Proteins are an essential component of the human diet. When expressed per kg body weight, the protein intake in young children is very high, typically two to three times higher than in adults and also two to three times higher than the physiological requirement (Rolland-Cachera et al. Reference Rolland-Cachera, Deheeger and Bellisle1999). It is unknown whether this has positive or negative long-term effects. It has been outlined that a diet rich in animal protein lacks alkalizing effects and therefore may have a negative effect on Ca balance and bone mineral density (Demigné et al. Reference Demigné, Sabboh, Puel, Rémésy and Coxam2004) in some studies, but dietary protein is also positively associated with bone mineral density in children (Hoppe et al. Reference Hoppe, Mølgaard and Michaelsen2000) and adults (Bonjour et al. Reference Bonjour, Schurch and Rizzoli1997; Teegarden et al. Reference Teegarden, Lyle and McCabe1998). However, in general the metabolic effects of a high-protein diet are poorly understood, especially in children. In addition, the knowledge of the significance of protein source is also very limited. It was demonstrated that a milk protein diet increases serum insulin-like growth factors (Hoppe et al. Reference Hoppe, Mølgaard, Juul and Michaelsen2004) as well as insulin and insulin resistance in prepubertal boys (Hoppe et al. Reference Hoppe, Mølgaard, Vaag, Barkholt and Michaelsen2005). However, it has not been studied any further to what extent a high-protein diet and protein source may also have other biochemical effects.
Proton NMR provides concurrent detection of all hydrogen-containing molecules in a sample without pretreatment, and it is widely recognized that NMR-based metabonomics is an excellent tool in pharmaceutical studies for detection of biochemical effects without any a priori knowledge (Lindon et al. Reference Lindon, Holmes and Nicholson2006). Within the field of nutrition the use of NMR-based metabonomics is more rare, however, the technique has exciting opportunities for elucidating how diet elicits changes in metabolite profile (Gibney et al. Reference Gibney, Walsh, Brennan, Roche, German and van Ommen2005), and applications of NMR-based metabonomic for elucidating the biochemical effects of dietary bioactive compounds in both animals (Solanky et al. Reference Solanky, Bailey, Holmes, Lindon, Davis, Mulder, Van Duynhoven and Nicholson2003a ; Bertram et al. Reference Bertram, Bach Knudsen, Serena, Malmendal, Nielsen, Fretté and Andersen2006) and man (Solanky et al. Reference Solanky, Bailey, Holmes, Beckwith-Hall, Davis, Bingham, Holmes, Nicholson and Cassidy2003b ; Daykin et al. Reference Daykin, Van Duynhoven, Groenewegen, Dachtler, Van Amelsvoort and Mulder2005; Wang et al. Reference Wang, Tang, Nicholson, Hylands, Sampson and Holmes2005) have been demonstrated. Thus, NMR-based metabonomics could be an effective tool for investigating possible biochemical effects associated with protein source and a high-protein diet, as the technique enables a screening without any a priori knowledge of expected outcome.
The aim of the present study was to investigate the ability of NMR-based metabonomics to identify exogenous and endogenous biochemical effects of a milk protein and a meat protein diet given to prepubertal children in an explorative way. We therefore present a proton NMR-based metabonomics approach applied on urine and serum samples from an intervention study, in which 8-year-old boys were given a high amount of animal protein as either milk or meat for 1 week (Hoppe et al. Reference Hoppe, Mølgaard, Juul and Michaelsen2004).
Materials and methods
Subjects
As previously described (Hoppe et al. Reference Hoppe, Mølgaard, Juul and Michaelsen2004), twenty-four 8-year-old boys were asked to take about 53 g protein daily, twelve boys as 1·5 l skimmed milk, and twelve other boys as 250 g low-fat meat. The milk and meat supplements were provided to the children from the Department of Human Nutrition. In addition, they were asked to eat ad libitum their normal diet.
The subjects were recruited in the following manner: boys born between October and December 1992 were drawn at random from the Central Personal Register, and were invited to participate in the study. Children with a habitual milk intake of 500 ml/d or less, who were willing to increase their intake of milk or meat considerably during a week, were eligible for the study. Children with chronic illnesses, and children who suffered from any condition likely to affect their protein metabolism or growth were excluded from the study. From a total of 313 invited subjects, thirty agreed to participate, and twenty-eight were eligible for the study. Of the twenty-five boys who were the first to agree to participate in the study, fourteen were eligible and twelve completed the intervention with milk, and of the remaining fifteen boys, fourteen were eligible and twelve completed the intervention with meat. All participants were Caucasian. Subjects and their parents received oral and written information about the study, and the parents gave their written consent. The participants (with their parents) kept a 3 d weighed food record prior to the intervention day (day − 3 to day − 1) and during the last 3 d of the intervention (day 4 to day 7). Blood serum samples were drawn from venepuncture after an overnight fast at baseline and after 7 d of intervention. The serum was kept frozen at − 80°C for 6 years until analysis in 2006. Urine was collected for 24 h at baseline and after 7 d of intervention. Urine was collected in containers with 10 ml 1 m-HCl added, and urine samples were subsequently kept frozen at − 20°C for 6 years until analysis in 2006.
