Metabolic profiling and targeted lipidomics reveals a disturbed lipid profile in mothers and fetuses with intrauterine growth restriction

Fetal growth may be impaired by poor placental function or maternal conditions, each of which can influence the transfer of nutrients and oxygen from the mother to the developing fetus. Large-scale studies of metabolites (metabolomics) are key to understand cellular metabolism and pathophysiology of human conditions. Herein, maternal and cord blood plasma samples were used for NMR-based metabolic fingerprinting and profiling, including analysis of the enrichment of circulating lipid classes and subclasses, as well as the number of sub-fraction particles and their size. Changes in phosphatidylcholines and glycoproteins were prominent in growth-restricted fetuses indicating significant alterations in their abundance and biophysical properties. Lipoprotein profiles showed significantly lower plasma concentrations of cholesterol-intermediate density lipoprotein (IDL), triglycerides-IDL and high-density lipoprotein (HDL) in mothers of growth-restricted fetuses compared to controls (p < 0.05). In contrast, growth-restricted fetuses had significantly higher plasma concentrations of cholesterol and triglycerides transporting lipoproteins [LDL, IDL, and VLDL, (p < 0.005; all)], as well as increased VLDL particle types (large, medium and small). Significant changes in plasma concentrations of formate, histidine, isoleucine and citrate in growth-restricted fetuses were also observed. Comprehensive metabolic profiling reveals that both, mother and fetuses of pregnancies complicated with fetal growth restriction have a substantial disruption in lipid metabolism.

SCIENTIFIC REPORts | (2018) 8:13614 | DOI: 10.1038/s41598-018-31832-5 to controls (p = 0.02). Similarly, triglycerides transporting lipoproteins, IDL and HDL, were significantly lower in mothers of fetuses with sub-optimal fetal growth compared to AGA mothers (p = 0.01, 0.03 and 0.02, respectively) ( Table 2). In terms of lipoproteins properties, there were lower concentrations of large HDL lipoproteins in mothers of FGR fetuses (p = 0.04). Changes in glycoproteins and phosphatidylcholines suggested a disruption in their biophysical properties among study groups. Specifically, the width of one of the phosphatidylcholines peaks was significantly lower in FGR cases (p = 0.02). With regards to LMW metabolites, maternal plasma concentrations of alanine, citrate, 2-oxoisovaleric acid and pyruvate were significantly lower in mothers of small fetuses compared to controls (all p values < 0.05) ( Table 2).
Interestingly, the behavior of triglycerides and cholesterol lipoproteins was the opposite in affected fetuses compared to their mothers. FGR fetuses had significantly higher concentrations of cholesterol-VLDL (+56%), -IDL (+24%) lipoproteins (p = 0.008 and 0.04, respectively), as well as higher concentrations of triglycerides-VLDL (+24%), and -IDL (+18%) compared to controls (p = 0.008 and 0.04, respectively) ( Table 3). Importantly, we found a gradient of higher concentrations of cholesterol (VLDL and IDL) and triglycerides (VLDL, IDL and LDL) lipoproteins in SGA and FGR cases compared to controls (Jonckheere-Terpstra test p < 0.05). FGR cases also showed a significant increase in large HDL, medium LDL and all VLDL particle types (large, medium and small: average + 37%) vs. controls (all p values < 0.05). When analyzed as a trend, the Jonckheere-Terpstra test also showed that the increase in particles size and distribution was present in the two clinical phenotypes following a severity ascendant gradient, higher in FGR fetuses. Likewise, changes in phosphatidylcholines and glycoproteins

Figure 1.
Maternal and cord blood nuclear magnetic resonance profiles among the study groups. The more evident differences in the Diff spectral vectors were observed at 1.29 and 0.89 ppm, suggesting lower relative plasma concentrations of lipids in mothers of small fetuses (both, in SGA and FGR cases), while in cord blood, the relative concentrations of lipids were higher in cases, being greater in FGR compared to SGA cases, as well as compared to controls.

