 0.395
∑ n-3 FA70.76 1.29 0.49 1.25 0.079 0.031 0.001 0.128
∑ n-6 FA82.34 3.07 2.47 2.02 0.585 0.098 0.001 0.092
1LSM.
2ANOVA F-test for the effects of basal ration (B), linseed supplementation (L), or their interaction (B × L).
3Sum of C4:0, C6:0; C8:0, C10:0, C11:0, C12:0, C13:0, C14:0, C14:0 iso, C15:0, C15:0 iso, C15:0 anteiso, C16:0, C16:0 iso, C17:0, C17:0 iso,
C17:0 anteiso, C18:0, C18:0 iso, C20:0, C22:0, and C24:0.
4Unsaturated fatty acids (FA) represent the sum of MUFA (C10:1, C12:1, C14:1 cis-9, C16:1 cis-9, C16:1 trans-9, C17:1 cis-9, C18:1 cis-9,
C18:1 cis-11, C18:1 cis-12, C18:1 trans-9, C18:1 trans-11, and C20:1 cis-11) and PUFA (C18:2 cis-9,trans-11, C18:2 cis-9,cis-12, C18:3 cis-6,cis-
9,cis-12, C18:3 cis-9,cis-12,cis-15, C20:2 n-6, C20:3 n-6, C20:4 n-6, C20:5 n-3, and C22:5 n-3).
5Sum of C18:1 cis-9, cis-11, and cis-12.
6Sum of C18:1 trans-9 and trans-11.
7Sum of C18:3 n-3, C20:5 n-3, and C22:5 n-3.
8Sum of C18:2 n-6, C18:3 n-6, C20:2 n-6, C20:3 n-6, and C20:4 n-6.

8ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
concentrations were higher with LS-containing diets (P
< 0.05; Table 3). A basal ration × linseed interaction
was found for 4 of the 43 individual MFA (P < 0.05;
C17:0, C18:1 cis-9, C20:3 n-6, and C22:5 n-3; Table 3).
The concentrations of the MFA groups UFA, MUFA,
PUFA, sums of C18:1 cis or trans isomers, and the sum
of n-3 FA were higher with LS compared with L0 diets
(P < 0.01; Table 4).
For combined data of all diets collectively, positive
correlations were found between CH4 production and
concentrations of selected individual SFA, the sum of
all SFA, as well as C18:3 n-6 (P ≤ 0.05; Figure 1 and
Supplemental Table S4; https: / / doi .org/ 10 .3168/ jds
.2018 -14911). We observed negative correlations be-
tween CH4 production and the levels of individual MFA
of C15:0 iso, C17:1 cis-9, C18:1 cis-9, C18:1 cis-11,
C18:1 cis-12, C18:1 trans-9, C18:3 n-3, and C22:0 and
groups of UFA, MUFA, C18:1 cis and trans isomers,
and n-3 MFA (P ≤ 0.05). Correlations of CH4 produc-
tion with the levels of individual MFA differed for data
of the various pooled diets (Figure 1 and Supplemental
Table S4). For example, correlations between C10:0
and CH4 production ranged from 0.07 (P = 0.76) for
the L0 diets to 0.63 (P = 0.006) for the LS diets, and
for C18:3 n-3 from −0.17 (P = 0.509) for the GS diets
to −0.70 (P = 0.002) for the LS diets; correlations for
C24:0 ranged from 0.14 (P = 0.605) for the LS diets to
0.83 (P = 0.001) for the CS diets.
Methane Prediction Models Using the Complete
Set of MFA Variables
The prediction of CH4 production based on data sets
1 to 3 resulted in coefficients of determination of the
models (R2
Model) between 0.36 and 0.91 (P < 0.01),
R2Adj between 0.32 and 0.89, R2CV between 0.47 and 0.92,
and root mean square error between 38.78 and 97.41,
whereas CCC was between 0.53 and 0.95 (Table 5). The
R2Adj values were slightly but systematically lower than
that for R2Model. The additional inclusion of ECM (data
set 2) resulted in higher R2Model values, ranging from
0.61 to 0.91, and R2CV, from 0.71 to 0.92, as compared
with data set 1. Nevertheless, for all diets combined,
the R2Model (0.69) and R2CV (0.74) of the equations for
data set 2 were lower compared with that of data set
1 (R2Model = 0.81 and R2CV = 0.85). The inclusion of
DMI (data set 3) instead of ECM (data set 2) resulted
in comparable prediction values, with an R2Model range
from 0.62 to 0.91 and R2CV from 0.70 to 0.92 (Table 5).
With the exception of the GS diets, the R2CV and CCC
values based on data sets 1 to 3 were at least 0.74 and
0.82, respectively, or higher. The R2CV based on GS
diets was always lower compared with that of CS diets
Figure 1. Heatmap of correlations between individual and groups
of milk fatty acids (% of total lipids) and methane production (L/d)
calculated from data of the combined basal diets [grass silage (GS),
corn silage (CS)], each basal diet with (L0) or without (LS) linseed
supplementation, and all diets collectively.

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 9
(R2
CV from 0.47 to 0.71 vs. 0.89 to 0.92). We found no
discernible pattern of predictive MFA as explanatory
variables in the prediction equations. However, C18:1
cis-9 and C24:0 appeared more often in the prediction
equations than others (Supplemental Table S5; https: /
/ doi .org/ 10 .3168/ jds .2018 -14911).
