Plasma and milk metabolomics in lactating sheep divergent for feed efficiency

Enhancing the ability of animals to convert feed into meat or milk by optimizing feed efficiency (FE) has become a priority in livestock research. Although untargeted metabolomics is increasingly used in this field and may improve our understanding of FE, no information in this regard is available in dairy ewes. This study was conducted to (1) discriminate sheep divergent for FE and (2) provide insights into the physiological mechanisms contributing to FE through high-throughput metabolomics. The ultra-high-performance liquid chromatography-quadrupole time-of-flight mass spectrometry (UHPLC-Q/TOF-MS) technique was applied to easily accessible animal fluids (plasma and milk) to assess whether their metabolome differs between high-and low-feed efficient lactating ewes (H-FE and L-FE groups, respectively; 8 animals/ group). Blood and milk samples were collected on the last day of the 3-wk period used for FE estimation. A total of 793 features were detected in plasma and 334 in milk, with 100 and 38 of them, respectively, showing differences between H-FE and L-FE. The partial least-squares discriminant analysis separated both groups of animals regardless of the type of sample. Plasma allowed the detection of a greater number of differential features; however, results also supported the usefulness of milk, more easily accessible, to discriminate dairy sheep divergent for FE. Regarding pathway analysis, nitrogen metabolism (either anabolism or catabolism) seemed to play a central role in FE, with plasma and milk consistently indicating a great impact of AA metabolism. A potential influence of pathways related to energy/lipid metabolism on FE was also observed. The variable importance in the projection plot revealed 15 differential features in each matrix that contributed the most for the separation in H-FE and L-FE, such as l-proline


INTRODUCTION
Livestock farming faces the challenge of maintaining profitability while reducing the environmental impact.Feed represents the largest share of total costs of production (approximately 66% in representative dairy sheep farms in Spain; RENGRATI, 2021), and its efficiency of use is a key determinant of the farm sustainability (Herrero et al., 2013).For this reason, livestock systems have increasingly focused on the identification of animals capable of optimizing the utilization of feed (Hegarty et al., 2007;Connor et al., 2012;Løvendahl et al., 2018).Estimation of feed efficiency (FE) may use several metrics (i.e., conversion ratios, energy balances, and residual-based traits) and it is the subject of extensive discussion (Hurley et al., 2016;Cantalapiedra-Hijar et al., 2018;Dorji et al., 2021).In any case, FE estimations involve recording individual feed intake, which requires time-consuming procedures and expensive equipment (Karisa et al., 2014), fostering the search for alternative means.
Ideally, a set of biomarkers predictive of FE could greatly reduce the cost of direct measurements, while making the detection process more feasible.As a complex phenotype, different major physiological mechanisms may contribute to FE variation, e.g., feed intake behavior, rumen fermentation and digestion, thermoregulation, or cell anabolism/catabolism (Herd and Arthur, 2009).
Metabolomics allows for the analysis of end products of the complex genetic, epigenetic, and environmental interactions, representing a valuable tool to improve our understanding of physiological processes with influence on livestock production (Wu et al., 2018;Xu et al., 2020;Yanibada et al., 2021).In recent years, increasingly research effort has also been devoted to describe a distinct metabolite signature of highly efficient animals (Shabat et al., 2016;Novais et al., 2019;Clemmons et al., 2020).This is particularly challenging in ruminants due to the complexity of the host-microbiome interactions through many metabolic pathways (Shabat et al., 2016;McLoughlin et al., 2020).Artegoitia et al. (2019) combined analyses of multiple tissues or fluids (i.e., duodenal digesta, liver, subcutaneous fat, and longissimus dorsi) in beef animals with divergent ADG at similar DMI, and they suggested that metabolomics may offer a powerful approach for defining the molecular basis of differences in FE.However, to include FE metrics in dairy breeding programs, samples need to be easily accessible and, if possible, obtained through a noninvasive and cheap procedure.In dairy ruminants, milk and blood would represent convenient alternatives (Sun et al., 2017;Xu et al., 2020;Yanibada et al., 2021).
To date most studies conducted using metabolomics have focused on beef cattle (Karisa et al., 2014;Clemmons et al., 2020) and very few on lamb (Goldansaz et al., 2020).Reports in dairy cows (Xi et al., 2017;Wang and Kadarmideen, 2019;Yanibada et al., 2021) are scarce, and to our knowledge, there is no publication applying this technique in lactating sheep.
We have recently reported (Toral et al., 2021) variations in milk lactose and fatty acid composition in dairy ewes divergent for FE.Here we build on these findings by using high-throughput metabolomics applied to plasma and milk to assess whether their metabolite profiles differ between high-and low-efficient dairy sheep, which would help to (1) discriminate animals divergent for FE and (2) provide new insights into the physiological mechanisms contributing to FE variation.

