The dietary fibre (DF) fraction represents a very diverse group of polysaccharides organised as a three-dimensional network that makes up the major part of plant cell walls(Reference Bach Knudsen1). While the DF level of conventional feeds for sows typically varies between 150 and 250 g/kg DM, very high DF diets (400–500 g/kg DM) may be applied for pregnant sows to prolong the feeling of satiety and reduce the stereotypic behaviour of loose reared pregnant sows(Reference Danielsen and Vestergaard2, Reference de Leeuw, Bolhuis and Bosch3). DF has also shown numerous health benefits in human studies; for example, evidence has been obtained that DF reduces risk factors for CHD such as hyperlipidaemia(Reference Salas-Salvado, Farres and Luque4, Reference Trowell5), a high intake of soluble DF improves glycaemic control in type 2 diabetes(Reference Chandalia, Garg and Lutjohann6) and improves gastrointestinal health(Reference Roberfroid7). Consequently, both from an animal and human nutrition point of view, obtaining a better understanding of the metabolic consequences of DF consumption is of interest.
Metabonomics is an emerging tool in nutritional physiology(Reference Solanky, Bailey and Holmes8, Reference Wishart9). It is referred to as the quantitative measurement of the dynamic multiparametric metabolic response of living systems to pathophysiological stimuli or genetic modification(Reference Nicholson, Lindon and Holmes10). This can be seen as metabolic profiles of the metabolic network measured in tissue or more typically in biofluids. Accordingly, metabonomics in combination with classical animal models, e.g. the catheterised pig model, can provide new insight into the absorption and metabolism of nutrients. The catheterised pig model with permanent catheters placed in the portal vein and the mesenteric artery and a flow probe around the portal vein is a well-established model for studying the uptake of water-soluble compounds to the portal vein system(Reference Rerat, Vaissade and Vaugelade11, Reference Bach Knudsen, Laerke and Steenfeldt12).
NMR spectroscopy is a commonly used tool for metabonomics and has been used to study the metabolic response in toxicology and pharmacology(Reference Nicholson, Lindon and Holmes10, Reference Bock13–Reference Nicholson, Buckingham and Sadler15) but also to identify specific components in nutritional studies(Reference Nicholson, Connelly and Lindon16–Reference Solanky, Bailey and Holmes22). NMR is an inherently quantitative and non-destructive technique that requires little or no sample preparation. In principle, liquid-state 1H NMR spectroscopy detects all mobile proton-containing metabolites. Thus, 1H NMR spectroscopy is a powerful tool for holistic and explorative investigations. The measurements, however, suffer from low sensitivity compared with other methods, e.g. MS. In a recent study, we investigated the quantitative uptake of glucose, lactate and SCFA and the apparent insulin production in sows fed a concentrated control diet and two high-fibre diets (approximately 440 g DF/kg DM) composed of co-products from the vegetable food and agro industries by using traditional analytical methods(Reference Serena, Jorgensen and Bach Knudsen23, Reference Serena, Jorgensen and Bach Knudsen24). The results from 1H NMR spectroscopy of blood samples from the present study have been presented in a short paper(Reference Yde, Bertram and Bach Knudsen25) that indicated a relationship between betaine uptake and elevated creatine plasma concentrations. In the present study, we further expanded the previous findings to provide a more complete insight into the metabolites that are altered as a result of DF ingestion and the dynamic aspect of absorption. Consequently, the main objective of the present study was to investigate whether 1H NMR-based metabonomics can provide new and a more complete insight into nutritive and non-nutritive components released during the digestion processes and the metabolism thereof. For this purpose, we used plasma samples from the mesenteric artery and the portal vein of six catheterised sows. The principle of the method that has been used in the present study is as follows: (1) blood samples that have been simultaneously collected from the mesenteric artery and the portal vein; (2) establishment of 1H NMR arterial–venous differences (ΔAV); (3) identification of biomarkers by a multivariate approach using the multilevel structured cross-over study design; (4) use of biomarkers in an ANOVA model for time profiles and determination of significant variation.