NMR measurements
Prior to the measurements the serum samples were thawed and 200 μl aliquots were mixed with 400 μl D2O. Urine samples were thawed, pH of the urine was adjusted to 4 by adding the appropriate amount of concentrated HCl or NaOH, and 400 μl aliquots were mixed with 200 μl D2O. The NMR measurements were performed at 298 K on a Varian Inova 800 spectrometer (Varian, Palo Alto, CA, USA) equipped with a 5 mm 1H observe probe. 1H NMR spectra were obtained using a standard 1D pulse sequence (relaxation delay–90°–t1–90°–tm–90°–acquire free induction decay). Water suppression was applied using the SWET scheme, which suppresses the water magnetization by spatial scrambling (Wu & Otting, Reference Wu and Otting2005). A total of 128 transients were collected into 32 000 data points using a pulse recycle time of 2 s. A spectral width of 13 000 Hz was used. All spectra were referenced to the methyl group of creatine at 3·04 ppm. The spectra were subdivided into 0·005 ppm integral regions and integrated, reducing each spectrum into 1840 independent variables in the region 0·18–4·4 and 5·1–10·0 ppm. The reduced spectra in the region 0·18–4·4 and 5·1–10·0 ppm were normalized to the whole spectrum to remove any concentration effects, and further analysis was performed using the Unscrambler software version 9·2 (Camo, Oslo, Norway). Principal component analysis was applied to the centred data to explore any clustering behaviour of the samples, and partial least square regression discriminant analysis (PLS-DA) was performed to explore intrinsic biochemical dissimilarities between baseline and after 7 d of intervention. During all regressions, Martens uncertainty test (Martens & Dardenne, Reference Martens and Dardenne1998) was used to eliminate noisy variables, and all models were validated using full cross-validation (Martens & Martens, Reference Martens and Martens2000).
To aid spectral assignment 2D 1H–1H total correlation (TOCSY), 2D 1H–1H NOESY and 2D 1H–13C HSCQ spectra were recorded on selected urine samples using water suppression. The TOCSY spectra were acquired with a spectral width of 8000 Hz in both dimensions, 4000 data points, 512 increments with thirty-two transients per increment and 80 ms spinlock. The NOESY spectra were acquired with a similar size and number of data points as the TOCSY and a mixing time of 600 ms. The HSCQ spectra were acquired with a spectral width of 8000 Hz in the F2 dimension and 28 149 Hz in the F1 dimension, a data matrix with a size of 2400 × 512 data points and sixty-four transients per increment.
Results
Urine
Figure 1 shows a typical 1H NMR spectrum acquired on a urine sample. Several resonances are observed, which were assigned combining the information provided by the 2D experiments with literature values (Lehnert & Hunkler, Reference Lehnert and Hunkler1986; Fan, Reference Fan1996; Lindon et al. Reference Lindon, Nicholson and Everett1999). In order to investigate possible metabolic differences between sample classes, a principal component analysis was performed on the NMR spectra. A scatter plot of score 1 versus score 2 for mean-centred data (Fig. 2) shows a clear clustering into samples taken before and after the 7 d of meat and milk diet interventions, respectively.
Fig. 1 800 MHz 1H NMR spectra of a representative baseline urine sample. For details of procedures, see p. 759. HDO, residual water signal; Ileu, isoleucine; Leu, leucine; TMAO, trimethylamine-N-oxide; Val, valine.
Fig. 2 Principal component analysis score plot showing the two first principal components for urine samples taken before intervention (○), and after the meat (●) and milk (*) diet interventions. For details of procedures, see p. 759.