Ethics Statements
All experimental procedures were approved by the Research Ethics Committees of the Instituto de Ganadería de Montaña, the Spanish National Research Council (CSIC), and the Junta de Castilla y León (Spain), following proceedings described in Spanish and European Union legislation (Royal Decree 53/2013 and Council Directive 2010/63/EU).

Animals, Experimental Diet, and Management
Details of the experimental design and methodology are described in Toral et al. (2021).Briefly, we conducted the trial with 40 lactating Assaf ewes in the first half of the lactation period (BW = 73.9± 0.21 kg; DIM = 61.6 ± 0.10; parity = 2.7 ± 0.03; milk yield = 2.6 ± 0.01 kg/d).
Ewes were housed in individual pens and fed ad libitum a TMR formulated from alfalfa hay and concentrate (50:50).The TMR contained 182 g of CP, 130 g of starch, and 288 g of NDF/kg of DM [see Toral et al. (2021) for further details about chemical composition and ingredients].
Sheep were milked twice daily (at approximately 0830 and 1830 h).

Feed Efficiency Estimation and Selection of Ewes
As described in Toral et al. (2021), feed efficiency was estimated in the 40 ewes to select 8 of the least (L-FE group) and 8 of the most efficient (H-FE group) animals to carry out the metabolomics analysis in plasma and milk.For feed efficiency estimation, individual feed intake and milk yield were recorded over a 3-wk period.Feed intake was calculated by weighing the amounts of DM offered and refused by each animal, and milk yield by weighing the total milk produced by each animal at morning and evening milking.The BW was recorded in 2 consecutive days per week, and then averaged per week.
The selection was based on the feed efficiency index (FEI), which was calculated as the difference between recorded and predicted DMI: DMI R − DMI P .The DMI R is the mean value of recorded DMI over the experimental phase, and the DMI P is the mean value of predicted DMI for the same period.Mean values were used because no differences in DMI, milk production and composition, and energy requirements were found (P > 0.10) over the monitoring period.Predicted DMI (DMI P ) was computed as ME mp /ME TMR .The ME mp are the metabolizable energy requirements for maintenance, production, and BW change (MJ/d), and were estimated using the equations proposed by the Agricultural and Food Research Council (AFRC, 1993) for nonpregnant lactating sheep.Similarly, the ME TMR or ME of the TMR (MJ/kg of DM) was estimated using TMR formulation and tables of nutritional value of AFRC (1993).
The higher the feed efficiency, the lower the value of the FEI, which averaged 0.81 ± 0.084 for L-FE and −0.29 ± 0.046 for H-FE ewes.A good agreement between this index and residual feed intake (RFI, estimated as the residual term from the regression of feed intake on various energy sinks: 0.16 ± 0.084 for L-FE, and -0.18 ± 0.082 for H-FE ewes) was reported in our previous study (r = 0.69, P < 0.01; Toral et al., 2021).Data of dairy performance are also detailed in Toral et al. (2021).In brief, DMI was similar in L-FE and H-FE (3.10 vs. 3.18 kg/d, respectively, P = 0.66), but milk yield was 34% greater in H-FE (P < 0.01), with increases in lactose, protein, and fat yield (+38, +30, and +33%, respectively; P < 0.05).In addition, H-FE sheep gained more BW over the experiment than L-FE (3.15 vs. 0.58 kg, P < 0.01).