Materials and methods
Diets
For the purpose of the study, three diets were prepared from whole wheat and barley supplemented with different co-products from the vegetable food and agro industries: potato pulp, KMC (Brande, Denmark); sugarbeet pulp, Danisco Sugar A/S (Assens, Denmark); pectin residue, CPKelco ApS (Lille Skensved, Denmark); brewers' spent grain, Agro-korn A/S (Videbæk, Denmark); pea hulls and seed residue, DLF Trifolium A/S (Roskilde, Denmark). The diets were formulated to contain different types and levels of DF (see Table 1 for composition). A low-DF diet (LF; 177 g DF/kg DM) was prepared from wheat and barley as the main carbohydrate source and two high-DF diets (approximately 440 g DF/kg DM) were prepared by substituting wheat and barley with sugarbeet pulp, potato pulp and pectin residue (high-soluble fibre, HFS) and with approximately one-third of sugarbeet pulp, potato pulp and pectin residue and two-thirds of brewers' spent grain, pea hulls and seed residue (high-fibre insoluble diet, HFI). The diets were formulated to meet the Danish recommendations for essential macro- and micronutrients and were milled to pass through a 2 mm screen. The daily feeding level was 2000 g feed (as-fed basis) once per d and the sows had free access to water. In our diet formulation, proper adjustment was made to ensure a sufficient and similar supply of digestible amino acids in all three diets, which was the reason for the greater total protein concentration of the two high-DF diets compared with the LF diet (Table 2). The amount of carbohydrates was greatest in the LF diet (698 g/kg DM) and similar in the HFS and HFI diets (628–622 g/kg DM). The LF diet had the greatest starch content, whereas it was much less in the two high-DF diets. The two high-DF diets were similar in total NSP and similar in cellulose and non-cellulosic polysaccharides but with variable proportions of water-soluble-to-insoluble non-cellulosic polysaccharides (30:70 in the HFS diet and 20:80 in the HFI diet). The different chemical compositions translate into differences in the physico-chemical properties of the diets. Swelling and water-binding capacity were greatest in the HFS diet and least in the LF diet. The daily intake of metabolisable energy was 24·9 MJ for the LF diet, 24·0 MJ for the HFS diet and 20·7 MJ for the HFI diet (data obtained from Serena et al. (Reference Serena, Jorgensen and Bach Knudsen26)). The intake of metabolisable energy was significantly lower for the HFI diet than for the two other diets.
Table 1 Dietary ingredients (g/kg diet) of the experimental diets
LF, low-fibre diet; HFS, high-fibre soluble diet; HFI, high-fibre insoluble diet.
* Provided per kg of final diet: 3·03 mg retinol acetate; 25 μg cholecalciferol; 60 mg all rac dl-α-tocopherol acetate; 2·2 mg menadione; 2·2 mg thiamin; 5·5 mg riboflavin; 3·3 mg pyridoxine; 16·5 mg d-pantothenic acid; 22 mg niacin; 1·65 mg folic acid; 220 μg biotin; 22 μg cyanocobalamin; 60 mg butylated hydroxytoluene 100 mg Fe as FeSO4·7H2O; 150 mg Zn as ZnO; 28 mg Mn as MnO; 20 mg Cu as CuSO4·5H2O; 304 μg I as KI; 300 μg Se as Na2SeO3.
Table 2 Chemical composition and physico-chemical properties of the experimental diets
LF, low-fibre diet; HFS, high-fibre soluble diet; HFI, high-fibre insoluble diet.
Animals
Experiments complied with the guidelines of the Danish Animal Experiments Inspectorate, Ministry of Justice, Copenhagen, Denmark, with respect to animal experimentation and care of the animals under study. The experiment was carried out as a 3 × 3 repeated cross-over design with six sows fed the three different diets: LF, HFS and HFI, 7 d/diet. Non-pregnant sows with an initial average body weight of 202 (sd 28) kg were selected after weaning of their first litter (Peter Bøjlesen, Vammen, Denmark). After 10 d of adaptation, each sow was surgically fitted with two catheters, one in the portal vein (1·2 mm in inner diameter and 2·3 mm in outer diameter; Cole-Parmer, Vernon Hills, IL, USA) and the other in the mesenteric artery (1·0 mm in inner diameter and 1·8 mm in outer diameter; Cole-Parmer), and with an ultrasonic blood flow probe (28A probe, 28 mm; Transonic System, Inc., Ithaca, NY, USA) around the portal vein. A flowmeter (Transonic T206 flowmeter with P-option; Transonic System, Inc.) was used for measuring the blood flow rate. After a 7 d recovery period from the surgery, the sows were introduced to one of the three experimental diets. Sows were placed in farrowing pens during the collection of blood from the portal vein and the mesenteric artery. Blood samples were collected the last day in each treatment period at 0, 1, 4 and 10 h after feeding. The blood was collected into heparinised plastic tubes and centrifuged (1800 g in 10 min at 8 °C). Plasma was frozen until further analysis. On the days of blood sampling, any feed leftovers were collected. For further details on the experimental protocol, see Serena et al. (Reference Serena, Jorgensen and Bach Knudsen24).
