Reduced serum acetyl-L-carnitine is associated with hypertensive disorders in women with gestational diabetes mellitus: a multi-omics study

Abstract

Background

Gestational diabetes mellitus (GDM) and hypertensive disorders of pregnancy (HDP) are common metabolic complications during pregnancy. GDM is an independent risk factor for HDP, while HDP is a major contributor to maternal and perinatal morbidity and mortality. Reliable early tools for identifying women with GDM who are at risk of subsequently developing HDP remain limited. This study aimed to integrate metabolomic, lipidomic, and clinical data to identify reproducible candidate biomarkers associated with subsequently diagnosed HDP in women with GDM and to characterize the accompanying metabolic and lipidomic alterations.

Methods

In this prospective exploratory biomarker study, a discovery cohort (n = 128) and a validation cohort (n = 55) were recruited. Serum samples collected at 24–28 weeks of gestation were analyzed using untargeted and targeted metabolomics. Lipidomics was performed to complement the metabolomic analysis and characterize lipid-related alterations. A combined predictive model was constructed using logistic regression, and its predictive performance was evaluated using receiver operating characteristic (ROC) analysis.

Results

Untargeted metabolomics identified lower serum acetyl-L-carnitine (ALC) in the GDMHDP group than in the GDM group, and this difference was confirmed by targeted LC–MS/MS in the validation cohort. The combined ALC–diastolic blood pressure (DBP) model achieved an area under the curve (AUC) of 0.907 in the discovery cohort and 0.886 in the validation cohort. Adding ALC to DBP improved discrimination and reclassification in both cohorts. Multi-omics analysis showed complementary and partly distinct lipid associations with ALC and DBP. Lower ALC was associated with selected ceramide and monolysocardiolipin species, as well as with lipid patterns consistent with altered fatty-acid handling, whereas DBP showed a partly distinct pattern of lipid correlations.

Conclusions

Reduced serum ALC at 24–28 weeks of gestation was reproducibly associated with GDM complicated by HDP. Combining ALC with DBP improved discrimination between GDMHDP and uncomplicated GDM, suggesting complementary metabolic and hemodynamic information. Given the modest number of GDMHDP cases and the single-center design, these findings should be interpreted as exploratory and hypothesis-generating; they support further evaluation of ALC as a candidate metabolic marker but require confirmation in larger independent cohorts.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12933-026-03347-1.

Research insight

What is currently known about this topic?

• Women with GDM are at increased risk of hypertensive disorders of pregnancy.

• Conventional glycemic and blood pressure measures alone do not reliably identify women with GDM who are at risk of subsequently developing HDP.

• Metabolomic studies have investigated GDM or HDP separately, but few have focused on metabolic divergence within GDM pregnancies.

What is the key research question?

• Can circulating metabolic signatures at 24–28 weeks identify women with GDM at risk of subsequent hypertensive complications?

What is new?

• The combination of ALC and DBP showed better discrimination of GDMHDP than either marker alone in both the discovery and validation cohorts.

• Multi-omics analysis showed that lower ALC was associated with lipotoxic ceramide species and a mitochondrial stress-related lipid signature.

How might this study influence clinical practice?

• ALC may provide a metabolically informed approach to identify women with GDM who require closer vascular monitoring during pregnancy.

Introduction

Gestational Diabetes Mellitus (GDM), defined as any degree of glucose intolerance first occurring or identified during pregnancy [1], has garnered significant attention due to its high global prevalence, estimated at approximately 16.7% by the International Diabetes Federation (IDF) [2]. GDM not only raises the risk of gestational complications but is also increasingly recognized as a major contributor to hypertensive disorders of pregnancy (HDP) [3]. HDP, characterized by new-onset hypertension (with or without proteinuria) after 20 weeks of gestation and encompassing severe complications such as preeclampsia [4], represent a leading cause of maternal mortality worldwide and a major risk factor for adverse fetal outcomes [5, 6]. Gestational diabetes mellitus complicated by hypertensive disorders of pregnancy (GDMHDP) is associated with significantly worse maternal and neonatal outcomes than either condition alone, including higher rates of preterm birth and fetal growth restriction, as well as an increased long-term risk of maternal cardiovascular disease [7, 8]. Despite its considerable clinical impact, GDMHDP develops in only a subset of women with GDM. Recent large cohorts from China, Japan, and Canada have reported an incidence of preeclampsia or broader hypertensive disorders of approximately 3.6–9.6% among pregnancies affected by GDM, although the estimates vary according to the study population and outcome definition [9–12]. This relatively limited incidence makes early identification of high-risk individuals within the broader GDM population both challenging and clinically important. Impaired placental function in HDP patients may lead to maternal systemic vascular endothelial dysfunction [3], thereby exacerbating insulin resistance and glucose metabolic abnormalities [13, 14]. Currently, the clinical management of GDM primarily focuses on glycemic control. However, blood pressure abnormalities often become clinically apparent and HDP is diagnosed only later in pregnancy, by which time placental function has already sustained significant damage [15]. Consequently, early identification of high-risk individuals, close gestational monitoring, and prophylactic aspirin administration could reduce adverse pregnancy outcomes and improve maternal–fetal health.

