Introduction

Obesity and overweight have become a global epidemic [1]. In women of reproductive age, obesity poses a risk to maternal health, with consequences from gestational diabetes (GDM) [2] and adverse pregnancy outcomes to later type-2 diabetes (T2D) and cardiovascular diseases [3]. It is also associated with adverse outcomes for child at birth and during neonatal, infant and later life periods [4].

GDM is defined as glucose intolerance that is first recognized during pregnancy. Risk factors have been identified (including obesity, age and family history); however, the underlying mechanisms remain an enigma. GDM might be due to a pancreatic β-cell defect similar to that in type-1 diabetes (T1D), or to a dormant, pre-existing insulin resistance (IR) clinically manifesting as GDM during the diabetogenic state of late pregnancy [5]. Women developing GDM show impaired ability to stimulate glucose disposal and to suppress both glucose production and fatty acid (FA) levels [5]. Maternal obesity and GDM, each and combined, predispose to adverse short- and long-term infant outcomes [4, 6, 7].

To provide an understanding of the connection between early metabolic programming and the increased incidence of metabolic diseases resulting from the disruption of the intrauterine environment, associations between maternal prepregnancy Body Mass Index/GDM and maternal–cord blood metabolic profiles were studied.

Experimental section

Participants

The study was performed on data from the PREOBE study, a prospective observational cohort study (NCT01634464). Study procedures, group classification and subject inclusion/exclusion criteria have been described by Berglund et al. [8].

The study was approved by the Bioethical Committees for Clinical Research of the University Hospital San Cecilio, the Mother-Infant University Hospital of Granada, Spain.

Sample and data collection

Maternal venous blood samples were drawn into EDTA-containing tubes at delivery. Vein umbilical cord blood samples were obtained after clamping the cord. Blood samples were processed and plasma aliquots were stored at − 80 °C until analysis. Samples for metabolomics analyses were available for 200 mothers and 124 newborns, with an overlap for 119 mother/children pairs. At recruitment, information was collected on maternal anthropometry and used to calculate pBMI [8]. Gestational weight gain (GWG) at 34 weeks and data regarding clinical outcome at delivery were collected from medical records. Biochemical analysis of serum samples was performed as previously reported [8].

Since the maternal fasting status at time of blood collection was unknown, we used a fasting blood glucose threshold (126 mg/dl) previously validated for diabetes to identify non-diabetic, non-fasting mothers. Namely, if a non-diabetic mother presents a plasma glucose value above the threshold, this indicates non-fasted status. 27 non-GDM mothers were defined as non-fasted and excluded from the analysis. Such a distinction was obviously not possible for GDM mothers [9].

Targeted metabolomics assays

Targeted metabolomics involved analysis of 400 metabolites comprising: polar lipids [acylcarnitines (AC), diacyl-phosphatidylcholines (PCaa), acyl-alkyl-phosphatidylcholines (PCae), sphingomyelines (SM), acyl-lysophosphatidylcholines (LPCa), alkyl-lysophosphatidylcholines (LPCe)], sum of hexoses (H1), amino acids (AA), keto acids, non-esterified FA (NEFA), and Krebs cycle metabolites. NEFA analysis was performed by liquid chromatography with tandem mass spectrometry (LC–MS/MS) as previously reported [10]. Polar lipids were analysed with flow-injection analysis tandem mass spectrometry (FIA–MS/MS) [11]. A formula CX:Y was assigned for polar lipids and NEFA where X: length of carbon chain, Y: number of double bonds, OH: indicates presence of hydroxyl group. Letters ‘a’ and ‘e’ indicate that the acyl chain is bound via an ester or ether bond to the backbone, respectively. AA analysis was done by derivatization and separation by ion-pair LC–MS/MS [12]. AA were coded according to IUPAC abbreviations.

Six plasma quality control (QC) samples per batch were consistently measured with the samples. Concentrations were calculated in µmol/l; the analytical process was controlled and post-processed by Analyst 1.6.1 and R software (R version 3.4.3).