1H NMR spectroscopic analysis
All NMR spectra were recorded on a Bruker 600 MHz NMR spectrometer (Bruker Biospins, Rheinstetten, Germany) operating at a frequency of 600·13 MHz for 1H and equipped with a 5 mm TXI (triple resonance inverse) probe. Then, 200 μl aliquots of the plasma samples were mixed with a solution of 400 μl of 0·9 % saline (NaCl) and 20 % 2H-labelled water. A Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence(Reference Meiboom and Gill27) with water suppression was applied for the acquisition of 1H NMR spectra, with the Carr–Purcell–Meiboom–Gill delay added to attenuate signals from macromolecules. The total spin–spin relaxation delay was 100 ms (2nτ), the spin-echo delay was 1 ms and the recycle delay was 2 s. The spectra were acquired by sixty-four scans, 32 000 data points, a spectral width of 17·34 parts per million (ppm) and a temperature of 310 °K. A fixed receiver–gain value was used for recording all samples. An exponential line broadening of 0·3 Hz was applied before Fourier transformation. Each spectrum was manually phased, baseline corrected and referenced to the lactate doublet signal at 1·33 ppm (the reference was checked by the α-glucose anomeric doublet at 5·23 ppm). The ΔAV spectra were the difference between the 1H spectra of simultaneously collected plasma samples from the portal vein and the mesenteric artery. NMR spectroscopy is a quantitative technique and the ΔAV spectra reflected this by showing a mainly positive absorption profile. These positive signals were the result of nutrient uptake from the gastrointestinal tract. The region 0·5–9·0 ppm of the NMR spectra and the ΔAV difference was segmented into bins of 0·013 ppm and the integral of each bin was determined. Semi-quantitative absorption values for the bins were calculated by multiplying portal flow measurements to the ΔAV according to absorption determination described by Rerat et al. (Reference Rerat, Vaissade and Vaugelade11). To exclude the water signal, the region 4·4–5·0 ppm was not included in the statistical analysis. The betaine signal at 3·26 ppm, the β-glucose signal at 3·24 ppm and the creatine signal at 3·93 ppm were manually integrated.
To aid spectral assignment, two-dimensional (2D) 1H–1H correlation spectroscopy with double-quantum filter and 2D 13C–1H heteronuclear single quantum coherence experiments were performed on a representative venous plasma sample using water suppression. The correlation spectroscopy spectra were acquired with a spectral width of 6127 Hz in both dimensions, 4000 data points, 512 increments with sixty-four transients per increment and a recycle delay of 1·5 s. The heteronuclear single quantum coherence spectra were acquired with a spectral width of 6250 Hz in the 1H dimension and 21 128 Hz in the 13C dimension, a data matrix with a size of 2048 × 512 data points, 256 transients per increment and a recycle delay of 2 s. The 13C chemical shift was referenced to the α-glucose anomeric carbon at 94·9 ppm.
Amino acid analysis
Amino acid analysis was carried out using an amino acid analysing kit (Phenomenex®EZ:faastTM kit; phenomenex, Torrance, CA, USA) and analysed by GC. This analysis was carried out on blood samples collected at − 1, 0, 0·5, 1, 2, 3, 4, 6, 8 and 10 h after feeding. The statistical analysis was the same as for the analytical methods described in Serena et al. (Reference Serena, Jorgensen and Bach Knudsen24).
Multivariate data analysis
In the analysis of the NMR data from the plasma samples and the ΔAV, we used the paired structure in the cross-over design to increase the statistical power. This also allows determining the effect of diet from the within-subject variation by multilevel partial least-squares discriminant analysis (MLPLSDA)(Reference van Velzen, Westerhuis and van Duynhoven28, Reference van Velzen, Westerhuis and van Duynhoven29). This approach takes the advantage of the cross-over design study to separate the between-subject from the within-subject variation. Before the multivariate data analysis, the data were Pareto-scaled by dividing each variable by the square root of its standard deviation. For each MLPLSDA model, we used twenty cross-model validations (CMV), which is a double cross-validation where 25 % of the multilevel pairs are randomly selected and used as a test set for the final model for each of the CMV(Reference Westerhuis, Hoefsloot and Smit30). Variable selection was applied during CMV for model optimisation corresponding with 100, 50, 20, 5 and 2 % of all variables(Reference Anderssen, Dyrstad and Westad31). To test the significance of the models, permutation testing was performed. In a permutation test, the samples are randomly assigned a label while keeping the number of the two classes the same(Reference van Velzen, Westerhuis and van Duynhoven28, Reference van Velzen, Westerhuis and van Duynhoven29, Reference Smit, van Breemen and Hoefsloot32). For each MLPLSDA model, a thousand permutations were performed to determine the H 0 distribution of no-effect. The effect was considered significant if the P value was < 0·05. The rank product (RP) was used to select the most discriminating variables(Reference Breitling, Armengaud and Amtmann33). The variables were ranked according to the highest absolute regression coefficient for all CMV and multiplied for each of the twenty CMV. To select biomarkers that are favoured more than just by chance, only bins with an RP value below the 20 % significance limit of the RP values from the 1000 permutations were considered to be important bins and therefore discriminating between the two classes.