Figure 2.
Loading weight of each spectral variable extracted from an orthogonal partial least square discriminant analysis model (ASCLAN) of the data set among the study population. The color-coded loading plot shows the discriminatory variables, non-discriminatory signal variables and the noise variables, which are indicated by red, blue and grey, respectively. The loading plot shows the most influential regions of the spectrum that are responsible for discrimination between cases and controls.

Discussion
Principal findings of the study. In this comprehensive metabolomic study, we expanded previous observations that lipid profiles are disturbed in small fetuses and their mothers. Yet, our innovative approach allowed quantifying the number and size of the particles in the lipoprotein fractions as well as low-molecular-weight metabolites, describing the complex maternal and fetal metabolic response in pregnancies with fetal smallness. However, the study failed to observe any remarkable difference between the two clinical phenotypes of fetal smallness. Thus, SGA and FGR presented similar findings, supporting the notion that both phenotypes might be exposed to different degrees of undernutrition and thus, both might be at risk of metabolic fetal programming. Future studies assessing placenta metabolism might delineate potential mechanisms in both phenotypes.

Why are circulating lipids lower in mothers of growth-restricted fetuses?.
Our results describing an overall decrease in circulating lipids in mothers of small fetuses are in line with previous studies [42][43][44][45][46][47] . However, this study revealed further details, providing qualitative and quantitative measurements of lipoproteins, their cholesterol and triglyceride concentration, as well as the number of sub-fraction particles and their size. Compared to controls, there was a triglyceride-impoverishment in lipoproteins such as HDL (decrease in large particles) and IDL as well as low concentration of cholesterol in other lipoproteins, namely VLDL and IDL in mothers of both SGA and FGR cases. A maternal metabolic adaptation during pregnancy 47-53 is essential to increase availability of cholesterol and triglycerides for fetal development and the synthesis of steroid hormones by the placenta [54][55][56] . When maternal plasma cholesterol is low, birthweight is lower than normal. Indeed, term infants of mothers with low total cholesterol weigh are on average 150 g less than those born to mothers with normal cholesterol concentrations 57 . The first trimester of gestation is considered an anabolic period, in which the mother increases her deposit of fat, through a reduction in the activity of lipoprotein lipase, and increasing in the insulin responsiveness 58 . Interestingly, this fat accumulation is maintained even under conditions of severe maternal malnutrition 59 .
In the third trimester, a switch to catabolism occurs in order to supply fetal demands and to synthesize large amounts of steroids hormones by the placenta 54 . The products of an enhanced lipolytic activity reach the liver to be used for triglycerides synthesis and to increase circulating availability of triglycerides by 208% (transported mainly by VLDL), and total, HDL-and LDL-cholesterol in 65, 26 and 64%, respectively 50 . It seems plausible that in the third trimester of pregnancy, placental insufficiency generates a specific fetal and maternal response that is reflected in the metabolomic profile. Low concentrations of lipids in the maternal circulation may reflect an increased uptake into the placenta. In normal gestation, support of lipids, and particularly essential fatty acids, to the fetus is ensured by increased maternal liver production of VLDL rich in triglycerides (build upon peripheral deposits of adipose tissue) 60 , peripheral enrichment of HDL and LDL and lipoprotein lipase activities in the placenta 61,62 . The findings of this and previous studies suggest (1) failure of the normal signaling leading to increased lipids in pregnancy, and/or (2) a reduction in lipid reserve in the mothers.
Why are there higher cord blood lipids in SGA and FGR fetuses?. Our findings confirmed previous studies reporting higher circulating triglycerides and cholesterol lipoproteins in growth-restricted fetuses [63][64][65][66] . Importantly, our metabolic fingerprinting approach allowed to describe that VLDL and IDL were the most deregulated lipoproteins in FGR fetuses, suggesting that VLDL rich in triglycerides may be an alternative fuel mobilized by the growth-restricted fetus. This is a novel finding in humans, since VLDL is mostly synthesized in the fetal liver, implying an altered hepatic synthesis of lipoproteins in adverse intrauterine conditions. Lipids are fundamental molecules for life. During fetal life, triacylglycerols provide energy for metabolic processes, while fatty acids, cholesterol and phospholipids are required to develop the fetal brain and central nervous system, to build cell membranes, and as a precursor of bile acids and steroid hormones 67 . Both, the placenta and the fetus have the capacity of denovo cholesterol synthesis. It has been estimated that up to 20% of cholesterol  [73][74][75][76][77] . Therefore, important challenges for future research are to ascertain the mechanistic pathways whereby fetal abnormal lipid profile might influence the higher risk of growth restricted fetuses to cardiovascular disease in adulthood.