Metabolomics quality control (QC) and preprocessing

First, the QC for metabolomics measurements was done using a threshold of 20% and 30% for the intra- and inter-batch coefficient of variation, respectively (with allowance of max. 1 outlier measurement > 2 IQR from the next measurement). In total, 202 metabolites passed the QC. Measurements lying away than 1.5 standard deviations (SD) from the next closest measurement were removed. We corrected for batch effects by dividing metabolite concentrations by the ratio intra-batch median/inter-batch median. Then, the measurements were split into two datasets corresponding to maternal and cord blood.

Within each dataset, analytes with > 40% missing values were excluded. Sums were computed: Ʃ branched chain AA (BCAA), ƩLPCa, ƩPCaa, ƩPCae. Also, ratios were computed: ΣPCaa/ΣPCae, reflecting oxidative stress [13]; ΣLPCa/ΣPCaa, as a lipid biomarker of inflammation; (LPCa16:0 + LPCa18:0)/ΣPCaa as a proinflammatory biomarker [14]; (LPCa18:1 + LPCa 18:2)/ΣPCaa as an anti-inflammatory biomarker [15]; AC ratios (AC16:0/free carnitine (Carn) and AC2:0/AC16:0) as markers of carnitine palmitoyl transferase-1 activity (CPT1) and FA β-oxidation, respectively [16]. Moreover, FA ratios were used to estimate activities of stearoyl-CoA desaturase-1 (SCD-1; 16:1/16:0, SCD-16 and 18:1/18:0, SCD-18) [17] and AA ratios Asn/Asp and Gln/Glu as indicators for anaplerosis or replenishing of Krebs cycle metabolites. Analytes, sums and ratios were log2-transformed after inspection of boxplots and quantile–quantile plots. Outliers were defined as points lying further away than 1 SD from the next measurement and were excluded.

Statistical analysis

Baseline characteristics

The ‘overweight’ and ‘obese’ groups were merged into one category (pBMI was considered a continuous variable). Differences between the covariates in the three groups were evaluated via Kruskal–Wallis and Fisher tests for continuous and categorical covariates, respectively.

Associations of maternal and cord blood metabolome with BMI and GDM

The associations between metabolite concentrations in maternal/cord blood and the exposures of interest were determined using multiple linear regression models with covariates adjustment. The final model used log2 of the metabolite concentration as outcome and log2(pBMI) and GDM as independent variables, GWG at 34 weeks, maternal age, mode of delivery, and infant sex as covariates. Parity was not included in the model due to high missingness (> 50%). A sensitivity analysis including gestational age showed no differences in the effect size and was not included in the final models. For these models, median values of 157 and 111 observations were used for maternal and cord blood, respectively. Results were depicted in Manhattan plots with|log10 (P)|-values on y-axis and metabolites on x-axis, with the direction on y-axis and the magnitude indicating the sign and strength of the association, respectively. We used false discovery rate (FDR) to minimize the occurrence of false positives (type I errors), a common issue in multiple testing. FDR controls for the expected proportion of false predictions relative to the total number of predictions at the level of significance. Nevertheless, because of the exploratory nature of the analysis, we also inspected associations with uncorrected p < 0.05 (‘trends’).

Results

Baseline characteristics of the two populations used in the analyses are summarized in Table 1A, B.

Table 1 Demographics characteristics of study participants
Full size table

In a nutshell, our analyses of maternal and cord blood metabolites showed highly significant associations of GDM with several maternal and, to a lesser extent, cord blood metabolites, yet weaker associations with pBMI (Figs. 1, 2, 3, 4).

Fig. 1
figure 1

Manhattan plot showing association of metabolites in maternal plasma with maternal prepregnancy BMI (pBMI). The y-axis represents the log10 p value of pBMI; the direction (positive or negative) corresponds to the sign of the estimate for pBMI. p value and estimate were calculated in the linear model with the log2 metabolite concentration as dependent variable and log2 pBMI, GDM, GWG at 34 weeks, fetal sex and mode of delivery as independent variables. The dashed lines represent the uncorrected 0.05 significance level. AA amino acid, AC acylcarnitines, BCAA branched chain amino acid, H1 sum of hexoses, LPCa acyl-lysophosphatidylcholines, LPCe alkyl–lysophosphatidylcholines, NEFA non-esterified fatty acid, PCaa diacyl-phosphatidylcholines, PCae acyl-alkyl-phosphatidylcholines, TCA tricarboxylic acid, SM sphingomyelines

