Serum and clinical metadata for 160 pregnant women, 51 with HDP and 109 without HDP (frequency matched by gravidity), were selected from the Global Alliance to Prevent Prematurity and Stillbirth (GAPPS) repository for women who enrolled between 2011 and 2016. Study participants were enrolled during pregnancy in outpatient clinics from the University of Washington Medical Center, Seattle; Swedish Medical Center, Seattle; and Yakima Valley Memorial Hospital, Yakima, WA. Pregnant women ≥14 years of age were eligible to be enrolled in GAPPS Repository. Women who received narcotics in the previous 12 h, were pregnant with multiples, or were in active labor at enrollment were excluded. Participants were followed up to 10 weeks after delivery. For this study, we excluded 6 participants whose BMI was missing, which resulted in a total of 154 (105 non-HDP participants and 49 HDP participants).

Diagnosis of HDP was collected from the medical records directly by GAPPS study personnel (17). HDP cases included cases of gestational hypertension and pre-eclampsia. Gestational hypertension was defined by the new onset of hypertension at ≥20 weeks of gestation; pre-eclampsia was defined by either pre-existing hypertension with superimposed proteinuria and/or significant end-organ dysfunction, or gestational hypertension with proteinuria or/and significant end-organ dysfunction.

BMI (measured by healthcare personnel) and serum specimens from pregnant women were obtained at the 1st prenatal care (PNC) visit during the 1st trimester of pregnancy (6⁺¹–13⁺⁶ weeks). Since the previous study shows that the pre-pregnancy BMI and the first 1^st trimester BMI are generally identical regarding the BMI category for most women (25), in this study, we used the adult BMI cutoff criteria (26) from the Center for Disease Control and Prevention to define normal weight (NW, with BMI between 18 and 24.9 kg/m²), overweight (OW, with BMI between 25 and 29.9 kg/m²) and obese (OB, with BMI ≥30 kg/m²). The average gestational week, when BMI and serum specimens were obtained, was at 10.51 ± 1.80 weeks and was not different between the HDP participants (10.64 ± 1.73 weeks) and the non-HDP participants (10.45 ± 1.84 weeks).

The participants’ demographics, health history, and lifestyle information before and during early pregnancy were obtained using the questionnaires designed by the GAPPS repository. The covariates in this study included gravidity, maternal age, the gestational age at the time the BMI was measured, and the blood sample collected, maternal race, and whether participants had smoked more than 100 cigarettes in their lifetime.

Untargeted metabolomics data was acquired on the 1st trimester serum samples collected from 2011 to 2016. Metabolomics data were acquired using a Vanquish UHPLC system coupled with a Q Exactive^™ HF-X Hybrid Quadrupole-Orbitrap^™ Mass Spectrometer (UHPLC-HRMS; Thermo Fisher Scientific). Procedures for sample preparation, data acquisition, preprocessing, metabolite identification, and annotation have been published in Harville et al. (17). Briefly, serum (50 μL) was mixed with 400-μL methanol (containing 500 ng/mL L-tryptophan-d5) by vortex at 5,000 rpm for 2 min. Quality control samples (QC pools) were made by pooling 7-μL serum from each study sample. Study samples and QC pools were processed with identical procedures. All samples were centrifuged at 16,000 rcf for 5 min at 4°C. The supernatant (350-uL) was dried and reconstituted with 100 μL water–methanol (95:5, v/v) for data acquisition. Study samples were randomized before sample preparation and acquisition with QC pools interspersed. Metabolites were separated via an HSS T3 C18 column (2.1 × 100 mm, 1.7 μm, Waters Corporation) at 50°C with mobile phases of water (A) and methanol (B), each containing 0.1% formic acid (v/v). The untargeted data were acquired from 70 to 1,050 m/z under the data dependent acquisition mode.

The acquired data was processed by Progenesis QI (version 2.1, Waters Corporation) for peak picking, alignment, and normalization. Signals that were highly varied (RSD > 50%) across QC pools, significantly differed amongst running QC pools batches (ANOVA, with FDR correction q < 0.05), or missing in all QC pool samples, were excluded for further analysis.