PLS-DA, which is more focused on discriminating between pre-defined classes than principal component analysis, was performed on the 1H NMR spectra to identify the metabolic differences between baseline and after 7 d intervention. The first two components in the PLS-DA could explain 67 and 71 % of the variance in Y, for the milk and meat diets, respectively, and for both milk and meat diets the PLS-DA score plots of the first and second component show a clear separation of baseline samples and samples taken after intervention along the first component (Fig. 3(a, c)). The regions of the NMR spectrum that most strongly influence separation between baseline and after diet intervention are evident in the first X-loading (Fig. 3(b, d)). The discrimination between baseline samples and samples taken after milk intervention can mainly be ascribed to a decrease in the spectral regions at 4·03, 7·56, 7·65 and 7·84 ppm (Fig. 3(b)), which has been identified as hippurate. For the meat diet, the main differences observed before and after intervention is in the spectral regions at 3·13, 3·52 and 4·28 ppm, which decrease in intensity upon intervention, and the spectral regions at 3·04, 3·88, 3·96, 6·74, 7·43 and 8·69 ppm, which increase in intensity upon intervention (Fig. 3(d)). The resonance at 3·52 ppm has been identified as trimethylamine-N-oxide (TMAO), and the baseline sample positioned in the upper left corner of the scoreplot (Fig. 3(c)) is characterized by a high intensity of this signal. The spectral regions at 3·13 and 4·28 ppm, which decrease slightly upon meat intervention, have been identified as creatinine, while the spectral regions at 3·04 and 3·96 ppm, which increase upon meat intervention, have been identified as creatine. The positive value of the resonance at 6·74 ppm reveals an increase in urea in the urine upon the meat intervention, while the positive values of the resonances at 7·43 and 8·69 can be ascribed to an increase in histidine in the urine upon the meat intervention. It has not been possible to identify the increase in the 3·88 ppm signal, which originates from CH or CH3 protons.
Fig. 3 (a), Partial least square regression discriminant analysis (PLS-DA) score plot from analysis of NMR spectra obtained on baseline urine samples (●) and after milk intervention (○). (b), The first X-loading of the PLS-DA shown in (a). (c), PLS-DA score plot from analysis of NMR spectra obtained on baseline urine samples (●) and after meat intervention (○) (↖, outlier described on p. 760, which is characterized by a high intensity of the trimethylamine-N-oxide signal at 3·52 ppm). (d), The first X-loading of the PLS-DA shown in (c). For details of procedures, see p. 759.
Blood serum
A representative 1H NMR spectrum acquired on a serum sample is shown in Fig. 4. In order to investigate possible metabolic differences between sample classes, a principal component analysis was performed on the serum spectra. While a scatter plot of score 1 v. score 2 does not show any evident clustering (data not shown), a scatter plot of score 2 v. score 3 for mean-centred data shows no separation of samples taken before intervention and after meat intervention but a tendency for a separation of samples taken after the milk intervention (Fig. 5). PLS-DA was performed on the 1H NMR spectra acquired on serum to identify the metabolic differences between baseline and after 7 d intervention. For the meat diet, PLS-DA could not discriminate between baseline samples and samples taken after intervention (Fig. 6(a)) and the first two components could only explain 26 % of the variance in Y. For the milk diet the first two components could explain 40 % of the variance in Y, and a better discrimination is obtained (Fig. 6(b)). The first X-loading reveals that the discrimination between baseline samples and samples taken after 7 d of the milk diet can be ascribed to increases in the intensity of spectral regions that can be assigned to lipid protons (Fig. 6(c)).
Fig. 4 800 MHz 1H NMR spectra of a representative baseline blood serum sample. For details of procedures, see p. 759. Cho, choline; HDO, residual water signal; PCho, phosphocholine; TMAO, trimethylamine-N-oxide.
Fig. 5 Principal component analysis score plot showing the two first principal components for serum samples taken before intervention (○), and after the meat (●) and milk (*) diet interventions. For details of procedures, see p. 759.
Fig. 6 (a), Partial least square regression discriminant analysis (PLS-DA) score plot from analysis of NMR spectra obtained on baseline serum samples (●) and after milk intervention (○). (b), PLS-DA score plot from analysis of NMR spectra obtained on baseline urine samples (●) and after meat intervention (○). (c) The first X-loading of the PLS-DA shown in (b). For details of procedures, see p. 759.
Discussion
NMR-based metabonomics enables a screening for perturbations in molecular concentrations in biofluids without any a priori knowledge, and thereby without any hypothesis-induced restrictions. While several NMR-based metabonomic studies have been reported for toxicological or pharmaceutical applications (Lindon et al. Reference Lindon, Holmes and Nicholson2006), applications within the field of nutrition are more rare. Although other dietary changes might contribute to the findings, the present study for the first time investigated the ability of NMR-based metabonomics to identify exogenous and endogenous biochemical effects of a milk protein and a meat protein diet given to prepubertal children by measurements on blood serum and urine samples from an intervention study, in which 8-year-old boys were given animal protein as either milk or meat for 1 week. The aim of the study was not absolute quantification but to compare relative differences, and pronounced effects of the two diets were observed on the urine profile. The milk diet reduced the urinary excretion of hippurate. Hippurate is mainly derived via gut microfloral breakdown of plant phenolics and aromatic amino acids (Phipps et al. Reference Phipps, Stewart, Wright and Wilson1998; Williams et al. Reference Williams, Eyton-Jones, Farnworth, Gallagher and Provan2002; Mulder et al. Reference Mulder, Rietvald and van Amelsvoort2005). Accordingly, the present findings suggest changes in the gut microfloral metabolism upon an intensive milk diet. The origin of this change remains unknown, as the milk diet was not associated with a major change in dietary intake of aromatic amino acids. Nevertheless, studies on rats have shown a complex relationship between diet and urinary hippurate excretion (Phipps et al. Reference Phipps, Stewart, Wright and Wilson1998), and the interaction between available substrates and the metabolic capability of the intestinal microflora to produce hippurate is yet not understood. Further studies elucidating the implications of urinary hippurate excretion would be of great interest.