Advancements in high-throughput methodologies have significantly enhanced our understanding of the pathogenesis of complex diseases, including GDM and HDP during pregnancy [16–18]. Metabolomics and lipidomics offer powerful tools for revealing the intricate changes in metabolite levels within the human body [19]. The chronic inflammatory state associated with HDP triggers metabolic alterations that can serve as valuable biomarkers for diagnosis, disease progression monitoring, and treatment response assessment [20]. Liquid chromatography-mass spectrometry (LC–MS)-based metabolomics and lipidomics represent robust platforms for discovering potential biomarkers and elucidating metabolic pathway remodeling in diseases [21]. Previous lipidomic studies have reported disturbances in glycerolipid, glycerophospholipid, sphingolipid, and ceramide metabolism in GDM and preeclampsia [22–25]. Recent multi-omics evidence has further demonstrated distinct lipidomic alterations in pregnancies complicated by the coexistence of GDM and preeclampsia [26]. While these technologies have been widely used, previous studies have predominantly compared GDM or HDP separately with healthy controls, thereby overlooking the specific metabolic transition associated with the development of GDMHDP within the GDM population. Early identification of women at risk of subsequently developing HDP is crucial to prevent irreversible damage and improve maternal and perinatal outcomes [27].

Existing studies lack reliable early predictive markers for assessing the risk of HDP specifically within the GDM population. To address this gap, this study employed a multi-omics approach combining untargeted metabolomics, targeted validation, and lipidomics with machine learning algorithms. Untargeted metabolomics and targeted validation were used to identify and confirm reproducible circulating candidate biomarkers, whereas lipidomics was used to provide complementary information on lipid-associated metabolic alterations. We aimed to characterize the serum metabolic and lipidomic alterations associated with GDM complicated by subsequently diagnosed HDP, identify reproducible candidate biomarkers, and provide a broader metabolic context for the development of hypertensive complications in women with GDM.

Methods and materials

Study participants

This prospective study enrolled pregnant women undergoing a routine 75-g oral glucose tolerance test (OGTT) at 24–28 weeks of gestation at the Women's Hospital, School of Medicine, Zhejiang University. A total of 183 eligible participants were included in the final analysis and allocated into two cohorts: a discovery cohort (n = 128) and a validation cohort (n = 55). Specifically, the discovery cohort consisted of 58 women with normal pregnancy (Normal), 50 patients with GDM, and 20 patients with GDMHDP, whereas the validation cohort comprised 25 Normal controls, 22 GDM patients, and 8 GDMHDP patients. This prospective biomarker study enrolled eligible participants according to predefined criteria and followed them until delivery. All eligible GDMHDP cases identified during follow-up and meeting the predefined clinical and sample-quality criteria were included in the omics analyses. The final analytical sample comprised participants with complete follow-up, confirmed pregnancy outcomes, and adequate serum samples for the planned omics analyses. Exclusion criteria included multiple gestation, pre-existing diabetes or hypertension, metabolic disorders, renal disease, infections, or incomplete clinical records. GDM was diagnosed in line with the following International Association of Diabetes and Pregnancy Study Groups (IADPSG) criteria, based on OGTT thresholds: fasting plasma glucose ≥ 5.1 mmol/L, 1 h plasma glucose ≥ 10.0 mmol/L, 2 h plasma glucose ≥ 8.5 mmol/L. HDP, encompassing gestational hypertension and preeclampsia, was defined according to the International Society for the Study of Hypertension in Pregnancy (ISSHP) criteria as systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg on two occasions, with or without proteinuria or other maternal organ dysfunction. Participants were classified as GDMHDP when a formal HDP diagnosis was documented during follow-up before delivery. Blood pressure values used in the analyses were obtained after 20 weeks of gestation, and the gestational age at measurement was recorded. For participants who subsequently developed HDP, the gestational age at formal HDP diagnosis and the corresponding systolic and diastolic blood pressure (SBP and DBP) values were additionally extracted.

HDP subtype and severity were determined according to the final clinical diagnoses documented in the medical records. Pregnancy and neonatal outcomes included gestational age at delivery, birth weight, small for gestational age (SGA), fetal growth restriction (FGR), large for gestational age (LGA), preterm birth, cesarean delivery, and neonatal intensive care unit (NICU) admission. SGA and LGA were defined as birth weight below the 10th percentile and above the 90th percentile, respectively, for infant sex and gestational age. FGR was defined according to the final obstetric diagnosis documented in the medical record, and preterm birth was defined as delivery before 37 completed weeks of gestation. Information on aspirin prophylaxis and smoking status was also extracted from the medical records.

The study was approved by the Ethics Committee of the Obstetrics and Gynecology Hospital Affiliated to Zhejiang University School of Medicine (Approval No. IRB-20210112-R), and all participants provided written informed consent. Fasting venous blood samples were collected at the time of the 24–28-week OGTT assessment. Serum was immediately separated by centrifugation and stored at −80 °C for subsequent metabolomic and lipidomic assays. Sociodemographic characteristics, baseline variables, and clinical data were systematically extracted from electronic health records, and all participants were followed until delivery.

Reagents and materials

LC–MS grade methanol, acetonitrile, and isopropanol were purchased from Merck KGaA (Darmstadt, Germany). Formic acid and ammonium acetate (NH4OAc) were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. Ketoprofen and Sulfamerazine for untargeted metabolomics were purchased from Cambridge Isotope Laboratories (Tewksbury, MA, United States). Lysophosphatidylcholine LPC(19:0) for untargeted lipidomics was purchased from Avanti Polar Lipids (Alabaster, AL, United States).

