Application of Metabolomics to Identify Hepatic Biomarkers of Foie Gras Qualities in Duck

Foie gras is a traditional dish in France that contains 50 to 60% of lipids. The high-fat content of the liver improves the organoleptic qualities of foie gras and reduces its technological yield at cooking (TY). As the valorization of the liver as foie gras products is strongly influenced by the TY, classifying the foie gras in their potential technological quality before cooking them is the main challenge for producers. Therefore, the current study aimed to identify hepatic biomarkers of foie gras qualities like liver weight (LW) and TY. A group of 120 male mule ducks was reared and overfed for 6–12 days, and their livers were sampled and analyzed by proton nuclear magnetic resonance (¹H-NMR). Eighteen biomarkers of foie gras qualities were identified, nine for LW and TY, five specific to LW, and four specific to TY. All biomarkers were strongly negatively correlated to the liver weights and positively correlated to the technological yield, except for the lactate and the threonine, and also for the creatine that was negatively correlated to foie gras technological quality. As a result, in heavy livers, the liver metabolism was oriented through a reduction of carbohydrate and amino acid metabolisms, and the plasma membrane could be damaged, which may explain the low technological yield of these livers. The detected biomarkers have been strongly discussed with the metabolism of the liver in nonalcoholic steatohepatitis.

Introduction

Foie gras is one of the flagship products of French gastronomy. It is the product of hepatic steatosis due to the overfeeding of ducks with an energy-rich feed based on corn. The foie gras is the result of the increase in triglyceride storage in the liver. Actually, in the small intestine, the degradation of starch from the corn feed led to an accumulation of carbohydrates absorbed into the hepatic portal vein and carried to the liver. There, the de novo lipogenesis process converts carbohydrate precursors into fatty acids. During the overfeeding period, the imbalance between the lipid neosynthesized in the liver and their export in the hepatic vein causes lipids to accumulate as triglycerides in the hepatocytes, conducting to hepatic steatosis. Furthermore, some exported triglycerides go back to the liver and are stored in this organ. The first mechanism consists of exporting the newly formed triglycerides to the peripheral tissues like muscles and abdominal or subcutaneous tissues for storage or energy utilization. This process is mediated by very low-density lipoprotein (VLDL; Goodridge, 1987). Then, a part of these exported triglycerides is returned to the liver via high-density lipoprotein (HDL) and stored in the liver (Tavernier et al., 2018).

The mule duck is the main represented species for foie gras production because it has the best ability to hepatic steatosis (Baéza et al., 2005). The duration of fasting before slaughtering and the conditions of evisceration also play important roles in the foie gras quality (Auvergne et al., 1998). Indeed, studies have shown that if the liver evisceration occurs 20 min after slaughtering, and if the cooling of livers is quick, the livers have a high technological yield (TY; Bouillier-Oudot et al., 2004). The TY of foie gras that is the opposite of the melting rate has strong repercussions on both the organoleptic qualities of foie gras and on the performances of the industrial production units. Thus the TY has already been widely studied (Theron et al., 2013). In ducks, there is also a positive correlation between the liver weight (LW) and the melting rate at cooking, especially above 600 g (Blum et al., 1990; Marie-Etancelin et al., 2011). Similarly, the lipid level correlates with the melting rate at cooking (Rousselot-Pailley et al., 1992). Zootechnical factors, pre-, post-mortem conditions, and liver characteristics have already been analyzed to control better the liver melting at cooking, but some individual variabilities persist (Theron et al., 2012).

In a recent study, plasmatic biomarkers of foie gras quality (LW and TY) were identified by proton nuclear magnetic resonance (¹H-NMR; Mozduri et al., 2021). The plasmatic biomarkers are of main interest because they can bring information of foie gras quality before slaughtering the animals. In a second step, hepatic biomarkers of foie gras qualities are detected on the same experimental materials. Hepatic biomarkers of crude livers might be very useful to choose the cooking program to apply to livers like pasteurization, sterilization, or emulsion to optimize the foie gras products in the industry. Moreover, metabolomics studies with ¹H-NMR were already performed on livers at the end of the overfeeding period to distinguish the signaling of livers with high fat loss corresponding to low TY from livers with low-fat loss that corresponded to high TY (Bonnefont et al., 2014). However, the livers weighed around 570 g to 582 g. The strength of the present study was that male mule ducks were overfed for 6 to 12 days which provided a wide variety of liver weights from 300 g to over 900 g.

This study aimed to identify hepatic biomarkers specific to LW and TY by metabolomics approach using ¹H-NMR by analyzing the duck livers after 6 to 12 days of overfeeding. This study can also provide information to understand nonalcoholic fatty liver disease in humans better.

Materials and Methods

Animal Experimental Design and Liver Characteristics

The animal design was clearly described previously (Bonnefont et al., 2019; Mozduri et al., 2021). Briefly, 120 male mule ducks (Cairina moschata × Anas platyrhynchos) were reared until 12 weeks and overfed twice a day during 6 to 12 days corresponding to 11 to 23 meals. The feed was composed of 97% corn (38% of grain and 62% of flour) supplemented with 3% of a commercial premix diluted into water. The amount of feed increased gradually from 265 to 420 g by meal (Bonnefont et al., 2019). A total of 30 ducks were slaughtered every other day in the second half of the overfeeding period from the 11th meal on day 6 to the 23rd meal on day 12 (Figure 1A). The duck body weight was registered before slaughtering. Then the liver was eviscerated and weighed to obtain LW. Then the livers were cooked as described in Rémignon et al. (2018), and the TY was determined as the ratio between cooked liver weights trimmed of all visible fat and raw liver weights (TY = 100 – % fat loss). At each time point (D6 to D12), 16 ducks were selected among the 30 ducks for liver analyses (n = 64 in total). They were chosen to obtain equivalent means and variabilities of LW and TY in the initial and subgroup groups. The selected samples corresponded to the samples used for identifying plasmatic biomarkers (Mozduri et al., 2021). The body weight, LW, and TY of the selected samples are represented in Figures 1B–D.

FIGURE 1

Figure 1. Description of the duck experimental design and the duck characteristics. Experimental design and liver sampling (A), evolution during the overfeeding period of duck body weight before slaughtering (B), of liver weight after eviscerating (C) and liver technological yield after cooking (D; n = 16 or 17 at each time point).

Liver Sampling and ¹H-NMR Analysis

At 20 min post-mortem, a sample of 20 g was taken off in the upper part of the main lobe of the liver. All samples were dropped into liquid nitrogen and stored at −80°C for ¹H-NMR analyses. They were ground into fine powder. Then, their polar metabolites were extracted with a method adapted from Beckonert et al. (2007) from 0.25 g of the crushed liver with methanol and dichloromethane (Beckonert et al., 2007) and carefully described in Mozduri et al. (2021). The upper phases composed of water and methanol with hydrophilic metabolites were collected in new polypropylene tubes and evaporated with a vacuum concentrator (Concentrator Plus, Eppendorf, Hamburg, Germany) and stored at −80°C until ¹H-NMR analysis. Then all samples were diluted into 650 μl of NMR pH7 phosphate buffer in deuterated water (D₂O) with sodium trimethylsilyl propionate (17.2 g TMSP for 100 ml). The tubes were vortexed and then centrifuged for 15 min at 5,350 g. Finally, 600 μl were sampled in NMR tubes of 5 mm. The ¹H-NMR analyses were done by Bruker Avance III HD NMR spectrometer operating at 600 MHz for a proton resonance frequency. The first step consisted of acquiring all spectra. For this purpose, the NOESYPR1D spin-echo pulse sequence was used to attenuate signals from water. The spectra were acquired at 300 K with time domain: 32 k, 16 dummy scans, and 512 scans for all samples. After Fourier transformation, using Topspin (V2.1, Bruker, Biospin, Munich, Germany), they were manually phased, corrected for the baseline, and calibrated with chemical shifts of TMSP at 0 ppm.