The algorithms for data pretreatment, MLPLSDA, CMV and permutation tests have been used previously in a study by van Velzen et al. (Reference van Velzen, Westerhuis and van Duynhoven28), and they were performed using Matlab (version 2009a; The MathWorks, Inc., Natick, MA, USA) and in-house written Matlab routines. These routines are available via the Internet at http://www.bdagroup.nl/.
Univariate data analysis
ANOVA was performed on the important bins and their absorption values selected as biomarkers in the MLPLSDA models. ANOVA is a univariate method that does not take covariance between the bins into account. However, a linear mixed model with PROC MIXED in the SAS statistical software package version 9.1 (SAS Institute, Cary, NC, USA) was carried out to obtain a model with all diet and time effects and the interaction between diet and time in one model. First, model 1 including the blood compartment (arterial or venous) as a fixed effect was performed on all samples for each of the selected important bins, which tested whether there was a significant difference between the arterial and venous samples (i.e. absorption). Second, model 2 was used to analyse the portal levels of all selected important bins, and whether model 1 showed a significant difference between the venous and arterial samples, and then the important bin absorption values were also analysed.
where i = 1,…, 6 for the sows; j is an important bin; k = 1,…,4 for the time points; l = 1, 2, 3 for the diets; m = 1, 2 for the blood compartments. μj is the overall offset for the bin j; αjk is the overall time effect; βjl is the diet effect; γjm is the blood compartment; αβ jkl is a two-factorial interaction between time and diet; ɛijklm is the error term. X ijklm represents the NMR signal for the sow i, at the bin j, for the time point k, for the diet l and for the blood compartment m.
where i = 1,…, 6 for the sows; j is an important bin or its absorption value; k = 1,…,4 for the time points; l = 1, 2, 3 for the diets. μj is the overall offset for the bin j; αjk is the time effect; βjl is the diet effect; αβ jkl is a two-factorial interaction between time and diet; ɛijkl is the error term. X ijkl represents the NMR signal or the absorption value for the sow i, at the bin j, for the time point k and for the diet l.
Random effects for both models were sow, interaction of diet and week, and the interaction of sow and time. The longitudinal data were modulated using a heterogeneous autoregressive covariance(Reference Martens and Dardenne34) structure when the first model showed a significant difference between the arterial and venous samples (blood compartment). With no difference, indicating no absorption, a uniform covariance structure (compound symmetry(Reference Littell, Milliken and Storup35)) was used for the plasma levels because we assume that the correlations between the sampling points are the same. The sampling point 0 h after feeding was presumed to be a 24 h after feeding sample in the statistical analysis because of the higher correlation between 10 and 24 h samples than between 0 and 1 h. Least square means were compared by applying the probability of difference procedure of SAS. All P values reported are for two-sided tests, and significance level was set at α < 0·05.
Results
1H NMR spectra of plasma samples collected from the portal vein had in general higher intensities than spectra from simultaneously collected mesenteric artery plasma samples, as illustrated in Fig. 1(A) and (B) for sampling of a HFI-fed pig 1 h after feeding (see also Table 3). This demonstrates that differences in absorption can be seen from the ΔAV spectra with positive intensities for most metabolites (Fig. 1(C)). The negative resonance at 3·54 ppm arises from glycine and is due to shift in resonance frequency between the arterial and venous samples contributing to different bins. Fig. 2 shows the region 3·19–3·30 ppm with assignments of a betaine (singlet), β-glucose (doublet) and signals of choline-containing compounds. A slight shoulder is seen on the betaine signal from protons of three methyl groups. This was assigned to one of the lines from a doublet of doublets of β-glucose with J couplings of 8·7 and 8·0 Hz.
Fig. 1 1H NMR spectra of plasma samples taken from the (A) portal vein and the (B) mesentery artery of a representative sow 1 h after feeding on a high-fibre insoluble diet. The spectral difference between (B) mesenteric artery and (A) portal vein plasma samples was determined by (C) subtraction. Assignments are shown in spectra A as follows (multiplicity in parenthesis): 1, formate (singlet, s); 2, α-glucose (doublet, d); 3, lactate (quartet, q); 4, creatinine (s); 5, creatine (s); 6, betaine (s); 7, β-glucose; 8, overlapping peaks for creatine (s) and creatinine (s); 9, unassigned peak (s); 10, propionate (q); 11, acetate (s); 12, alanine (d); 13, lactate (d); 14, CH2 in lipids (multiplet, m); 15, propionate (triplet); 16, isoleucine, leucine and valine region; 17, CH3 in lipids (m). ppm, Parts per million.
Table 3 Resonance assignments of the compounds listed in Fig. 1 and as detailed in the text
ppm, Parts per million; s, singlet; d, doublet; q, quartet; dd, double doublet; m, multiplet; t, triplet.
Fig. 2 Expanded region in the 1H NMR spectra of Fig. 1(A) from a representative plasma sample showing signals from betaine, glucose and choline. s, singlet; dd, double doublet; ppm, parts per million.