Methane Prediction Models Using a Reduced
Number of MFA Variables
Three data sets (data sets 4 to 6) comprising the con-
centrations of a reduced number of MFA and biochemi-
cally related groups of summed MFA (Supplemental
Table S2; https: / / doi .org/ 10 .3168/ jds .2018 -14911)
were used to develop multiple regression equations. Us-
ing data set 4, prediction of CH4 production resulted in
R2Model between 0.26 and 0.57, R2Adj between 0.21 and
0.52, CCC between 0.42 and 0.72, and a R2CV between
0.31 and 0.67 (Table 5). The additional inclusion of
ECM (data set 5) or DMI (data set 6) increased the
R2Model and R2CV values, respectively, compared with
data set 4. The R2Adj values were similar to R2Model but
systematically lower. Models for CS diets always re-
sulted in a higher prediction (R2CV from 0.56 to 0.84)
and higher CCC values (from 0.59 to 0.89) compared
with GS diets. Similarly, models of data from L0 diets
(R2CV from 0.67 to 0.80, CCC from 0.72 to 0.85) gave
higher predictions as compared with LS diets (Table 5).
Table 5. Summary of quality parameters and validation results of the multiple regression equations predicting
methane production (CH4, L/d) using the complete data set of 43 milk fatty acid (MFA) variable (data sets
1–3), or a reduced number of MFA variables and groups of MFA (data sets 4–6)1
Pooled diets N R2Model PModel R2Adj CCC RMSE R2CV
Data set 1
GS 17 0.36 0.011 0.32 0.53 97.41 0.47
CS 20 0.86 0.001 0.84 0.93 45.54 0.89
L0 20 0.85 0.001 0.83 0.92 60.74 0.87
LS 17 0.78 0.001 0.75 0.88 41.77 0.84
All diets 37 0.81 0.001 0.78 0.90 56.38 0.85
Data set 2
GS 17 0.61 0.001 0.58 0.75 77.84 0.71
CS 20 0.91 0.001 0.89 0.95 41.79 0.91
L0 20 0.91 0.001 0.89 0.95 48.25 0.92
LS 17 0.78 0.001 0.75 0.88 41.77 0.84
All diets 37 0.69 0.001 0.67 0.82 75.28 0.74
Data set 3
GS 17 0.62 0.001 0.60 0.77 79.74 0.70
CS 20 0.91 0.001 0.89 0.95 39.55 0.92
L0 20 0.90 0.001 0.88 0.95 50.25 0.91
LS 17 0.81 0.001 0.79 0.90 38.78 0.87
All diets 37 0.74 0.001 0.73 0.85 66.86 0.80
Data set 4
GS 17 0.26 0.035 0.21 0.42 106.50 0.31
CS 20 0.42 0.002 0.39 0.59 83.24 0.56
L0 20 0.57 0.001 0.52 0.72 93.39 0.67
LS 17 0.49 0.002 0.46 0.66 60.61 0.62
All diets 37 0.30 0.001 0.28 0.46 93.75 0.48
Data set 5
GS 17 0.61 0.001 0.58 0.75 77.67 0.71
CS 20 0.74 0.001 0.71 0.85 61.84 0.79
L0 20 0.70 0.001 0.69 0.82 74.30 0.80
LS 17 0.49 0.002 0.46 0.66 60.61 0.62
All diets 37 0.69 0.001 0.67 0.82 64.57 0.80
Data set 6
GS 17 0.62 0.001 0.60 0.77 79.56 0.70
CS 20 0.80 0.001 0.78 0.89 54.31 0.84
L0 20 0.73 0.001 0.72 0.85 73.34 0.80
LS 17 0.56 0.001 0.53 0.72 57.01 0.68
All diets 37 0.73 0.001 0.71 0.84 62.45 0.81
1Each row summarizes validation results for a regression equation specific to data of diets based on combina-
tions of basal ration [grass silage (GS), corn silage (CS)] and linseed supplementation [without (L0), with (LS)]
or all diets collectively. Data set 1 included MFA only. Data sets 2 and 3 included individual MFA and ECM
or DMI, respectively, as independent variables. Data set 4 included MFA variables (reduced number of vari-
ables) only, whereas data sets 5 and 6 additionally included ECM or DMI, respectively. The R2Model and PModel
values, the adjusted R2Adj, the concordance correlation coefficient (CCC) of the model, the root mean square
error (RMSE) of cross-validation as well as the cross-validation coefficient of determination (R2CV) are given

10 ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
However, for the LS diets, the inclusion of ECM (data
set 5) did not improve the model. Predictions based on
data from all 4 diets collectively including ECM as an
explanatory variable in the model enhanced the predic-
tive power by 130% (R2Model = 0.69) and 69% (R2CV
= 0.80) compared with data set 4 (R2Model = 0.30 and
R2CV = 0.48). The dominant predictive MFA variable in
the regression equations of data set 4 to 6 was the MFA
group MUFA (Supplemental Table S5; https: / / doi .org/
10 .3168/ jds .2018 -14911).
DISCUSSION
Animal Performance, Methane Production
Parameters, and MFA Composition
The CS diets resulted in a higher DMI and ECM
yield than the GS diets, which was comparable to what
was reported previously for diets with higher propor-
tions of corn silage than grass silage (Kliem et al., 2008;
Sterk et al., 2011; Livingstone et al., 2015). In contrast
to earlier studies, the type of the basal ration did not
affect CH4 yield (L/kg of DMI) of cows (Beauchemin et
al., 2008; Hart et al., 2015; van Gastelen et al., 2015).
This was likely due to similar crude fiber, ADF, and
NDF intakes with grass- and corn silage-based diets.
To make the diets isoenergetic, wheat and barley were
added to the GS-L0 diet, and this added starch, diluted
the fiber content, and aligned the fiber intake of the 2
basal diets. Furthermore, effects of starch and sugar
in the GS-L0 diets might have reduced CH4 produc-
tion, which otherwise was expected to be higher with
diets containing higher proportions of grass silage (van
Gastelen et al., 2015). Although the NDF/DMI value
was still higher in GS-based diets, the DMI and energy
intake of cows fed GS diets was, on average, 3 kg and
17% MJ of NEL less per day, respectively, as compared
with the CS-based diets. This is a limitation of our
study and possibly affected the width of range of CH4
production. Methane production (L/d) tended to be
higher with CS compared with GS diets due to the
higher DMI, which is known to explain 52 to 64% of
CH4 production (Knapp et al., 2014). In contrast, CH4
intensity tended to be lower in the CS than in the GS
group due to higher ECM yield, which illustrates the
importance of a high milk performance to mitigate CH4
intensity (Yan et al., 2010; Zehetmeier et al., 2012;
Gerber et al., 2013).