Untargeted metabolomics analysis

In the discovery cohort, serum samples were thawed at 4℃, and metabolites were extracted by adding methanol containing internal standards (ketoprofen and sulfamerazine). After vortexing for 30 s, the samples were incubated at −20 °C for 1 h and centrifuged at 14,000 g for 20 min at 4 °C. The supernatant was collected, and quality control (QC) samples were prepared by pooling equal aliquots of all individual samples. The supernatants were vacuum-dried and stored at −80℃ until analysis. Prior to LC–MS analysis, samples were redissolved by 66.6% methanol and 0.1% formic acid in water.

The data acquisition was performed on an Orbitrap Exploris 120. Metabolites were separated on a Waters HSS T3 column (100 × 2.1 mm, 1.8 μm) using a Thermo Vanquish UPLC system and the column temperature was kept at 40 °C. Mobile phase A was 0.1% formic acid in water, and B was acetonitrile for both the positive (ESI +) and negative (ESI –) modes. The flow rate was 0.35 mL/min, and the gradient was set as follows: 0–3 min, 5–20% B; 3–5 min, 20–40% B; 5–9 min, 40–60% B; 9–16 min, 60–65% B; 16–18 min, 65–80% B; 18–21 min, 80–95% B; 21–23 min, 95–5% B; 23–25 min, 5% B. The injection volume was 5 μL. QC samples were deployed at random with every 20 experimental samples for the evaluation of LC–MS stability. The full MS resolution was set as 60,000 and the mass range was set at m/z 100–1500. For the dd-MS2 settings, MS resolution was set as 15,000.

Raw MS data (.raw) files were acquired by Xcalibur (version 4.4.16.14) and processed with Compound Discoverer 3.3 software. Metabolite annotation was performed by matching MS/MS spectra against KEGG (https://www.genome.jp/kegg/), mzCloud (https://www.mzcloud.org/), and LIPID MAPS (https://lipidmaps.org/) databases.

Targeted metabolomics for validation

Targeted quantification of the candidate metabolites was performed using an Agilent 1260 Infinity II Prime UPLC system coupled with an Agilent 6495 triple quadrupole mass spectrometer (Agilent Technologies, USA). Sample preparation followed the same protocol as that for untargeted metabolomics. Chromatographic separation was achieved on a Waters ACQUITY UPLC BEH C18 column (1.7 μm, 100 × 2.1 mm) maintained at 40 °C. The mobile phase consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B), delivered at a flow rate of 0.3 mL/min with an injection volume of 1 μL. Mass spectrometric detection was conducted using an Agilent Jet Stream (AJS) electrospray ionization (ESI) source. The source parameters were optimized as follows: capillary voltage was set at 3000 V for positive mode and − 3000 V for negative mode; nozzle voltage was 1500 V; ion transfer tube and vaporizer temperatures were maintained at 200 °C; and the sheath gas temperature was set to 250 °C with a flow rate of 11 L/min. Data acquisition was performed in multiple reaction monitoring (MRM) mode. To ensure data quality and system stability, blank (BK) and QC samples were injected after every 10 experimental samples.

Untargeted lipidomics analysis

Serum lipids were extracted using a modified Bligh-Dyer protocol [28, 29]. Briefly, 50 μL of serum samples were mixed with 300 μL methanol and 1 mL methyl tert-butyl ether, followed by shaking at 4 °C for 2 h. After adding 300 μL water, the mixture was centrifuged at 14,000 g for 15 min. The supernatants were dried and reconstituted in acetonitrile/isopropanol/water (65:30:5, v/v/v) before LC–MS/MS.

The data acquisition was performed using an UPLC system Thermo Scientific Vanquish UPLC coupled to a Thermo Scientific Orbitrap Exploris 120. A Waters ACQUITY UPLC BEH C18 column (1.7 μm, 100 × 2.1 mm) was used for the LC separation and the column temperature was kept at 55 °C. The mobile phases were (A) acetonitrile/water (6:4, v/v) and (B) isopropanol/acetonitrile (9:1, v/v), both containing 10 mM ammonium acetate. The gradient elution was programmed as follows: 0–1.5 min, 32% B; 1.5–15.5 min, increased to 85% B, 15.6–18.0 min, ramped to 97% B; 18.1–21.0 min, re-equilibrated to 32% B. The flow rate was 300 μL/min and QC samples were injected every 20 runs.

MS detection utilized in both positive and negative ESI modes with spray voltages of + 3250 V and – 3000 V. The ion transfer tube and vaporizer temperatures were set to 300 °C and 275 °C, respectively. Sheath gas was set as 40 arb. Aux gas was set as 10 arb. The full MS resolution was set as 60,000 and the mass range was set at m/z 100–1500. For the dd-MS2 settings, MS resolution was set as 15,000, with an isolation window of 1.5 m/z and stepped HCD collision energies of 30%, 50%, and 70%.

Lipid identification was performed via LipidSearch 4.1. The LipidSig database (http://chenglab.cmu.edu.tw/lipidsig/) and LIPID MAPS database (https://lipidmaps.org/) were used to annotate lipids.

Statistical analysis

Statistical analyses were performed using R version 4.5.0, SPSS Statistics version 27.0, and SIMCA version 14.1. Unsupervised principal component analysis (PCA) was performed on the processed omics data to visualize the overall sample distribution and the clustering of quality-control samples. Orthogonal partial least squares discriminant analysis (OPLS-DA) was used to evaluate supervised group separation and derive variable importance in projection (VIP) values. The permutation test in SIMCA software (14.1 version, Umetrics, Malmo, Sweden) was used to evaluate the overfitting risk of OPLS-DA models. Differential metabolites and lipids were screened using the criteria of VIP > 1 and FDR < 0.05. MetaboAnalyst 6.0 (https://www.metaboanalyst.ca/) and KEGG database (https://www.genome.jp/kegg/) were used for pathway enrichment analysis of differential metabolites and differential lipids.