MLPLSDA was performed on Pareto-scaled within-subject data to elucidate the biochemical dissimilarities of feeding the different experimental diets. In Fig. 3, the mean CMV prediction error based on twenty CMV is compared with 1000 permutations under the H 0 distribution for all venous samples and the ΔAV when the sows are fed the HFS or HFI diet at all sampling points. Only one venous sample was missing; thus, there were twenty-three multilevel pairs out of twenty-four multilevel pairs. A CMV prediction error of 19·4 misclassifications and an average of the permutation of 22·9 misclassifications were found for the MLPLSDA model in Fig. 3(A) of the venous samples, and the classification result was significant (P = 0·018). The ordinary PLSDA model did not result in a significant diet effect between the two diets (Fig. 3(C); P = 0·057). To calculate a MLPLSDA model of the ΔAV data, thirty-eight out of forty-eight samples were used because of one missing venous sample and four missing arterial samples and the need for a corresponding sample to determine the ΔAV difference. Data analysis of the ΔAV did also show a significant diet effect with the MLPLSDA model (P = 0·025), whereas the ordinary PLSDA model did not indicate a significant diet effect (P = 0·19). When using MLPLSDA to compare the diet effect between the LF and the high-fibre diets, all models for venous samples and the ΔAV were highly significant (P < 0·001; data not shown).
Fig. 3 Number of misclassifications of the mean cross-model validation and the H 0 distribution of no-effect determined by permutation testing with 1000 permutations. (A) Multilevel partial least-squares discriminant analysis (PLSDA) model for all venous samples and the (B) arterial–venous difference (ΔAV) when feeding the high-fibre soluble (HFS) and high-fibre insoluble (HFI) diets. (C) Ordinary PLSDA model for all venous samples and the (D) ΔAV when feeding the HFS and HFI diets.
The rank product was used to identify valid biomarkers in the plasma between sows fed the experimental diets. The variables with the lowest rank are the most discriminating between the two classes. To eliminate false-positive biomarkers, only the RP values below the 20 % significance limit in comparison with the RP values obtained by the permutations were selected from Figs. 4 and 5. The most discriminating rank products in Fig. 4(A) show that formate, glucose, lactate, creatine, betaine, creatine/creatinine (3·04 ppm), pyruvate, propionate, acetate, alanine, the isoleucine/leucine/valine region and lipid signal at 0·86 ppm differentiate between feeding the LF and the HFS diets in the venous plasma. The metabolites in venous samples found to vary between HFI- and LF-fed sows were lactate, creatine, betaine, creatine/creatinine (3·04 ppm), acetate, lipid signal at 0·82 and 1·21 ppm, and the isoleucine/leucine/valine region (Fig. 4(B)). As shown in Fig. 4(C), pyruvate, propionate, acetate, lipid signal at 0·87 and 1·27 ppm, and the isoleucine/leucine/valine region discriminate between the two high-fibre diets, HFS and HFI, in the venous samples. Fig. 5 illustrates the variation in ΔAV between feeding the different diets. The rank product for variables ascribed to propionate, acetate and the isoleucine/leucine/valine region was found to discriminate between the ΔAV of the HFS- and the LF-fed sows (Fig. 5(A)). Fig. 5(B) indicates that glucose, acetate, alanine and lactate are biomarkers for the classification between the ΔAV when feeding the HFI and LF diets. Metabolites differentiating in ΔAV between the two high-fibre diets, HFS and HFI, are creatine, creatine/creatinine (3·04 ppm), propionate and acetate (Fig. 5(C)). The bin at 3·04 ppm contains the signal from two partly overlapping singlets of creatine and creatinine. The MPLSDA of Figs. 3 and 4 did indicate variation for the creatine singlet at 3·93 ppm and no variation for the creatinine singlet at 4·06 ppm. Therefore, variation of the bin at 3·04 ppm can be ascribed to creatine. The region between 3·3 and 3·9 ppm also contains important variables in the MLPLSDA models that correspond to overlapping signals from protons in glucose and amino acids (mainly α-protons).
Fig. 4 Rank product (RP1/20) from multilevel partial least-squares discriminant analysis models of venous samples from comparison (A) between the high-fibre soluble and low-fibre (LF) diets, (B) between the high-fibre insoluble and LF diets and (C) between the two high-fibre diets. The nomenclature is the same as in Fig. 1: 1, formate; 2, α-glucose; 3, lactate; 5, creatine; 6, betaine; 8, creatine/creatinine; 9, unassigned peak; 10, propionate; 11, acetate; 12, alanine; 13, lactate; 14, CH2 in lipids; 15, propionate; 16, isoleucine, leucine and valine region; 17, CH3 in lipids. Only bins with an RP value below the 20 % significance limit of the RP values from the 1000 permutations have been numbered. ppm, Parts per million.