We observed only few differences in MFA concentra-
tions between basal rations, although others reported
more diverse MFA patterns with different silage types
(Kliem et al., 2008; Livingstone et al., 2015; van Gas-
telen et al., 2015). For example, diets with different
proportions of grass and corn silage differed in the MFA
concentrations of C18:1 isomers, total CLA, C18:2 n-6,
and C18:3 n-3 (Kliem et al., 2008; van Gastelen et al.,
2015). The small differences in concentrations of only
a few MFA among basal diets found here were due to
similar crude fiber, ADF, and NDF intakes with the
basal diets, as discussed above. This can limit the range
of MFA concentrations, and thus may affect CH4 pre-
diction.
In the present study, dietary supplementation of
linseed decreased the level of all CH4 emission param-
eters by 6, 10 and 13% for CH4 yield, intensity, and
production, although the reductive effects were greater
with CS diets, as was observed earlier (Benchaar et al.,
2015; Martin et al., 2016). Others found reductions in
CH4 yield between 8 and 20% by feeding various diets
supplemented with extruded linseed (5 to 10% of di-
etary DM) or linseed oil (4% of dietary DM; Benchaar
et al., 2015; Martin et al., 2016; Bayat et al., 2018).
A 10 g/kg of DM increase in dietary fat resulted in a
lower CH4 yield by 1 g/kg of DMI in cattle (Grainger
and Beauchemin, 2011). An increase of 1% in dietary
lipid content led to a 4 to 5% reduction of CH4 produc-
tion (Clark, 2013). In our study, the LS diets contained
28 g/kg of DM more crude fat than the L0 diets, which
corresponds to a lower CH4 yield by 2.8 g/kg of DM,
or 12 to 15%. The actual reduction of CH4 yield by LS
supplementation of GS and CS diets was 0.7 and 2.1
g/kg of DM or 2.7 and 9.4%, respectively. Thus, in our
study, the linseed supplementation reduced CH4 yield
less than predicted by Grainger and Beauchemin (2011)
or Clark (2013), but the reduction was larger with the
corn silage-based diet. The lesser mitigation observed
in our study compared with Grainger and Beauchemin
(2011) might be due to the fact that we did not add
linseed oil but linseeds to the diets. This might reduce
or slow down the release of linseed oil in the rumen and
is thus less suppressive to methanogens. As argued by
Benchaar et al. (2015), the type of basal diet might
influence the effect of linseed oil on CH4 mitigation.
For example, Chung et al. (2011) observed a mitigation
effect of linseed only in combination with feeding barley
silage (33% of CH4 yield) but not with grass hay. This
is similar to what was found in our study, which might
be due to the opposite effect of linseed and grass, grass
hay, or grass silage on CH4 production.
Linseed supplementation tended to reduce DMI.
Others observed no decrease of DMI when linseed sup-
plementation was moderate (≤60 g of fat/kg of DM) as
compared with a control diet (Ferlay et al., 2013 Kliem
et al., 2017; Meignan et al., 2017). In the present study,
the higher concentrate content in L0 diets was used to
compensate for the higher energy density of linseed,

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 11
which might have led to a slightly higher DMI because
concentrate is known as a favored feedstuff of cows
when forage is provided ad libitum (Reynolds, 2006;
Allen, 2014). The LS-supplemented diets contained
twice the amount of fat than the nonsupplemented
diets, which affected the concentrations of the majority
of MFA. Almost all SFA were decreased with linseed
supplementation through the direct inhibitory effect of
long-chain SFA and UFA on fiber digestibility and de
novo synthesis of MFA (Maia et al., 2007; Glasser et
al., 2008; Shingfield et al., 2013). In contrast, C4:0 in
milk fat was not affected by LS-containing diets. Ef-
fects of linseed on C4:0 concentrations are inconsistent
in the literature (Ferlay et al., 2013; van Lingen et al.,
2014; Kliem et al., 2017). Concentrations of off- and
branched-chain MFA, which are derived from micro-
bial synthesis (Vlaeminck et al., 2015), decreased when
diets were supplemented by linseed. Lower levels of
off- and branched-chain MFA are presumably related
to the toxic effects of dietary PUFA on rumen micro-
biota (Enjalbert et al., 2017). The intake of increased
levels of C18:3 n-3 MFA with LS-supplemented diets
increased the proportion of C18 MFA in milk fat due
to biohydrogenation (Buccioni et al., 2012; Meignan
et al., 2017). Diets containing linseed decreased SFA
and increased UFA, trans MFA, n-3 MFA, and C18:0
(Chilliard et al., 2009; Ferlay et al., 2013; Meignan et
al., 2017); the magnitude of changes in MFA pattern
depended on the amount and form of linseed supply
(Chilliard et al., 2009; Shingfield et al., 2013). Ferlay
et al. (2013) showed that the C18:1 cis/trans isomer
profile was specific for the forage type. This was simi-
lar to our study, where the C18:1 trans isomers were
forage type-dependent, whereas the addition of linseed
increased both C18:1 cis and trans isomers.
Diet Effects on the Relationship Between MFA
and CH4 Production and CH4 Prediction
Each diet produces a characteristic MFA pattern
(Kliem et al., 2008; Ferlay et al., 2013; Shingfield et
al., 2013). We compared the correlations of individual
MFA with CH4 production among pooled diets (GS,
CS, L0, LS, all diets) and found—depending on the
dietary group—positive, negative, or no correlations.