Spectra Preprocessing and Statistical Analysis

The ¹H-NMR spectra were analyzed by two methods: (i) a bucket method and (ii) a metabolite method, both of which were carefully described previously (Mozduri et al., 2021). (i) Briefly, the traditional bucket method consisted in converting the ¹H-NMR spectra into a bucket value table with the Workflow4Metabolomics 3.3 online platform (Giacomoni et al., 2015).¹ After spectra preprocessing (solvent suppression from 5.1 to 4.5 ppm and 3.35 to 3.2, zero-filling, apodization, application of Fourier transform, phasing, baseline correction, and calibration with TSP at 0.0 ppm) and spectra alignment, the spectra were split into buckets with a 0.01 ppm interval from 0.5 to 10 ppm. The raw bucket values were calculated as the integration of the spectrum curves for the corresponding buckets. Then the bucket values were normalized with the integration of the whole spectrum curves as following:

$\begin{array}{l}\mathrm{Normalized\; bucket\; value}=\mathrm{Raw}\mathrm{bucket\; value}/\mathrm{Whole\; spectrum\; integration.}\end{array}$

A table of bucket values was obtained with 64 rows corresponding to the animals and 714 columns corresponding to the buckets identified by their chemical shifts. (ii) Briefly, the metabolite method converted the ¹H-NMR spectra into metabolite relative concentration tables with the ASICS R package (R package version 4.0.2).² ASICS package performed an automatic approach to identify and quantify metabolites in complex ¹H-NMR spectra from their unique peak pattern (fingerprint; Tardivel et al., 2017; Lefort et al., 2019). The metabolite database used consisted of the spectra of 176 pure metabolites described in Tardivel et al. (2017). A total of 80 metabolites were identified and quantified in at least one sample. The methanol was removed as it was used to extract the metabolites. Then only 41 metabolites were kept for further analyses as they were present in at least 50% of the samples at one time point. Thus, the final table of metabolite relative concentrations contained 64 rows corresponding to the animals and 41 columns corresponding to the metabolites.

The bucket and metabolite relative concentration tables were analyzed with SIMCA P+ software (version 12, Umetrics, AB, Umea, Sweden) for carrying out the multivariate statistical analysis as previously described (Mozduri et al., 2021). Briefly, the variables were preprocessed with Pareto normalizations. Principal component analysis (PCA) was performed for finding outliers. Then partial least square analyses (PLS) were performed to explain Y variables (LW and TY) by the X variables (bucket or metabolite data). The PLS scatter plots were drawn, but as only one latent variable was created, the PLS scatter plot represented the scores (t1) on the vertical axis vs. sample identification on the horizontal axis. The goodness-of-fit of the models were estimated by the proportion of cumulative explained variance (R²) for both the X variables (X = buckets or metabolites) and the Y variable (Y = LW or TY) and by the predictive ability of the model (Q²). The root mean square error of estimations (RMSEE) were computed and indicated the fits of the observations to the model. The root mean square error after cross-validations (RMSECv) were also computed. The plot of the Y observed vs. Y predicted values were drawn for each PLS model. The validation of the PLS model was evaluated by comparing the goodness of fit (R²Y and Q²) of the original model with the goodness of fits of 500 models based on data where the ranks of the Y-observations have been randomly permuted, while the X-matrix (bucket or metabolite) has been kept intact. The permutation plots were drawn. The latent variables associated with interesting axes were analyzed using the variable importance in the projection (VIP) method. The variable (bucket or metabolite) with a VIP superior to 1 was considered as “important.” Then a one-by-one regression with either the LW effect or the TY effect was performed on the whole datasets. The p-values were corrected for multiple tests with the Benjamini–Hochberg correction using the R software (version 3.6.1) and named “BH p-value.” A variable was “significant” when the BH p-value was inferior to 0.05 and “tended to be significant” when the BH p-value was between 0.05 and 0.1.

For the buckets with VIP superior to 1, the corresponding metabolites were identified manually by importing the chemical shift lists into the Human Metabolome Database (Wishart et al., 2009).³ All carbohydrates identified were D-carbohydrates, and all amino acids were L-amino acids. To simplify the names of the metabolites, the “D-” and the “L-” were removed before the names of the carbohydrates and the amino acids, respectively. To confirm the identification of the metabolites, the ¹H-NMR peaks of these metabolites were manually checked on the sample spectra with TopSpin software (version 4.0, Bruker BioSpin, Rheinstetten, Germany). For each metabolite, all ¹H-NMR peaks were listed. For each ¹H-NMR peak, the VIP values and the BH p-values of the corresponding buckets were summarized by the number of buckets with VIP superior to 1 and by the range of BH p-values, respectively. The relative concentrations (RC) of a metabolite with the bucket data were estimated with a method adapted from Kostidis et al. (2017) by the following formula (Kostidis et al., 2017):

$\mathrm{Metabolite}{\mathrm{RC}}_j=\mathrm{mean}\left[\left({\mathrm{intensity\; Peak}}_{\mathrm{ij}}\right)/\left(H{\mathrm{number\; Peak}}_{\mathrm{ij}}\right)\right]$

where j represented a specific metabolite, i represented each proton peak of the ¹H-NMR spectrum of the j metabolite, “intensity Peak_ij” was computed as the sum of the bucket intensity of the i peak for the j metabolite, and “H number Peak _ij” was the number of protons corresponding to the i peak for the j metabolite. Then the lists obtained by the bucket method and the metabolite method were compared with Venn diagrams.⁴

All the biomarkers identified by the bucket method and/or the metabolite method were considered as biomarkers. Network analysis based on the correlation of the biomarker RC and the variable (LW or TY) was performed with the functions pls and network of the MixOmics R package (Lê Cao et al., 2009; Rohart et al., 2017; R package version 4.0.2.).⁵

Results

The overfeeding of male mule ducks from 6 to 12 days enabled to obtain animals with large variability of performances. The body weights of the ducks were between 5 and 7.5 kg, their LW between 302.3 and 914.9 g, and the liver TY between 54.8 and 99.5% (Figures 1B–D). This experimental design enabled to obtain livers with strong variabilities like in the foie gras industry, and it was suitable to detect hepatic biomarkers of foie gras quality by ¹H-NMR analysis.

Identification of Hepatic Biomarkers of the Liver Weight of Foie Gras

First, a PCA was implemented to observe the data and to find outlier samples, but no outlier was detected (not shown). The partial least squares (PLS) analysis scatter plot had only one latent variable, and the parameters were cumulative R²X = 0.506 and R²Y = 0.657. The projection of the samples highlighted an evolution of LW with the first latent variable on the vertical axis (Figure 2A). The prediction of the model was Q² = 0.633. The original R²Y and Q²-values were higher than those obtained after 500 permutations, and the regression line of Q²-points intersected the vertical axis below zero (−0.126; Figure 3A). The RMSEE and the RMSECv were close (109.4 and 111.5, respectively, Figure 4A).