Fig. 5 Rank product (RP1/20) from multilevel partial least-squares discriminant analysis models of the arterial–venous difference (ΔAV) from comparison (A) between the high-fibre soluble and low-fibre (LF) diets, (B) between the high-fibre insoluble and LF diets and (C) between the two high-fibre diets. The nomenclature is the same as in Fig. 1: 2, α-glucose; 3, lactate; 5, creatine; 7, β-glucose; 8, creatine/creatinine; 10, propionate; 11, acetate; 12, alanine; 13, lactate; 15, propionate; 16, isoleucine, leucine and valine region. Only bins with an RP-value below the 20 % significance limit of the RP values from the 1000 permutations have been numbered. ppm, Parts per million.
MPLSDA revealed numerous biomarkers of the biochemical differences in venous samples and the ΔAV between the diets. These biomarkers were tested in an ANOVA model for the venous plasma levels and the absorption value as the ΔAV multiplied by the blood flow. Isoleucine, leucine and valine were not tested in the ANOVA model because of an overlap with the lipid signal and some chemical shifting of these peaks in the NMR spectra that can cause variation. In Table 4, the significance of these bins and their contrasts are shown, together with the ratio of the intensity:absorption value divided by the lowest intensity:absorption value. The ratio reflects the quantitative relationship between protons from a specific metabolite in a bin, i.e. for venous samples the least square means of acetate show that LF-fed sows have the lowest acetate concentrations and that HFS- and HFI-fed sows have 129 and 79 % higher concentrations, respectively (P < 0·0001). The least square means of acetate venous plasma concentrations were significantly different for all three experimental diets. If no significant difference between the arterial and venous samples exists, only the data for the venous plasma are included in Table 4, as this suggests that no uptake of nutrients is likely to occur. A significant diet effect on venous plasma samples between the two high-fibre- and the low-fibre-fed sows was found for betaine (P = 0·008), creatine (P = 0·020) and glucose (P = 0·010). A significant difference between 0 h and the rest of the sampling points was found for both betaine (P = 0·0014) and creatine (P = 0·048). The betaine time profile showed a significant higher absorption in the absorptive phase (P = 0·0005). Creatine did not have a significant difference between the venous and arterial samples, indicating no absorption of creatine. Glucose absorption was highest in the absorptive phase (1 and 4 h; P < 0·0001) and when the sows were fed the LF diet (P = 0·010). The SCFA acetate (P < 0·0001) and propionate (P < 0·017) were in the order HFS>HFI>LF in the venous samples, and the numerical values of the absorption values did reflect this order, but only the absorption values for HFS-fed sows were significantly different from the other experimental diets (P = 0·027 for acetate; P = 0·0011 for propionate). The SCFA were not affected by time. Significant diet effects on venous plasma levels were found for alanine, with a high concentration when feeding the LF diet, intermediate when feeding the HFI diet and low when feeding the HFS diet (P = 0·046). Venous alanine concentrations were high in the absorptive phase (1 and 4 h) and low in the post-absorptive phase (10 h) and before feeding (0 h; P = 0·0079). Absorption was not significantly affected by diet (P = 0·098); however, the time profile was similar to concentration in the venous samples (P = 0·0021). There was no significant diet effect for lactate (P = 0·26). However, lactate did vary between the sampling points with a high venous plasma concentration at 1 h and a low concentration at 0 and 10 h (P = 0·0083). Absorption of lactate was significantly different between 0 h and the rest of the sampling points (P = 0·014). There was no indication of the absorption of formate. Formate concentrations in venous samples were high at 4 h, intermediate at 10 h and low at 0 and 1 h (P < 0·0001), and no significant diet effect was found for formate venous concentrations (P = 0·16). Also, no significant effects were found for pyruvate.
Table 4 ANOVA of important bins and absorption values*
ppm, Parts per minute.
a,b,c Least-square mean values within a row with unlike superscript letters were significantly different for the diet or time effects (P < 0·05).
* The effects are determined as the intensity:absorption value divided by the lowest intensity:absorption value.
The amino acid analysis revealed a significant time effect on alanine (P < 0·0001) and glycine (P = 0·025) but no significant diet or diet × time interaction effects. The time profiles for alanine, glycine and methionine plasma venous concentrations and methionine absorption, as determined by GC, can be found in Table 5. For venous plasma concentrations, P values of methionine were P = 0·0028 for diet and P < 0·0001 for time and for the absorption of methionine, P = 0·0012 for time and the diet effect was non-significant. There was too much overlap from an unsaturated lipid signal to determine the variation of the methionine singlet at 2·14 ppm by NMR spectroscopy. Due to overlapping in the glucose region, it was also not possible to determine the variation of glycine at 3·54 ppm.
Table 5 Alanine and glycine portal vein concentrations, and methionine portal vein concentration and absorption*
a,b,c Least-square mean values within a row with unlike superscript letters were significantly different for the diet effect (P < 0·05). Only least square means are shown for the sampling points included in the NMR analysis.