When using MFA data of all 4 diets collectively, we ob-
served moderate correlations between CH4 production
and individual MFA (maximum r = 0.52 vs. −0.62).
In a meta-analysis combining data from studies with
a variety of different diets, weak to moderate correla-
tions between CH4 yield and intensity and MFA were
detected (maximum r = 0.36 vs. −0.56; van Lingen et
al., 2014). This suggests that direction and strength of
correlations between individual MFA and CH4 emis-
sion parameters can be explained by the interaction
of different dietary constituents on MFA patterns and
CH4 production. Thus, when MFA data derived from
different diets were combined to determine the overall
relationship between individual MFA and CH4 produc-
tion, the average correlation was weak to moderate. In
general, our results confirm earlier observations that
correlations between CH4 production and some of the
de novo-synthesized MFA and groups of SFA are posi-
tive, but negative for C18:1 cis isomers, UFA, MUFA,
and n-3 FA (Chilliard et al., 2009; Mohammed et al.,
2011; van Lingen et al., 2014).
When comparing CH4 prediction among the tested
diets, the coefficients of determination were lower for
GS diets compared with CS diets. The effect of basal
diet (GS and CS) on the CH4 prediction potential was
likely limited in our study because of few differences in
MFA concentrations between basal rations, as discussed
above. The prediction equations based on MFA data of
diets containing linseed (LS vs. L0) produced 15 to 23%
lower R2CV values compared with data from diets with
no linseed, irrespective of the data set used. That linseed
supplementation deteriorates CH4 prediction confirms
earlier observations (Williams et al., 2014; Dijkstra et
al., 2016; Rico et al., 2016). Williams et al. (2014) con-
cluded that the CH4 prediction model by Chilliard et
al. (2009) is a result of linseed oil supplementation and
cannot accurately predict CH4 when cows are fed other
diets. Furthermore, Dijkstra et al. (2016) adopted the
CH4 prediction equations by van Lingen et al. (2014)
and concluded that MFA profile and CH4 yield of cows
fed grass- or grass silage-based diets differ from those
fed other diet types, especially for diets containing fat
additives. In this context, Rico et al. (2016) pointed
out that interaction effects of basal rations and linseed
supplementation may influence prediction equations.
Thus, as outlined above, we concluded that the high
intake of n-3 FA affected interrelated biochemical
pathways on several tissue and metabolic levels but in
opposite directions for CH4 production and MFA con-
centrations. Furthermore, we cannot exclude that tis-
sue FA synthesis, desaturation, and elongation, which
are unrelated to changes in rumen fermentation and
thus CH4 production, play a more dominant role when
higher amounts of n-3 FA were fed.
Methane Prediction Models Using Full and Reduced
Data Sets for MFA Variables
Our second objective was to compare CH4 prediction
models based on differently sized and composed data
sets of MFA. Therefore, we compared CH4 prediction

12 ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
equations based on all 43 quantified MFA (full set; data
sets 1 to 3) to equations constructed from a reduced
number of MFA variables containing groups of biochemi-
cally related MFA (data sets 4 to 6). In previous studies
it has been shown that the selected MFA groups as
well as the individual MFA are positively (SFA, C16:0)
or negatively (UFA, MUFA, PUFA, n-3 MFA, C18:0)
associated with CH4 emission (Castro-Montoya et al.,
2016a; van Gastelen and Dijkstra, 2016). Models based
on all 43 MFA showed a better prediction than those
based on the reduced number of MFA group variables.
In other studies, the prediction for CH4 emission based
on 24 to 83 MFA variables ranged between coefficients
of determination of 0.47 and 0.95 (Chilliard et al., 2009;
van Lingen et al., 2014; Rico et al., 2016), which is in
the same range as with our equations, except for the
GS diets in data sets 1 and 4.
The MFA entered in the equations as explanatory
variables using data sets 1 to 3 differed within the data
sets. Explanatory individual MFA frequently (4 to 7
times) occurring in our prediction equations were also
found in other CH4 prediction equations [i.e., C17:1
cis-9 (Mohammed et al., 2011; Castro-Montoya et al.,
2016b; Rico et al., 2016), C18:1 cis-9 (van Lingen et
al., 2014), and C24:0 (van Gastelen et al., 2017)]. van
Gastelen and Dijkstra (2016) pointed out that only
C17:1 cis-9 and C18:1 cis-11 appeared in several pub-
lished equations, but in our study C18:1 cis-11 did not
play a major role. When ECM or DMI were addition-
ally included, C18:1 cis-11 was not considered. The
explanatory MFA variable C17:1 cis-9 is derived from
rumen microbial membrane lipids or from propionate
(Vlaeminck et al., 2006) and is negatively associated
with CH4 production (g/d; Mohammed et al., 2011;
Castro-Montoya et al., 2016b; Rico et al., 2016). In
contrast, C18:1 cis-9 is derived directly from plant feed
or is a desaturation product of C18:0, one of the major
products when diets rich in forage and plant oils are fed
(Alves et al., 2013; Meignan et al., 2017). Long-chain
MFA (≥C20) are scarcely reported because they seem
to be less frequently analyzed in CH4-prediction studies
(Dijkstra et al., 2011; van Lingen et al., 2014). Only a
few publications reported long-chain MFA (C20:1 cis-9,
C20:1 cis-11, C20:4 n-3, C22:0, C22:6 n-3, C24:0) as
explanatory variables for CH4 prediction (Mohammed
et al., 2011; Castro-Montoya et al., 2016b; van Gastelen
et al., 2017). Models based on a reduced number of
MFA variables using concentration sums of groups of
biochemically related MFA showed MUFA as dominat-
ing explanatory variable. The group of MUFA includes
biohydrogenation intermediates and was therefore re-
ported to be negatively associated with CH4 production
(Mohammed et al., 2011; Castro-Montoya et al., 2016a;
Vanrobays et al., 2016). Although SFA could not be
identified as important explanatory variable for CH4
production in our study, Weill et al. (2009) proposed
SFA <C16 as predictors for CH4 intensity.
