Estradiol (E2), estrone (E1) and estriol (E3) were purchased from China National Institutes for Food and Drug Control (Beijing, China). 2-Hydroxyestrone (2OHE1), 2-hydroxyestradiol (2OHE2), 2-methoxyestrone (2MeOE1), 2-methoxyestradiol (2MeOE2), 4-methoxyestrone (4MeOE1), 4-methoxyestradiol (4MeOE2), 16α-hydroxyestrone (16OHE1), 16-epiestriol (16EpiE3), 17-epiestriol (17EpiE3), aldosterone (ALD) and internal standards (deuterated steroid hormones) were purchased from TRC (Toronto, Canada). Progesterone (P4), 17α-hydroxyprogesterone (17α-OHP4), testosterone (T), dihydrotestosterone (DHT), androstenedione (AD), corticosterone (CORT) and cortisol (F) were purchased from Dr. Ehrenstorfer (Augsburg, Germany). Dehydroepiandrosterone (DHEA) was purchased from Cerilliant (Round Rock, TX, USA). Dehydroepiandrosterone sulfate (DHEA-S) was purchased from ISOscience (Ambler, PA, USA). 4-Hydroxyestrone (4OHE1), dansyl chloride, L-ascorbic acid, and β-glucuronidase/arylsulfatase were obtained from Sigma-Aldrich (St. Louis, MO). Methanol (HPLC grade) and formic acid were purchased from Merck (Darmstadt, Germany) or Fisher (Waltham, MA, USA). Deionized water was prepared using a Milli-Q system (Millipore, Milford, MA, USA). Acetone was obtained from Nanjing Chemical Reagent CO., LTD.

This study was approved by the ethical committee of Drum Tower Hospital Affiliated to Medical School of Nanjing University (No.2021-021-01) and was registered to ChiCTR (identifier: ChiCTR2100048675). Written consent has been obtained from each patient or subject after full explanation of the purpose and nature of all procedures used. All women in this study had singleton pregnancies and underwent routine 75-g oral glucose tolerance test (OGTT) screening at 24-28 gestational weeks. GDM was diagnosed based on IADPSG guideline to meet one or more of the following diagnostic criteria (28): (1) fasting blood glucose≥5.1 mmol/L; (2) 1 h≥10.0 mmol/L; (3) 2 h≥8.5 mmol/L. The final study population was determined after excluding pregnant women with any of the following characteristics: (1) <18 years of age; (2) severe maternal or fetal illness (malignancy, decompensated liver disease, heart disease, hypertension, congenital anomaly, etc.); (3) pre-existing diabetes mellitus or overt diabetes; (4) polycystic ovarian syndrome; (5) presence of drug use that interferes with steroid metabolism (progestin drugs); (6) non-singleton pregnancies; (7) newborns diagnosed with congenital anomaly (heart disease, neurological disease, metabolic diseases, etc.) or missing health assessment information for the newborns. Finally, a total of 40 GDM women and 70 healthy pregnant women aged 26–42 were recruited for analysis.

A detailed medical history including age, gestational age, height and other basic information was recorded for each participant. Maternal weight at 24-28 weeks of gestation was measured by the standard method. The pre-pregnancy weight was recorded according to patient statements. BMI value was calculated by dividing the weight (kilograms) by the square of height (meter²). Each participant underwent a 75-g OGTT after an overnight fast of 10–12 hours. Blood lipids (triglycerides (TG), total cholesterol (TC), high-density lipoprotein (HDL), and low-density lipoprotein (LDL) and liver enzymes, including aminotransferase (ALT), aspartate aminotransferase (AST), g-glutamyl transferase (GGT) and plasma glucose were measured by using a biochemical autoanalyzer (Beckman, CA, USA).

