Abstract

Background

   Pregnant women with gestational diabetes mellitus (GDM) or type 2
   diabetes mellitus (T2DM) are at increased risks of pre-term labor,
   hypertension and preeclampsia. In this study, metabolic profiling of
   blood samples collected from GDM, T2DM and control pregnant women was
   undertaken to identify potential diagnostic biomarkers in GDM/T2DM and
   compared to pregnancy outcome.

Methods

   Sixty-seven pregnant women (21 controls, 32 GDM, 14 T2DM) in their
   second trimester underwent targeted metabolomics of plasma samples
   using tandem mass spectrometry with the Biocrates MxP^® Quant 500 Kit.
   Linear regression models were used to identify the metabolic signature
   of GDM and T2DM, followed by generalized linear model (GLMNET) and
   Receiver Operating Characteristic (ROC) analysis to determine best
   predictors of GDM, T2DM and pre-term labor.

Results

   The gestational age at delivery was 2 weeks earlier in T2DM compared to
   GDM and controls and correlated negatively with maternal HbA1C and
   systolic blood pressure and positively with serum albumin. Linear
   regression models revealed elevated glutamate and branched chain amino
   acids in GDM + T2DM group compared to controls. Regression models also
   revealed association of lower levels of triacylglycerols and
   diacylglycerols containing oleic and linoleic fatty acids with pre-term
   delivery. A generalized linear model ROC analyses revealed that that
   glutamate is the best predictors of GDM compared to controls (area
   under curve; AUC = 0.81). The model also revealed that
   phosphatidylcholine diacyl C40:2, arachidonic acid,
   glycochenodeoxycholic acid, and phosphatidylcholine acyl-alkyl C34:3
   are the best predictors of GDM + T2DM compared to controls
   (AUC = 0.90). The model also revealed that the triacylglycerols
   C17:2/36:4 and C18:1/34:1 are the best predictors of pre-term delivery
   (≤ 37 weeks) (AUC = 0.84).

Conclusions

   This study highlights the metabolite alterations in women in their
   second trimester with diabetes mellitus and identifies predictive
   indicators of pre-term delivery. Future studies to confirm these
   associations in other cohorts and investigate their functional
   relevance and potential utilization for targeted therapies are
   warranted.

   Keywords: Metabolomics, Pregnancy, Gestational diabetes, Type 2
   diabetes, Pre-term delivery

Background

   Gestational diabetes mellitus (GDM) represents any degree of glucose
   intolerance with onset during pregnancy, regardless of whether treated
   by insulin or diet modification, or whether the condition persists
   after pregnancy or not [[43]1]. It does not exclude the possibility
   that unrecognized glucose intolerance may have antedated or begun
   concomitantly with the pregnancy. GDM represents one of the most
   frequent complications in pregnancy [[44]2]  with a prevalence range of
   1–14% based on the diagnostic criteria, study population, ethnicity and
   geographical location [[45]3]. Its increasing prevalence has been
   attributed to the obesity epidemic among women of reproductive age
   [[46]4]. GDM is diagnosed when glucose levels are elevated in the late
   second trimester [[47]5]. Postpartum, 20% of women with GDM develop
   impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT),
   causing a 7.4 times higher risk of type 2 diabetes mellitus (T2DM)
   later in life compared to matching controls [[48]6] and have an
   increased risk of cardiovascular disease (CVD) [[49]7]. GDM has been
   associated with adverse outcomes including pre-term delivery,
   preeclampsia, macrosomia, perinatal mortality and neonatal metabolic
   complications [[50]8]. Spontaneous pre-term delivery has been linked to
   poor glycemic control and parity [[51]9–[52]12], and is associated with
   a higher risk for neonatal intensive care unit admission due to
   respiratory failure and hypoglycemia [[53]13]. It is also associated
   with chronic respiratory disease, ischemic heart disease and metabolic
   disorders [[54]14–[55]16]. These adverse outcomes of GDM have led
   clinicians to implement various strategies including fetal surveillance
   and induction of labor [[56]17, [57]18]. A number of risk factors for
   GDM have already been identified, including maternal age, family
   history of diabetes, pre-pregnancy obesity, and multiple pregnancies
   [[58]19]. However, the metabolic pathways in GDM and/or pre-gestational
   T2DM in pregnancy and their relationship to pregnancy outcomes is
   poorly understood. The identification of novel biomarkers may,
   therefore, have clinical diagnostic and therapeutic applications.