Additional Inclusion of ECM and DMI
in Prediction Models
Recently, it was suggested that the combination of
MFA data sets with other CH4 proxies could improve
CH4 prediction (van Gastelen and Dijkstra, 2016;
Negussie et al., 2017). Rico et al. (2016) predicted CH4
production (g/d) for cows with models including diet
components, MFA, and DMI as variables in the data
sets, but it turned out that in the best-fit model diet
components as explanatory variables did not play a
role; this indicates that the inclusion of DMI or ECM
could be more meaningful as explanatory variables for
the prediction models than dietary composition. The
best equation reported by Chilliard et al. (2009) also
included forage DMI as a parameter, indicating its
potential importance. Thus, our third objective was
to test the effect of additional inclusion of ECM and
DMI. Dry matter intake is the major determinant for
CH4 production (Knapp et al., 2014) and ECM can
reflect DMI (Hristov et al., 2013). Inclusion of ECM
or DMI improved CH4-prediction equations, resulting
in an increase of the R2CV by approximately 30% when
based on pooled data from all diets. This was only true
for the data set 5 and 6 with a reduced number of
MFA variables, showing that CH4 prediction by the
complete set of 43 MFA variables could explain more
variability of CH4 production than DMI or ECM. van
Gastelen et al. (2017) reported improved prediction of
CH4 production by 29% when using equations contain-
ing MFA plus volatile and nonvolatile metabolites of
milk as compared with equations based on 42 MFA
variables alone, indicating that accuracy of prediction
can be gained from inclusion of CH4 proxies other than
only MFA as explanatory variables. Interestingly, with
MFA data based on the GS diet with addition of ECM
(data sets 2 and 5) or DMI (data sets 3 and 6) as a
variable excluded MFA as explanatory variable from
the equation. This might explain why the predictions
for GS diets resulted in the lowest-quality parameters,
because GS diets combine feedstuffs and nutrients with
opposite effects on CH4 production (Maia et al., 2007;
Shingfield et al., 2013; Castro-Montoya et al., 2016a).
In contrast, based on data of the LS diets, inclusion of
ECM (data sets 2 and 5) did not improve the prediction
as compared with predictions with data sets 1 and 4,

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 13
respectively, containing MFA variables only, suggesting
a dominant effect of linseed (i.e., n-3 MFA) on the MFA
pattern and thus CH4 prediction.
CONCLUSIONS
Our study shows that the effect of the 2 basal diets
containing either a large proportion of grass silage or
corn silage on CH4 production and MFA profile was
small due to only moderate intake differences in crude
fiber, ADF, and NDF, nutrients known to have a large
effect on both CH4 production and MFA profile. How-
ever, with CS diets DMI was higher and resulted in a
tendency for higher CH4 production. In contrast, the
supplementation of the basal diets with linseed strongly
reduced CH4 emission parameters, and modified the
MFA profile considerably. The developed CH4 predic-
tion models for dairy cows based on a data set of 43
individual MFA provided good prediction (R2Adj 0.32
to 0.89; R2CV 0.47 to 0.92), which was reduced by 10
to 15% for models based on a reduced data set with 9
MFA parameters (biochemically related MFA groups
and C16:0 and C18:0 MFA). Prediction of CH4 produc-
tion based on a reduced data set can be improved when
ECM or DMI are added as variables, indicating a gain
of prediction accuracy when combining CH4 proxies.
In this study, prediction equations based on GS diets
resulted in the lowest prediction accuracy, which might
be due to fiber and starch as TMR ingredients in our
GS diets with opposite effects on methane production.
Prediction equations based on diets with linseed supple-
ments had a lower prediction accuracy compared with
equations based on data from diets without linseed.
Therefore, caution is needed when prediction equations
are derived from data sets including fat-supplemented
diets or when diet-specific prediction equations are ap-
plied to cows fed nonmatching diets.
ACKNOWLEDGMENTS
The authors express their gratitude to K.-D. Witt, B.
Stabenow, D. Oswald, T. Lenke, R. Gaeth, A. Schulz,
and K. Pilz, staff members at the EAR and the Tiertech-
nikum of the Leibniz Institute for Farm Animal Biology
(FBN), for preparation of TMR, feeding, and animal
care as well as assistance with sample collection and
indirect calorimetry measurements. The assistance of
F. Schultz (RinderAllianz GmbH, Woldegk, Germany)
in the selection of cows is gratefully acknowledged. We
thank DANONE GmbH, Haar, Germany, for providing
milk composition data and V. Krüger and B. Göschl
of DANONE GmbH for their involvement and fruit-
ful discussions. This work was part of the project “In-
novation potential to reduce greenhouse gas emissions
in the dairy supply chain” (INNO MilCH4) and was
supported by funds of the Federal Ministry of Food,
and Agriculture (BMEL, Berlin) based on a decision
of the Parliament of the Federal Republic of Germany
via the Federal Office for Agriculture and Food (BLE,
Bonn) under the innovation support programme (grant
no. 2817501011).
REFERENCES
Allen, M. S. 2014. Drives and limits to feed intake in ruminants. Anim.
Prod. Sci. 54:1513–1524.
Alves, S. P., J. Santos-Silva, A. R. J. Cabrita, A. J. M. Fonseca, and
R. J. B. Bessa. 2013. Detailed dimethylacetal and fatty acid com-
position of rumen content from lambs fed lucerne or concentrate
supplemented with soybean oil. PLoS One 8:e58386.