Binomial logistic least absolute shrinkage and selection operator (LASSO) regression was performed among the GDM and GDMHDP participants in the discovery cohort using the 25 overlapping differential metabolites as candidate predictors and GDMHDP status as the binary outcome. Predictors were standardized, and an L1 penalty was applied (α = 1). The regularization parameter λ was selected by 10-fold cross-validation based on binomial deviance. The one-standard-error criterion (lambda.1se) was used, and metabolites with non-zero coefficients at lambda.1se were retained for subsequent evaluation.

The ability of metabolites and clinical variables to differentiate between GDM and GDMHDP patients was evaluated using ROC curve analysis and the AUC. Binary logistic regression was used to construct the combined ALC–DBP model. Net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to assess the incremental predictive value of adding ALC or DBP to the corresponding single-variable model. Additional sensitivity analyses were performed to assess the temporal robustness of the ROC results. Spearman correlation analysis was used to evaluate relationships among ALC, blood pressure parameters, other clinical variables, and differential lipid species.

Continuous variables were assessed for normality and are presented as mean ± standard deviation (SD) or median (interquartile range [IQR]), as appropriate. Comparisons among the N, GDM, and GDMHDP groups were performed using one-way analysis of variance (ANOVA) for normally distributed variables or the Kruskal–Wallis test for non-normally distributed variables. Direct comparisons between the GDM and GDMHDP groups were performed using Student’s t-test or the Mann–Whitney U test, as appropriate. Categorical variables are presented as n (%) and were compared using the chi-square test or Fisher’s exact test, as appropriate. All tests were two-sided, and P < 0.05 was considered statistically significant.

Results

Clinical characteristics of the study population

The demographic and clinical characteristics of the discovery and validation cohorts are presented in Table 1. In both cohorts, no significant differences were observed in maternal age, pre-pregnancy BMI, or lipid profiles among the three groups (P > 0.05), indicating comparable baseline characteristics. While glucose metabolic parameters (OGTT FPG, 1-h, 2-h plasma glucose, and HbA1c) were significantly elevated in the GDM and GDMHDP groups compared to the Normal group (P < 0.001), none of these indicators could distinguish between the GDM and GDMHDP groups (P > 0.05). DBP values were higher in the GDMHDP group than in the GDM group in both cohorts, with a similar direction and magnitude of the between-group difference, although statistical significance was reached only in the discovery cohort. SBP and hemoglobin also showed directionally concordant differences across cohorts, with higher values in the GDMHDP group (Table 1). None of the participants received aspirin prophylaxis or had a history of smoking.

Table 1.

Baseline characteristics of the study cohorts

Group	N	GDM	GDMHDP	P1	P2
Discovery cohort	(n = 58)	(n = 50)	(n = 20)
Age (years)	32.72 ± 4.33	32.76 ± 3.84	32.95 ± 5.50	0.980	0.889
Pre-BMI (kg/m2)	23.24 (22.50–24.73)	23.21 (22.24–24.23)	24.62 (22.85–26.70)	0.223	0.100
OGTT_FPG (mmol/L)	4.40 (4.27–4.59)	4.74 (4.55–4.89)	4.82 (4.52–5.07)	<0.001	0.439
OGTT_1H (mmol/L)	7.88 (6.67–8.67)	10.40 (9.65–10.73)	10.49 (9.66–11.27)	<0.001	0.301
OGTT_2H (mmol/L)	6.71 (6.14–7.61)	8.77 (8.30–9.78)	9.21 (8.79–9.85)	<0.001	0.156
SBP (mmHg)	110.24 ± 10.25	112.86 ± 9.87	120.30 ± 10.36	<0.001	0.009
DBP (mmHg)	65.57 ± 7.17	68.08 ± 6.83	74.90 ± 10.85	<0.001	0.015
TG (mmol/L)	2.15 (1.77–2.65)	2.39 (1.98–2.94)	2.55 (2.06–2.92)	0.061	0.682
TC (mmol/L)	6.17 ± 0.90	6.25 ± 1.20	5.88 ± 1.10	0.415	0.223
HDL (mmol/L)	2.02 ± 0.34	2.00 ± 0.36	1.88 ± 0.36	0.294	0.216
LDL (mmol/L)	3.39 ± 0.60	3.43 ± 0.76	3.23 ± 0.78	0.571	0.350
HbA1c (%)	5.30 (5.10–5.50)	5.50 (5.30–5.70)	5.50 (5.30–5.73)	<0.001	0.675
HB (g/L)	115.43 ± 7.65	115.94 ± 9.78	123.32 ± 11.93	0.005	0.023
Aspirin prophylaxis, n (%)	0 (0)	0 (0)	0 (0)	–	–
Smoking history, n (%)	0 (0)	0 (0)	0 (0)	–	–
Validation cohort	(n = 25)	(n = 22)	(n = 8)
Age (years)	32.08 ± 3.90	33.27 ± 3.41	33.00 ± 4.63	0.553	0.882
Pre-BMI (kg/m2)	23.88 (22.31–24.61)	23.63 (22.62–24.47)	23.30 (22.03–26.12)	0.987	0.888
OGTT_FPG (mmol/L)	4.44 (4.17–4.55)	4.65 (4.49–4.86)	4.81 (4.60–5.04)	<0.001	0.291
OGTT_1H (mmol/L)	7.76 (7.16–8.46)	10.23 (9.70–10.58)	10.04 (9.47–10.64)	<0.001	0.909
OGTT_2H (mmol/L)	6.55 (6.07–7.52)	8.71 (8.20–9.25)	8.83 (8.48–8.99)	<0.001	0.743
SBP (mmHg)	111.28 ± 11.36	113.18 ± 11.86	118.88 ± 10.48	0.273	0.225
DBP (mmHg)	64.32 ± 5.73	67.09 ± 9.02	73.12 ± 11.09	0.032	0.196
TG (mmol/L)	2.28 (1.83–2.78)	2.63 (2.14–2.81)	3.14 (2.12–3.42)	0.176	0.164
TC (mmol/L)	6.02 ± 1.00	6.31 ± 1.10	5.70 ± 1.21	0.369	0.241
HDL (mmol/L)	1.92 ± 0.35	2.10 ± 0.42	1.83 ± 0.29	0.143	0.069
LDL (mmol/L)	3.33 ± 0.64	3.48 ± 0.64	3.07 ± 0.81	0.335	0.229
HbA1c (%)	5.20 (5.00–5.30)	5.50 (5.30–5.60)	5.65 (5.47–5.82)	<0.001	0.100
HB (g/L)	114.04 ± 7.74	116.81 ± 6.87	119.88 ± 9.17	0.150	0.411
Aspirin prophylaxis, n (%)	0 (0)	0 (0)	0 (0)	–	–
Smoking history, n (%)	0 (0)	0 (0)	0 (0)	–	–