Full size image
Fig. 2
figure 2

Manhattan plot showing association of cord blood metabolites with maternal pBMI. For abbreviations and explanation of the plot, see Fig. 1 legend

Full size image
Fig. 3
figure 3

Manhattan plot showing association of metabolites in maternal plasma with gestational diabetes mellitus (GDM). The y-axis represents the log10 p value of GDM; the direction (positive or negative) corresponds to the sign of the estimate for GDM. For abbreviations and explanation of the plot, see Fig. 1 legend. Associations significant after FDR correction are marked with bold names for the metabolite

Full size image
Fig. 4
figure 4

Manhattan plot showing association of metabolites in cord blood with GDM. For abbreviations and explanation of the plot, see Fig. 1 legend. Associations significant after FDR correction are marked with bold names for the metabolite

Full size image

Maternal prepregnancy BMI

For pBMI, none of the maternal metabolites were significant after correction for multiple testing. All BCAA, SM32:2, 34:2, PCaa38:4 and alpha-aminoadipic acid (AAA) showed positive trends with pBMI, while PCae36:1 showed a negative trend. Also, scatterplot inspection revealed positive trends for AC3:0, 4:0, 5:0, and 9:0 (Fig. 1).

In cord blood, BCAA behaved similarly except Val, while Cys showed a negative trend. Similarly, β-hydroxybutyric acid (BHBA), was positively associated with pBMI concomitant with a negative tendency for α-ketoglutaric acid. There were positive tendencies with NEFA22:4 and ΣPCaa/ΣPCae ratio with pBMI, the latter due to reduced PCae levels, while Σ(LPCa18:1 + LPCa18:2)/ΣPCaa, an anti-inflammatory biomarker, showed a negative tendency (Fig. 2).

Maternal gestational diabetes mellitus

H1 (about 90–95% glucose, 5% other hexoses) was significantly higher in GDM in both maternal and, especially, cord blood (Figs. 3, 4). In both maternal and cord blood, Asn/Asp and Gln/Glu ratios were elevated in GDM subjects, with decrease in Asp and Glu. Another common finding is the overall decrease in the majority of phospholipids (PL). Most maternal LPCs and PCaa were decreased in GDM mothers, with strongly significant associations for LPC16:0, PCaa38:3,38:5, and some SM, especially SM32.2. PCae were not affected by GDM, thus resulting in a negative association for ΣPCaa/ΣPCae ratio.

In cord blood, LPC concentrations showed no difference with maternal GDM status, while there was a tendency for PCae and some PCaa to be decreased in GDM babies (PCae38:0 being significant after FDR correction). Interestingly, Carn was significantly lower in GDM babies with the same tendency for all AC, especially short-chain AC and acetyl carnitine. Markers for CPT1 activity and β-oxidation were higher and lower, respectively, in GDM babies. Also, an overall decrease in cord blood long-chain NEFAs and Krebs metabolites was observed in association with GDM, most notably NEFA26:1, malic, and succinic acids with an elevation of 3-methyl-2-oxobutanoic acid.

Similar and contrasting associations in GDM and obesity

In cord blood, Asn/Asp ratio was positively associated with pBMI and elevated (though not statistically significant) in GDM subjects. Ile was positively associated with pBMI and negatively with GDM.

Some maternal PCaa and SM were slightly elevated in overweight/obese and significantly reduced in GDM mothers.

Discussion

Influence of obesity on maternal and cord blood metabolome

In high pBMI, the most characteristic common features between maternal and cord blood were the elevated BCAA levels, which are essential AA promoting protein synthesis and turnover and glucose metabolism. In cord blood, BCAA are transplacentally transported from maternal to fetal side to be utilized for fetal growth and protein metabolism, and their transport is highly enhanced in late pregnancy to provide for the increased fetal nutrient demands [18]. This reflects other observations linking elevated BCAA and their metabolic by-products to obesity and metabolic syndrome [19]. Explanations were given showing BCAA as a consequence of obesity, others as a cause. Some authors attributed elevated BCAA levels to increased dietary protein intake, or excessive protein breakdown in skeletal muscle due to obesity-related IR [20]. Another hypothesis involves downregulation of BCAA oxidation enzymes, especially those involved in first pass BCAA metabolism, and was supported by animal and human studies [20]. There is also a speculation about role of gut microbiome in BCAA de-novo synthesis, thus contributing to their circulating levels [20].