Metabolite identification/annotation was conducted by Progenesis QI via matching against the in-house experimental standards library (IESL) that was built by acquiring data from over 2,400 compounds under identical conditions to study samples. Signals were also matched to public database, including HMDB, NIST, and METLIN. The evidence supporting each identification or annotation was labeled with the Ontology level (e.g., OL1-PDd), which were shown and further explained in Supplementary Tables S1, S2.

All statistical analyses were performed using R (version 3.6.1) within the R Studio (version 1.3.1093) platform. For normally distributed continuous variables, we calculated the mean and standard deviation. For continuous variables not normally distributed, we calculated median, 25th percentile, and 75th percentile. The student’s t-test or analysis of variance (ANOVA) test was used for parametric data depending on the number of comparison groups, while the Wilcoxon rank-sum test or Kruskal-Wallis test was used for nonparametric data. The number of participants and percentage for categorical variables and compared across groups using the Chi-square test.

We used 3 modeling approaches to determine signals associated with BMI and HPD (Figure 1). For approach 1, we used linear regression models to determine the BMI (continuous) associated signals in women who developed HDP (n = 49) and who did not develop HDP (n = 105), respectively. Multiple linear regression models were used to adjust for maternal age, whether they smoked more than 100 cigarettes in their lifetime and gestational week when serum samples and BMI were obtained. The False Discovery Rate (FDR) adjustment was used to correct for multiple testing. A two-sided FDR corrected p-value (or q-value) of <0.05 was considered significant.

For approach 2, we first stratified the 154 participants into 3 BMI categories: NW (BMI < 25 kg/m², n = 45); OW (25 kg/m² ≤ BMI < 30 kg/m2, n = 48), and OB (BMI > 30 kg/m², n = 61). Within each BMI subgroup, we used logistic regression models to determine signals associated with HDP by adjusting gravidity (frequency-matching variable) and maternal age. Because of the small sample size, the original nominal p-value < 0.05 (without FDR adjustment) was considered significant for this discovery investigation.

For approach 3, we first used linear regression models to determine signals significantly associated with BMI among all 154 participants (105 without HDP, 49 with HDP), adjusting for maternal age, whether they smoked more than 100 cigarettes in their lifetime, and gestational week when serum samples and BMI were obtained. Within the list of signals associated with BMI, we used logistic regression models adjusting for gravidity (frequency-matching variable) to determine signals associated with BMI and HDP. A two-sided FDR corrected p-value (or q-value) of <0.05 was considered significant.

Multivariable linear or logistic regression models adjusting for potential confounders were used to estimate the adjusted effect sizes. Covariates associated with both the predictor and the outcome variable, but not on the causal pathway, in each model were considered potential confounders. All peaks meeting the significance level were considered in the analysis. A complete case analysis approach was used. Covariate missing data were minimal: maternal age (1%) and whether they smoked more than 100 cigarettes in their lifetime (4%).

Pathway analysis was conducted using MetaboAnalyst 5.0 via the Functional Analysis (MS Peaks module), built on the Mummichog algorithm (27–30). All normalized signals (m/z) that remained after preprocessing and quality control were uploaded to the pathway analysis tool, together with the p-value calculated by the regression models (without adjusting covariates and FDR correction) to indicate the association between signal and outcomes (e.g., BMI and/or HDP). An original nominal p-value cut-off of 0.05 was used to determine the size of the permutation group that the algorithm used for selecting significant signals to match all possible metabolites. A 3-ppm mass accuracy tolerance was used in signal annotations to identify candidate pathways. All possible metabolites matched by m/z were searched in the Homo sapiens (human) [MFN] pathway library.

To further understand the direction of change impacted by BMI and HDP status for specific pathways, we verified the enriched empirical annotations made via the Mummichog algorithm by matching the corresponding signals against our in-house physical standard library or public bases (NIST, HMDB). We then used box plots to demonstrate the up- and down of these key metabolites in the enriched pathway according to different pairwise comparisons using peak intensity.