* The effects are determined as the intensity:absorption value divided by the lowest intensity:absorption value
The betaine peak at 3·26 ppm and the creatine peak at 3·93 ppm were manually integrated and the betaine integral was subtracted from the integral of the upfield β-glucose doublet of doublets line at 3·24 ppm. Fig. 6 shows the results of the venous betaine and creatine levels for all sows and each of the three diets. The venous betaine levels seem to indicate a slow increase of betaine in the portal vein and higher levels when feeding the LF diet. The venous creatine integrals clearly show a general trend for higher creatine concentration when feeding the LF diet without any sign of a time effect.
Fig. 6 Plasma (A–C) betaine and (D–F) creatine venous levels for sows fed a (A, D) low-fibre diet, (B, E) high-fibre soluble diet and (C, F) high-fibre insoluble diet. The six sows are marked with different symbols (○, ●, Δ, ▲, □ and ■).
Discussion
In the present study, the use of NMR spectroscopy for studying absorption dynamics of dietary nutrients was investigated. Previously, the concentrations and absorption of acetate, glucose, lactate and propionate have been measured by traditional analytical methods from the same study(Reference Serena, Jorgensen and Bach Knudsen24). These data were carried out on seventeen sampling points, whereas the NMR method was only applied to four sampling points. The 1H NMR data showed the same major trends in the time profiles as the traditional analytical methods. Accordingly, determination of the absorption from the ΔAV by NMR spectroscopy and the portal blood flow proved to be an effective way of determining the uptake of nutrients from the gastrointestinal tract. The biochemical changes of these metabolites will not be discussed here but we refer to the studies using the traditional analytical methods(Reference Serena, Jorgensen and Bach Knudsen23, Reference Serena, Jorgensen and Bach Knudsen24). Other metabolites that showed a large variation in plasma concentration and/or absorption and were well resolved in the NMR spectra included formate and alanine. Good agreement was found between alanine levels determined by the NMR-based method and by GC, respectively. The plasma concentration and absorption profiles of alanine showed uptake that is characteristic for the small intestine (Fig. 5). Formate is a by-product of carbon metabolism but it can also be a product of fermentation of DF by the gut microbes. There was no observed absorption, suggesting that the time effect of formate was a result of carbon metabolism.
Betaine is known to be an effective osmolyte. Another function of betaine is as a methyl donor in the methylation of homocysteine in the methionine cycle, as shown in Fig. 7(Reference Finkelstein and Martin36). It is believed that betaine decreases the concentrations of homocysteine in blood, which is a risk factor for CVD(Reference Olthof and Verhoef37). Thus, betaine can be considered as an important metabolite. The absorption profile of betaine followed basically that of glucose, with a rapid increase in absorption in the absorptive phase at 1 h after feeding, followed by a steady decline to 4 and 10 h after feeding for all three diets. There was a diet effect on concentration in plasma with the highest levels when feeding the LF diet (28 % higher concentration in the portal vein when feeding the LF diet than the HFS diet; the HFS and HFI diets had similar portal vein concentrations). This LF diet had a high content of wheat, which is known to have a relatively high amount of betaine(Reference Sakamoto, Nishimura and Ono38). LF-fed sows had a higher numerical absorption of betaine, but it was not significant. However, based on the results from the venous samples, a fair assumption is that betaine plasma levels accumulate from the start of the period the sows are fed the LF diet due to increased uptake of betaine.
Fig. 7 Pathway for creatine synthesis and related transmethylation. BHMT, betaine-homocysteine methyltransferase; SAM, S-adenosylmethionine; SAH, S-adenosylhomocysteine; GAMT, guanidinoacetate methyltransferase.
Creatine is used in vertebrates for energy supply to muscle as stored creatine phosphate. There was no significant ΔAV in creatine between the diets in the ANOVA model, indicating no uptake into the portal vein. A plant-based diet would not contain creatine and the creatine levels in the blood of the sows are synthesised de novo in the liver mainly by the use of arginine, glycine and methionine(Reference Wyss and Kaddurah-Daouk39). An interesting question is whether the elevated levels of creatine seen in the plasma of LF-fed sows increase the creatine phosphate levels in creatine kinase-containing tissues as well, and thereby promoting an increased tissue pool of creatine and creatinine maybe caused by the higher dietary level of starch in the LF diet compared with the two high-DF diets. This could have negative effects when feeding the high-fibre diets. Because the high-fibre soluble and insoluble diets have the same time profile of creatine levels, it would be reasonable to leave out the physico-chemical properties of the diet as the main reason.
Methionine was not well resolved in the NMR spectra (at 2·14 ppm). However, the GC amino acid analysis pointed at significant elevated levels of methionine in the venous samples of the LF diet compared with the high-DF diets. Absorption was not significantly different between the diets. Hence, we found an endogenous methionine variation as an effect of diet.