We hypothesize that increased lipid availability associated with obesity favours FAO for satisfying energy needs. This reduces catabolism of other fuels as AA and glucose, sparing them for transfer to the fetus for fetal growth, and consequently leads to BCAA accumulation in maternal blood [21]. However, such an interpretation fails to explain the encountered elevated AC3:0 and AC5:0 levels, end products of BCAA catabolism, as well as the elevated AC4:0 and AC9:0, which were in agreement with previous reports from multiple cohorts of obese and IR subjects [19, 20].

In overweight/obese women, elevated levels of maternal hormones such as leptin, insulin, and IL-6 were discovered to play a key role in activating the mammalian target of rapamycin complex-1 (mTORC1) signalling and AA transporter activity [22], causing enhanced fetal growth. Moreover, evidence was found that placental BCAA uptake is further increased in pregnancies with enhanced maternal ketogenesis and β-oxidation rate [23].

In addition, we confirmed associations between pBMI and metabolites commonly related to inflammation, such as alpha-aminoadipic acid (AAA), LPC, and Cys. Low-grade inflammation, a common finding in obesity and diabetes, happens when excessive fat accumulation induces the release of adipokines which, in turn, lead to reactive oxygen species (ROS) production [24]. In maternal blood, mothers with high pBMI showed elevated AAA levels, whose excessive production was suggested to result from increased breakdown of Lys through oxidative stress and ROS [25]. AAA has also been identified as a potential modulator of glucose homeostasis in humans and an important factor mediating central obesity and diabetes [26]. In line with this picture, cord blood Σ(LPCa18:1 + LPCa18:2)/ΣPCaa levels, an anti-inflammatory marker, negatively associated with maternal pBMI [15]. Decreased LPCa18:1 and LPCa18:2 levels were previously associated with obesity and obesity-related factors [11, 27]. It is believed that not only the absolute concentration, but also the LPC acyl composition could be linked to inflammation, obesity, and atherogenesis, with a higher saturated-to-unsaturated LPC ratio being observed in inflammation [28]. We focused on four LPC metabolites (LPC containing 16:0, 18:0, 18:1, and 18:2 FA) and their ratios to PC because they were demonstrated to be among the most abundant serum metabolites [28]. The anti-inflammatory effect of LPC18:1 may be linked to decreased superoxide production and platelet aggregation [29] and was previously reported to be decreased among other unsaturated LPC in studies involving obese versus normal subjects [27]. Cord blood Cys was reduced suggesting that the existing oxidative stress drives Cys towards glutathione biosynthesis to fight ROS, thus enhancing its consumption and decreasing its cord blood levels [28].

Interestingly, maternal SM32:2 showed a weak positive trend with pBMI. Our group found the same tendency in another cohort of pregnant women [30] as well as a strong positive association with BMI and waist circumference in young adults [11] and children (unpublished results).

BHBA, a ketone body produced from the oxidation of fatty acids and utilized as an energy source particularly in the neonatal period and under starving conditions, was elevated only in cord blood in association with pBMI. Ketone bodies produced in maternal circulation are readily transported to the fetus as a source of energy. There are reports supporting free flow of ketone bodies from maternal to fetal circulation [31].

Influence of GDM on maternal and cord blood metabolome

High pBMI and GDM share some common associations with some metabolites in maternal/cord blood which can be explained on the same basis, since obesity also presents increased IR and altered glucose tolerance, typical of GDM [8]. Trends for higher Asn/Asp levels and lower levels of some Krebs cycle metabolites were encountered in cord blood, which agrees with previous reports on these metabolites in association with IR and hyperglycaemia during pregnancy and reduced use of these metabolites for Krebs cycle replenishment [20, 32].