   Discovery of metabolic mediators underlying disease progression in
   obesity-associated insulin resistance and T2DM [[59]20] were
   facilitated by advancing metabolomic tools such as mass spectrometry
   (MS) technologies, providing a better understanding of the etiology of
   the disease. The metabolic signature differentiating healthy controls
   from individuals at higher risk of T2DM included various carbohydrates
   (e.g. glucose and fructose), lipids (e.g. phospholipids,
   sphingomyelins, and triglycerides), and amino acids (branched-chain
   amino acids, aromatic amino acids, glycine, and glutamate)
   [[60]20–[61]22]. The pathophysiological changes that occur in
   pre-gestational T2DM and GDM are similar [[62]23]. However, metabolomic
   studies aimed at predicting risk of GDM in pregnant women have shown
   inconsistent findings, as some reported lower blood creatinine,
   trimethylamine-N-oxide, and betaine, while others reported elevated
   acetylcarnitines, bile acids, ketones, creatinine, carbohydrate, and
   other lipids and organic acids [[63]24]. However, few studies have
   investigated the association between these metabolic markers and risk
   of pre-gestational T2DM and GDM-associated pathologies including
   pre-term labor [[64]25].

   The aim of this study was to perform targeted metabolomics analysis of
   blood samples from pregnant women in their second trimester with
   pre-gestational T2DM, GDM or matching healthy controls to investigate
   the metabolic pathways underlying these pathologies and identify
   potential predictors of increased risk of pre-term labor.

Methods

Study design

   This was a cross sectional study in 67 pregnant women (21 controls, 32
   GDM, 14 T2DM) who were recruited during their second trimester at the
   antenatal clinic at The Women Wellness and Research Center of Hamad
   Medical Corporation (GDM and pre-gestational T2DM) in Doha, Qatar.
   Protocols were approved by Institutional Review Boards (IRBs) of the
   Hamad Medical Corporation (15101/15) and Weill Cornell Medical College
   in Qatar (15-00016). Demographics, anthropometrics and medical history
   data were collected including age, ethnicity, socio-economic
   background, vital signs, height, weight, menstrual cycle, period of
   infertility, medications, complications, comorbidities and family
   medical history. All pregnant women are screened in the first antenatal
   care visit using fasting blood glucose (FBG). If the FBG at the first
   visit is < 5.1 mmol/l (92 mg/dl); 75 g oral glucose tolerance test
   (OGTT) is performed at 24 weeks’ gestation. The world health
   organization (WHO) criteria [FBG ≥ 5.1 mmol/L (92 mg/dl), 1 h post
   OGTT ≥ 10.0 mmol/L (180 mg/dl) or 2 h post
   OGTT ≥ 8.5 mmol/L(153 mg/dl)] is used to diagnose GDM. GDM patients
   were started on diet for 2 weeks with the aim of a FBG ≤ 5.3 mmol/l
   (95 mg/dl) and the 2 h post prandial glucose being ≤ 6.8 mmol/l
   (120 mg/dl) in ≥ 80% of the readings. If more than 20% of the readings
   were above target then Metformin therapy was implemented and increased
   incrementally followed by insulin supplementation when glucose targets
   were not achieved. Women with Type 2 diabetes were all treated with
   Metformin and basal-bolus insulin.

   Laboratory tests included second trimester full blood count,
   biochemical profile and thyroid function tests. Blood samples were
   collected for the metabolomics analysis. All patients gave their
   written informed consent and the conduct of the study was in accordance
   with the International Council for Harmonisation Good Clinical Practice
   and the Declaration of Helsinki. Pregnancy outcomes of gestational age
   at delivery, birthweight, maternal weight, blood pressure and foetal
   outcome were recorded and collated with the metabolomic profile for all
   subjects who participated in the study.