Sample Collection and Processing
Plasma.On the last day of the experimental period, before the morning milking and the morning administration of the feed, blood samples were collected into lithium heparin tubes (BD Vacutainer) and centrifuged (3,000 rpm) at 4°C for 10 min.Aliquots of 1 mL of plasma were stored at −80°C.Before metabolomics analysis, samples were thawed on ice and homogenized.Then, 400 μL of cold methanol was added to 100 μL of plasma and mixed with a vortex for 1 min.Samples were kept on ice for 10 min and centrifuged (14,800 rpm) at 4°C for 20 min.Supernatant (300 μL) was dried in a SpeedVac (Thermo Fisher Scientific) concentrator, reconstituted with 100 μL of 80% methanol, mixed with a vortex for 1 min, and centrifuged (14,800 rpm) at 4°C for 5 min.The upper phase was stored at −80°C until MS analysis was performed.
Milk.On the last day of the trial, aliquots of composited milk (10 mL) from each ewe were collected and stored at −80°C.Before metabolomics analysis, milk samples were thawed on ice and centrifuged (4,500 rpm) at 4°C for 15 min.Then, 50 μL of the aqueousintermedium phase was added to 350 μL of methanol: methyl tert-butyl ether (50:50, vol/vol) and mixed for 1 min with a vortex.Samples were kept on ice for 10 min and centrifuged (14,800 rpm) at 4°C for 15 min.The upper phase (300 μL) was stored at −80°C until MS analysis.

Untargeted Metabolomics Analysis
Untargeted metabolomics analyses of milk and plasma samples were performed using the Metabolomics Platform at the Institute of Food Science Research (CIAL, CSIC-Universidad Autónoma de Madrid), following the methodology described by Ballesteros-Vivas et al. ( 2019) and using an Agilent 1290 Infinity ultrahigh performance liquid chromatography system hyphenated to an Agilent 6540 UHD accurate-mass quadrupole time-of-flight mass spectrometer (UHPLC-Q/ TOF-MS).Chromatographic separation was performed using a Zorbax Eclipse Plus C18 column (1.8 μm, 2.1 × 100 mm).Mobile phase A was 0.1% formic acid in water, and the mobile phase B was 0.1% formic acid in acetonitrile.Mass detection was run in the MS scan mode from m/z 25 to 1,100 in electrospray ionization (+).Samples were analyzed in randomized order and in triplicate.Relative quantification was conducted using the area of the chromatographic peak.

Metabolomics Data Analysis
Raw data sets from each type of sample were processed simultaneously with Agilent MassHunter Profinder (v.10.0) and Mass Profiler Professional (v.15.1), as described previously (Ballesteros-Vivas et al., 2019).Univariate and multivariate statistical analysis was performed with MetaboAnalyst (v5.0;Pang et al., 2021).Only features that were present in 6 out of 8 samples in each group of sheep (H-FE and L-FE) were further used in the analyses.Then, data were first subjected to nonsupervised principal component analysis (PCA) and subsequently to 2 different analyses.First, nonparametric Wilcoxon-Mann-Whitney test was performed to identify the features with concentrations that statistically differed between high and low efficient animals (i.e., between H-FE and L-FE; P-value <0.05 and fold change >1.2).Second, supervised multivariate data analysis (namely partial least-squares discriminate analysis; PLS-DA) was conducted to further differentiate the contributions of particular features to the separation of H-FE and L-FE sheep.The corresponding variable importance in the projection (VIP) scores were calculated in the PLS-DA model.The differential features with VIP ≥1.5 were selected.Univariate analysis was performed applying the Wilcoxon rank-sum test between H-FE and L-FE.Features with VIP ≥1.5 and with statistical significant difference (P-value adjusted using false discovery rate < 0.05) were submitted to annotation, which was carried out using the exact mass, isotopic pattern, and MS/MS spectra of the detected features, within a ±10 ppm mass range, and online Bos taurus metabolomics databases [Massbank, Metlin, and Kyoto Encyclopedia of Genes and Genomes (KEGG); Horai et al., 2010;Guijas et al., 2018;Kanehisa et al., 2021].
Metabolic pathway analysis was conducted using the MetaboAnalyst (v5.0;Pang et al., 2021).Only plasma or milk metabolites significantly (P < 0.05) divergent in concentrations when comparing H-FE and L-FE sheep were used.The P-values from enrichment analysis were further adjusted for multiple testing, and the pathway impact value calculated from pathway topology analysis.