The intervention with the meat diet increased the urinary excretion of creatine, which is in agreement with the fact that meat is the primary dietary creatine source (Barr & Rideout, Reference Barr and Rideout2004). Lately many studies have been conducted with the aim of evaluating the possibility that creatine could function as an ergogenic aid (Lemon, Reference Lemon2002), and it has been observed that adopting a vegetarian diet reduced the muscle creatine concentration (Lukaszuk et al. Reference Lukaszuk, Robertson, Arch, Morre, Yaw, Kelley, Rubin and Moyna2002). However, there has been little focus on the significance of creatine in children's diet. Urinary content of histidine also increased upon the intervention with the meat diet, which is consistent with the fact that meat is the primary histidine source (Hipkiss, Reference Hipkiss2005). It has been suggested that the dipeptide carnosine, which contains histidine, possesses beneficial effects such as an anti-glycating effect, a regulatory effect on blood glucose level and protection against oxidative damage (Hipkiss, Reference Hipkiss2005).
TMAO also appeared in the loadings discriminating between baseline urine samples and samples taken after meat intervention. However, this can be ascribed to the fact that a single subject contained a high amount of TMAO in his urine baseline sample (subject indicated with an arrow in Fig. 2(c)). TMAO is often seen after a high intake of fish (Lenz et al. Reference Lenz, Bright, Wilson, Huges, Morrison, Lindberg and Lockton2004), and the food record prior to the intervention verified a high fish intake at the time of the baseline sampling which explains the elevated concentration of TMAO in the urine from this boy.
In contrast to urine samples, minor differences were observed in the blood serum metabolite profile, and for the meat diet no effects of the intervention could be spotted. This should probably be ascribed to the fact that blood is narrowly controlled in terms of composition and less susceptible to perturbations due to short interventions and that the samples were taken in the morning after at least 10 h fasting. The spectra on serum were dominated by signals from lipoproteins and lipids, and for the milk diet increases in signal intensities of lipid signals (CH 3, (CH 2) n , CH = CH-CH 2 and CH 2-CO) were observed upon the intervention when compared with baseline. The effect on CH 2-CO suggests a higher number of fatty acids in serum upon the milk diet, which probably could be ascribed to the presence of additional SCFA in which milk is known to be particularly rich (Garnsworthy et al. Reference Garnsworthy, Masson, Lock and Mottram2006). Further studies compiling a complete fatty acid profile of serum in relation to milk consumption could be of great interest for a further understanding of the role of milk intake on serum lipid profile.
It has been claimed that diets low in carbohydrate and rich in protein and lipid are effective in weight loss, and numerous books on the topic have followed, the most well known being Dr. Atkins' New Diet Revolution (Atkins, Reference Atkins1998). According to Atkins' book, weight is lost on the ad libitum low-carbohydrate diet because of increased energy expenditure. However, this has yet not been proven, and mechanisms that could explain weight loss as a result of a diet low in carbohydrate still need to be proven (Astrup et al. Reference Astrup, Larsen and Harper2004). In the present study we did no observations that could attribute directly to understanding a possible weight loss as a result of increasing the protein proportion of the diet.
In conclusion, the present study for the first time demonstrated how information on the overall biochemical effects of consumption of different animal proteins can be identified using proton NMR-based metabonomics. The milk diet-induced reduction in urinary excretion of hippurate demonstrates alterations in gut microflora, and further studies are needed to elucidate the significance of the present finding.
Acknowledgements
The Danish Technology and Production Research Council (FTP) is thanked for financial support through the project ‘NMR-based Metabonomics on Tissues and Biofluids’ (project no. 274-05-339). The 800 MHz spectra were obtained using the Varian Unity Inova spectrometer of the Danish Instrument Center for NMR Spectroscopy of Biological Macromolecules. Professor Rasmus Bro is greatly acknowledged for his guidance on the multivariate data analysis.