Metabolomics

   Targeted metabolomics of plasma samples was performed using tandem mass
   spectrometry with the Biocrates MxP^® Quant 500 Kit (Biocrates,
   Innsbruck, Austria) at the Fraunhofer Institute for Toxicology and
   Experimental Medicine (ITEM). Lipids were measured by Flow Injection
   Analysis Tandem Mass Spectrometry (FIA-MS/MS) using a 5500 QTRAP^®
   instrument (AB Sciex, Darmstadt, Germany) with an Electrospray
   ionization (ESI) source, and small molecules were measured by Liquid
   chromatography–mass spectrometry (LC–MS/MS) using the same 5500 QTRAP^®
   instrument as previously described [[65]26]. Briefly, a 96-well based
   sample preparation device was used to quantitatively analyze the
   metabolite profile in the samples. This device consists of inserts that
   have been spotted with internal standards, and a predefined sample
   amount was added to the inserts. Next, a phenylisothiocyanate solution
   was added to derivatize some of the analytes (e.g. amino acids), and
   after the derivatization was completed, the target analytes were
   extracted with an organic solvent, followed by a dilution step. The
   obtained extracts were then analyzed by FIA-MS/MS and LC–MS/MS methods
   using multiple reaction monitoring (MRM) to detect the analytes. Data
   were quantified using appropriate MS software (Sciex Analyst^®) and
   imported into Biocrates MetIDQ™ software for calculating analyte
   concentrations, data assessment and compilation.

Statistical analysis

   Demographics traits analysis: Statistical analyses were carried out
   using IBM SPSS version 25, R version 3.2.1 and SIMCA 14 software
   (Umetrics, Sweden). Variables with skewed distributions were log
   transformed to ensure normality [[66]27]. Comparisons were performed
   with t-test, Wilcoxon–Mann–Whitney and 1-way ANOVA as appropriate.
   Significance was defined as P ≤ 0.05. Non-parametric tests were used
   for comparing ordinal or non-normal variables. Metabolomics data
   analysis: Principle component analysis (PCA) was performed using R
   version 2.14, [67]www.r-project.org/. PCA revealed two main components
   (PC1 and PC2) that together captured 24% of the variance in the data.
   Orthogonal partial least square discriminant analysis (OPLS-DA),
   implemented as part of the software SIMCA, was used to compare
   controls, GDM and T2DM groups. OPLS-DA is recommended in cases of
   regression where the number of explanatory variables is high, and where
   it is likely that the explanatory variables are correlated as it is the
   case in our data. All metabolites with a percentage of missing values
   greater than 50% were excluded from SIMCA analysis. Linear regression
   was performed to identify significant metabolites differentiating study
   groups (controls vs GDM and T2DM) and (Controls vs GDM + T2DM) using
   the R statistical package (version 2.14, [68]www.r-project.org/) after
   correcting for age, BMI and principle components (PC1 and PC2).
   Additionally, the gender interaction effect was evaluated in ANOVA
   model that featured the same confounders. Contrast analysis was
   conducted using R package Emmeans to pinpoint the significance of
   effect per gender group. Function enrichment analysis was performed
   using Fisher’s exact test by considering metabolites with a nominal
   p-value less than 0.1 from linear regression analysis. For a given
   biological function, the test assesses the probability of observing the
   associated nominally-significant metabolites from the linear model by
   pure chance. The biological categories tested for enrichment were
   provided by Biocrates and expanded manually by reference to the Human
   Metabolome Database [[69]28]. The Elastic net regularization of linear
   models, implemented in R package GLMNET, was used for selection of best
   predictors of clinical traits of interest to this study.

Results

General characteristics of participants

   67 young (31.9 ± 5.7 years) overweight/obese (31.9 ± 6.9 kg/m^2)
   pregnant women were included in this study. All included pregnancies
   resulted in single births. The T2DM women were older, had higher
   triacylglycerols (TG), insulin and glycated haemoglobin (HbA1c), than
   body mass index (BMI)-matched controls and GDM (p < 0.05)
   (Table [70]1). Gestational age (GA) at delivery was on average 2 weeks
   earlier in T2DM than controls and GDM (p = 0.01) (Fig. [71]1a). GA at
   delivery was negatively correlated with HbA1c at baseline (R = − 0.34,
   p = 0.01, Fig. [72]1b) and with systolic blood pressure (SBP) at
   baseline (R = − 0.34, p = 0.005, Fig. [73]1c), but positively
   correlated with serum albumin (R = 0.39, p = 0.002, Fig. [74]1d). A
   regression model, incorporating explanatory variables (study group,
   HbA1c, SBP and serum albumin), indicated independent significant
   association of all four covariates with GA at delivery (study group:
   Beta = − 1.7, HbA1c: Beta: − 0.85, p = 0.01, SBP: Beta = − 0.04,
   p = 0.03, serum albumin: Beta = 0.06, p = 0.07) (Fig. [75]1). There was
   no difference in the proportion of male to female offspring was
   identified in the control group (11F vs 10 M). However, a significant
   interaction with offspring gender in the GDM and T2DM groups (p = 0.02)
   was identified, with a higher proportion of female than male offspring
   in GDM participants (21F vs 11 M), whilst the opposite was true with
   T2DM participants (3F vs 11 M).