Steroid hormones in serum samples collected at the at 24-28 weeks of gestation were determined using a Waters UPLC I-Class system interfaced with a Xevo TQ-S triple quadrupole mass spectrometer (Waters Corp., Milford, MA, USA). This method validation was performed according to the Guidelines for Bio-analytical Method Validation. For the determination of estrogens, mass spectrometry was performed in positive mode and multiple reaction monitoring (MRM). Liquid chromatography separation was achieved on a Waters CORTECS C18 column (2.1 × 150 mm, 1.6 μm) at 40°C. The flow rate was 0.3 mL/min. The mobile phase consisted of solvent A, deionized water (containing 0.05% formic acid and 2 mM ammonium acetate) and solvent B, methanol. The elution gradient was as follows: 0.0-4.1min, isocratic 78% B; 4.1-4.2 min, linear gradient 78-81% B; 4.2-8.0 min, isocratic 81% B; 8.0-8.01min, linear gradient 81-78% B; 8.01-10.0min, isocratic 78% B. The total run time was 10.0 min. For the determination of corticosteroids, progestins and androgens, mass spectrometry was performed in a positive and negative ion switching mode and MRM. Liquid chromatography separation was achieved on a Phenomenex Kinetex XB-C18 (3.0 × 50 mm, 2.6 µm) at 40°C. The flow rate was 0.3 mL/min. The mobile phase consisted of solvent A, deionized water (containing 100 μM ammonium fluoride) and solvent B, methanol. The elution gradient was as follows: 40% B increased to 98% B from 0.0 min to 3.0 min, maintained for 0.5 min, retuned to initial conditions, and then re-equilibrated for 2.5 min. The total run time was 6.0 min.

Sample preparation procedure was performed as previously described (29, 30). Briefly, charcoal-stripped human serum with no detectable levels of any steroid hormones was used for preparation of calibration standards and quality control samples. Free estrogens, conjugated (-glucuronide/sulfate, -G/S) estrogen metabolites and other steroid hormones were directly or indirectly measured through preparation procedures with or without β-glucuronidase/sulfatase hydrolysis step. Serum samples and enzymatically hydrolyzed serum samples underwent liquid-liquid extraction with methyl tert-butyl ether. After extraction, the organic solvent portion was transferred and evaporated to dryness at 60°C under nitrogen gas. For the determination of estrogens, the dried samples then underwent the derivatization process using dansyl chloride. All residue samples were reconstituted with methanol and analyzed by UPLC-MS/MS.

The normality of values was analyzed by Shapiro-Wilk test. Values with normal distribution were summarized as the mean ± standard deviation (SD), and the corresponding significance was examined by using the Student’s t-test. Values with non-normality were summarized as the median (interquartile range, IQR), and significance was examined by Mann-Whitney U test. P < 0.05 was considered significant. Volcano graph was drawn by GraphPad Prism 8.0 to screen for the obvious changes of serum steroid hormones in GDM women compared to healthy pregnant women. A mixed-effects Least Absolute Shrinkage Selection Operator (LASSO) logistic regression, implemented by STATA/MP version 16.0, was used for multicollinearity elimination and variable selection. Receiver operating characteristic (ROC) curves were plotted to classify the performance of biomarkers. The most relevant steroid hormones are grouped into quartiles to determine the linear trend relationship between the independent variables and the prevalence of GDM.

A combined sensitive and rapid UPLC-MS/MS method was developed to profile the steroid network in GDM women. 23 kinds of steroid hormones, including 5 kinds of upstream C21 steroid hormones (corticosteroids and progestins), 5 kinds of C19 steroid hormones (androgens) and 13 kinds of downstream C18 steroid hormones (estrogens), were directly measured through the UPLC-MS/MS method. 13 kinds of conjugated Phase II metabolites were indirectly measured by the β-glucuronidase/sulfatase hydrolysis method. As shown in Figure 2 , we optimized the elution gradient of chromatographic mobile phases to achieve rapid separation of multiple isomers. The peaks for each steroid appeared sharp and symmetrical. The total running time kept within 6 mins and 10 mins. This method validation was performed according to the guidelines for Bio-analytical Method Validation published by the US FDA and fulfilled the acceptance criteria, as shown in Supplementary Tables S1–4 . The linear range of the measured steroid hormones was as follows: estrogens (0.001-2 ng/mL), T (0.05-10 ng/mL), DHT (0.025-5 ng/mL), 17α-OHP4(0.05-10 ng/mL), P4 (0.05-10 ng/mL), CORT (0.1-20 ng/mL), DHEA (0.1-20 ng/mL), AD (0.05-10 ng/mL), F (1-200 ng/mL), ALD (0.01-2 ng/mL), DHEA-S (50-10000 ng/mL).

A total of 110 pregnant women participated in this study, including 40 GDM women and 70 healthy pregnant women. The baseline characteristics of the GDM group and corresponding controls are shown in Table 1 . There was no statistical significance in age and weight between GDM and healthy pregnant women. GDM women had higher BMI values and higher levels of serum fasting and post-load glucose, TG and HDL compared with healthy pregnant women. The liver function index and other serum lipids were comparable between the two groups.