Angulo, J., B. Hiller, M. Olivera, L. Mahecha, D. Dannenberger, G.
Nuernberg, B. Losand, and K. Nuernberg. 2012. Dietary fatty acid
intervention of lactating cows simultaneously affects lipid profiles
of meat and milk. J. Sci. Food Agric. 92:2968–2974.
Bayat, A. R., I. Tapio, J. Vilkki, K. J. Shingfield, and H. Leskinen.
2018. Plant oil supplements reduce methane emissions and improve
milk fatty acid composition in dairy cows fed grass silage-based di-
ets without affecting milk yield. J. Dairy Sci. 101:1136–1151.
Beauchemin, K. A., M. Kreuzer, F. O’Mara, and T. A. McAllister.
2008. Nutritional management for enteric methane abatement: A
review. Aust. J. Exp. Agric. 48:21–27.
Benchaar, C., F. Hassanat, R. Martineau, and R. Gervais. 2015. Lin-
seed oil supplementation to dairy cows fed diets based on red clo-
ver silage or corn silage: Effects on methane production, rumen
fermentation, nutrient digestibility, N balance, and milk produc-
tion. J. Dairy Sci. 98:7993–8008.
Bielak, A., M. Derno, A. Tuchscherer, H. M. Hammon, A. Susenbeth,
and B. Kuhla. 2016. Body fat mobilization in early lactation influ-
ences methane production of dairy cows. Sci. Rep. 6:28135.
Buccioni, A., M. Decandia, S. Minieri, G. Molle, and A. Cabiddu.
2012. Lipid metabolism in the rumen: New insights on lipolysis
and biohydrogenation with an emphasis on the role of endogenous
plant factors. Anim. Feed Sci. Technol. 174:1–25.
Castro-Montoya, J. M., S. De Campeneere, B. De Baets, and V.
Fievez. 2016a. The potential of milk fatty acids as biomarkers for
methane emissions in dairy cows: A quantitative multi-study sur-
vey of literature data. J. Agric. Sci. 154:515–531.
Castro-Montoya, J. M., N. Peiren, J. Veneman, B. De Baets, S. De
Campeneere, and V. Fievez. 2016b. Predictions of methane emis-
sion levels and categories based on milk fatty acid profiles from
dairy cows. Animal 7:1153–1162.
Chilliard, Y., C. Martin, J. Rouel, and M. Doreau. 2009. Milk fatty
acids in dairy cows fed whole crude linseed, extruded linseed, or
linseed oil, and their relationship with methane output. J. Dairy
Sci. 92:5199–5211.
Chung, Y. H., M. L. He, S. M. McGinn, T. A. McAllister, and K. A.
Beauchemin. 2011. Linseed suppresses enteric methane emissions
from cattle fed barley silage, but not from those fed grass hay.
Anim. Feed Sci. Technol. 166–67:321–329.
Clark, H. 2013. Nutritional and host effects on methanogenesis in the
grazing ruminant. Animal 7:41–48.
Derno, M., H. G. Elsner, E. A. Paetow, H. Scholze, and M. Schwei-
gel. 2009. Technical note: A new facility for continuous respiration
measurements in lactating cows. J. Dairy Sci. 92:2804–2808.
Dijkstra, J., S. van Gastelen, E. C. Antunes-Fernandes, D. Warner, B.
Hatew, G. Klop, S. C. Podesta, H. J. van Lingen, K. A. Hettinga,
and A. Bannink. 2016. Relationships between milk fatty acid pro-
files and enteric methane production in dairy cattle fed grass- or
grass silage-based diets. Anim. Prod. Sci. 56:541–548.

14 ENGELKE ET AL.
Journal of Dairy Science Vol. 102 No. 2, 2019
Dijkstra, J., S. M. van Zijderveld, J. A. Apajalahti, A. Bannink, W.
J. J. Gerrits, J. R. Newbold, H. B. Perdok, and H. Berends. 2011.
Relationships between methane production and milk fatty acid
profiles in dairy cattle. Anim. Feed Sci. Technol. 166–167:590–595.
Engelke, S. W., G. Daú, M. Derno, A. Tuchscherer, W. Berg, B. Kuhla,
and C. C. Metges. 2018. Milk fatty acids estimated by mid-infra-
red spectroscopy and milk yield can predict methane emissions in
dairy cows. Agron. Sustain. Dev. 38:27.
Enjalbert, F., S. Combes, A. Zened, and A. Meynadier. 2017. Rumen
microbiota and dietary fat: A mutual shaping. J. Appl. Microbiol.
123:782–797.
Ferlay, A., M. Doreau, C. Martin, and Y. Chilliard. 2013. Effects of
incremental amounts of extruded linseed on the milk fatty acid
composition of dairy cows receiving hay or corn silage. J. Dairy
Sci. 96:6577–6595.
Firl, N., H. Kienberger, and M. Rychlik. 2014. Validation of the sen-
sitive and accurate quantitation of the fatty acid distribution in
bovine milk. Int. Dairy J. 35:139–144.
Gerber, P. J., B. Henderson, and H. P. S. Makkar. 2013. Mitigation of
greenhouse gas emissions in livestock production: A review of tech-
nical options for non-CO2 emissions. Pages 37–38 in FAO Animal
Production and Health Paper. FAO, Rome, Italy.
German Federal Law Gazette. (Bundesgesetzblatt (BGBl). 1981. Fut-
termittelverordnung (FuttMV) in der Fassung der Bekanntmac-
hung vom 29. August 2016 (BGBl. I S. 2004), die zuletzt durch
Artikel 2 der Verordnung vom 18. Juli 2018 (BGBl. I S. 1219)
geändert worden ist/ Feed regulation last updated 18 July 2018.
Berlin, Germany.