Table 1.

   General characteristics of participants
   Time Variables Total Controls GDM T2DM F P value DM F P value
   (N = 67) (N = 21) (N = 32) (N = 14) ANOVA (N = 46) Controls vs DM
   Second trimester Age (years) 31.9 (5.7) 29.2 (5.1) 32.8 (5.6) 34.2
   (5.5) 4.243 0.019 33.2 (5.6) 0.187 0.007
   BMI (kg/m2) 31.9 (6.8) 29.9 (8.4) 33.2 (6.1) 32.3 (5.0) 1.545 0.221
   32.9 (5.7) 2.462 0.089
   SBP (mmHg) 111.6 (12.4) 109.9 (12.0) 109.7 (11.9) 118.5 (12.5) 2.913
   0.062 112.3 (12.6) 0.017 0.459
   DBP (mmHg) 63.7 (7.7) 66.5 (9.3) 62.3 (5.7) 62.4 (8.2) 2.223 0.117 62.3
   (6.5) 0.110 0.037
   Cholesterol (mmol.L) 4.8 (1.1) 5.0 (1.0) 4.7 (0.8) 4.9 (1.6) 0.449
   0.783 4.7 (1.1) 4.220 0.552
   Triglycerides (mmol.L) 1.5 (0.8) 1.2 (0.6) 1.4 (0.9) 2.1 (0.5) 4.384
   0.024 1.6 (0.8) 0.019 0.090
   HDL (mmol.L) 1.4 (0.4) 1.6 (0.4) 1.3 (0.3) 1.1 (0.5) 3.101 0.058 1.2
   (0.4) 0.682 0.030
   LDL (mmol.L) 2.8 (0.9) 2.9 (0.9) 2.8 (0.7) 2.8 (1.4) 0.004 0.996 2.8
   (1.0) 1.757 0.933
   Insulin (uIU.L) 0. 5 (0.4) 0.3 (0.3) 0.5 (0.3) 0.7 (0.5) 4.026 0.023
   0.5 (0.4) 8.120 0.048
   HbA1C 5.4 (0.7) 5.1 (0.4) 5.3 (0.4) 6.0 (0.8) 11.976 0.000 5.5 (0.7)
   2.943 0.02
   Vitamin D3 15.2 (6.0) 15.8 (5.6) 15.1 (6.8) 14.2 (4.3) 0.220 0.804 14.9
   (6.2) 0.727 0.600
   Urea (nmol.L) 3.1 (1.2) 3.2 (1.0) 3.2 (1.5) 2.6 (0.4) 2.081 0.241 3.0
   (1.2) 4.835 0.445
   Creatinine (nmol.L) 50.1 (9.1) 49.9 (8.1) 49.4 (10.1) 51.7 (8.9) 0.302
   0.741 50.2 (9.7) 0.058 0.890
   ALT 15.2 (10.1) 13.2 (10.7) 17.0 (11.4) 14.6 (5.0) 0.894 0.414 16.2
   (9.7) 0.609 0.269
   AST 16.8 (7.1) 16.3 (5.8) 18.1 (7.9) 14.9 (7.2) 1.010 0.370 17.0 (7.7)
   0.248 0.700
   Bilirubin (umol.L) 7.6 (4.4) 7.8 (2.4) 7.9 (4.9) 6.8 (5.6) 0.305 0.721
   7.5 (5.1) 2.879 0.813
   ALP (iu.L) 83.2 (35.5) 81.5 (41.7) 86.4 (34.9) 79.6 (27.7) 0.204 0.817
   84.1 (32.4) 3.058 0.791
   Albumin (g.L) 34.7 (6.7) 36.5 (6.7) 35.0 (7.0) 31.7 (5.2) 2.251 0.114
   33.9 (6.6) 2.328 0.145
   TSH (mU.L) 2.4 (3.4) 2.8 (3.5) 2.6 (4.0) 1.5 (0.7) 0.604 0.579 2.3
   (3.4) 1.698 0.624
   Thyroxine (pmol.L) 12.6 (1.8) 12.9 (2.0) 12.4 (1.7) 12.6 (1.9) 0.283
   0.666 12.4 (1.7) 0.065 0.388
   TP (g.L) 66.3 (5.4) 65.1 (5.2) 67.0 (5.7) 66.8 (4.8) 0.839 0.437 67.0
   (5.4) 0.023 0.199
   Gestational age (weeks) 20.9 (4.4) 20.9 (4.9) 20.6 (4.3) 21.6 (3.9)
   0.258 0.774 20.9 (4.2) 0.753 0.947
   At birth SBP (mmHg) 119.9 (11.8) 118.7 (11.8) 119.4 (9.5) 122.9 (16.4)
   0.566 0.571 120.4 (11.8) 2.875 0.576
   DBP (mmHg) 72.1 (9.6) 70.4 (8.5) 72.4 (8.1) 74.1 (14.3) 0.618 0.542
   72.9 (10.1) 6.333 0.328
   HbA1c 5.6 (0.9) N/A 5.2 (0.4) 6.0 (1.3) 5.831 0.02 5.6 (0.9) 1.991 N/A
   Weight (Kg) 82.8 (14.0) 80.9 (17.0) 83.5 (12.7) 84.2 (12.6) 0.296 0.745
   83.7 (12.6) 0.588 0.450
   Gestational age (weeks) 38.2 (1.6) 38.8 (1.2) 38.4 (1.4) 36.8 (1.6)
   9.276 0.000 37.9 (1.6) 0.229 0.020
   Newborn weight (grams) 3073.9 (527.7) 3088.8 (492.2) 3110.3 (561.6)
   2968.2 (523.1) 0.358 0.700 3067.0 (548.3) 0.356 0.877
   [76]Open in a new tab

   BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic
   blood pressure; LDL, low density lipoprotein; HDL, high density
   lipoprotein; HbA1c, glycated haemoglobin; TP, total protein; ALP,
   alkaline phosphatase; ALT, alanine transaminase; AST, aspartate
   aminotransferase. Data are presented as mean (SD). Differences between
   Controls, GDM and T2DM were tested by ANOVA. Differences between
   controls and all DM (GDM + T2DM) were tested by independent sample t
   test (normally distributed variables) or Mann–Whitney U (variables with
   skewed distribution) test. A p-value significance level of 0.05 was
   used

Fig. 1.

   [77]Fig. 1
   [78]Open in a new tab

   Differences in gestational age (GA) at delivery in controls,
   gestational diabetes (GDM) and type 2 diabetes (T2DM) groups (a).
   Correlation of gestational age at delivery with baseline HbA1c (b)
   systolic blood pressure (c) and serum albumin (d). GA at delivery was
   compared among controls, GDM and T2DM groups using ANOVA. Pearson’s
   correlation was used to assess the relationship between GA at delivery
   and serum HbA1c, albumin and systolic blood pressure measured at third
   trimester. p-value significance level of 0.05 (*) was used

A holistic view of the metabolic profile of GDM and T2DM

   OPLS-DA comparing the multivariate metabolic profile of controls, GDM
   and T2DM was used. OPLS-DA showed one class-discriminatory components
   accounting for 25% of the variation in the study group Y-variable. The
   Variable Importance in Projection (VIP) list emphasized the
   discriminatory effects of various glycerophospholipids
   (phosphatidylcholine acyl-alkyl, PC.ae) such as PC.ae.C34.3,
   PC.ae.C34.2 and PC.ae.C42.3, triacylglycerols (TG16.0/34.1, TG16.0/34.0
   and TG18.1/32.0) and the amino acids, such as glutamate (Glu),
   aspartate (Asp), isoleucine (Ile) and arachidonic acid (AA) (Additional
   file [79]1: Table S1). The score plot in Fig. [80]2a indicates an
   x-axis that separates the controls from the GDM and T2DM. The
   corresponding loading plot, shown in Fig. [81]2b, indicates the
   aforementioned metabolites from the VIP list responsible for the
   groups’ separation. When GDM + T2DM (DM) were combined into one group,
   OPLS-DA showed one discriminatory component accounting for 92% of the
   variation in the control/combined DM group (Fig. [82]2c). The VIP list
   highlighted amino acids (Glu, leucine (Leu), valine (Val), Ile and
   Asp), triacylglycerols (TG16.0/34.2, TG16.0/34.1, TG18.2/32.0 and
   TG18.1/32.0) and the glycerophospholipid (C34:3) (Additional file
   [83]1: Table S2) [[84]29], also indicated in the loading plot
   (Fig. [85]2d).

Fig. 2.

   [86]Fig. 2
   [87]Open in a new tab

   Orthogonal partial least square discriminate analysis (OPLS-DA)
   comparing metabolites from controls, GDM and T2DM individuals (top) and
   controls versus combined GDM + T2DM (bottom). Score plots from the two