Changes in low molecular weight metabolites. While major changes were found in lipids, analysis of LMW metabolites revealed lower cord blood concentrations of acetate and formate in late-onset FGR and maternal concentrations of citrate, pyruvate and 2-oxoisovaleric acid in both, SGA and FGR pregnancies. Biophysical abnormalities in phosphatidylcholines and glycoproteins were also detected. Phosphatidylcholine is a major membrane phospholipid, made in mammalian cells from choline via the CDP-choline pathway and has a key role in neuronal differentiation and cell fate determination 78 . Importantly, phosphatidylcholine biosynthesis is required for normal VLDL secretion from hepatocytes 79 . Metabolomic analysis showed some differences in comparison with previous studies. For instance, in this study we did not replicate our previous finding of lower concentrations of valine and leucine in late-onset FGR 38 , nor we could demonstrate differences in other essentials aminoacids in SGA fetuses as reported in other studies 36,37,80 . Favretto et al. reported up regulation of phenylalanine, tryptophan, and glutamate in FGR fetuses 37 , while Ivorra et al. reported significant differences cord blood concentrations of five aminoacids in FGR fetuses (proline, glutamine and alanine were reduced, while phenylalanine and citrulline were increased) 36 . In previous studies, early-onset FGR showed significant decrease in glucose 63,81 , while no differences have been found in late-onset FGR 63 . It is known that glucose levels are inversely correlated to the clinical severity of FGR and the capability of transplacental glucose gradient, which might differ between the two phenotypes. Thus, these discrepancies suggest differences in case selection, platforms and protocols used among studies assessing metabolites in placental dyfunction 36,37 . Metabolic differences between small-for-gestational age and growth-restricted fetuses. FGR has two main clinical presentations, early-and late-onset disease. Early-onset FGR is highly associated with severe placental insufficiency and chronic fetal hypoxia 82,83 . In early-onset FGR, the typical scenario is a progressive deterioration of fetal well-being, accompanied with higher rates of preeclampsia and adverse perinatal outcomes, often requiring preterm delivery 4 . On the other hand, late-onset FGR represents 70-80% of FGR, has less signs of placental disease and the association with preeclampsia is minimal 84 . Despite a more benign nature as compared with early FGR, and that chronic hypoxia seems to be mild, the increase demands of oxygen and nutrients by vital organs such as brain and heart towards term increase the risk of acute fetal deterioration before labor, as suggested by the high contribution to late-pregnancy mortality, and a high association with intra-partum fetal distress and neonatal acidosis 85 . One of the aims was to evaluate whether late-onset SGA and FGR are associated with differential metabolomic patterns. Although there is a widely shared assumption that SGA are "constitutionally small", but otherwise normal, fetuses, more recent studies reported that SGA display similar, albeit milder, neurodevelopmental and cardiovascular changes as those observed in FGR fetuses 13,14,25,86,87 . In line with this, the present study found that SGA and FGR fetuses had remarkably similar pattern of alterations in maternal and fetal lipid profiles, although there was a gradation of metabolic disruption according to the severity of the disease. These findings support that, at least a proportion of SGA fetuses represent a milder clinical form of true growth restriction not associated with Doppler changes or increased obstetrical risks. It could be hypothesized that SGA fetuses suffer undernutrition and its long-term consequences, but the respiratory function of the placenta is still largely preserved. Consequently, while the clinical distinction between "low risk" SGA and FGR is relevant for obstetrical management, fetal smallness represents a (potential) high-risk situation for long-term quality of life since metabolic adaptations occurs in utero regardless of its clinical prenatal presentation.