Gesellschaft für Ernährungsphysiologie. 2001. Empfehlungen zur En-
ergie- und Nährstoffversorgung der Milchkühe und Aufzuchtrinder
(Recommended energy and nutrient supply for dairy cows and
growing cattle). German Society of Nutrition Physiology/ Aus-
schuss für Bedarfsnormen, No. 8 ed. DLG-Verlag, Frankfurt am
Main, Germany.
Glasser, F., A. Ferlay, and Y. Chilliard. 2008. Oilseed lipid supple-
ments and fatty acid composition of cow milk: A meta-analysis. J.
Dairy Sci. 91:4687–4703.
Grainger, C., and K. A. Beauchemin. 2011. Can enteric methane emis-
sions from ruminants be lowered without lowering their produc-
tion? Anim. Feed Sci. Technol. 166–167:308–320.
Hammond, K., L. A. Crompton, A. Bannink, J. Dijkstra, D. R. Yanez-
Ruiz, P. O’Kiely, E. Kebreab, M. A. Eugene, Z. Yu, K. J. Shin-
gfield, A. Schwarm, A. N. Hristov, and C. K. Reynolds. 2016. Re-
view of current in vivo measurement techniques for quantifying
enteric methane emission from ruminants. Anim. Feed Sci. Tech-
nol. 219:13–30.
Hart, K. J., J. A. Huntington, R. G. Wilkinson, C. G. Bartram, and L.
A. Sinclair. 2015. The influence of grass silage-to-maize silage ratio
and concentrate composition on methane emissions, performance
and milk composition of dairy cows. Animal 9:983–991.
Hook, S. E., A. D. G. Wright, and B. W. McBride. 2010. Methano-
gens: Methane producers of the rumen and mitigation strategies.
Archaea 2010:945785.
Hristov, A. N., J. Oh, C. Lee, R. Meinen, F. Montes, T. Ott, J. Firkins,
A. Rotz, C. Dell, A. Adesogan, W. Yang, J. Tricarico, E. Ke-
breab, G. Waghorn, J. Dijkstra, and S. Oosting. 2013. Mitigation
of Greenhouse Gas Emissions in Livestock Production: A Review
of Technical Options for non-CO2 Emissions. P. J. Gerber, B. Hen-
derson, and H. P. S. Makkar, ed. FAO, Rome, Italy.
Kaps, M., and W. Lamberson. 2004. Biostatistics for Animal Science.
CABI Publishing, Wallingford, United Kingdom.
Kliem, K. E., D. J. Humphries, C. K. Reynolds, R. Morgan, and D. I.
Givens. 2017. Effect of oilseed type on milk fatty acid composition
of individual cows, and also bulk tank milk fatty acid composition
from commercial farms. Animal 11:354–364.
Kliem, K. E., R. Morgan, D. J. Humphries, K. J. Shingfield, and
D. I. Givens. 2008. Effect of replacing grass silage with maize si-
lage in the diet on bovine milk fatty acid composition. Animal
2:1850–1858.
Kliem, K. E., and K. J. Shingfield. 2016. Manipulation of milk fatty
acid composition in lactating cows: Opportunities and challenges.
Eur. J. Lipid Sci. Technol. 118:1661–1683.
Knapp, J. R., G. L. Laur, P. A. Vadas, W. P. Weiss, and J. M. Tri-
carico. 2014. Invited review: Enteric methane in dairy cattle pro-
duction: Quantifying the opportunities and impact of reducing
emissions. J. Dairy Sci. 97:3231–3261.
Lanier, J. S., and B. A. Corl. 2015. Challenges in enriching milk fat
with polyunsaturated fatty acids. J. Anim. Sci. Biotechnol. 6:26–
35.
Livingstone, K. M., D. J. Humphries, P. Kirton, K. E. Kliem, D.
I. Givens, and C. K. Reynolds. 2015. Effects of forage type and
extruded linseed supplementation on methane production and
milk fatty acid composition of lactating dairy cows. J. Dairy Sci.
98:4000–4011.
Maia, M. R. G., L. C. Chaudhary, L. Figueres, and R. J. Wallace.
2007. Metabolism of polyunsaturated fatty acids and their tox-
icity to the microflora of the rumen. Antonie van Leeuwenhoek
91:303–314.
Martin, C., A. Ferlay, P. Mosoni, Y. Rochette, Y. Chilliard, and M.
Doreau. 2016. Increasing linseed supply in dairy cow diets based
on hay or corn silage: Effect on enteric methane emission, rumen
microbial fermentation, and digestion. J. Dairy Sci. 99:3445–3456.
Meignan, T., C. Lechartier, G. Chesneau, and N. Bareille. 2017. Ef-
fects of feeding extruded linseed on production performance and
milk fatty acid profile in dairy cows: A meta-analysis. J. Dairy Sci.
100:4394–4408.
Mohammed, R., S. M. McGinn, and K. A. Beauchemin. 2011. Predic-
tion of enteric methane output from milk fatty acid concentrations
and rumen fermentation parameters in dairy cows fed sunflower,
flax, or canola seeds. J. Dairy Sci. 94:6057–6068.
Moraes, L. E., A. B. Strathe, J. G. Fadel, D. P. Casper, and E. Ke-
breab. 2014. Prediction of enteric methane emissions from cattle.
Glob. Chang. Biol. 20:2140–2148.
Naumann, C., R. Bassler, R. Seibold, and C. Barth. 1976. Methoden-
buch III: Die chemische Untersuchung von Futtermitteln/ Method
book III: Chemical analysis of feedstuffs. VDLUFA, Verlag, Darm-
stadt, Germany.
Negussie, E., Y. de Haas, F. Dehareng, R. J. Dewhurst, J. Dijkstra, N.