   models respectively showing one dimensional separation of all three
   classes (control, GDM, T2DM) (a) and two classes (controls and
   GDM + T2DM) (c). The corresponding loading plots showing top associated
   metabolites differentiating controls, GDM and T2DM (b) or GDM + T2DM vs
   Controls (d) groups

Metabolites differentiating controls from GDM + T2DM

   Following multivariate analysis, univariate regression models were used
   to characterize metabolite-level associations with GDM and DM
   (GDM + T2DM). The analysis revealed significant differences (false
   discovery rate (FDR) ≤ 0.05) in the levels of five metabolites between
   controls and T2DM (Additional file [88]1: Table S3). These include four
   amino acids (Asp, Glu, citrulline (Cit) and Ile) and the
   glycerophospholipid (PC.ae.C34.3). A number of metabolites
   differentiated controls from GDM, but none reached FDR level of
   significance (Additional file [89]1: Table S4). When comparing control
   vs DM, the linear model revealed a list of metabolites differentiating
   the two groups, with only Glu reaching FDR level of significance
   (FDR = 0.04) (Additional file [90]1: Table S5). Table [91]2 lists the
   top 10 metabolites differentiating control and DM with their sub and
   super-pathways. Enrichment analysis indicated that branched chain amino
   acids (BCAA) were elevated in DM compared to controls (p = 0.006)
   (Fig. [92]3).

Table 2.

   Top ten metabolites differentiating controls from DM (GDM + T2DM) in
   women in their second trimester
   Metabolite Biochemical ID Super pathway Sub pathway Estimate SE P value
   FDR
   Glu Glutamate Amino Acids Glutamate Metabolism 15.8 3.8 < 0.001 0.04
   PC.ae.C34.3 Phosphatidylcholine C34:3 Lipid Glycerophospholipids − 1.5
   0.4 < 0.001 0.13
   Ile Isoleucine Amino Acids Leucine, Isoleucine and Valine Metabolism
   10.9 3.3 0.002 0.25
   Asp Aspartate Amino Acids Alanine and Aspartate Metabolism 1.3 0.5
   0.005 0.29
   Leu Leucine Amino Acids Leucine, Isoleucine and Valine Metabolism 14.8
   5.7 0.011 0.29
   Val Valine Amino Acids Leucine, Isoleucine and Valine Metabolism 23.5
   8.9 0.011 0.29
   ProBetaine Proline betaine Amino-Acid-related Xenobiotic − 2.5 1 0.011
   0.29
   AconAcid Aconitate [cis or trans] Carboxylic acids TCA Cycle 0.3 0.1
   0.009 0.29
   DG.16.0/18.1 Palmitoyl-oleoyl-glycerol Lipid Diacylglycerol 1.7 0.6
   0.009 0.29
   DG.16.0/18.2 Palmitoyl-Linoleic acid-glycerol Lipid Diacylglycerol 1.1
   0.4 0.006 0.29
   AA Arachidonate (20:4n6) Fatty Acids Long Chain Polyunsaturated Fatty
   Acid 0.4 0.1 0.003 0.29
   [93]Open in a new tab

   Linear regression was performed to identify significant metabolites
   differentiating Controls from GDM + T2DM (DM) using the R statistical
   package after correcting for age, BMI and principle components (PC1 and
   PC2). Estimate (beta value), SE, standard error; FDR, False Discovery
   Rate

Fig. 3.