Metabolomics Analysis
A total of 793 features were detected in plasma and 334 in milk samples (see Supplemental Material for archived raw, processed, and normalized data from each animal: https: / / doi .org/ 10 .20350/digitalCSIC/ 14524, Toral et al., 2022).
Neither the PCA plot of plasma results nor that of milk were able to discriminate L-FE and H-FE (Figure 1).
The PLS-DA of plasma and milk samples displayed a clear clustering between both groups of ewes (Figure 4).For the first 5 components of PLS-DA, accuracy and estimates of the goodness of fit (R 2 ) and goodness of prediction (Q 2 ) were 0.69, 0.99, and 0.26 for plasma, and 0.63, 0.99, and 0.19 for milk, respectively.
Multivariate statistical analyses produced a VIP plot (Figure 5) with 15 features in each matrix (milk and plasma).Using MS/MS fragmentation data, it was possible to annotate 9 of them in the plasma and 9 in the milk.The plasma samples contained 10 features that were more abundant (P < 0.05) in the H-FE group (e.g., l-proline and trans-cinnamic acid) and 5 that were less abundant (e.g., Leu-Ala-Pro-Leu-Glu pentapeptide and PC 20:4e; P < 0.05).In the milk samples, 8 features were found to have significantly (P < 0.05) higher concentrations in H-FE (e.g., PE 18:2 and lpipecolic acid) and 7 were more abundant (P < 0.05) in milk from L-FE (e.g., 5-pyrimidinecarboxamide).

Pathway Analysis
The most relevant metabolic pathways involved in the divergence between H-FE and L-FE ewes in plasma and milk are shown in Figure 6.
Regarding milk, the results showed that significant metabolites in the milk (P < 0.05) were mainly involved in lysine degradation (l-pipecolate), porphyrin and chlorophyll metabolism (porphobilinogen), glycerophospholipid metabolism (acetylcholine), and tryptophan metabolism (l-tryptophan).The analysis revealed some other relevant pathways, such as aminoacyl-tRNA biosynthesis (l-tryptophan; P < 0.05), which was also detected in plasma.

DISCUSSION
Enhancing the ability of animals to convert feed into meat or milk has become a priority in livestock research, but our knowledge of the factors involved is still limited, particularly in dairy ruminants (Connor et al., 2012;Løvendahl et al., 2018).Based on the hypothesis that FE may be related to a range of different metabolic processes (Archer et al., 1999;Herd and Arthur, 2009), metabolomics may represent a valuable tool to improve our understanding of this complex aspect of performance.To our knowledge, this is the first study using high-throughput metabolomics in relation to FE in lactating ewes.

