**3. Results**

The PCA score plot in Figure 3 presents the overall NMR spectroscopic relations between all the available urine samples, with lines between consecutive samples from the same infant. Urine samples from the first week of life occupy the lower right quadrant of the plot, and mostly progress towards the middle left with increasing age of the infant. There was no obvious difference in distribution or temporal development between the intervention and control group.

Several samples in the upper right quadrant deviated from this general trend and are marked as outliers. The deviation was characterized by strong NMR signals, predominantly in the aromatic region of the NMR spectrum, which could not be identified as known metabolites. Whereas the 10 infants with outlier samples had a somewhat lower gestational age than the others, there were no significant differences with respect to the nutritional intervention, SGA status, sex or infections or any other of the clinical parameters (data not shown).

The PCA loadings (not shown) revealed that the first principal component (PC1, the *x*-axis) of Figure 3 corresponded to increasing levels of citrate, betaine, glycine and hydroxyproline from right to left, along with decreasing unidentified spectral signals at 0.57 and 5.50 ppm. The second (PC2, the *y*-axis) and the third principal component (not shown) were dominated by the unidentified signals of the outlier samples mentioned above.

**Figure 3.** PCA score plot of NMR spectra of all available urine samples, presented as points marked with infant age in weeks and color-coded as earlier. PCA: Principal component analysis, NMR: Nuclear magnetic resonance, PC: Principal component with percent of explained total variation. Lines connect consecutive samples from one infant; line color red for intervention, gray for control group. Outlier samples marked with a dashed line in the upper right quadrant. Inset: Cumulative explained variation (black) and cross-validation (red) of the first five PCs.

The metabolites were studied in relation to changes over time in a linear mixed model for repeated measures (Table 3). There was no significant interaction between intervention group and time for any of the metabolites, thus interaction terms were not included in the final models. There was no significant effect of the intervention, but for most of the metabolites, there was a significant effect of time. The levels of amino acids and many other metabolites increased between weeks 1 and 3, whereas gluconate and two strong, unidentified signals at positions 0.57 and 5.50 ppm disappeared. The results were similar irrespective of whether the outliers were kept or removed in the analyses (data not shown).


**Table 3.** Mixed model and change of mean metabolite levels between sampling weeks 1, 3, 5 and 7 (*n* = 48).

Linear mixed model for intervention (diet) and week of life (time) with adjustment for gestational age (GA) at birth and small for gestational age (SGA) status was used. Statistical significance was assumed for *p* < 0.002. Increase or decrease of log-transformed pseudo-concentrations, presented as fold-change (FC; ratios below 1 are presented as í1/ratio). The FC is based on available paired urine samples from the same child at the respective weeks of age. Metabolites marked "x" disappeared entirely; FC is therefore not applicable. Histidine is an uncertain assignment, based on a narrow doublet at 7.9–8.0 ppm. Total integral of the urine spectra was determined early relative to the added internal standard trimethylsilylpropionate-d4 (TSP) and then used to normalize all specified compounds.

Multivariate PLS regression analyses were carried out between the NMR spectra and clinical variables (Table 4). The nutritional intervention, presence of infections, and infants' sex did not influence the urine spectra, whereas SGA status did show an effect on the metabolite profiles: The PLS model of the infants' SGA status based on all spectra except those of the outliers reached a Q2 /R<sup>2</sup> ratio of 0.40. This increased to 0.53 by focusing on spectra from the first week of life, whereas spectra of the urine samples at 36 weeks PMA could not be linked to SGA status. Inclusion of the outliers in the analysis did not change these results (data not shown). PLS regression analysis also indicated that PMA as well as chronological age were associated with the urine spectra.


**Table 4.** Partial least squares regression of variables to sample spectra.

R2 : Endpoint variation contained in regression model. Q<sup>2</sup> : Variation reproduced in cross-validation. Higher numbers, or at least high Q<sup>2</sup> /R2 ratios mean reliable models. <sup>a</sup> Samples: All except outliers, first urine sampled from subject, and sample from 36 weeks PMA; <sup>b</sup> Number of PLS components resulting in best (highest) Q2 /R2 ratio; results with Q2 /R2 ratio below 0.3 not shown.

Observations linking the first-week urine samples to SGA status as well as PMA were also studied by linear regression analyses for selected metabolites (Table 5). In simple linear regression analyses, SGA status was associated with increased levels of glycine, histidine and threonine (8 × 10<sup>í</sup><sup>4</sup> *p* 0.003), as well as creatinine, succinate and *trans*-4-hydroxy-L-proline (hydroxyproline) (0.016 *p*  30\$ ZDV DVVRFLDWHG ZLWK D EURDGHU UDQJH RI PHWDEROLWHV (3 × 10í *p* IRUDOOYDULDEOHVLQ7DEOH7KH6\*\$LQIDQWVKDGDKLJKHUPHDQJHVWDWLRQDO age at birth than AGA infants (29.9 *vs.* 27.5 weeks, *p* = 0.003). When adjusting for PMA in the multiple linear regression analyses, there is an indication that SGA was associated with glycine (*p* = 0.027) and threonine (*p* = 0.033), although not significant at the adjusted significance level.