Strengths and limitations.
The strengths of this study include the prospective design and the inclusion of a well-defined cohort including samples from mothers and fetuses of both, SGA and FGR cases with a relatively large sample size, meaning that the possibility of selection bias was minimized. From a methodological standpoint, the platforms used are highly reproducible and standardized. NMR and mass spectrometry are not mutually exclusive but rather complementary. They have advantages and disadvantages, and the method of choice depends on the aim of the study. One of the main advantages of NMR is its extraordinary reproducibility, which makes this technique a very good option for fingerprinting analysis, such as the study reported herein. Besides identification, we had the interest of quantify specific lipids, therefore we have used NMR spectroscopy since the peak area of a compound in the NMR spectrum is directly related to the concentration of specific nuclei, making quantification of compounds in a complex mixture very precise 88,89 . On the other hand, NMR is a non-destructive technique that allows samples to be recovered after analysis, which is not trivial considering the incalculable value of these samples. While basal contamination with macromolecules and lipids in the plasma are usually overlooked, herein, were discriminated and quantified and in addition, the resolution of spectral process described (buckets within 0.003 ppm) is significantly higher to that reported before (buckets within 0.05 ppm) [36][37][38] . Moreover, the meticulous deconvolution process applied to each metabolite, is by far more rigorous, than the "blinded" quantification previously used [36][37][38] . This is the first time that Liposcale is used in pregnancy. This test was originally developed to rapidly produce a lipoprotein profile of an individual's plasma as part of cardiovascular disease risk work up 90,91 . The positive correlations describe between Liposcale and enzymatic colorimetric methods suggests that these variations are Figure 3. Heat map following metabolomic profiling on the annotated and significant (among the three groups) identified in plasma samples of mothers and neonates. A colored heat map of significant LMW metabolites and the Liposcale analysis along with their p-values, and percentage of fold change was obtained to visualize and characterize the differences between adequate for gestational age (AGA), all cases with a birthweight below the 10 th centile (BW <10 th ) and separately small-for-gestational age (SGA) as well as fetal growth restriction (FGR) groups from maternal and umbilical cord blood. Fold change of significant differences (p < 0.05) for each metabolite are on a graded color scale from green (lower value) to black (no differences) to red (high values).
SCIENTIFIC REPORts | (2018) 8:13614 | DOI:10.1038/s41598-018-31832-5 biologically relevant. Yet, we acknowledge several limitations. First, our sample size may cover potential associations in other metabolites and hide interaction of confounders (clinical characteristics similar among populations but non-obvious confounders could have affected). Secondly, as we focused in late-onset FGR (by far, the most common clinical presentation), we might have missed the most extreme and severe cases occurring before 32 weeks of gestation, in which fetal metabolic adaptations might be of uttermost relevance. Third, it is likely that our classification has overlapping between SGA and FGR, and for instance it is likely that constitutional small fetuses were included. Fourth, because of the design of the study, we only had few maternal samples of elective cesarean sections (maternal fasting conditions) and thus, non-fasting blood samples were collected. Typically, there is marginal effect of fasting on total cholesterol, HDL-C, and LDL-C, although triglycerides have a tendency to increase approximately 15% in non-fasting vs. fasting samples [92][93][94] . Fifth, although questionable, other more difficult to control confounding factors such as differences in the length of labor, time lapse difference between maternal and fetal sampling 95 or seasonal variation of lipid profile 96 might influence maternal lipid profile in the same form as fetal gender might influence the metabolomic profile of the fetuses included 97 . Future studies with a larger sample size might include these confounding factors into the analysis.