Data are presented as mean ± SD or median (interquartile range [IQR]), as appropriate

Pre-BMI pre-pregnancy body mass index, OGTT oral glucose tolerance test, FPG fasting plasma glucose, SBP systolic blood pressure, DBP diastolic blood pressure, TG triglycerides, TC total cholesterol, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, HbA1c glycated hemoglobin, HB hemoglobin.

P₁: P value for comparison among N, GDM, and GDMHDP groups, calculated using one-way ANOVA for normally distributed variables or Kruskal–Wallis test for non-normally distributed variables. P₂: P value for comparison between GDM and GDMHDP groups, calculated using Student’s t-test for normally distributed variables or Mann–Whitney U test for non-normally distributed variables.

Among women with GDMHDP, the discovery cohort included 9 cases of gestational hypertension, including 1 case of severe gestational hypertension, and 11 cases of preeclampsia, including 5 with severe features. The validation cohort included 3 cases of gestational hypertension and 5 cases of preeclampsia, including 1 with severe features; no cases of eclampsia occurred in either cohort (Table S3). Across both cohorts, the GDMHDP group generally showed earlier delivery, lower birth weight, and numerically higher frequencies of SGA, FGR, preterm birth, and NICU admission than the GDM group. Complete cohort-specific maternal and neonatal outcome data are provided in Table S4.

Metabolomics profile of GDM and GDMHDP

To characterize the metabolic signatures associated with GDM complicated by HDP, we performed untargeted metabolomics on serum samples from the N, GDM, and GDMHDP groups. A total of 2,337 metabolites were identified, of which 442 were annotated as endogenous metabolites. Representative total ion chromatograms (TIC) are shown in Fig. 1A. QC assessment demonstrated high reproducibility, with Pearson correlation coefficients exceeding 0.91 among QC samples (Fig. S1A), indicating robust instrument stability throughout the analysis.

Fig. 1.

Untargeted metabolomic profiling of N, GDM, and GDMHDP groups. A Representative TICs of serum samples obtained from the N, GDM, and GDMHDP groups. B Unsupervised PCA score plot showing the global metabolic profiles of the N, GDM, and GDMHDP groups. C–E OPLS-DA score plots for pairwise group comparisons: C N vs. GDM, D N vs. GDMHDP, and E GDM vs. GDMHDP. F Venn diagram illustrating the overlap of differential metabolites identified among the three pairwise comparisons. G Hierarchical clustering heatmap of the 25 shared differential metabolites across the N, GDM, and GDMHDP groups. H KEGG pathway enrichment analysis of differential metabolites identified in the GDM vs. GDMHDP comparison

The PCA score plot revealed a distinct separation between the normal control (N) group and the two pathological groups (Fig. 1B). Notably, the GDM and GDMHDP groups exhibited a degree of overlap in the PCA model, reflecting their shared underlying metabolic perturbations. To maximize class discrimination, supervised OPLS-DA was applied. Clear segregation was achieved in all pairwise comparisons (N vs. GDM, N vs. GDMHDP, and GDM vs. GDMHDP), as shown in Fig. 1C–E. Permutation tests (n=200) confirmed the validity of the OPLS-DA models, with all Q² intercepts below zero, indicating no overfitting (Fig. S1B-D).

A total of 188 differential metabolites were identified across the comparisons. Venn diagram analysis revealed 25 overlapping differential metabolites shared among the three pairwise comparisons (Fig. 1F). These overlapping metabolites were used as a conservative candidate set for subsequent biomarker screening, while the direct GDM vs. GDMHDP comparison remained central to the pathway-level interpretation. Hierarchical clustering heatmap analysis further visualized the distinct abundance patterns of these 25 intersection metabolites across the groups (Fig. 1G), highlighting their potential relevance to the metabolic differences associated with GDMHDP.