**Table 5.** Linear regression results for selected metabolites at week 1 with respect to SGA status and PMA.

Histidine is an uncertain assignment, based on a narrow doublet at 7.9–8.0 ppm. Significance assumed for *p* < 0.002. a Only selected contributions are shown; <sup>b</sup> Simple linear regression analyses of log-transformed pseudo-concentrations and SGA status; results are presented as fold-change (FC), e.g., glycine levels were 1.8-fold higher in the SGA group and 1.1-fold higher for each week's difference in PMA at week 1 (ratios below 1 are presented as í1/ratio); c Corresponding analyses for PMA; FC is per one week difference in PMA; <sup>d</sup> Multiple linear regression including both SGA status and PMA.

The previous considerations are summarized for the urinary metabolites glycine and threonine in Figure 4. There were similar levels of glycine and threonine in the intervention and control group (Figure 4a,b). Glycine and threonine levels appeared to differ between SGA and AGA children in the first week of life, but not at later time points (Figure 4c,d). The same applied when the infants' age was defined as PMA instead of chronological age (Figure 4e,f).

Finally, the metabolite pseudo-concentrations were examined with respect to growth velocity from birth to four weeks of life [9]. In the initial linear regression models, first week glycine and threonine levels and third week glycine and hydroxyproline levels correlated positively with growth velocity. However, when adjusting for PMA in the models, these relations disappeared.

**Figure 4.** Temporal development of glycine and threonine log-pseudo-concentrations (means and 95% CIs) related to nutritional intervention, SGA status and age. (**a**) Glycine levels by nutritional intervention (intervention red, control gray); (**c**) Glycine levels by SGA status (SGA orange, AGA green) for samples from weeks 1, 3, 5 and 7; (**e**) As above, but samples selected by PMA instead of weeks of life; (**b**, **d**, **f**) Corresponding figures for threonine.

#### **4. Discussion**

Due to proposed risks of overfeeding, we investigated the impact of enhanced nutrition on the urinary excretion profile during the first weeks of life in premature infants. We did not observe significantly different metabolic trajectories between the intervention group receiving nutritional support in the upper range of recent recommendations as compared to the infants on standard nutrient supply; neither in the first-week urine profiles nor in the change over time. Furthermore, infants in the intervention group exhibited better overall growth [9]. Together, this suggests that premature infants handle enhanced nutrient supply similarly to a standard diet because the urinary profiles in the intervention group did not indicate an overload of the renal function as compared to the controls.

Our study also revealed that all infants exhibited substantial changes in their urinary profiles during the early postnatal period (Figure 3, Table 3), and these changes correlated with gestational age at birth and with chronological age (Tables 3 and 5). The correlation between PMA and urinary metabolite profiles has been reported previously [17,31,32], and may reflect the degree of organ development and metabolic maturity. Between the first and the third week of life the glucogenic amino acids glycine, threonine, hydroxyproline and tyrosine increased along with metabolites of the tricarboxylic acid cycle like 2-oxoglutarate, citrate, fumarate and succinate. In most mammals, the prenatal-postnatal transition is accompanied by important adaptations in carbohydrate metabolism due to the abrupt change from the placental supply of nutrients to a cyclic supply of nutrients via the breast milk [7]. In rodents this period is characterized by the appearance of gluconeogenic enzymes to maintain glucose homeostasis in the newborn [33]. Thus, the metabolite changes observed in our premature infants might reflect similar metabolic adaptations. The presence of metabolites linked to the tricarboxylic acid cycle may be due to the high metabolic turnover in premature infants. The tricarboxylic acid cycle is important in energy metabolism, providing intermediates for the synthesis of glucose and some amino acids [34].

Hydroxyproline showed a threefold increase in concentration during the initial postnatal period. Urinary hydroxyproline reflects collagen metabolism and is considered a marker of infant growth [35,36]. Although we observed a positive correlation between third week hydroxyproline levels and growth velocity, this correlation disappeared when PMA was introduced as a covariate, suggesting that urinary hydroxyproline is closely related to PMA.

In parallel with the increase of the other metabolites during the first month of life, gluconate and two unidentified metabolites with an NMR signals at 5.50 and 0.57 ppm disappeared. The latter unidentified metabolite has also been observed in the urine of pregnant women [26,37]. Its disappearance shortly after birth suggests that this substance was transferred from the mother to the infant and may reflect a sulfate- or glucuronide-conjugate of pregnanediol or estrogen [26].