Conclusion
In summary, metabolic profiling combined with clinical phenotyping demonstrates the potential to enrich stratified medicine research, revealing that mothers of small fetuses present substantial reductions in lipid metabolites, suggesting a failure in the maternal metabolic adaptation to pregnancy. While, both, SGA and growth-restricted fetuses have a substantial increase in lipids, indicating a similar metabolic response to undernutrition, there was a gradation of the metabolic disruption according to the severity of the phenotype.

Materials and Methods
Subjects and sample collection. This study is part of a larger prospective research program on FGR at the Department of Maternal-Fetal Medicine in Hospital Clinic Barcelona, involving maternal, fetal, and perinatal outcomes, as well as short-and long-term postnatal follow-up. A prospective cohort of singleton gestations with antenatal suspicion of fetal smallness and diagnosis established later than 32 weeks of gestation, were prospectively followed; those who delivered neonates with a birthweight below the 10 th centile at term (≥37 weeks of gestation) were included as cases (n = 52). According to our clinical protocol, small fetuses were subdivided: those with a birthweight <3 rd centile and/or abnormal uterine artery Doppler and/or abnormal cerebroplacental ratio were termed fetal growth restriction (FGR), while those with a birthweight between the 3 rd and the 9 th centile and normal fetoplacental Doppler were considered small-for-gestational age (SGA) cases. A control group of 28 pregnancies with adequate-for-gestational ages (AGA) fetuses were selected among low-risk pregnancies attending third trimester routine pregnancy care and were included if they delivered term neonates with a normal birthweight (between 20 th and 90 th centile). Exclusion criteria included multiple gestations, spontaneous preterm labor or delivery, premature rupture of membranes, chromosomal abnormalities or major structural abnormalities. All the patients included in this study were delivered at term and did not receive steroids for fetal lung maturity. The Institutional Research and Ethics Committee approved the study protocol (review board 2014/7154), all parents gave their written informed consent and all experiments were performed in accordance with relevant guidelines and regulations.
Clinical and Ultrasound Data. Maternal, perinatal and neonatal data were prospectively recorded in all patients. Gestational age was calculated on the basis of fetal crown-rump length, measured at 11-13 weeks. Transabdominal ultrasound with Doppler evaluation was performed in both, cases and controls, at recruitment, using 6-4-MHz probes (Siemens Sonoline Antares, Siemens Medical Systems, Malvern, PA, USA) and a Voluson 730 Expert Machine (GE Medical systems, Zipf, Austria). Estimated fetal weight (EFW) was calculated using the Hadlock formula 98 , and adjusted according to fetal gender and gestational age using local standards 99 . Feto-placental Doppler included: umbilical artery-pulsatility index (PI), mean uterine artery PI and middle cerebral artery PI. The cerebroplacental ratio was calculated as middle cerebral artery PI/umbilical artery PI 100 , and defined as abnormal when <5 th centile for gestational age 100 . The mean uterine arteries PI value was considered abnormal when >95 th centile 101 . Cases of FGR were delivered electively between 37 and 38 weeks of gestation, while SGA cases were allowed to deliver up to 40 weeks of gestation 5,6,11 . Controls were allowed to have a spontaneous onset of labor but, if they reached 41 weeks of gestation, induction of labor was offered.
Biological samples collection and storage. Maternal blood samples were drawn 2-4 hours after delivery (non-fasting conditions but at least eight hours after their last meal and blood sampling). Umbilical vein cord blood samples were obtained from the clamped umbilical cord immediately after delivery of the fetus. All blood samples were collected in EDTA-treated tubes and processed within one hour. Plasma was separated by centrifugation at 3000 rpm for 10 minutes at 4 °C, and stored at −80 °C until further use. NMR data acquisition. Plasma samples were thawed overnight and prepared for nuclear magnetic resonance (NMR) analyses according to the Bruker-specific metabolomics protocol 102 . Aliquots of each sample (300 μl) were mixed with 300 μl of sodium phosphate buffer for immediate analysis. High-resolution 1 H-NMR spectroscopy data were acquired on a Bruker 600 MHz Spectrometer (Bruker Biospin, Rheinstetten, Germany) equipped with an Avance III console and a TCI CryoProbe Prodigy: 1D Nuclear Overhauser Effect SpectroscopY (NOESY), Carr-Purcell-Meiboom-Gill (CPMG), and 2D j-resolved spectroscopy (JRES), all with pre-saturation to suppress the residual water peak, to characterize small molecules such as amino acids and sugars; and 1D Diffusion (Diff, TR/TE, b values, NS, ACQ), to detect larger molecules such as lipoproteins, glycoproteins and choline compounds 103,104 . All the sequences were run at 37 °C in quantitative conditions (systematic pre-calibration of radio frequency pulses and sample temperature, and same receiver gain adjustment). CPMG and Diff data were preprocessed at the NMR console (TopSpin 3.2, Bruker Biospin, Rheinstetten, Germany) for basic corrections, such as phase correction and exponential line broadening (0.5 Hz for CPMG; 1.0 Hz for Diff).