FIGURE 2

Figure 2. PLS score plots (A) for liver weight with the bucket method (R²X = 0.506, R²Y = 0.657, Q² = 0.633) and (B) with the metabolite method (R²X = 0.534, R²Y = 0.65, Q² = 0.626), (C) for liver technological yield with the bucket method (R²X = 0.506, R²Y = 0.514, Q² = 0.471), (D) for liver technological yield with the metabolite method (R²X = 0.533, R²Y = 0.522, Q² = 0.418). The numbers correspond to the identification of the samples and the colors to the liver weight value. The legend is indicated on the right of the figure. The numbers correspond to the identification of the samples on the X axis and the colors to the liver weight or technological yield values.

FIGURE 3

Figure 3. Permutation plots (A) for liver weight with the bucket method, (B) for liver weight with the metabolite method, (C) for liver technological yield with the bucket method, (D) for liver technological yield with the metabolite method. 500 permutations were performed. The permutation plot shows, for a selected Y-variable, on the vertical axis the values of R²Y and Q² for the original model and of the Y-permuted models. The horizontal axis shows the correlation between the permuted Y-vectors and the original Y-vector for the selected Y. The original Y has a correlation of 1.0 with itself.

FIGURE 4

Figure 4. Plots of predicted vs. observed data (A) for liver weight with the bucket method, (B) for liver weight with the metabolite method, (C) for liver technological yield with the bucket method, and (D) liver for technological yield with the metabolite method. The Root Mean Square Error of Estimation (RMSEE) and the Root Mean Square Error after cross validation (RMSECv) were indicated.

A group of 64 buckets with a VIP > 1 explained the first latent variable (Supplementary Data 1). For the buckets with VIP > 1 and BH p < 0.05, the involved metabolites were identified manually by importing chemical shift lists into the Human Metabolome Database (Wishart et al., 2009; See Footnote 3). They corresponded to 14 metabolites summarized in Table 1. There were seven carbohydrates identified as biomarkers for LW. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was respectively, seven out of 10 peaks for glucose-6-phosphate (HMDB0001401), six out of eight peaks for glucuronic acid (HMDB0000127), all the two peaks for glyceric acid (HMDB0000139), all two peaks for glycogen (HMDB0000757), all the two peaks for lactate (HMDB0000190), two out of four peaks for malic acid (HMDB0000156), and 12 out of 13 peaks for maltose (HMDB0000163; Table 1). There were five amino acids identified as biomarkers of LW. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was, respectively, two out of four ¹H-NMR peaks for arginine (HMDB0000517), one out of two peaks for N-acetylglycine (HMDB0000532), three out of five peaks for proline (HMDB0000162), all the two peaks for taurine (HMDB0000251) and six out of seven peaks for trans-4-hydroxy-L-proline (HMDB0000725; Table 1). There were also two other organic compounds identified as biomarkers of LW. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was the one peak for allantoin (HMDB0000462) and three out of four peaks for glycerophosphocholine (HMDB0000086; Table 1).

TABLE 1

Table 1. List of the 14 biomarkers of foie gras liver weight identified with the bucket method.

The relative concentrations (RC) of all the 14 metabolites were computed. The BH p-values were summarized in Table 1. A total of nine metabolites were statistically significant (BH p < 0.05) and two tended to be significant (p < 0.10). All these 11 metabolites were further considered as LW biomarkers identified by the bucket method. There were six out the seven carbohydrates: glucuronic acid (p < 0.001), glyceric acid (p < 0.001), glycogen (p = 0.070), lactate (p < 0.001), malic acid (p < 0.001) and maltose (p < 0.001) for carbohydrates, three out of five amino acids: arginine (p < 0.001), taurine (p < 0.001), and trans-4-hydroxy-L-proline (p = 0.005) and also allantoin (p = 0.020) and glycerophosphocholine (p = 0.070; Table 1).

In parallel, the metabolite method was applied, and the 64 spectra were converted into a table of 41 metabolite values with the ASICS R package. No outlier was detected by PCA. The PLS scatter plot had only one latent variable, and the parameters of the models were cumulative R²X = 0.534 and R²Y = 0.650 (Figure 2B). The prediction of the model was low Q² = 0.626. The evolution of LW was well represented on the vertical axis corresponding to the first latent variable (Figure 2B). The original R²Y and Q²-values were higher than those obtained after 500 permutations, and the regression line of Q²-points intersected the vertical axis below zero (−0.119; Figure 3B). The RMSEE and the RMSECv were close (110.6 and 112.5, respectively, Figure 4B). The evolution of LW was well represented on the vertical axis corresponding to the first latent variable (Figure 2B). Only five metabolites had a VIP > 1 to explain this axis, of which all had a BH p < 0.05 (Table 2). Including lactate (VIP = 4.11, BH p < 0.001), glucose (VIP = 2.87, BH p < 0.001), threonine (VIP = 1.72, BH p < 0.001), alanine (VIP = 1.55, BH p < 0.001) and taurine (VIP = 1.29, BH p < 0.001; Table 2).

TABLE 2

Table 2. List of the five biomarkers of foie gras liver weight identified with the metabolite method.

In conclusion, for LW, there were 14 biomarkers. Two biomarkers were identified by the bucket method and the metabolite method (lactate, taurine), three were identified only by the metabolite method (glucose, alanine, and threonine), and nine metabolites were only identified by the bucket method (glucuronic acid, glyceric acid, glycogen, malic acid, maltose, arginine, trans-4-hydroxy-L-proline, allantoin, and glycerophosphocholine; Figure 5A). For the 14 biomarkers, their RCs were computed with the bucket data, and the plots of their RCs in the function of LW were presented in Figure 6A. The correlation network between LW and the biomarkers was presented in Figure 7A. LW was strongly negatively correlated with glucose (−0.95), glycogen (−0.97), glucuronic acid (−0.93), glyceric acid (−0.89), malic acid (−0.61), and maltose (−0.97) and also to alanine (−0.83), arginine (−0.92), taurine (−0.78), trans-4-hydroxy-L-proline (−0.81), and to allantoin (−0.80) and glycerophosphocholine (−0.84). On the contrary, TY was strongly correlated with lactate (0.98) and threonine (0.98; Figure 7A).

FIGURE 5

Figure 5. Comparisons of biomarker lists with Venn diagram. (A) Biomarkers of liver weight identified by the bucket method and by the metabolite method (with VIP > 1 and BH p-value < 0.1). (B) Biomarkers of technological yield identified by the bucket method and by the metabolite method. (C) Biomarkers of liver weight and technological yield identified by at least one method.

FIGURE 6

Figure 6. Plots of biomarker relative contents in function of liver weight (A) or liver technological yield (B). The metabolite relative contents were computed with the bucket data and had no unit. The regression curves were in red.

FIGURE 7

Figure 7. Correlation networks of foie gras biomarkers (A) of liver weight (Y represents the liver weight) and (B) of liver technological yield (Y represents the technological yield). The relative concentration used for the correlations are calculated with the bucket data.