Discrimination of Dairy Ewes Divergent for FE
Previous studies in beef have shown that metabolomics allows discrimination of animals diverging in FE, estimated through ratio and residual traits (Clemmons et al., 2017;Cantalapiedra-Hijar et al., 2018;Novais et al., 2019).In our study, we were not able to discriminate L-FE and H-FE dairy ewes through PCA, and only the results from the PLS-DA provided a clear separation of both groups, regardless of the type of sample (Figure 4).Similarly, PLS-DA led to good discrimination by FE level using other biological samples (e.g., ruminal or duodenal digesta) and metrics (e.g., feed conversion ratio or RFI; Clemmons et al., 2017;Artegoitia et al., 2019;Novais et al., 2019).It is probably worth mentioning here that different FE estimators have frequently been used in the literature, either as ratio traits, residual traits, or energy balances (Hurley et al., 2016;Cantalapiedra-Hijar et al., 2018;Dorji et al., 2021), which might explain some apparently contradictory results, although most FE metrics are well correlated in dairy sheep (Hervás et al., 2021;Toral et al., 2021).
Furthermore, unraveling the physiological basis of FE may be affected by some confounding factors, such as diet composition and intake level (Cantalapiedra-Hijar et al., 2018;Hervás et al., 2021;Fischer et al., 2022).Thus, although differences in DMI are related to individual variations in FE in meat production, feeding behavior and digestive related mechanisms do not seem to be determinants of FE, as the relationship would mostly be explained by co-variation (Cantalapiedra-Hijar et al., 2018).To avoid this problem, some metabolomics studies have focused on examining beef cattle with divergent ADG and similar DMI (Artegoitia et al., 2017(Artegoitia et al., , 2019).An analogous situation was observed in our trial (i.e., different production level at similar DMI; Toral et al., 2021), but the lack of information on the metabolome of dairy ewes divergent in FE or merely in milk yield precludes attributing the differences observed between H-FE and L-FE to one or the other factor.
Concerning findings in plasma and milk, our results indicate that there is not a single type of sample to be used as standard, in line with Ilves et al. (2012) or Sun et al. (2017), and multiple biological matrices might be useful to discriminate animals by FE.Under our conditions, plasma has a higher potential because it can better capture the degree of individuality than milk samples (Ilves et al., 2012) and it allowed the detection of a greater number of features.This is in agreement with other authors reporting that metabolomics analysis of this matrix is able to predict FE with high accuracy in beef cattle (e.g., Karisa et al., 2014;Wang and Kadarmideen, 2019).Nonetheless, our results also validate the usefulness of milk samples to discriminate H-FE and L-FE ewes, as observed previously using a detailed milk fatty acid profile (Toral et al., 2021).The use of milk samples would be of particular interest under practical conditions, due to the easy access from regular milk recording in farms.In any event, some matrices may be more closely linked to different physiological processes than others (e.g., milk to mammary metabolism and plasma to ruminal fermentation, digestion, and hepatic metabolism).A potential confounding effect of milk yield in our trial might be speculated to have a greater impact on milk than plasma metabolome.Thus, complementarity might exist between them, which could provide more thorough information.This is consistent with reports in dairy cows (Ilves et al., 2012;Sun et al., 2015Sun et al., , 2017)).