The average maternal age for all 154 participants was 29.7 ± 5.3 years old, and the average gestational age for BMI measurement and serum samples collection was 10.5 ± 1.8 weeks. There was no significant difference between participants with and without HDP in maternal and gestational age when BMI and serum samples were collected (Table 1). About 71% of the participants were white, and there was no significant difference in demographic composition between groups with- and without HDP. The overall percentage of participants who smoked more than 100 cigarettes in their lifetime was 22.3%, and the percentage was significantly higher in the HDP group (33.3%) than in the group without HDP (17%) (p = 0.03). The mean gestational age of delivery was 39.3 weeks for all participants, while participants who developed HDP delivered earlier on average compared to the non-HDP participants [median (q1–q3): 38.6 (37.9–40.0) vs. 39.3 (38.9–40.3), p < 0.001].

Participants who developed HDP showed significantly higher 1st PNC BMI than those without HDP [median (q1–q3): 34.29 (28.1–40.4) vs. 27.27 (23.8–31.5), p < 0.001] (Table 1). Among different BMI categories classified according to the 1st PNC measurements, 6 out of 45 (13%) women in NW developed HDP, 14 out of 48 (29%) women in OW developed HDP, and 29 out of 61 (48%) participants in OB developed HDP.

We stratified the 154 participants into 2 groups: with- (n = 49) and without (n = 105) HDP, and then used linear regression models to determine metabolic profiles and pathways that are associated with BMI (continued variance). After adjusting for confounding factors, including maternal age, tobacco usage (lifetime smoking ≥100 cigarettes), and the gestational week of the blood draw and BMI measurement, 241 signals were determined to associate with BMI in women without HDP, with 38/241 signals being identified or annotated via matching against the in-house library and public database (Figure 2A; Supplementary Table S1). In contrast, 10 signals were associated with BMI in women with HDP, with 6/10 signals identified or annotated. Three BMI-associated signals overlapped between participants with and without HDP (Figure 2A), and 2/3 were annotated as steroids/hormones and vitamin D3-related metabolites (Supplementary Table S1).

Pathway enrichment was performed using all 3,300 normalized signals, regardless of being identified/annotated, together with the p-value that indicated the statistical strength of association with BMI. As shown in Figure 2B and Table 2i, 5 BMI-associated pathways were enriched for women without HDP, while 11 BMI-associated pathways were enriched for women with HDP. The shared BMI-associated pathways for women with and without HDP were vitamin D3 and lysine metabolism. Three BMI-associated pathways unique to women without HDP related to energy balance and mitochondrion function, including the TCA cycle, vitamin B5-CoA biosynthesis from pantothenate, and ubiquinone biosynthesis. Comparatively, more perturbed pathways were found to associate with BMI in women with HDP, involving lipid and fatty acid metabolism (carnitine shuttle, glycerophospholipids metabolism), one-carbon metabolism (methionine and cysteine metabolism), chronic-low grade inflammation (tryptophan and tyrosine metabolism, glutathione metabolism), and xenobiotics (cytochrome P450) (Table 2i).

We stratified the 154 participants into 3 BMI categories: NW (n = 45, BMI < 25 kg/m²), OW (n = 48, 25 kg/m² ≤ BMI < 30 kg/m²), and OB (61 subjects, BMI > 30 kg/m²) and used logistic regression models to determine the HDP-associated metabolic profiles and pathways in each BMI category.

The number of signals/metabolites associated with HDP was 69, 40, and 138 for the NW, OW, and OB sub-group, respectively, after covariate adjustment (Figure 3A; Supplementary Table S2). The HDP-associated metabolites/signals in each of the BMI sub-group seemed quite unique, as no shared signal/metabolite was determined across the three categories, only 2 signals/metabolites overlapped between the NW and OB, and only 4 signals/metabolites were overlapped between OW and OB (Figure 3A; Supplementary Table S2). The identified or annotated HDP-associated metabolites in the NW sub-group include tryptophan derivatives (e.g., 5-hydroxyindoleacetic acid, OL1), polyphenols derivatives (e.g., 2,6-Di-tert-butyl-4-hydroxymethylphenol, PDa), and arachidonic acid metabolites. In contrast, the HDP-associated metabolites in the OB sub-groups include nucleosides (e.g., cytidine, OL1), tryptophan derivatives (e.g., acetylserotonin, OL2b, indoleacetic acid, OL2b, 3,4-dihydroxyphenylacetic acid, OL2b), cholic acid metabolites (e.g., dehydrolithocholic acid, OL2b, glycylcholic acid, OL2b), hormones (e.g., 4-androstene-3,17-dione, PDa, and 5-androstene-3.beta.-ol-17-one, PDa), and polyphenol derivatives (e.g., 3,4,5-trimethoxybenzaldehyde, OL1). Compared to the NW and OB subgroups, relatively fewer signals/metabolites are associated with HDP in the OW subgroup, and many are lipid and fatty acid-related compounds.