Identification of Hepatic Biomarkers of Foie Gras Technological Yield

First, the spectra were analyzed with the bucket method. A PCA was first performed, but no outlier was detected (not shown). The PLS scatter plot to explain TY had only one latent variable, and the parameters were cumulative R²X = 0.506 and R²Y = 0.514. The prediction of the model was Q² = 0.471. The original R²Y and Q²-values were higher than those obtained after 500 permutations, and the regression line of Q²-points intersected the vertical axis below zero (−0.114; Figure 3C). The RMSEE and the RMSECv were close (8.9 and 9.2, respectively; Figure 4C). The projection of the samples highlighted an evolution of TY with the first latent variable on the vertical axis (Figure 2C). A total of 64 buckets had a VIP value superior to 1 (Supplementary Data 2). They corresponded to 14 metabolites. All the buckets corresponding to each 1H-NMR peak were identified for each metabolite, and their VIP values and their BH p-values were summarized in Table 3. There were six carbohydrates identified as biomarkers for TY. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was, respectively, 21 out of 22 peaks for glucose (HMDB0000122), eight out of 10 peaks for glucose-6-phosphate (HMDB0001401), all of the two peaks for glyceric acid (HMDB0000139), all the two peaks for glycogen (HMDB0000757), all the two peaks for lactate (HMDB0000190) and 12 out of 13 peaks for maltose (HMDB0000163). There were also five amino acids identified as biomarkers of TY of foie gras. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was, respectively, three out four peaks for arginine (HMDB0000517), eight out of nine peaks for glutamic acid (HMDB0000148), one out of two peaks for N-acetylglycine (HMDB0000532), all the five peaks for proline (HMDB0000162) and all the seven peaks for trans-4-hydroxy-L-proline (HMDB0000725). There were also three other organic compounds identified as biomarkers of TY. For each metabolite, the number of peaks that contained at least one bucket with VIP > 1 was, respectively, the only one peak for allantoin (HMDB0000462), all the two peaks for creatine (HMDB0000064), and all the two peaks for ethanolamine (HMDB0000149; Table 3).

TABLE 3

Table 3. List of the 14 biomarkers of foie gras technological yield identified with the bucket method.

The RCs of all the 14 metabolites were computed. The BH p-values were summarized in Table 3. A total of nine metabolites were considered as statistically significant (BH p < 0.05) whose glucose, glyceric acid, glycogen, lactate, maltose for carbohydrates, only glutamic acid for the amino acids and allantoin, creatine, and ethanolamine (Table 2). All these nine metabolites were further considered as biomarkers of TY.

In parallel, the metabolite method was performed. The individuals were less well represented on the scatter plot than with the bucket data (1 latent variable, cumulative R²X = 0.533, R²Y = 0.522; Figure 2D). The prediction of the model was Q² = 0.418. The original R²Y and Q²-values were higher than those obtained after permutation, and the regression line of Q²-points intersected the vertical axis below zero (−0.104; Figure 3D). The RMSEE and the RMSECv were close (8.9 and 9.1, respectively; Figure 4D). The latent variable on the vertical axis explained the evolution of TY (Figure 2D), and six metabolites had VIP values superior to 1 (Table 4). Including glucose (VIP = 2.82, BH p < 0.001), lactate (VIP = 4.06, BH p < 0.001), alanine (VIP = 1.43, BH p < 0.001), taurine (VIP = 1.51, BH p < 0.001), threonine (VIP = 1.72, BH p < 0.001) and guanidinoacetic acid (VIP = 1.04, BH p < 0.001).

TABLE 4

Table 4. List of the 6 biomarkers of foie gras technological yield identified with the metabolite method.

In conclusion, for TY, there were 13 biomarkers: two biomarkers were identified by the bucket method and the metabolite method (glucose and lactate), four were identified only by the metabolite method (alanine, taurine, threonine, and guanidino acetic acid), and seven metabolites were identified by the bucket method (glyceric acid, glycogen, maltose, glutamic acid, allantoin, creatine, and ethanolamine; Figure 5B).

For the 13 biomarkers of TY, their RC was computed with the bucket data, and the plots of their RC in the TY function were presented in Figure 6B. The correlation network between TY and the biomarker RC was presented in Figure 7B. TY was positively correlated with glucose (0.92), glycogen (0.98), glyceric acid (0.9), maltose (0.9), and also to alanine (0.89), glutamic acid (0.93), taurine (0.75), and to allantoin (0.75), ethanolamine (0.95), and guanidinoacetic acid (0.46), whereas TY was negatively correlated with lactate (−0.98), threonine (−0.96) and creatine (−0.57; Figure 7B).

Consequently, the results of the ¹H-NMR analysis identified 14 hepatic biomarkers for the foie gras liver weight and 13 for its technological yield (Table 5 and Figures 5A,B). As the phenotypic correlation between LW and TY was strong (−0.80, p < 0.001), nine biomarkers were common to LW and TY (Figure 5C), of which five carbohydrates: glucose, glyceric acid, glycogen, lactate, maltose, three amino acids: alanine, taurine, threonine, and allantoin. There were five biomarkers specific to LW: glucuronic acid, malic acid, arginine, trans-4-hydroxy-L-proline, glycerophosphocholine, and four biomarkers specific to TY: glutamic acid, creatine, ethanolamine, and guanidinoacetic acid (Figure 5C).

TABLE 5

Table 5. List of the biomarkers of liver weight and technological yield of foie gras.

Discussion

In this experimental design, the ducks were overfed for 6 to 12 days and received 11 to 23 meals. As a result, large variations in duck body weights, LW, and liver TY occurred, explaining the high correlations observed between the biomarker relative quantities and the liver characteristics.

During the overfeeding, the feed was based on corn. As corn is a cereal, it is rich in starch (around 63% as fed). The ingested starch is converted into glucose in the small intestine. Then the glucose is absorbed into the hepatic portal vein and carry to the liver. There, it can be stored like glycogen or converted into fatty acids by de novo lipogenesis. In the study, the liver weight was highly increased (from 80 g on day 0 to more than 750 g on day 12). Thus, the plasmatic glucose contents measured in the carotid artery corresponded to the result of the glucose transfer into the portal vein and the glucose uptake by the liver and other organs. However, the plasmatic glucose content decreased during the overfeeding while LW was increased (correlation of −0.94 in Mozduri et al., 2021) although the ingested starch was strongly increased as the corn quantity meal was increased. That means that the plasmatic glucose content was up-taken more efficiently by the liver during the overfeeding (Pioche et al., 2019). On the contrary, the glucose content in the liver was decreased when LW was increased with a negative correlation of −0.95 between them. So, the glucose was highly metabolized in the liver. At the beginning of the overfeeding period, the liver metabolism was more oriented through glycogenogenesis with strong glycogen storage (until 106 mg of glycogen by grams of the liver after three meals at day 2). In contrast, after seven meals on day 4, the liver metabolism shifted to lipogenesis with a strong triglyceride accumulation (29.5% of lipids at day 4 vs. 4.6 at day 0; Bonnefont et al., 2019). The glycogen content estimated in the liver in the second half of the overfeeding (from day 6 to day 12) was strongly negatively correlated with liver weight (−0.97). Furthermore, in other animal models as rats with hepatic steatosis, the liver glycogen content was lower than in control rats (Kusunoki et al., 2002).

In addition, LW was negatively correlated with TY (−0.82). Thus, the glucose content and the glycogen content were strongly correlated with TY (+0.92 and +0.98). That is consistent with previous results on foie gras with livers weighing around 580 g (Bonnefont et al., 2014).

