Metabolic dysfunction in pregnancy: Fingerprinting the maternal metabolome using proton nuclear magnetic resonance spectroscopy

Abstract Aims Maternal metabolic disorders place the mother at risk for negative pregnancy outcomes with potentially long‐term health impacts for the child. Metabolic syndrome, a cluster of features associated with increased risk of metabolic disorders, such as cardiovascular disease, diabetes and stroke, affects roughly one in five Canadians. Metabolomics is a relatively new technique that may be a useful tool to identify women at risk of metabolic disorders. This study set out to characterize urinary metabolic biomarkers of pregnant women with obesity and of pregnant women who later developed gestational diabetes mellitus (pre‐GDM), compared to controls. Methods and Materials Second trimester urine samples were collected through the Alberta Pregnancy Outcomes and Nutrition (APrON) cohort and examined with 1H nuclear magnetic resonance (NMR) spectroscopy. Multivariate analysis was used to examine group differences, and machine learning feature selection tools identified the metabolites contributing to separation. Results Obesity and pre‐GDM metabolomes were distinct from controls and from each other. In each comparison, the glycine, serine and threonine pathways were the most impacted. Pantothenate, formic acid and glycine were downregulated by obesity, while formic acid, dimethylamine and galactose were downregulated in pre‐GDM. The three most impacted metabolites for the comparison of obesity versus pre‐GDM groups were upregulated creatine/caffeine, downregulated sarcosine/dimethylamine and upregulated maltose/sucrose in individuals who later developed GDM. Conclusion These findings suggest a role for urinary metabolomics in the prediction of GDM and metabolic marker identification for potential diagnostics and prognostics in women at risk.


| INTRODUC TI ON
An estimated 20% of Canadians are currently diagnosed with metabolic syndrome, a cluster of features that puts an individual at higher risk of developing diabetes and cardiovascular disease (CVD). 1 During pregnancy, metabolic disorders such as obesity and gestational diabetes mellitus (GDM) have potentially adverse long-term consequences for both mother and child, such as increasing the risk of preeclampsia, preterm birth, caesarean section and neonatal intensive care unit admissions. 2,3 Obesity can be a consequence of metabolic dysfunction, or a precursor to metabolic syndrome as it increases an individual's risk of developing other metabolic conditions, such as GDM or hypertension. 3 During pregnancy, the recommended weight gain for a woman of normal weight is between 25 and 35 pounds (11-16 kg), this amount decreases with higher pre-pregnancy body mass index (BMI). 4 Weight gain beyond these recommendations, or preexisting obesity, may lead to adverse pregnancy outcomes such as preeclampsia, miscarriage, congenital anomalies, preterm birth and/or foetal complications (eg, macrosomia). 5 Furthermore, poor post-partum weight loss serves as a predictor of future obesity. 6 Gestational diabetes mellitus is characterized by glucose intolerance with onset or first recognition during pregnancy, and it affects an estimated 3.7% of all pregnancies in Alberta. 7 Despite its prevalence, the initial pathogenic mechanisms of GDM are not fully understood. While GDM is strongly associated with the development of type 2 diabetes later in life, many individuals diagnosed with GDM have no prior known metabolic dysfunction. 8 Inter-and transgenerational inheritance of GDM risk, however, has previously been suggested. 9,10 While both GDM and obesity are individually associated with adverse pregnancy outcomes, in conjunction they have synergistic effects. Hence, the 2012 Hyperglycemia and Adverse Pregnancy Outcome (HAPO) study reported that mothers with both GDM and obesity had significantly increased birth weight, newborn body fat, caesarean delivery rates and prevalence of preeclampsia when compared to individuals with only one, or neither, risk factor. 11 Despite the increased risk of the said adverse health outcomes associated with metabolic syndrome, obesity or GDM, not all individuals with either condition go on to develop these disorders.
The underlying mechanisms that cause conditions to worsen in some individuals but not others are not fully understood; however, cross-sectional analysis of at-risk individuals may address this gap by identifying factors of risk that have not yet been considered. [12][13][14] In many cases, such as prediabetes, mitigation of these factors can improve health outcomes, and even prevent the disease state. 13,15 Metabolomics, the study of the metabolism and associated pathways, presents an effective tool for biomarker discovery to identify high-risk individuals for targeted interventions and disease prevention. 16,17 A recent study investigating the early pregnancy serum metabolomic profile of overweight and obese women identified several biomarkers that predict risk of developing GDM,these included small high-density lipid (HDL) particles, branched chain amino acids (BCAAs) and inflammatory markers. 18 The present study aims to identify robust biomarkers associated with obesity and GDM status in the urinary metabolome. Urinary metabolomics enables the noninvasive detection of metabolic diseases during pregnancy without the need for a blood test. This study uses 1 H nuclear magnetic resonance (NMR) spectroscopy to investigate metabolomic signatures linked to obesity and GDM in the urine of pregnant women collected in the second trimester from the Alberta Pregnancy Outcomes and Nutrition (APrON) study. 19 We hypothesized that women with obesity or GDM will show characteristic urinary metabolomic signatures that differ from controls.