Numerous metabolites are involved in creatine synthesis, but it is possible to determine the pathways because creatine plasma concentration is of endogenous origin. Increased absorption of betaine can lead to the elevated levels of creatine in plasma, because of the requirement of a methyl donor from S-adenosylhomocysteine to synthesis creatine by the methylation of homocysteine (Fig. 7). It has been found in a study on creatine metabolism in piglets that there are relatively high activities of guanidinoacetate methyltransferase in the liver(Reference Brosnan, Wijekoon and Warford-Woolgar40). Thus, it has been hypothesised that high betaine absorption when feeding the LF diet causes elevated creatine plasma concentrations. In agreement with the hypothesis, NMR analysis on urine samples from a separate study(Reference Serena, Jorgensen and Bach Knudsen26) using the same three diets indicated higher excretion of betaine and creatinine in the urine when feeding the LF diet compared with the high-fibre diets (CC Yde and KE Bach Knudsen, unpublished results). The high methionine plasma concentrations when feeding the LF diet reflected its contribution to methyl donation in the synthesis of creatine. Glycine is a precursor of creatine synthesis, but it showed no diet effect that can explain the diet effect on creatine.
In a study by Hoffman et al. (Reference Hoffman, Ratamess and Kang41), betaine supplementation to human subjects appeared to improve muscle endurance in a squat exercise, which was suggested to be related to increased muscle creatine concentrations. The relationship between the high content of betaine in the LF diet and the elevated plasma creatine concentration may indicate a performance relationship between the formulation of the diet and the endurance. Further studies are needed for determining a clear link. It has been found that a whole-grain diet compared with a non-whole-grain diet increases the dietary intake of betaine(Reference Bertram, Bach Knudsen and Serena17). This leads to increased excretion of creatinine without influencing the portal vein concentration of creatine. The present findings suggest that feeding a diet high in betaine may cause interesting metabolic effects. However, further investigations are needed to determine the importance of feeding different types and levels of DF with regard to creatine synthesis. There are also other factors to take into account. Different protein concentrations (low for LF) and daily energy intake (significant lower for HFI) could also influence the outcome. Consequently, a more controlled set-up is needed to directly access the importance of these factors.
The betaine signal at 3·26 ppm overlapped with the downfield line of the doublet of doublets from β-glucose at 3·25 ppm. The glucose levels were highest at 1 h after feeding, which influence the results by increasing the size of the bin assigned to betaine at 3·25 ppm. Consequently, the binning procedure overestimated the betaine levels at mainly 1 h after feeding. Determination of the venous betaine levels from the integral of the betaine and glucose signal indicated slower betaine absorption than found by the bins. Unfortunately, the other betaine signal from the CH2 protons at 3·89 ppm was heavily overlapped from the glucose signal. A special issue concerning the ΔAV was the low signal:noise ratio of the ΔAV spectra, which may reduce reliability. Improvement of the method could be achieved by removing the lipoproteins, as the interactions between metabolites and lipoprotein cause line broadening, and overlap from lipoprotein signals strongly reduces the resolution. An approach to deal with these difficulties is diffusion NMR(Reference Liu, Nicholson and London42). Diffusion NMR is a technique that enables distinguishing between the signals from low-molecular-weight compounds from large-molecular-weight compounds. Another method is the physical removal of the lipoprotein by extraction or ultrafiltration(Reference Tiziani, Einwas and Lodi43).
Conclusion
The main purpose of the present study was to determine the absorption of specific metabolites without bias towards a specific metabolite, which proved to complement the findings from traditional analytical methods. Furthermore, the absorption profiles determined from 1H NMR-based metabonomics revealed new insight into the effects of feeding different types and levels of DF. A low-fibre diet resulted in elevated concentrations of creatine, which was shown to be of endogenous origin. In addition, the elevated concentration of creatine was related to the uptake of betaine, which is known to have health-promoting homocysteine-lowering effects. Thus, by using this explorative approach, we could determine the biochemical changes from unexpected findings of variation in hundreds of metabolites by a multivariate technique.
Acknowledgements
The authors are grateful to the work carried out by Anja Serena. This includes animal handling, sampling and the traditional analytical measurements. The authors would like to thank the technical staff for assistance. The Danish Technology and Production Research Council (FTP) is thanked for financial support through the project ‘Advances in food quality and nutrition research through implementation of metabolomic technologies’ and the Danish Agency for Science, Technology and Innovation for co-funding the PhD project for C. C. Y. The authors' contributions are as follows: C. C. Y. performed the NMR measurements, data analysis and wrote the manuscript. J. A. W. contributed to the statistical work. H. C. B. contributed to the NMR measurements. K. E. B. K. was responsible for the project development. All authors contributed to the reviewing and editing of the manuscript. There are no conflicts of interest.