Gengler, D. P. Morgavi, H. Soyeurt, S. van Gastelen, T. Yan, and
F. Biscarini. 2017. Invited review: Large-scale indirect measure-
ments for enteric methane emissions in dairy cattle: A review of
proxies and their potential for use in management and breeding
decisions. J. Dairy Sci. 100:2433–2453.
Patra, A. K. 2016. Recent advances in measurement and dietary miti-
gation of enteric methane emissions in ruminants. Front. Vet. Sci.
3:39.
Pickering, N. K., V. H. Oddy, J. Basarab, K. Cammack, B. Hayes, R.
S. Hegarty, J. Lassen, J. C. McEwan, S. Miller, C. S. Pinares-Pa-
tino, and Y. de Haas. 2015. Animal board invited review: Genetic
possibilities to reduce enteric methane emissions from ruminants.
Animal 9:1431–1440.
Reynolds, C. K. 2006. Production and metabolic effects of site of starch
digestion in dairy cattle. Anim. Feed Sci. Technol. 130:78–94.
Rico, D. E., P. Y. Chouinard, F. Hassanat, C. Benchaar, and R. Ger-
vais. 2016. Prediction of enteric methane emissions from Holstein
dairy cows fed various forage sources. Animal 10:203–211.
Shingfield, K. J., M. Bonnet, and N. D. Scollan. 2013. Recent develop-
ments in altering the fatty acid composition of ruminant-derived
foods. Animal 7:132–162.
Spiekers, H., H. Nußbaum, and V. Potthast. 2009. Erfolgreiche Milch-
viehfütterung/ Successful feeding of dairy cattle. DLG-Verlag,
Frankfurt am Main, Germany.
Sterk, A., B. E. O. Johansson, H. Z. H. Taweel, M. Murphy, A. M. van
Vuuren, W. H. Hendriks, and J. Dijkstra. 2011. Effects of forage
type, forage to concentrate ratio, and crushed linseed supplemen-
tation on milk fatty acid profile in lactating dairy cows. J. Dairy
Sci. 94:6078–6091.
van Gastelen, S., E. C. Antunes-Fernandes, K. A. Hettinga, and J. Di-
jkstra. 2017. Relationships between methane emission of Holstein

Journal of Dairy Science Vol. 102 No. 2, 2019
MILK FATTY ACIDS TO PREDICT METHANE PRODUCTION 15
Friesian dairy cows and fatty acids, volatile metabolites and non-
volatile metabolites in milk. Animal 11:1539–1548.
van Gastelen, S., E. C. Antunes-Fernandes, K. A. Hettinga, G. Klop,
S. J. J. Alferink, W. H. Hendriks, and J. Dijkstra. 2015. Enteric
methane production, rumen volatile fatty acid concentrations, and
milk fatty acid composition in lactating Holstein-Friesian cows fed
grass silage- or corn silage-based diets. J. Dairy Sci. 98:1915–1927.
van Gastelen, S., and J. Dijkstra. 2016. Prediction of methane emis-
sion from lactating dairy cows using milk fatty acids and mid-
infrared spectroscopy. J. Sci. Food Agric. 96:3963–3968.
van Lingen, H. J., L. A. Crompton, W. H. Hendriks, C. K. Reynolds,
and J. Dijkstra. 2014. Meta-analysis of relationships between en-
teric methane yield and milk fatty acid profile in dairy cattle. J.
Dairy Sci. 97:7115–7132.
Van Soest, P. J., J. B. Robertson, and B. A. Lewis. 1991. Methods for
dietary fiber, neutral detergent fiber, and nonstarch polysaccha-
rides in relation to animal nutrition. J. Dairy Sci. 74:3583–3597.
Vanrobays, M. L., C. Bastin, J. Vandenplas, H. Hammami, H. Soyeurt,
A. Vanlierde, F. Dehareng, E. Froidmont, and N. Gengler. 2016.
Changes throughout lactation in phenotypic and genetic correla-
tions between methane emissions and milk fatty acid contents pre-
dicted from milk mid-infrared spectra. J. Dairy Sci. 99:7247–7260.
Vlaeminck, B., V. Fievez, A. R. J. Cabrita, A. J. M. Fonseca, and R.
J. Dewhurst. 2006. Factors affecting odd- and branched-chain fatty
acids in milk: A review. Anim. Feed Sci. Technol. 131:389–417.
Vlaeminck, B., R. Gervais, M. M. Rahman, F. Gadeyne, M. Gorniak,
M. Doreau, and V. Fievez. 2015. Postruminal synthesis modifies
the odd- and branched-chain fatty acid profile from the duodenum
to milk. J. Dairy Sci. 98:4829–4840.
Weill, P., G. Chesneau, Y. Chilliard, M. Doreau, and C. Martin,
inventors. 2009. Method for evaluating the amount of methane
produced by a dairy ruminant and method for decreasing and
controlling this amount. WO 2009/156453 A1. https: / / patents
.google .com/ patent/ WO2009156453A1/ en. Assignee: Valorisation
Par Extrusion.
Williams, S. R. O., P. J. Moate, M. H. Deighton, M. C. Hannah, and
W. J. Wales. 2014. Methane emissions of dairy cows cannot be
predicted by the concentrations of C8:0 and total C18 fatty acids
in milk. Anim. Prod. Sci. 54:1757–1761.
Yan, T., C. S. Mayne, F. G. Gordon, M. G. Porter, R. E. Agnew, D.
C. Patterson, C. P. Ferris, and D. J. Kilpatrick. 2010. Mitigation of
enteric methane emissions through improving efficiency of energy
utilization and productivity in lactating dairy cows. J. Dairy Sci.
93:2630–2638.
Zehetmeier, M., J. Baudracco, H. Hoffmann, and A. Heissenhuber.
2012. Does increasing milk yield per cow reduce greenhouse gas
emissions? A system approach. Animal 6:154–166.