   [94]Fig. 3
   [95]Open in a new tab

   Boxplot of metabolites differentiating Control from DM (GDM + T2DM)
   groups. A Function enrichment analysis was performed using Fisher’s
   exact test by considering metabolites with a nominal p-value less than
   0.1 from linear regression analysis. p-value significance level of 0.05
   (*) was used

   Furthermore, offspring gender interaction analysis revealed nominally
   significant association between metabolites and GDM + T2DM which were
   significantly different in pregnancies with male versus female
   offspring, including the triglycerides (18:2/31:0, 16:1/36:5,
   17:1/36:4, 16:0/35:3, 18:2/33:0), the amino acids/related (glutamine
   and ProBetaine) and the cholesterol ester CE18:3 (Additional file
   [96]2: Figure S1).

Metabolites associated with GA at delivery

   Similarly, a linear model was used to assess the significance of
   metabolites associated with GA at delivery. One hundred and eleven
   metabolites exhibited significant association at FDR level of
   significance ≤ 0.05. The list of metabolites and their associated
   pathways are shown in Additional file [97]1: Table S6. Among these, 22
   lipids were associated with gestational age at delivery at FDR ≤ 0.05
   level of significance, including triacylglycerols, diacylglycerols and
   bile acids (Table [98]3). Among these, triacylglycerols and
   diacylglycerols containing C18:1 and C18:2 were enriched in pre-term
   deliveries (p < 0.01). Gender interaction analysis revealed three FDR
   significant metabolites where the slope of the regression line was
   significantly different between males and females in relation to GA at
   delivery. These include the sphingolipids (SM) C24:1 and C16:1 and the
   amino acid-related Taurine (Fig. [99]4). No significant associations
   were identified between metabolites and other adverse pregnancy
   outcomes, including pre-eclampsia, intrauterine fetal death,
   macrosomia, and maternal blood pressure at delivery (data not shown).

Table 3.

   Metabolites associated with gestational age at delivery
   Metabolite Biochemical ID Super pathway Sub pathway Estimate SE P value
   FDR
   TG17.2/36.4 TG17:2/36:4 Lipid Triaclylgerol − 0.4 0.1 < 0.001 0.038
   DG.18.1/18.1 DG18:1/18:1 Lipid Diacylglycerol − 1.7 0.5 0.002 0.044
   DG.18.1/18.2 DG18:1/18:2 Lipid Diacylglycerol − 4.4 1.2 0.001 0.044
   DG.18.2/18.2 DG18:2/18:2 Lipid Diacylglycerol − 2.8 0.9 0.002 0.044
   TG16.0/36.3 TG16:0/36:3 Lipid Triaclylgerol − 157.6 43.0 0.001 0.044
   TG16.1/36.3 TG16:1/36:3 Lipid Triaclylgerol − 12.8 3.8 0.001 0.044
   TG18.0/36.3 TG18:0/36:3 Lipid Triaclylgerol − 7.9 2.3 0.001 0.044
   TG18.1/34.1 TG18:1/34:1 Lipid Triaclylgerol − 194.1 57.2 0.001 0.044
   TG18.1/34.2 TG18:1/34:2 Lipid Triaclylgerol − 132.1 39.4 0.001 0.044
   TG18.1/36.1 TG18:1/36:1 Lipid Triaclylgerol − 14.0 4.3 0.002 0.044
   TG18.1/36.2 TG18:1/36:2 Lipid Triaclylgerol − 53.1 15.5 0.001 0.044
   TG18.1/36.3 TG18:1/36:3 Lipid Triaclylgerol − 61.7 16.2 < 0.001 0.044
   TG18.2/34.1 TG18:2/34:1 Lipid Triaclylgerol − 125.3 36.6 0.001 0.044
   TG18.2/36.0 TG18:2/36:0 Lipid Triaclylgerol − 1.0 0.3 0.002 0.044
   TG18.2/36.1 TG18:2/36:1 Lipid Triaclylgerol − 9.7 3.0 0.002 0.044
   TG18.2/36.2 TG18:2/36:2 Lipid Triaclylgerol − 31.2 9.6 0.002 0.044
   TG18.2/38.4 TG18:2/38:4 Lipid Triaclylgerol − 1.9 0.5 0.001 0.044
   TG20.0/32.4 TG20:0/32:4 Lipid Triaclylgerol − 0.5 0.1 0.002 0.044
   TG20.4/36.2 TG20:4/36:2 Lipid Triaclylgerol − 3.7 1.1 0.001 0.044
   GCA Glycocholic acid Lipid Bile Acid − 0.2 0.1 0.002 0.046
   TG18.0/36.4 TG18:0/36:4 Lipid Triaclylgerol − 3.8 1.2 0.002 0.046
   [100]Open in a new tab