Physiological Mechanisms Contributing to FE Variation
The examination of differences in the metabolite profiles of H-FE and L-FE ewes provided insights into the physiological mechanisms that may underlie FE variation.Our results support the growing evidence that protein metabolism (either anabolism or catabolism) has a central role in FE, not only in beef and lamb, but also in dairy animals (Wang and Kadarmideen, 2019;Clemmons et al., 2020;Goldansaz et al., 2020).In particular, the pathway analysis of plasma and milk consistently indicated a great impact of AA metabolism pathways on FE, with a special relevance of tryptophan, phenylalanine, tyrosine, arginine, and proline.Previous studies have suggested that RFI and feed conversion ratio may be linked to metabolites and genes involved in tryptophan metabolism in beef (Artegoitia et al., 2017;Cantalapiedra-Hijar et al., 2018;Lima et al., 2019) and dairy cows (Wang and Kadarmideen, 2019).In addition, phenylalanine and tyrosine metabolism has been associated with higher milk yield in dairy cows (Sun et al., 2015;Wu et al., 2018), and l-phenylalanine with carcass merit and muscle development in sheep (Wester et al., 2000;Goldansaz et al., 2020).In our trial, VIP scores suggest that plasma l-proline might be useful to discriminate H-FE and L-FE ewes.The level of AA in milk, but not in plasma, has recently been related to energy balance in dairy cows (Xu et al., 2020).
Within nitrogen metabolism, protein turnover is a major contributor to between-animal variations in basal energy expenditures and it has been related to FE in beef, because lower heat production in high FE animals may result from a decreased protein turnover (Cantalapiedra-Hijar et al., 2018).A similar role might exist in lactating animals, which would contribute to explain the apparently conflicting relationship between FEI and both protein anabolism (aminoacyl-tRNA biosynthesis) and catabolism (lysine degradation) in dairy ewes.This is in line with metabolomics analysis reported in dairy cows and lambs divergent for RFI (Wang and Kadarmideen, 2019;Goldansaz et al., 2020).Lysine, an essential AA involved in stimulating protein synthesis, has also been associated with changes in the urea cycle linked to RFI, with a lower activity in more efficient growing heifers (i.e., showing lower RFI; Jorge-Smeding et al., 2019).Some products of lysine catabolism have been proposed as candidate biomarkers of FE, such as aminoadipic acid, which was found by Goldansaz et al. (2020) to be downregulated in efficient lambs.According to VIP scores (Figure 5), l-pipecolic acid was one of the features in milk that contributed the most to discrimination between H-FE and L-FE groups.This intermediate metabolite of lysine catabolism has not been reported in metabolomics studies of milk and blood in dairy cows with divergent FE or production level (Sun et al., 2015;Wu et al., 2018;Wang and Kadarmideen, 2019).Additional research would be required to elucidate if l-pipecolic acid may serve as a candidate biomarker of FE in lactating ewes.
Overall, these results would suggest a key role of protein metabolism in FE in dairy ewes.Furthermore, differences in protein metabolism might be speculated to derive, at least in part, from a better digestive utilization of dietary protein, as suggested by the 36% lower ruminal ammonia concentrations in H-FE than in L-FE (Toral et al., 2021).This lower ammonia might also be associated with variations in the urea cycle activity, as described in growing heifers (Jorge-Smeding et al., 2019).Targeted research would be recommended to unravel this relationship.
In addition to nitrogen metabolism, our results revealed other metabolic pathways with a putative impact on FE in dairy sheep, such as porphyrin and chlorophyll metabolism.This was one of the most altered pathway in milk samples and it has also been identified in metabolomics analysis of rumen fluid in efficient steers (i.e., with the greatest ADG for similar intake; Artegoitia et al., 2017) or in dairy cows with high production levels (Mu et al., 2019).Nevertheless, all these relationships derive from variation in single compounds, such as porphobilinogen in our trial.Information about this metabolite in ruminants seems limited to its role in health status (e.g., lead toxicity; Payne and Livesey, 2010), making it difficult to discern its role in FE.Porphobilinogen can be produced by some ruminal bacteria (Caldwell et al., 1965), which might suggest that its changes are related to observed differences in rumen microbiota between H-FE and L-FE ewes (Esteban-Blanco et al., 2020).
Surprisingly, no obvious variations were observed in fatty acid metabolism, in contrast with expectations based on our previous work (Toral et al., 2021), which suggested differences in both ruminal biohydrogenation and mammary lipogenesis between sheep divergent in FEI.In that study, we attributed alterations in milk de novo fatty acids/cis-9 18:1 ratio (3.28 vs. 3.96 for L-FE and H-FE, respectively; P = 0.02) to differences in metabolic status between H-FE and L-FE because increases in this ratio are used as a proxy of energy deficiency and body fat mobilization in cows (Dórea et al., 2017;Khiaosa-ard et al., 2020).Although changes in BW cannot be considered a good indicator of body reserve variation in dairy animals, such an argument seems consistent with changes in BW in H-FE and L-FE (Toral et al., 2021).
In any event, the present analysis supported changes in some pathways of energy/lipid metabolism, such as primary bile acid metabolism (Figure 6).This pathway might be involved in the discrimination between H-FE and L-FE in plasma samples due to the key role of bile acids in facilitating lipid digestion (Palmquist and Jenkins, 1980).An association with FE was also indicated in metabolomics analysis of ruminal and duodenal digesta in steers (Artegoitia et al., 2017(Artegoitia et al., , 2019)).
Within the same pathway, glycerophospholipid metabolism (including some annotated features with high VIP scores, such as PC 20:4e in plasma and PE 18:2 in milk) also showed differences between H-FE and L-FE.In cattle, phospholipid metabolism has broadly been related to feed conversion efficiency using metabolomics, transcriptomics and genome-wide association analyses (Alexandre et al., 2015;de Almeida Santana et al., 2016;Artegoitia et al., 2019).Nevertheless, the complexity of this metabolic pathway as well as the relatively unspecific changes make it difficult to elucidate the physiological mechanisms that link these results with divergences in FE in dairy ewes.Additional studies are therefore necessary to unravel this relationship.
Another metabolite within lipid metabolism would be the l-carnitine, which, according to the volcano plot, was one of the most significantly downregulated features in plasma from H-FE ewes (P < 0.05; Figure 3).This quaternary amine plays a relevant role in lipid/energy metabolism, being required for mitochondrial fatty acid oxidation (McGarry and Brown, 1997;Carlson et al., 2006).l-Carnitine also affects energy production from glucose through modulation of the available free CoA in the cell (Ringseis et al., 2018).In dairy cows, a positive correlation between carnitine and energy balance was detected applying metabolomics to plasma and milk samples (Xu et al., 2020).However, its influence on individual variations in FE remains uncertain, because both lower or higher levels have been detected in highly efficient steers (Karisa et al., 2014;Clemmons et al., 2017).These apparent inconsistencies might partly derive from the complex regulation of carnitine status in ruminants, which depends on the balance between its endogenous synthesis, absorption in the gut (from dietary sources or ruminal microbial production), and renal excretion (Ringseis et al., 2018).Thus, alterations in any of these processes may affect the relationship between FE and carnitine status, which could be a consequence rather than a cause of divergences.