To decipher the biological functions underlying these metabolic alterations, we performed KEGG pathway enrichment analysis. Most notably, the direct comparison between GDM and GDMHDP revealed that "Pantothenate and CoA biosynthesis" was the most significantly enriched pathway (Fig. 1H, P = 0.00035, Impact = 0.23). Given the central role of Coenzyme A (CoA) in fatty-acid metabolism and energy homeostasis, enrichment of Pantothenate and CoA biosynthesis in the GDM versus GDMHDP comparison was consistent with altered CoA-related fatty-acid handling in GDMHDP. In contrast, comparisons against the Normal group (GDM vs. N and GDMHDP vs. N) showed shared perturbations primarily in "Glycerophospholipid metabolism" (P < 0.002, Impact > 0.3) and "Amino acid biosynthesis" (Fig. S1E, F), reflecting the systemic lipid and amino acid dysregulation characteristic of the general gestational hyperglycemic state.

Screening and validation of candidate biomarkers

To identify candidate biomarkers capable of distinguishing women with GDM who subsequently developed HDP from those with uncomplicated GDM, we used the 25 overlapping differential metabolites as candidate predictors in the LASSO regression analysis. This machine learning strategy highlighted five potential metabolites: M495 (Acetyl-L-carnitine, ALC), M196 (Oleamide), M569 (5-Hydroxy-DL-tryptophan), M710 (Palmitamide), and M2026 (1-(hexadec-1-enyl)− 2-hexadecanoyl-sn-glycero-3-phosphoethanolamine) (Fig. 2A, B). The discriminatory performance of these five metabolites was initially evaluated using ROC analysis in the discovery dataset. ALC showed excellent performance with an AUC of 0.882, and its relative abundance was significantly decreased in the GDMHDP group (Fig. 2C, D). M196 and M710 exhibited perfect separation (AUC = 1.0) with increased levels. M569 showed an AUC of 0.905 with decreased levels. In contrast, M2026 demonstrated only moderate discrimination (AUC = 0.742) (Fig. S2A-H). To prioritize candidates for targeted validation, we applied an AUC threshold of >0.80. Consequently, M2026 was excluded, and the remaining four metabolites (M196, M495, M569, and M710) were advanced to the validation stage.

Fig. 2.

Machine learning–based biomarker screening and targeted validation. A LASSO coefficient profiles of the selected metabolites across different regularization parameters. B Feature selection using LASSO regression for discriminating GDMHDP from GDM patients. C ROC curve of ALC in the discovery cohort. D Relative abundance of ALC in the GDM and GDMHDP groups in the discovery cohort. E ROC curve of ALC based on targeted LC–MS/MS quantification. F Normalized relative abundance of ALC in the GDM and GDMHDP groups measured by targeted LC–MS/MS

To further assess the reproducibility of the candidate metabolites, we performed targeted LC–MS/MS quantification in the validation cohort. The correlation analysis of QC samples demonstrated high instrumental stability and data quality (Fig. S3A). The targeted validation provided an additional assessment of candidate reproducibility and revealed clear differences in performance. Specifically, M196, M569, and M710 failed to reproduce the differential patterns observed in the untargeted analysis. Their AUC values decreased markedly to 0.443, 0.534, and 0.545, respectively, and no statistically significant group differences were observed in normalized relative abundances (Fig. S3B-G). These discrepancies may reflect analytical differences, spectral interference, or cohort-related variability and highlight the importance of targeted validation of discovery-stage candidates. In contrast, the reduction in ALC was reproduced in the targeted validation cohort. ROC analysis based on the targeted quantification data demonstrated strong discriminative performance for ALC, with an AUC of 0.869 (Fig. 2E). Consistent with the untargeted metabolomics results, targeted validation further confirmed a significant reduction in ALC in the GDMHDP group (Fig. 2F), supporting its reproducibility as a candidate biomarker. Consequently, ALC was the only candidate biomarker that retained both significant differential abundance and good discriminatory performance in the targeted validation cohort and was therefore selected for subsequent model construction.

Construction and evaluation of the combined prediction model

We systematically evaluated routinely available clinical variables, including anthropometric, glycemic, lipid, hematological, and blood pressure parameters (Table S1). DBP showed the highest AUC in the discovery cohort (AUC = 0.724) and remained among the better-performing variables in the validation cohort (AUC = 0.685). Among blood pressure parameters, DBP outperformed SBP in both cohorts, while SBP and DBP showed partly overlapping hemodynamic information. SBP and DBP were moderately correlated in both cohorts (discovery: ρ = 0.574, P = 2.06 × 10⁻⁷; validation: ρ = 0.664, P = 6.21 × 10⁻⁵; Fig. S4A, B). Therefore, DBP was selected as the representative blood pressure parameter for integration with ALC in the primary model. A binary logistic regression model integrating ALC and DBP was constructed. The combined model achieved an AUC of 0.907 in the discovery cohort and 0.886 in the validation cohort, consistently outperforming individual predictors (Fig. 3A, B). The NRI analysis showed statistically significant improvements in risk classification with the addition of DBP to ALC (Discovery NRI = 0.70, P = 0.005; Validation NRI = 0.864, P = 0.018). Using DBP as the baseline clinical model, addition of ALC significantly improved reclassification and discrimination performance in both cohorts, as supported by the NRI and IDI estimates (discovery: NRI = 1.260, P = 2.00 × 10⁻¹⁰; IDI = 0.304, P = 8.89 × 10⁻⁹; validation: NRI = 1.136, P = 0.001; IDI = 0.398, P < 0.001) (Table S2). Review of the individual clinical records confirmed that serum collection preceded formal HDP diagnosis in all 28 GDMHDP participants; 27 were diagnosed on a subsequent date, whereas one was diagnosed later on the day of sampling after serum collection (Table S3). Sensitivity analyses excluding the same-day diagnosis case and, subsequently, a participant with an isolated elevated DBP but no formal HDP diagnosis at that assessment produced results comparable to the primary analysis (Fig. S5). To examine the basis of this complementary predictive value, we performed Spearman correlation analysis. The results revealed no significant correlation between ALC and DBP or other clinical parameters (P > 0.05) (Fig. 3C, D), indicating that ALC and DBP capture partly distinct dimensions of disease risk.