   Linear regression was performed to identify significant metabolites
   associated with pre-term delivery using the R statistical package after
   correcting for age, BMI and principle components (PC1 and PC2).
   Estimate (beta value), SE, standard error; FDR, False Discovery Rate

Fig. 4.

   [101]Fig. 4
   [102]Open in a new tab

   Gender specific associations with gestational age at delivery

Predictive biomarkers association with GDM, DM and GA at delivery

   We used an extension of the generalized linear model that features the
   elastic net regularization algorithm for variable selection, also known
   as GLMNET in R analysis software. The GLMNET model indicated that Glu
   (beta = 0.09, p = 0.001) was the best predictors of GDM compared to
   controls (AUC = 0.81) (Fig. [103]5a). The model also revealed that the
   phosphatidylcholine diacyl (PC.aa.C40.2) (beta = − 17.7, p = 0.05),
   arachidonic acid (AA) (beta = 3.2, pp = 0.003), glycochenodeoxycholic
   acid (GCDCA) (beta = 3.4, p = 0.02), and PC.ae.C34.3 (beta = 0.70
   p = 0.02) were the best predictors of all DM compared to controls
   (AUC = 0.90) (Fig. [104]5b). The model also revealed that TG17.2/36.4
   (beta = − 2.79, p0.03) and TG18.1/34.1 (beta = − 0.006, p = 0.01) were
   the best predictors of pre-term delivery (≤ 37 weeks) (AUC = 0.84)
   (Fig. [105]5c).

Fig. 5.

   [106]Fig. 5
   [107]Open in a new tab

   Predictors of GDM, DM and pre-term delivery. ROC curve indicates the
   best predictors of GDM (a), DM (b) and pre-term delivery (c) indicating
   the area under curve (AUC) for each model

Discussion

   The diabetes epidemic constitutes a global public health challenge.
   Considering the adverse effects of DM on pregnancy outcomes, perinatal
   morbidity, and development of chronic diseases later in life, a better
   understanding of the metabolic mediators underlying these adverse
   effects could potentially provide novel diagnostic and therapeutic
   targets. In this study, targeted metabolomics of plasma samples from 67
   pregnant women at second trimester was performed. Five metabolites
   exhibited significant differences between controls and T2DM, including
   four amino acids (Asp, Glu, Cit and Ile) and the glycerophospholipid
   (PC.ae.C34.3), whereas Glu was the only metabolite that significantly
   differentiated DM from controls. Glu was also found to be the best
   predictor of GDM by the GLMNET model confirmed by ROC analysis. The
   model also indicated that PC.aa.C40.2, AA, GCDCA, and PC.ae.C34.3 were
   the best predictors of all DM compared to controls. Our data also
   indicated that GA at delivery was on average 2 weeks earlier in T2DM
   than controls and GDM groups, which correlated negatively with HbA1c
   and SBP and positively with serum albumin at second trimester. GLMNET
   model confirmed by ROC analysis revealed that TG17.2/36.4 and
   TG18.1/34.1 were the best predictors of pre-term delivery (≤ 37 weeks).
   The potential functional relevance of identified metabolites in
   relation to diabetes and pre-term delivery is summarized in
   Table [108]4.

Table 4.

   The potential functional relevance of identified metabolites associated
   with diabetes and pre-term delivery
   Metabolite Association Potential relevance to pathophysiological
   aspects of diabetes and pre-term delivery References