In parallel, the hepatic lactate content was the most important metabolite to discriminate the livers in the function of their LW or TY (VIP values of the O-PLS models of 4.11 and 4.06, respectively). It was positively correlated with LW (+0.98) and negatively correlated with TY (−0.98). Previously, the lactate was the most discriminant metabolite between high TY livers and low TY livers with equivalent weights (Bonnefont et al., 2014). The lactate is the last metabolite of anaerobic glycolysis that converts glucose into pyruvate and then into lactate via the lactate dehydrogenase enzyme. In a recent mouse model, it was shown that glucose oxidation in the liver was central in the development of steatosis, as glycolysis metabolizes glucose into pyruvate, which can be anaerobically converted into lactate or aerobically converted into acetyl-CoA. The hepatic lactate in nonalcoholic steatohepatitis (NASH) was primarily diverted toward the production of acetyl-CoA for lipogenesis rather than the production of glucose (Zhu et al., 2018). In addition, Lo et al. (2020b) recently demonstrated in ducks undergoing overfeeding that low LW was associated with aerobic energy metabolism and high weight livers with anaerobic energy metabolism (Lo et al., 2020b). These results suggest that the efficiency of energy metabolism would influence both LW and TY. Therefore, it can be supposed that the increase in LW could lead to a decrease in the efficiency of this metabolism translated by the increase in hepatic lactate content, which would result in a decrease in TY.

Contrary to hepatic lactate content, the glucuronic acid hepatic content was negatively correlated with LW (−0.93). However, the glucuronic acid can be conjugated to lipophilic substrates via UDP-glucuronosyltransferases. These enzymes catalyze phase II biotransformation reactions and promote glucuronidation. Glucuronidation is a major detoxification pathway for endogenous and exogenous compounds, and it is involved in transporter-mediated excretion into bile and urine (Williams et al., 2004). In steatosis livers of mice, the UDP-glucuronosyltransferase expression increased with the increased hepatic triglyceride content and could have a significant impact on determining circulating hormone levels (Xu et al., 2012). Thus, the lower quantity of glucuronic acid in the high-weight liver could be explained using glucuronic acid for glucuronidation.

The amino acid metabolism of duck livers was strongly impacted by the overfeeding period at the same time as the carbohydrate metabolism. The alanine, taurine, and threonine were identified as biomarkers of both LW and TY, whereas arginine and trans-4-hydroxy-proline were only identified as biomarkers of LW and glutamic acid of TY. Briefly, the amino acid liver content was decreased when the LW was increased and/or when the TY was decreased except for the threonine. The reduction of amino acid metabolism in the liver with the enhanced hepatic steatosis was already underlined in overfed ducks (François et al., 2014; Lo et al., 2020b) and other hepatic steatosis models as obese mice (Du et al., 2012; Bruckbauer and Zemel, 2014). Thus, these amino acids were increased in low-weight livers. It was shown in rats and mice that a diet supplementation in taurine (Gentile et al., 2011) and arginine (Voloshin et al., 2014; Sellmann et al., 2017) reduced hepatic lipid accumulation. Thus, they were used as a preventative treatment against NASH. The taurine transporter-deficient mice showed strongly decreased taurine levels in various tissues like the liver and developed chronic hepatitis and liver fibrosis during adulthood, accompanied by severe augmentation of hepatocyte apoptosis (Warskulat et al., 2006). In addition, the restriction of lysine and threonine in diets increased the free fatty acid content in the liver of rats (Viviani et al., 1966).

Furthermore, the increase of the activities of the transaminase enzymes alanine aminotransferase (ALT) and aspartate aminotransferase (AST) with liver injury is strongly documented (Pratt and Kaplan, 2000). ALT catalyzes the transfer of an amino group from alanine to α-ketoglutarate to produce pyruvate and glutamate, and AST catalyzes the interconversion of aspartate and α-ketoglutarate to oxaloacetate and glutamate. Both reactions are reversible. Thus, the identification of alanine as a biomarker of both LW and TY and glutamic acid as a specific biomarker of TY should be directly linked to the ALT and AST activities in the liver. Also, the oxaloacetate issued from AST activity is a key intermediate in the citric acid cycle. Actually, in this cycle, the malate dehydrogenase converts the malate into oxaloacetate with the NAD⁺ cofactor, and after several reactions, the oxaloacetate is converted again into malate. However, the malate is known to play a role in lipogenesis by furnishing NADPH to reduce acetyl CoA to fatty acids (Wise Jr and Ball, 1964). This may explain the identification of malic acid as a biomarker of LW. In the Poland goose, the activity of ME was correlated positively with the weight of the fatty liver (Mourot et al., 2000).

In addition, glyceric acid was identified as a biomarker of LW and TY. The glyceric acid is a substrate for glycerol synthesis (Feraudi and Bert, 1977), and the glycerol is a substrate for triglyceride synthesis (Tidwell and Johnston, 1961). Thus, the negative correlation of glyceric acid with LW (−0.89) could be explained by using glyceric acid for lipogenesis.

Moreover, in the liver, creatine is synthesized from glycine and arginine. First, both amino acids are combined to form guanidinoacetate which is then methylated using S-adenosyl methionine to synthesize creatine (Rosenberg, 1959). The amount of arginine was negatively correlated with LW (−0.92). The arginine is involved in the process of creatine synthesis (Barcelos et al., 2016). Creatine plays a role in the antioxidant role against aqueous radical and reactive species ions (Lawler et al., 2002). The supplementation in creatine protects the liver from hepatotoxicity by attenuating oxidative stress (Aljobaily et al., 2021). Lo et al. (2020a) showed that, in overfed liver duck, an increase in liver weight resulted in a rise in the cellular oxidative stress level (Lo et al., 2020a). These results corroborated the positive correlation between the level of oxidative stress and LW.

The creatine phosphate serves as a dynamic reservoir of high-energy phosphate in exchange for ATP. The steatosis strongly impacts creatine metabolism as arginine is a specific biomarker of LW, and guanidinoacetate and creatine are specific biomarkers of TY. However, contrary to guanidinoacetate, creatine was negatively correlated with TY. Thus, the creatine content in the liver with low TY may be increased to control oxidative stress. Oxidative stress is defined as the presence of metabolic and radical substances or so-called reactive (oxygen, nitrogen, or chlorine) species (Elnesr et al., 2019; Elwan et al., 2019).

Furthermore, the positive correlation of hepatic amino acid content and TY in foie gras was previously underlined in association with a reduction of oxidative stress (Theron et al., 2011; Bonnefont et al., 2014; François et al., 2014). Actually in chronic liver disease there is increased reactive oxygen species production and decreased activity of antioxidant systems (De Minicis et al., 2006). Recently, hypoxic response to severe hypoxia was highlighted in heavy livers of ducks (Lo et al., 2020a).

Moreover, the modification of amino acid metabolism in livers during the overfeeding period corroborates the evolution of amino acid metabolism in plasma of the same ducks (Mozduri et al., 2021) and other hepatic steatosis models, as it was discussed in Mozduri et al. (2021).

Finally, the glycerophosphocholine was identified as a specific biomarker of LW (−0.84), and ethanolamine as a specific biomarker of TY (+0.95). Ethanolamine can be phosphorylated to form phosphoethanolamine. Glycerophosphocholine and phosphoethanolamine are cytosolic intermediates of phospholipids synthesis. In patients with cirrhosis that is considered the most severe stage of liver steatosis disease, the phosphoethanolamine liver content was increased whereas the glycerophosphocholine was decreased compared to controls but not in patients with a lower severe stage of fatty liver disease as NASH (Sevastianova et al., 2010). On the contrary, Kalhan et al. (2011) reported a significantly lower glycerophosphocholine content in steatohepatitis compared with nondiabetic healthy controls in humans (Kalhan et al., 2011). Furthermore, in overfed ducks, a higher glycerophosphocholine content in the low-fat-loss livers that corresponded to high TY livers was underlined (Bonnefont et al., 2014).