| Study design
Urine samples were collected from pregnant women in the Alberta Pregnancy Outcomes and Nutrition (APrON) study, a Canadian pregnancy cohort study. 20 The APrON study was created to examine the links between perinatal nutrition intake and birth outcomes, child development, and maternal mental health. 19 The full cohort consists of 2140 women, 2172 infants and 1417 biological fathers recruited in Calgary (population 1.1 million) and Edmonton (population 0.9 million), Alberta, Canada. The methodology details for the APrON study have been published elsewhere. 19,20 The present study used urine samples collected between 14 and 27 weeks of pregnancy from 29 women with obesity, 37 with GDM and 36 healthy controls. The three groups were identified as follows: (a) pregnant women (nondiabetic) with obesity were classified according to BMI status >30, or waist circumference >30); (b) pregnant women with GDM were extracted from hospital records whose diagnosis of GDM came from routine prenatal screening for GDM personalized medicine, pregnancy, preterm birth, urine GDM in the selected cases. Nine of the 37 GDM cases (23.7%) had a pre-pregnancy BMI >30. Controls were matched by age-, incomeand education level. Table 1 provides the characteristics of the participants in this study.

| Sample collection and preparation
Urine samples were obtained midstream during the first passage of the day following an overnight fasting period. The samples were stored at −80°C. Samples were thawed at room temperature, and The urine/buffer mixture was gently vortexed until homogenous and then centrifuged at 10 600 g for 5 minutes at 4°C. 550 μL of supernatant was pipetted into 5 mm NMR tubes before proceeding with 1 H-NMR spectroscopy.

| NMR data acquisition and processing
Nuclear magnetic resonance spectra of the urine samples were acquired at room temperature using a 700 MHz Bruker Avance III HD NMR Spectrometer (Bruker) equipped with a 5 mm triple resonance TBO-Z probe. The one-dimensional NOESY gradient water suppression pulse sequence (noesygppr1d) was utilized with the following parameters: mixing time of 10 ms; 128 k data points (TD); sweet width (SW) of 20.52 ppm, acquisition time (AQ) of 4.56 seconds, transmitter offset (o1p) of 4.6 ppm; recycle delay (D1) of 1 second; 128 scans (NS). Spectra were then processed using zero-filling to 256 k points, line broadening with a 0.3 Hz exponential multiplication, automatic phased and baseline-correction, and chemical shift referenced with respect to the TSP peak at 0 ppm. All spectra were converted to ascii files and exported to MATLAB (MathWorks) for further analysis. The spectra underwent dynamic adaptive binning, 21 followed by manual adjustment to correct for any errors in the algorithm. The bins containing the water and urea peaks were removed resulting in a total of 277 bins for all urine spectra. The spectra were normalized to the total area of all bins (the total metabolome), log transformed and Pareto scaled. 22

| Statistical analyses
For each comparison (obese vs control; GDM vs control; obese vs GDM), bins underwent both univariate and multivariate testing.
Univariate testing provides a statistical measure by which each of the spectral bins can be tested to determine whether it has been significantly altered across the comparison groups on an individual basis. Multivariate testing offers a method by which each of the bins can be statistically assessed with respect to their importance to class separation when considered as part of a complete set of variables.
Thus, univariate and multivariate testing provide complementary information about the importance of a bin, or metabolite, to observed group differences.

| Univariate testing
The decision tree algorithm outlined by Goodpaster et al 23

was utilized
to ensure that the appropriate univariate test was applied to the bins.
This decision tree algorithm first uses a Shapiro-Wilk test to determine whether the bins are normally distributed. If the bins follow a non-normal distribution, they undergo a Mann-Whitney U test (MW) for significance. In the case of this study, the data were determined to follow non-normal distributions and the Mann-Whitney U test was applied.

| Multivariate testing
Variable Importance Analysis based on random Variable Combination (VIAVC) 24 was utilized to assess variable significance when considered

| RE SULTS
Of the 277 total spectral bins, both VIAVC and MW tests identi- , and obesity vs GDM groups (C). A higher value on the y-axis indicates a lower P-value for the pathway. The x-axis gives the pathway impact, which indicates how affected the pathway is by the metabolites identified as significantly altered. Only metabolic pathways with P < .05 are labelled. This figure was created using the lists of metabolites identified as significantly altered between the obese and control groups by either MW or VIAVC testing (P < .05), starch and sucrose metabolism (P < .05), and the citrate cycle (P < .05).  While multiple metabolic pathways were impacted by the presence of obesity and GDM, altered glycine, serine and threonine metabolism pathway function was among the most distinct signatures for both conditions. In this pathway, serine is derived from glycolysis and in turn is converted into glycine. Threonine is an essential amino acid derived from diet, which is also converted into glycine. 37 Glycine deficiency in particular has been found to be associated with increased abdominal adipose tissue, 38 potentially contributing to the downregulation of the metabolite observed in the obese group.