Fig. 3.

Construction and performance of the combined ALC–DBP model. A ROC curves of ALC, DBP, and the combined ALC–DBP logistic regression model in the discovery cohort. B ROC curves of ALC, DBP, and the combined model in the validation cohort. C Spearman correlation analysis between ALC and DBP in the discovery cohort. D Spearman correlation analysis between ALC and DBP in the validation cohort

Lipidomic signatures associated with GDMHDP

To characterize lipidomic alterations associated with GDMHDP from a lipid metabolic perspective, untargeted lipidomics was performed on the N, GDM, and GDMHDP groups. A total of 1317 lipid metabolites were quantified. High correlations among QC samples confirmed robust data reproducibility and instrument stability (Fig. S3H). PCA showed separation between healthy controls and pathological groups (Fig. 4A), while OPLS-DA revealed distinct segregation between GDM and GDMHDP (Fig. 4B). The model robustness was validated by 200-permutation tests, ruling out overfitting (Fig. S3I). Based on the criteria of VIP > 1 and FDR < 0.05, specific differential lipid species were identified. Classification of the identified differential lipids revealed that lipid perturbations were predominantly clustered in membrane lipids and glycerolipids. Phosphatidylcholines (PC) accounted for the largest proportion (28.8%), followed by triglycerides (TG, 18.2%) and phosphatidylinositols (PI, 15.2%) (Fig. 4C).

Fig. 4.

Lipidomic alterations and multi-omics correlation analysis in GDMHDP. A PCA score plot of serum lipidomic profiles from N, GDM, and GDMHDP groups. B OPLS-DA score plot comparing lipidomic profiles between the GDM and GDMHDP groups. C Classification of differential lipid species between GDM and GDMHDP groups by lipid class. D Distribution of differential lipid species according to carbon chain length across major lipid classes. E Distribution of differential lipid species according to degree of unsaturation (number of double bonds). F Integrated correlation network linking ALC, DBP, and representative lipid species

To investigate the structural characteristics of the altered lipidome, we analyzed carbon chain length and double bond content. Accumulation of long-chain lipids (Fig. 4D): The GDMHDP group exhibited a distinct shift toward longer-chain lipid species. TG showed a dense cluster of upregulated species in the medium-to-long-chain range (C42–C54), a pattern consistent with enhanced de novo lipogenesis and energy overload. Notably, ceramides (Cer) were enriched at C39 and C42 lengths, whereas PI were depleted at C32. Changes in lipid unsaturation (Fig. 4E): Analysis of unsaturation revealed a shift toward a more saturated lipid profile, potentially associated with increased membrane rigidity. TGs displayed a divergent pattern: less unsaturated species with 2–3 double bonds were upregulated, whereas more highly polyunsaturated species with >4 double bonds were depleted. Consistent with this, lipid classes such as sphingomyelin (SM) and ceramides were enriched in saturated and monounsaturated forms (0–2 double bonds). This shift toward saturation may be associated with reduced membrane fluidity and oxidative loss of polyunsaturated lipids.

Multi-omics correlation network

To relate the lipid structural remodeling patterns to ALC and DBP, we performed an exploratory Spearman correlation analysis using the differential lipid species identified in the GDM versus GDMHDP comparison (Fig. 4F). Higher ALC levels were positively correlated with selected phosphatidylserine (PS) and PC species, including PS(42:4) (ρ = 0.54), and inversely correlated with selected Cer species, including Cer(d18:2/24:0) (ρ = −0.42), as well as with monolysocardiolipin [MLCL(54:10)] (ρ = −0.49). DBP was positively correlated with selected ether-linked phosphatidylethanolamine (PE-O) species, including PE(18:0e/22:5) (ρ = 0.41), and inversely correlated with selected PS and TG species. Among lipid species nominally associated with both ALC and DBP, the directions of association were opposite. These associations provided biological context for the complementary metabolic and hemodynamic information represented by ALC and DBP. A schematic summary integrating the observed multi-omics associations is shown in Fig. 5.

Fig. 5.

Schematic summary of multi-omics associations in GDMHDP

Discussion

GDM and HDP represent two major and often coexisting threats to maternal and fetal health. GDM primarily increases the risk of adverse perinatal outcomes through chronic hyperglycemia and metabolic dysregulation, whereas HDP remains a leading cause of maternal morbidity and mortality by impairing vascular function and placental perfusion [30]. Although traditionally regarded as distinct pregnancy complications, accumulating evidence suggests that shared metabolic and vascular pathways may contribute to their coexistence and mutual aggravation [31, 32]. In this prospective multi-omics study, we characterized metabolic differences associated with GDM complicated by HDP and evaluated an exploratory combined biomarker model within the GDM population.