The last biomarker of LW and TY was allantoin. It was shown that a supplementation in allantoin in NASH or diabetic mice inhibited the structural damage of the liver concerning fat accumulation (Movahhed et al., 2019; Ma et al., 2020) which explained its lower content in high weight livers.

Conclusion

To conclude, the analysis of the metabolism of male mule duck livers during the overfeeding period through ¹H-NMR analysis enabled the identification of eighteen liver biomarkers of foie gras qualities. Nine of them were identified for both LW and TY of foie gras: five were carbohydrates like glucose, glyceric acid, glycogen, lactate, and maltose, three were amino acids like alanine taurine and threonine, plus allantoin. Five of them were specific to the LW: two carbohydrates as glucuronic acid, malic acid, two amino acids as arginine, and trans-4-hydroxy-L-proline plus a glycerophosphocholine, a phospholipid. Furthermore, four biomarkers were specific to foie gras TY. Two of them were involved in creatine metabolism (creatine and guanidinoacetic acid). One could be an intermediate of phospholipid (ethanolamine) and may be involved in membrane stability. The last one was glutamic acid produced by ALT, whose activity in the liver is enhanced in hepatosteatosis. As a result, in heavy livers, the liver metabolism was oriented through a reduction of carbohydrate metabolism, and the plasma membrane could be damaged, which may explain the low technological yield of these livers.

These findings will be complete if an analysis of the correlation between the plasma and liver metabolisms is made to understand the co-evolution of these tissues during the overfeeding period.

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics Statement

The animal study was reviewed and approved by experimental approval A24-137-1.

Author Contributions

JA supervised the animal experimental rearing, overfeeding, and slaughtering. BL prepared the liver samples for NMR analysis with the supervision of NM-G. BL, NM-G, and CC performed the NMR analysis. BL and NM-G made the preprocessing of the NMR spectra. ZM, BL, and CB performed the statistical analyses to interpret the data that were helped by AB, MM, and AM, and wrote the paper. All authors contributed to the article and approved the submitted version.

Funding

This project was funded by Interprofession Foie gras (CIFOG) and Aquitaine Regional and Dordogne Departmental Councils.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Acknowledgments

The authors thank the staff of the Station d’Expérimentation Appliquée et de Démonstration sur l’oie et le Canard (Dordogne, France) for their excellent supervision of this study, especially Franck Lavigne and Cédric Mondoux. We also thank the staff of the UMR1388 Génétique Physiologie et Systèmes d’Elevage INRAe, in particular Michel Bouillier-Oudot and Caroline Molette for their help in designing the experimental project and getting funding, and Hélène Manse and Stéphane Seidlinger for liver sampling.

Supplementary Material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys.2021.694809/full#supplementary-material

Footnotes

1. ^https://workflow4metabolomics.org/

2. ^https://bioconductor.org/packages/ASICS/

3. ^http://hmdb.ca/

4. ^https://bioinfogp.cnb.csic.es/tools/venny/index.html

5. ^https://CRAN.R-project.org/package=mixOmics

References

Aljobaily, N., Viereckl, M. J., Hydock, D. S., Aljobaily, H., Wu, T.-Y., Busekrus, R., et al. (2021). Creatine alleviates doxorubicin-induced liver damage by inhibiting liver fibrosis, inflammation, oxidative stress, and cellular senescence. Nutrients 13:41. doi: 10.3390/nu13010041

PubMed Abstract | CrossRef Full Text | Google Scholar

Auvergne, A., Candau, M., Babile, R., Manse, H., and Bouillier-Oudot, M. (1998). Relation entre l’état de réplétion digestive et la composition hépatique du canard mulard en gavage. Reprod. Nutr. Dev. 38, 39–47. doi: 10.1051/rnd:19980104

CrossRef Full Text | Google Scholar

Baéza, E., Rideau, N., Chartrin, P., Davail, S., Hoo-Paris, R., Mourot, J., et al. (2005). Canards de Barbarie, Pékin et leurs hybrides: aptitude à l’engraissement. Prod. Anim. 18, 131–141. doi: 10.20870/productions-animales.2005.18.2.3516

CrossRef Full Text | Google Scholar

Barcelos, R. P., Stefanello, S. T., Mauriz, J. L., Gonzalez-Gallego, J., and Soares, F. A. A. (2016). Creatine and the liver: metabolism and possible interactions. Mini Rev. Med. Chem. 16, 12–18. doi: 10.2174/1389557515666150722102613

PubMed Abstract | CrossRef Full Text | Google Scholar

Beckonert, O., Keun, H. C., Ebbels, T. M., Bundy, J., Holmes, E., Lindon, J. C., et al. (2007). Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat. Protoc. 2, 2692–2703. doi: 10.1038/nprot.2007.376

PubMed Abstract | CrossRef Full Text | Google Scholar

Blum, J., Labie, C., and Raynaud, P. (1990). Influence du poids et de la composition chimique du foie gras d’oie sur la fonte mesurée après stérilisation à 104° C. Sci. Aliment. 10, 543–554.

Google Scholar

Bonnefont, C. M., Guerra, A., Théron, L., Molette, C., Canlet, C., and Fernandez, X. (2014). Metabolomic study of fatty livers in ducks: identification by 1 H-NMR of metabolic markers associated with technological quality. Poult. Sci. 93, 1542–1552. doi: 10.3382/ps.2013-03546

PubMed Abstract | CrossRef Full Text | Google Scholar

Bonnefont, C. M., Molette, C., Lavigne, F., Manse, H., Bravo, C., Lo, B., et al. (2019). Evolution of liver fattening and foie gras technological yield during the overfeeding period in mule duck. Poult. Sci. 98, 5724–5733. doi: 10.3382/ps/pez359

PubMed Abstract | CrossRef Full Text | Google Scholar

Bouillier-Oudot, M., Leprettre, S., Dubois, J., and Babilé, R. (2004). “Evolution de la composition hépatique lors du refroidissement post-mortem de foies gras d’oies dans la carcasse,” in 6. Journées de la Recherche sur les Palmipèdes à Foie Gras: Comité Interprofessionnel des Palmipèdes à Foie Gras; October 07–08, 2004; Arcachon, France.

Google Scholar

Bruckbauer, A., and Zemel, M. B. (2014). Synergistic effects of polyphenols and methylxanthines with leucine on AMPK/Sirtuin-mediated metabolism in muscle cells and adipocytes. PLoS One 9:e89166. doi: 10.1371/journal.pone.0089166

PubMed Abstract | CrossRef Full Text | Google Scholar

De Minicis, S., Bataller, R., and Brenner, D. A. (2006). NADPH oxidase in the liver: defensive, offensive, or fibrogenic? Gastroenterology 131, 272–275. doi: 10.1053/j.gastro.2006.05.048

PubMed Abstract | CrossRef Full Text | Google Scholar

Du, Y., Meng, Q., Zhang, Q., and Guo, F. (2012). Isoleucine or valine deprivation stimulates fat loss via increasing energy expenditure and regulating lipid metabolism in WAT. Amino Acids 43, 725–734. doi: 10.1007/s00726-011-1123-8

PubMed Abstract | CrossRef Full Text | Google Scholar

Elnesr, S. S., Elwan, H. A. M., Xu, Q. Q., Xie, C., Dong, X. Y., and Zou, X. T. (2019). Effects of in ovo injection of sulfur-containing amino acids on heat shock protein 70, corticosterone hormone, antioxidant indices, and lipid profile of newly hatched broiler chicks exposed to heat stress during incubation. Poult. Sci. 98, 2290–2298. doi: 10.3382/ps/pey609

PubMed Abstract | CrossRef Full Text | Google Scholar

Elwan, H. A., Elnesr, S. S., Xu, Q., Xie, C., Dong, X., and Zou, X. (2019). Effects of in ovo methionine-cysteine injection on embryonic development, antioxidant status, IGF-I and tlr4 gene expression, and jejunum histomorphometry in newly hatched broiler chicks exposed to heat stress during incubation. Animals 9:25. doi: 10.3390/ani9010025

PubMed Abstract | CrossRef Full Text | Google Scholar

Feraudi, M., and Bert, H. (1977). Role of rat-liver alcohol dehydrogenase in the glycerol pathway to L-lactate in homogenates. Arch. Int. Physiol. Biochim. 85, 91–100.