The levels of 36 steroid hormones measured by UPLC-MS/MS are presented in Table 2 . Briefly, in addition to androgens, there was a statistically significant increase in serum corticosteroid, progestin and some downstream estrogen levels of GDM women, compared with the healthy. The levels of serum androgens are roughly comparable. Changes in serum concentrations of progestins and corticosteroids were relatively mild between GDM and healthy women, except for 17-OHP4 (p<0.05, FC 1.6), an active intermediate derived from P4 via 17-hydroxylase.

Among all changed steroid hormones, estrogens showed relatively large changes. Among all kinds of estrogens measured, 3 kinds of conjugated metabolites, including E1-G/S, E3-G/S and 16OHE1-G/S, have absolute advantages in serum concentrations. The median concentrations of serum parent estrogens (E1, p<0.05; E2, p<0.001) and the conjugate metabolite (E1-G/S, p<0.001) in GDM women were higher than those in healthy women. E1-G/S was nearly 10 times higher in serum than its precursor E1. However, E2-G/S was not detected in serum samples. Interestingly, almost all metabolites via the 16-hydroxylation pathway from parent estrogens were higher in GDM women. Among these 16-hydroxylated metabolites, the median concentrations of 5 metabolites, including E3, E3-G/S, 16OHE1, 16OHE1-G/S, and 16EpiE3 (p<0.001), changed more than 1.5 times in the serum of GDM women compared with healthy women. In contrast, most estrogen metabolites via 4-hydroxylation pathway and more than half of the metabolites via 2-hydroxylation pathway were not significantly different in serum between the two groups. Besides, 2MeOE2-G/S and 4MeOE2 were not detected in all serum samples. 2MeOE1-G/S was not detected in more than 80% of serum samples.

From the perspective of metabolic flux, there was a significantly higher flux of hydroxylation metabolism of estrogens (p<0.001, FC 1.55) in GDM women, and this elevation of hydroxylation was mainly contributed by 16- hydroxylation pathway, as shown in Table 3 . No significant difference was observed in the total calculated concentration of metabolites via 2- or 4-hydroxylation pathways. In addition, the metabolic ratios of 2- and 4-pathways to total estrogens decreased significantly in GDM women.

As shown in Figures 3 , 4 , a total of 23 kinds of steroid indicators were screened out according to Student’s t-test results and fold change values (p<0.05, FC>1.2). ROC analysis was further employed to preliminarily evaluate the diagnostic performance of serum metabolites in the discrimination of GDM from controls. The area under the curve (AUC) values of 9 directly measured steroid hormones (16OHE1, 16OHE1-G/S, E3, E1-G/S, E2, 16-EpiE3-G/S, 17-EpiE3, 16-epiE3 and 2OHE2) and 4 indirectly calculated indicators (total estrogens, conjugated estrogens, total hydroxylated estrogens, and total 16-hydroxylated estrogens) achieved good AUC values (>0.7), which indicated good diagnostic efficacy and close correlation. Among these indicators, 16OHE1 achieved the highest AUC value of 0.85. Based on all quantitative results, a combined model was established through LASSO logistic regression considering the basic biochemical characteristics, directly measured and indirectly calculated steroid indicators to eliminate the multicollinearity and find the specific markers associated with GDM. With parameter lambda of 0.05 obtained by cross-validation, we identified a diagnostic panel of 5 elements, including BMI, TG, 16OHE1, E1-G/S and the ratio of total 2-pathway estrogens to total estrogens. The equation Logit (P) = 0.172*BMI+0.438*TG+0.104*E1-G/S+7.381*16OHE1-29.131* ratio (2-pathway: total)-10.172. The AUC-ROC of the model was 0.89, which indicated that the screened markers are closely related to the development of GDM.

To further explore the correlation between the screened steroid hormones and the risk of GDM, the concentrations of serum 16OHE1, E1-G/S and the ratio of 2-pathway metabolites to total estrogens were stratified into quartiles to perform binary logistic regression analysis, as shown in Table 4 . It was observed that 16OHE1 and E1-G/S were significantly positively correlated with GDM risk. This association was also enhanced in the age, BMI and TG–adjusted models. The age, BMI and TG adjusted odds ratios (ORs) of GDM for the highest quartile compared with the lowest were 72.22 (95% CI 11.27-462.71, P _trend <0.001) for 16OHE1 and 6.28 (95% CI 1.74-22.71, P _trend< 0.05) for E1-G/S. Besides, the ratio of total 2-pathway estrogens to total estrogens was negatively associated with the risk for GDM, whether with or without adjustment for potential confounders. After adjustment by potential confounders, the negative association between the ratio of total 2-pathway estrogens to total estrogens and the risk for GDM was not significant (adjusted OR for the highest versus the lowest quartile, 0.65; P _trend= 0.108).