In GDM, we found common metabolite associations in both maternal and cord blood. Expectedly, H1 was increased reflecting maternal hyperglycaemia and enhanced transplacental transfer of glucose. Glucose is transported down its concentration gradient from maternal to fetal circulation through “GLUT1-mediated transport” or facilitated glucose transporters in basal membrane (BM) of the syncytiotrophoblast (STB) epithelium of the human term placenta [33]. Increased GLUT1 expression and activity in the BM of STB in GDM women either treated with diet alone or with insulin therapy were reported [34]. This may lead to fetal hyperglycemia and hyperinsulinemia and was proposed as a mechanism causing fetal overgrowth in GDM [34]. Another important common finding is overall trend of reduced PL levels, especially maternal LPC and PCaa. Reduced synthesis of PC and, in turn, LPC, might explain this finding being a result of reduced hepatic de-novo synthesis of PC from phosphoethanolamines (PE) and S-adenosylmethionine (SAM), which is catalysed by phosphatidyl–ethanolamine methyl transferase (PEMT). To our understanding, the existing oxidative stress associated with GDM drives Cys metabolism towards glutathione biosynthesis as a defensive mechanism leading to a decrease in SAM and thus decreased PC and LPC [28, 35]. Decreased synthesis of PC, a major PL in the very low-density lipoprotein (VLDL) causes fat and cholesterol accumulation in the liver (hepatic steatosis) [35, 36]. These hypotheses, however, apply only to PCaa, since PCae did not behave similarly, therefore ΣPCaa/ΣPCae ratio was decreased in GDM mothers. This could be explained by considering that ether lipids represent only a small portion of the total PL (in our study, the total PCae concentration is < 10% of PCaa), and their intracellular levels are quite low in the liver [37], so their decreased synthesis might be too small to be detected.

Trends for negative association with SM might be explained by either reduced synthesis [20] or increased breakdown [38]. The reduced de-novo synthesis was explained by redirection of phosphatidic acid to triglycerides (TG) rather than PL synthesis [28]. As for the increased breakdown, it was reported that GDM is associated with increased breakdown of SM resulting in ceramides, which induce inflammation and β-cell apoptosis and are negatively correlated with insulin sensitivity [28, 38].

Only in cord blood, we observed a significant decrease in Carn levels and an overall decrease in AC levels. There were different reports on Carn levels. One study reported low Carn in children with T1D [39]. In GDM pregnancies, data on Carn are scarce, but increased Carn levels in GDM mothers [40] and their offspring have been reported without clear explanation [41].This correlation of fetal and maternal Carn levels was explained due to the fetus’s inability to synthetize Carn and its reliance on placental transfer [41]. Thus, one would expect that low fetal Carn would be secondary to low maternal levels. However, we found decreased fetal Carn with no differences for maternal Carn in the GDM group, along with a decrease in long-chain NEFA and fatty acid oxidation (FAO), as reflected in increased AC16:0/Carn and decreased AC2:0/AC16:0 ratios. A plausible explanation would be the reduced transport of Carn, and thus reduced initiation of FAO, as a consequence of fetal hyperglycemia. This matches the previously reported reduction in FAO, concomitant with elevated TG formation, in placentae of GDM mothers [42]. Thus, the whole picture suggests that GDM is associated with reduced placental transport of NEFAs and Carn along with incomplete or reduced FAO, which was reported in IR and diabetes [43].

Maternal blood lactic acid tended to be lower in GDM, although one might expect an elevation with increased IR, typical for GDM [44]. A plausible explanation would be the liver’s ability to utilize circulating lactate for reconversion to glucose through the Cori cycle [45].

The proposed mechanisms of metabolic alterations associated with obesity and GDM are depicted in Suppl. Fig. S1. Results from linear models for all investigated metabolites are listed in Suppl. Table S2.

Conclusion

Our study investigated the association of > 200 intermediates of energy metabolic pathways in maternal and cord blood with maternal BMI and GDM status. We observed consistent associations of BMI with BCAA as well as pronounced association of several markers related to inflammation and β-oxidation with GDM.