Google Scholar

François, Y., Marie-Etancelin, C., Vignal, A., Viala, D., Davail, S. P., and Molette, C. (2014). Mule duck “foie gras” shows different metabolic states according to its quality phenotype by using a proteomic approach. J. Agric. Food Chem. 62, 7140–7150. doi: 10.1021/jf5006963

PubMed Abstract | CrossRef Full Text | Google Scholar

Gentile, C. L., Nivala, A. M., Gonzales, J. C., Pfaffenbach, K. T., Wang, D., Wei, Y., et al. (2011). Experimental evidence for therapeutic potential of taurine in the treatment of nonalcoholic fatty liver disease. Am. J. Phys. Regul. Integr. Comp. Phys. 301, R1710–R1722. doi: 10.1152/ajpregu.00677.2010

PubMed Abstract | CrossRef Full Text | Google Scholar

Giacomoni, F., Le Corguille, G., Monsoor, M., Landi, M., Pericard, P., Pétéra, M., et al. (2015). Workflow4Metabolomics: A collaborative research infrastructure for computational metabolomics. Bioinformatics 31, 1493–1495. doi: 10.1093/bioinformatics/btu813

PubMed Abstract | CrossRef Full Text | Google Scholar

Goodridge, A. G. (1987). Dietary regulation of gene expression: enzymes involved in carbohydrate and lipid metabolism. Annu. Rev. Nutr. 7, 157–185. doi: 10.1146/annurev.nu.07.070187.001105

PubMed Abstract | CrossRef Full Text | Google Scholar

Kalhan, S. C., Guo, L., Edmison, J., Dasarathy, S., Mccullough, A. J., Hanson, R. W., et al. (2011). Plasma metabolomic profile in nonalcoholic fatty liver disease. Metabolism 60, 404–413. doi: 10.1016/j.metabol.2010.03.006

PubMed Abstract | CrossRef Full Text | Google Scholar

Kostidis, S., Addie, R. D., Morreau, H., Mayboroda, O. A., and Giera, M. (2017). Quantitative NMR analysis of intra-and extracellular metabolism of mammalian cells: A tutorial. Anal. Chim. Acta 980, 1–24. doi: 10.1016/j.aca.2017.05.011

PubMed Abstract | CrossRef Full Text | Google Scholar

Kusunoki, M., Tsutsumi, K., Hara, T., Ogawa, H., Nakamura, T., Miyata, T., et al. (2002). Correlation between lipid and glycogen contents in liver and insulin resistance in high-fat [ndash] fed rats treated with the lipoprotein lipase activator NO-1886. Metabolism 51, 792–795. doi: 10.1053/meta.2002.32732

PubMed Abstract | CrossRef Full Text | Google Scholar

Lawler, J., Barnes, W. S., Wu, G., Song, W., and Demaree, S. (2002). Direct antioxidant properties of creatine. Biochem. Biophys. Res. Commun. 290, 47–52. doi: 10.1006/bbrc.2001.6164

PubMed Abstract | CrossRef Full Text | Google Scholar

Lê Cao, K.-A., González, I., and Déjean, S. (2009). integrOmics: an R package to unravel relationships between two omics datasets. Bioinformatics 25, 2855–2856. doi: 10.1093/bioinformatics/btp515

PubMed Abstract | CrossRef Full Text | Google Scholar

Lefort, G., Liaubet, L., Canlet, C., Tardivel, P., Père, M.-C., Quesnel, H., et al. (2019). ASICS: an R package for a whole analysis workflow of 1D 1H NMR spectra. Bioinformatics 35, 4356–4363. doi: 10.1093/bioinformatics/btz248

PubMed Abstract | CrossRef Full Text | Google Scholar

Lo, B., Marty-Gasset, N., Manse, H., Bannelier, C., Bravo, C., Domitile, R., et al. (2020a). Cellular markers of mule duck livers after force-feeding. Poult. Sci. 99, 3567–3573. doi: 10.1016/j.psj.2020.03.048

PubMed Abstract | CrossRef Full Text | Google Scholar

Lo, B., Marty-Gasset, N., Pichereaux, C., Bravo, C., Manse, H., Domitile, R., et al. (2020b). Proteomic analysis of two weight classes of mule duck “foie gras” at the end of an overfeeding period. Front. Physiol. 11:569329. doi: 10.3389/fphys.2020.569329

PubMed Abstract | CrossRef Full Text | Google Scholar

Ma, J., Meng, X., Liu, Y., Yin, C., Zhang, T., Wang, P., et al. (2020). Effects of a rhizome extract of Dioscorea batatas and its bioactive compound, allantoin in high fat diet and streptozotocin-induced diabetic mice and the regulation of liver, pancreas and skeletal muscle dysfunction. J. Ethnopharmacol. 259:112926. doi: 10.1016/j.jep.2020.112926

PubMed Abstract | CrossRef Full Text | Google Scholar

Marie-Etancelin, C., Basso, B., Davail, S., Gontier, K., Fernandez, X., Vitezica, Z.-G., et al. (2011). Genetic parameters of product quality and hepatic metabolism in fattened mule ducks. J. Anim. Sci. 89, 669–679. doi: 10.2527/jas.2010-3091

PubMed Abstract | CrossRef Full Text | Google Scholar

Mourot, J., Guy, G., Lagarrigue, S., Peiniau, P., and Hermier, D. (2000). Role of hepatic lipogenesis in the susceptibility to fatty liver in the goose (Anser anser). Comp. Biochem. Physiol. B Biochem. Mol. Biol. 126, 81–87. doi: 10.1016/S0305-0491(00)00171-1

PubMed Abstract | CrossRef Full Text | Google Scholar

Movahhed, T. K., Moslehi, A., Golchoob, M., and Ababzadeh, S. (2019). Allantoin improves methionine-choline deficient diet-induced nonalcoholic steatohepatitis in mice through involvement in endoplasmic reticulum stress and hepatocytes apoptosis-related genes expressions. Iran. J. Basic Med. Sci. 22, 736–744. doi: 10.22038/ijbms.2019.33553.8012

PubMed Abstract | CrossRef Full Text | Google Scholar

Mozduri, Z., Marty-Gasset, N., Lo, B., Masoudi, A. A., Morisson, M., Canlet, C., et al. (2021). Identification of plasmatic biomarkers of foie gras qualities in duck by metabolomics. Front. Physiol. 12:628264. doi: 10.3389/fphys.2021.628264

PubMed Abstract | CrossRef Full Text | Google Scholar

Pioche, T., Skiba, F., Bernadet, M.-D., Seiliez, I., Massimino, W., Houssier, M., et al. (2019). Kinetic study of the expression of genes related to hepatic steatosis, global intermediate metabolism and cellular stress during overfeeding in mule ducks. bioRxiv [Preprint]. doi: 10.1101/690156