Most SGA infants have been exposed to a limited nutrient supply during fetal life, which may cause irreversible metabolic changes (fetal programming). Subsequent catch-up growth, both in early infancy and in childhood, is also associated with later obesity and cardiovascular disease risk [4,5,8]. The so-called mismatch hypothesis proposes that an obesogenic childhood environment increases later cardiovascular disease risk, whereas the postnatal programming or postnatal growth acceleration hypothesis links rapid weight gain in early infancy to later cardiovascular disease risk. In a recent review [3], growth during late infancy and childhood appeared to be the major determinant of later metabolic and cardiovascular disease risk, and not the early postnatal growth. It has also been shown that early postnatal growth has a significant impact on later neurodevelopment [3]. Both these findings support the aggressive nutritional approach in our intervention. We studied metabolic differences between SGA and AGA infants at birth

and over time, and identified glycine and threonine as potential biomarkers of an altered metabolic phenotype.

Glycine has been linked to nutrient restriction of pregnant baboons, where the fetal plasma levels more than doubled compared to control fetuses [38]. A similar increase in fetal glycine levels has been observed in human SGA fetuses [39,40]. Dessi *et al.*, profiled newborn urines one and four days after birth and reported that in addition to the glycine and threonine pathway, prenatal growth restriction also affected metabolic pathways involving hydroxyproline, creatinine and myo-inositol [19]. They interpreted these metabolites as potential early markers of the metabolic syndrome. In line with their study, we found similar differences in the first-week urine profiles between our SGA and AGA infants, although after adjusting for PMA, only threonine and glycine remained independently elevated in the SGA infants. Glycine and threonine are glucogenic amino acids, which may be converted to pyruvate during protein metabolism. Increased levels of plasma glycine may be caused by reduced amino acid oxidation or reduced gluconeogenesis as a strategy to conserve amino acids [38]. We observed that glycine and threonine were linked to SGA status in the first urine sample, but we were unable to detect a persistent difference during the course of time. Although our study did not exhibit similar results for all metabolites as compared to the study by Dessi *et al.* [19], and the elevation of glycine and threonine levels were insignificant after Bonferroni adjustment, it still highlights glycine, threonine and to some extent hydroxyproline as important targets for future research.

Our study has several limitations. In our original intervention trial [10], we planned to recruit 240 infants. The early termination due to increased occurrence of septicemia in the intervention group resulted in a reduced number of infants in our present study. In spite of the fact that the intervention group had a significantly lower mean birth weight and a higher proportion of SGA infants than the control group, we observed increased whole body growth [9], improved white matter maturation and motion perception in the intervention group (manuscript submitted). Moreover, we did not find any significant differences between the metabolic trajectories with regard to the two different diets. The occurrence of septicemia and electrolyte deviations did not seem to influence the urinary metabolite profiles, but the lack of significant differences must be interpreted with caution in view of the relatively small study sample. Similar electrolyte disturbances have been reported in other studies with early and enhanced nutrition to premature infants during the first week of life [41–46]. Thus, a difference in metabolic profile would probably occur during the early postnatal stay, and it raises the question as to whether we would have been able to identify an effect with a more frequent first week monitoring of the urines in our premature infants.

Multiple hypothesis testing was performed using Bonferroni correction. Although samples from such vulnerable patients are challenging to come by, the randomized design of the current trial as well as the strict adherence to the nutritional protocol reduced the number of confounding factors.

#### **5. Conclusions**

The urinary metabolite profiles were unaltered by the enhanced postnatal nutrition, suggesting that supply in the upper range of current recommendations did not overload renal function. Our data show that both gestational age at birth, *i.e.*, degree of maturation, and postnatal physiological adaptations, may influence metabolism in premature infants during the neonatal period. Several of the first-week urinary metabolites were associated to SGA status and postnatal growth and might be markers for long-term health outcomes.

### **Author Contributions**

D.S. acquired the NMR spectra of the urine samples. S.J.M. was responsible for the detailed planning of the nutritional protocol, recruitment and treatment of infants, and the collection of clinical parameters and endpoint data. D.S. and S.J.M. jointly carried out the statistical analysis and drafted the manuscript. E.W.B., K.S., B.N. and A.N.A. contributed to the conception and the design of the study, recruitment and treatment of infants, and the collection of clinical parameters and endpoint data. A.C.W., A.R., K.B., M.B.V., P.O.I. and C.A.D. contributed to the conception and the design of the study, and the interpretation of data, M.B.V., in addition to the statistical analysis. F.R. contributed to the design of the study, the metabolic profiling, and the analysis and interpretation of the results. J.P.B. planned the metabolic profiling and contributed to the statistical analysis and the interpretation of the results. All authors revised the manuscript critically for important intellectual comment.

#### **Conflicts of Interest**

The authors declare no conflict of interest.