CONCLUSIONS
Untargeted metabolomics provided valuable information into metabolic pathways that may underlie FE in dairy ewes and could represent a tool to discriminate high-and low-feed efficient lactating sheep.Although plasma would be the matrix of choice for metabolomics analysis because it allows the detection of a greater number of differential metabolites, our results also support the usefulness of milk samples, which are more easily accessible.Nitrogen metabolism (either anabolism or catabolism) seems to play a central role in FE in dairy sheep, with plasma and milk analyses consistently indicating a great impact of AA metabolism pathways.Metabolite profiles also reveal other metabolic pathways with a putative impact on FE in dairy sheep, such as lipid and porphyrin/chlorophyll metabolism.Further research is warranted to differentiate a potential confounding effect of variations in milk production level and to validate these findings under different dietary conditions.and Spanish Research State Agency (MCIN/ AEI/10.13039/501100011033,project PID2020-113441RB-I00).Cofunding by the European Regional Development Fund (ERDF/FEDER) is also acknowledged.The authors thank C. León Canseco (University Carlos III of Madrid, Spain) and H. Marina (University of León, Spain) for helpful assistance with metabolomics data analysis.The authors have not stated any conflicts of interest.

Figure 1 .
Figure 1.Score plots for a principal component analysis model with principal component 1 (PC1) plotted against principal component 2 (PC2), with 95% confidence ellipses around the high-and low-feed efficiency ewes (H-FE and L-FE) in (A) plasma and (B) milk.