The number of pathways associated with HDP was 5, 14, and 17 for the NW, OW, and OB, respectively (Figure 3B; Table 4). Tryptophan metabolism is associated with HDP for all 3 BMI subgroups. Six pathways were associated with HDP in both the OW and OB sub-groups (Table 2ii), mainly involving nutrients metabolism and energy balance (e.g., glycerophospholipids metabolism, ubiquinone biosynthesis, urea cycle/amino group metabolism, lysine- and tyrosine metabolism) and sex hormone hemostasis (androgen and estrogen biosynthesis and metabolism). There was no overlap of HDP-associated pathways between the OB and NW subgroup, and the only pathway shared between NW and OW subgroups is related to inflammation and immune response (prostaglandin formation from arachidonate). The HDP-associated pathways unique to the NW subgroup were all about arachidonic metabolism and inflammation (e.g., fatty acid activation, leukotriene metabolism, and arachidonic acid metabolism). In contrast, the pathways unique to OW or OB subgroups were mainly involved in nutrient metabolisms, such as vitamins, sugar, bile acids, lipids, amino acids, and proteins (Table 2ii).

We used linear and logistical regression models to determine signals/metabolites simultaneously associated with BMI and HDP based on the overall 154 participants. After adjusting for covariates, three signals/metabolites remained in the model. One of the three signals was identified as cytidine (OL1), and the other two were annotated as metabolites from glycerophospholipids and vitamin D3 metabolism (Table 3). Since there were only a few signals (n < 10) that showed significant association (p < 0.05) with the outcomes using this approach, regardless of whether under covariates adjustment or FDR correction or not, we did not receive meaningful pathway enrichment results.

In the enriched vitamin D3 metabolism (Figure 4), dihydroxylated [1,25(OH)2D3], tri-hydroxylated [1,24,25(OH)3D3], and the two oxidative metabolites [24-oxo-1,25 (OH)2D3, 1,25-(OH)2D3-26,23-lactone] were empirically annotated by the Mummichog algorithm, and the dihydroxylated- and tri-hydroxylated D3 were verified via matching the exact mass with public base. We observed a consistent trend of decrease in the vitamin D3 metabolites with BMI increment, especially for women without HDP (Figure 4), except for the C-24 oxidation metabolite 24-oxo-1,25 (OH)2D3, which was increased with the BMI increment, especially for the women with HDP. Within the category of OB, we observed a significant increase of 24-oxo-1,25 (OH)2D3 levels in women with HDP compared to those without HDP (p < 0.05).

As shown in Figure 5A, the majority of the empirically annotated tryptophan metabolites in the enriched pathway were verified by the in-house library within OL1 (RT, MS, MS/MS) or OL2b (MS, MS/MS) or by public databases within PDA (MS, MS/MS) or PDc (MS). Metabolites in the tryptophan-serotonin pathways, including serotonin, acetyl serotonin, 5-hydroxyindoleacetate, and 5-methoxyindoleacetate (5-MIAA), were found to be significantly varied with the BMI or HDP outcomes (Supplementary Table S2; Figure 5B; Supplementary Figure S1). Serotonin and aceylserotonin showed a trend of increase with BMI, and aceylserotonin was significantly increased in obese women with HDP compared to obese women without HDP (p < 0.05) (Figure 5B). In contrast, 5-methoxyindoleacetate (5-MIAA) presented an opposite trend compared to serotonin and acetyl serotonin and decreased with the increment of BMI, especially for women without HDP (p < 0.001). This metabolite was significantly decreased in women with HDP vs. without HDP in the OW (p < 0.05) and overall participants (p < 0.01) (Figure 5B).