CrossRef Full Text | Google Scholar

Pratt, D. S., and Kaplan, M. M. (2000). Evaluation of abnormal liver-enzyme results in asymptomatic patients. N. Engl. J. Med. 342, 1266–1271. doi: 10.1056/NEJM200004273421707

PubMed Abstract | CrossRef Full Text | Google Scholar

Rémignon, H., Yahia, R. B. H., Marty-Gasset, N., and Wilkesman, J. (2018). Apoptosis during the development of the hepatic steatosis in force-fed ducks and cooking yield implications. Poult. Sci. 97, 2211–2217. doi: 10.3382/ps/pey054

PubMed Abstract | CrossRef Full Text | Google Scholar

Rohart, F., Gautier, B., Singh, A., and Lê Cao, K.-A. (2017). mixOmics: An R package for ‘omics feature selection and multiple data integration. PLoS Comput. Biol. 13:e1005752. doi: 10.1371/journal.pcbi.1005752

PubMed Abstract | CrossRef Full Text | Google Scholar

Rosenberg, H. (1959). The occurrence of guanidinoacetic acid and other substituted guanidines in mammalian liver. Biochem. J. 72:582. doi: 10.1042/bj0720582

PubMed Abstract | CrossRef Full Text | Google Scholar

Rousselot-Pailley, D., Guy, G., Gourichon, D., Sellier, N., and Blum, J. (1992). Influence des conditions d’abattage et de réfrigération sur la qualité des foies gras d’oie. INRA Prod. Animal. 5, 167–172. doi: 10.20870/productions-animales.1992.5.3.4230

CrossRef Full Text | Google Scholar

Sellmann, C., Degen, C., Jun Jin, C., Nier, A., Engstler, A. J., Alkhatib, D. H., et al. (2017). Oral arginine supplementation protects female mice from the onset of nonalcoholic steatohepatitis. Amino Acids 49, 1215–1225. doi: 10.1007/s00726-017-2423-4

PubMed Abstract | CrossRef Full Text | Google Scholar

Sevastianova, K., Hakkarainen, A., Kotronen, A., Corner, A., Arkkila, P., Arola, J., et al. (2010). Nonalcoholic fatty liver disease: detection of elevated nicotinamide adenine dinucleotide phosphate with in vivo 3.0-T 31P MR spectroscopy with proton decoupling. Radiology 256, 466–473. doi: 10.1148/radiol.10091351

PubMed Abstract | CrossRef Full Text | Google Scholar

Tardivel, P. J., Canlet, C., Lefort, G., Tremblay-Franco, M., Debrauwer, L., Concordet, D., et al. (2017). ASICS: an automatic method for identification and quantification of metabolites in complex 1D 1 H NMR spectra. Metabolomics 13:109. doi: 10.1007/s11306-017-1244-5

CrossRef Full Text | Google Scholar

Tavernier, A., Ricaud, K., Bernadet, M.-D., Gontier, K., and Davail, S. (2018). Pre-and post-prandial expression of genes involved in lipid metabolism at the end of the overfeeding period of mule ducks. Mol. Cell. Biochem. 438, 111–121. doi: 10.1007/s11010-017-3118-6

PubMed Abstract | CrossRef Full Text | Google Scholar

Theron, L., Cullere, M., Bouillier-Oudot, M., Manse, H., Dalle Zotte, A., Molette, C., et al. (2012). Modeling the relationships between quality and biochemical composition of fatty liver in mule ducks. J. Anim. Sci. 90, 3312–3317. doi: 10.2527/jas.2011-4945

PubMed Abstract | CrossRef Full Text | Google Scholar

Theron, L., Fernandez, X., Marty-Gasset, N., Chambon, C., Viala, D., Pichereaux, C., et al. (2013). Proteomic analysis of duck fatty liver during post-mortem storage related to the variability of fat loss during cooking of “foie gras”. J. Agric. Food Chem. 61, 920–930. doi: 10.1021/jf302979q

PubMed Abstract | CrossRef Full Text | Google Scholar

Theron, L., Fernandez, X., Marty-Gasset, N., Pichereaux, C., Rossignol, M., Chambon, C., et al. (2011). Identification by proteomic analysis of early post-mortem markers involved in the variability in fat loss during cooking of mule duck “foie gras”. J. Agric. Food Chem. 59, 12617–12628. doi: 10.1021/jf203058x

PubMed Abstract | CrossRef Full Text | Google Scholar

Tidwell, H. C., and Johnston, J. M. (1961). Glyceride glycerol utilization in triglyceride formation. Arch. Biochem. Biophys. 93, 546–549. doi: 10.1016/S0003-9861(61)80050-7

CrossRef Full Text | Google Scholar

Viviani, R., Sechi, A., and Lenaz, G. (1966). Lipid metabolism in fatty liver of lysine-and threonine-deficient rats. J. Lipid Res. 7, 473–478. doi: 10.1016/S0022-2275(20)39256-7

PubMed Abstract | CrossRef Full Text | Google Scholar

Voloshin, I., Hahn-Obercyger, M., Anavi, S., and Tirosh, O. (2014). L-arginine conjugates of bile acids-a possible treatment for nonalcoholic fatty liver disease. Lipids Health Dis. 13, 1–11. doi: 10.1186/1476-511X-13-69

PubMed Abstract | CrossRef Full Text | Google Scholar

Warskulat, U., Borsch, E., Reinehr, R., Heller-Stilb, B., Mönnighoff, I., Buchczyk, D., et al. (2006). Chronic liver disease is triggered by taurine transporter knockout in the mouse. FASEB J. 20, 574–576. doi: 10.1096/fj.05-5016fje

PubMed Abstract | CrossRef Full Text | Google Scholar

Williams, J. A., Hyland, R., Jones, B. C., Smith, D. A., Hurst, S., Goosen, T. C., et al. (2004). Drug-drug interactions for UDP-glucuronosyltransferase substrates: a pharmacokinetic explanation for typically observed low exposure (AUCi/AUC) ratios. Drug Metab. Dispos. 32, 1201–1208. doi: 10.1124/dmd.104.000794

PubMed Abstract | CrossRef Full Text | Google Scholar

Wise, E. M. Jr., and Ball, E. G. (1964). Malic enzyme and lipogenesis. Proc. Natl. Acad. Sci. U. S. A. 52, 1255. doi: 10.1073/pnas.52.5.1255

PubMed Abstract | CrossRef Full Text | Google Scholar

Wishart, D. S., Knox, C., Guo, A. C., Eisner, R., Young, N., Gautam, B., et al. (2009). HMDB: a knowledgebase for the human metabolome. Nucleic Acids Res. 37, D603–D610. doi: 10.1093/nar/gkn810

PubMed Abstract | CrossRef Full Text | Google Scholar

Xu, J., Kulkarni, S. R., Li, L., and Slitt, A. L. (2012). UDP-glucuronosyltransferase expression in mouse liver is increased in obesity-and fasting-induced steatosis. Drug Metab. Dispos. 40, 259–266. doi: 10.1124/dmd.111.039925

PubMed Abstract | CrossRef Full Text | Google Scholar

Zhu, L., Baker, S. S., Shahein, A., Choudhury, S., Liu, W., Bhatia, T., et al. (2018). Upregulation of non-canonical Wnt ligands and oxidative glucose metabolism in NASH induced by methionine-choline deficient diet. Trends Cell Mol. Biol. 13, 47–56.

PubMed Abstract | Google Scholar