| D ISCUSS I ON
The phenylalanine metabolism was found in the top three significantly impacted pathways when comparing both obesity and GDM to control groups. Phenylalanine is an aromatic amino acid that acts as a precursor to tyrosine, along with multiple catecholamines including epinephrine, norepinephrine and dopamine. Metabolic disorders such as obesity and GDM have been found to lead to elevations in phenylalanine and several of its metabolic products. 39 Elevated levels of aromatic amino acids have been associated with obesity 12 and the development of insulin resistance in nondiabetic individuals. 40 BCAAs, particularly valine, leucine and isoleucine, are often presented as indicators of risk for insulin resistance alongside aromatic amino acids 41,42 and have been implicated in the development of GDM in overweight and obese pregnant women. 18 In this study's comparison between individuals with obesity and GDM, the BCAA leucine was significantly upregulated in the GDM group alongside phenylalanine (Table S3), and valine was found to be higher in individuals with GDM versus their control counterparts (Table S2), supporting the role of BCAAs in the development of insulin resistance.
Thus, phenylalanine, leucine and valine provide valuable urinary biomarkers of GDM risk in obese individuals.
It should be noted that caution must be taken when comparing studies using different biofluids, as metabolite expression can depend on a variety of cohort factors such as circadian rhythms, diet and physical activity. In the present study, the downregulation of galactose in individuals with GDM seemingly conflicts with previous findings that serum d-galactose is upregulated in response to the disease. 43 In addition, pantothenate, the most impacted individual metabolite in the obesity vs control comparison, represents another case of discrepancy between serum and urinary expression. While the present downregulation of urinary pantothenate may serve as a urinary indicator of obesity, serum pantothenate has been found to be upregulated in response to obesity. 44 As mentioned above, these discrepancies may be due to differences between the study cohorts.
It cannot be ruled out, however, that excreted galactose and pantothenate in the urine may reflect serum levels. For example, a metabolite being upregulated in blood may indicate increased demand for the metabolite, which would result in reduced excretion of the metabolite in the urine. As mentioned earlier, urinary metabolomics enables the noninvasive detection of metabolites during pregnancy without the need for a blood test. In addition, NMR-based urinary metabolomics provides information on 209 metabolites compared to only 49 metabolites via serum metabolomics. 45,46 The regulation of sucrose expression appears to also vary between the serum and urinary metabolomes. Increased intake of sucrose has been linked to insulin resistance in mice. 47 However, both the obese and GDM groups of this study experienced reduced urinary sucrose levels when compared to control individuals. This may be partially explained by sex differences,sucrose-induced insulin-resistant models have only been successfully created in male animals, and females appear to be resistant to sucrose-induced insulin resistance. 48 When comparing individuals with obesity to those with GDM, sucrose was the third most impacted metabolite, being increased in GDM individuals. This finding indicates that starch and sucrose metabolism does indeed affect females and potentially contributes to the development of GDM, which highlights the need for a better understanding of sex differences in diabetes-related health outcomes, such as coronary heart disease. 49 The most impacted pathway distinction between obesity and GDM concerned starch and sucrose metabolism. High starch diets appear to have the opposite effect of sucrose on insulin resistance, reducing its severity and decreasing adipose tissue weight. 50 Variations in starch and sucrose metabolism may reflect dietary variations in the subjects, which in turn may correlate with the presence of obesity and GDM states. 51 It should also be noted that, while this study controlled for age, income and education level, subjects underwent no dietary restrictions prior to sampling, so impacts of diet on the metabolome act as a potential confound.
The results of this study allow for further analysis of the comorbidity of obesity and GDM with their associated health risks.
Preeclampsia is one of the most severe pregnancy complications Spontaneous preterm birth, when the infant is delivered prior to 37 weeks of gestation, is another pregnancy complication that is associated with metabolic syndrome, obesity and GDM obesity. 2,54,55 Previous urinary metabolic analysis of preterm birth revealed an association between spontaneous preterm birth and decreased levels of formic acid. 56 This trend was also observed in the obese and GDM groups of this study when compared to control individuals.
The mechanisms by which this trend is associated with preterm birth are not yet known, but it was suggested that diminished urinary formic acid raises the risk of hypertension, 57 which in turn is positively associated with preterm birth. 2

| Synthesis and conclusion
This exploratory study contributes to the understanding of the bio- children, but also assist in the transition to preventative and precision medicine.

ACK N OWLED G EM ENTS
We are extremely grateful to all the families who took part in this study and the APrON team (https://apron study.ca/team-partn ers/ apron-team/), investigators, research assistants, graduate and undergraduate students, volunteers, clerical staff and managers. This

CO N FLI C T O F I NTE R E S T
The authors have no conflicts or competing interests to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.