NMR fingerprinting analysis.
For the metabolic fingerprinting analysis, spectral vectors were generated from the CPMG and Diff data by binning (0.003 ppm), alignment (alanine, 1,475 ppm) and region suppression (to minimize the influence of water and EDTA signals, specific to the CPMG data). This was carried out with MestReNova v11.0.3 (Mestrelab Research S.L., Santiago de Compostela, Spain). These spectral vectors were used for automatic spectroscopic data categorization by clustering analysis (ASCLAN), a supervised method to quickly assess discriminatory spectral regions (and indirectly metabolites) in the study groups 105 . The Diff NMR data was further used for lipoprotein profiling, based on the Liposcale test 90,91 . This test provides information about the size, lipid concentration (cholesterol and triglycerides) and number of particles for the main classes of lipoproteins [very-low density lipoprotein (VLDL), low-density lipoprotein (LDL), intermediate-density lipoprotein (IDL) and high-density lipoprotein (HDL)], as well as the concentration of particles in their subclasses (large, median, small) 90,91 . The Diff NMR data was also used to quantify choline compounds (3.3-3.18 ppm) and glycoproteins (2.15-1.9 ppm), based on peak deconvolutions. Details of the method for the Liposcale for lipoprotein characterization and phosphatidylcholines compounds and glycoprotein peak deconvolution are explained in the supplemental material. Finally, CPMG data was used for the profiling of 20 metabolites, based on a new, fully automated version of the software package Dolphin 106,107 . Signal annotation was based on templates prepared in previous studies with the help of available databases 108 and bibliography [109][110][111] . Validation of metabolite identification was assisted by STOCSY 112 and JRES data.
Statistical analysis. Categorical data are presented as n (%) and continuous data as mean (±SD) or median [interquartile range (IQR)] according to their distribution. To assess the categorical variables, proportions were compared with Fisher's exact test or the chi-square test. Distributions of continuous variables were examined for normality using the Kolmogorov-Smirnov test. When there was normality of continuous variables, the one-way ANOVA test and unpaired t-tests were used to compare differences. Otherwise, the Kruskal-Wallis one-way analysis of variance and Mann-Whitney U-test were used. Correlation between selected metabolites concentrations by 1 H-NMR and enzymatic colorimetric methods were assessed with Pearson coefficients (or Spearman for non-normal distributions). To initially evaluate the ability of metabolites in distinguishing cases from controls, orthogonal partial least square discriminant analysis (OPLS-DA) was performed based on multivariate, supervised, discriminant analysis of spectral vectors -Automatic Spectroscopic data Categorization by cLustering ANalysis (ASCLAN) 105 . Finally, the Jonckheere-Terpstra test was also used to test if the concentration of metabolites followed a ordered alternative hypothesis across severity groups (controls-SGA-FGR) 113,114 . Statistical analysis was performed using STATA 14 (Stata Corp LP, 2015, College Station, Texas) and MATLAB (MathWorks Inc, US). A p-value < 0.05 was considered to be statistically significant.

Data Availability Statement
Clinical information of the patients included, spectrum data and associated protocols are promptly available to readers without undue qualifications in material transfer agreements.