The principal finding of this study is that lower serum ALC levels at 24–28 weeks of gestation emerged as a candidate marker for the early prediction of subsequently diagnosed HDP in women with GDM. Based on this observation, we constructed a combined model integrating ALC with DBP, which showed good discriminatory performance in the discovery and validation cohorts, with AUCs of 0.907 and 0.886, respectively. Although the primary clinical question focused on distinguishing GDMHDP from uncomplicated GDM, the Normal group was included in the discovery analysis to provide a reference for the broader metabolic alterations associated with GDM. The 25 metabolites shared across all three pairwise comparisons were then carried forward as a stringent candidate set for LASSO and ROC-based prioritization, followed by targeted LC–MS/MS validation. This strategy was intended to prioritize candidate signals within the broader metabolic context rather than to define GDMHDP-specific metabolic alterations. Consistent with emerging evidence, metabolomics-derived biomarkers have been shown to enhance risk prediction and provide mechanistic insights beyond conventional clinical parameters; for example, the incorporation of metabolomic risk scores into traditional cardiovascular prediction models has been reported to yield incremental prognostic value [33].

From a clinical perspective, routine blood pressure monitoring after 20 weeks of gestation alone may be insufficient for identifying women with GDM who are at risk of subsequently developing HDP. As observed in our study, blood pressure profiles showed a concordant pattern across cohorts, with higher values in the GDMHDP group and comparable between-group differences, although statistical significance was reached only in the discovery cohort. Importantly, even in the cohort in which the difference in DBP reached statistical significance, the mean DBP remained within the clinically normal range. These findings suggest that metabolic dysregulation may be detectable before hypertension is clinically established and may provide information beyond routine mid-gestational blood pressure assessment. In this context, ALC and DBP provide complementary information: ALC reflects the metabolic state, whereas DBP may reflect early hemodynamic susceptibility to subsequent HDP, even when values remain below the diagnostic threshold for hypertension. The lack of correlation between ALC and DBP, together with the improved performance observed when ALC was added to the DBP-only model, supports the complementary value of integrating metabolic and hemodynamic information.

The pronounced reduction in serum ALC may reflect alterations in mitochondrial bioenergetics and fatty-acid handling associated with the development of HDP in women with GDM. In our pathway analysis, enrichment of pantothenate and coenzyme A (CoA) biosynthesis in the GDMHDP group was consistent with altered CoA-related metabolism, which is closely linked to mitochondrial β-oxidation. Supporting this interpretation, previous studies have reported downregulation of key placental CoA-synthetic enzymes, such as pantothenate kinase 1 (PANK1), in preeclampsia, accompanied by reduced CoA pools and altered downstream lipid metabolism [34]. Reduced ALC may suggest a relative limitation in carnitine-dependent fatty-acid transport, potentially affecting the efficient entry of fatty acids into mitochondria for β-oxidation. Such a limitation could contribute to metabolic backlog and the accumulation of circulating medium- and long-chain triglycerides, potentially reflecting altered de novo lipogenesis secondary to impaired fatty-acid utilization. Excess non-oxidized fatty acids may then be diverted into alternative pathways, favoring the formation of lipotoxic intermediates. Consistent with this hypothesis, we observed a marked elevation of long-chain ceramides in the GDMHDP group, which were inversely correlated with ALC levels. Prior evidence is also consistent with such metabolic coupling. Previous studies in high-altitude pregnancies have reported abnormal accumulation of medium- and long-chain acylcarnitines, intermediates of incomplete fatty-acid oxidation, in women with preeclampsia, a pattern consistent with impaired mitochondrial lipid metabolism in hypertensive pregnancy disorders [35]. In parallel, oxidative stress–associated alterations in ceramide metabolism, including increased synthesis and reduced degradation, have been shown to elevate ceramide levels in both maternal circulation and placental tissue in preeclampsia [36]. Taken together, these findings support a potential association between reduced ALC, altered mitochondrial fatty-acid handling, and lipid remodeling in GDMHDP.

Lipidomic structural analysis suggested a more saturated lipid profile in the GDMHDP group, characterized by increased saturated and monounsaturated fatty acyl chains and reduced PUFAs. Such compositional changes have been shown to reduce membrane fluidity and increase membrane rigidity, thereby affecting endothelial cell function. In vitro studies have demonstrated that saturated fatty acids, such as stearic acid, decrease endothelial membrane fluidity and promote apoptosis, whereas PUFAs, including eicosapentaenoic acid (EPA), can reverse membrane stiffening and preserve endothelial viability [37]. Reduced membrane fluidity may interfere with endothelial nitric oxide synthase (eNOS) activity, potentially impairing nitric oxide (NO) production and vasodilatory capacity. Together with the correlation patterns described above, these structural lipid changes are consistent with previous reports of oxidative stress–related alterations in placental ceramide metabolism [36] and changes in plasmalogen-derived ether lipids that may represent a compensatory antioxidant response in preeclampsia [38]. Collectively, these findings suggest that reduced ALC and altered lipid composition may be associated with mitochondrial lipid stress, oxidative stress, and vascular dysfunction in GDMHDP.

Taken together, these findings illustrate the distinct but complementary roles of metabolomics and lipidomics in the present study. Metabolomics identified ALC as a reproducible circulating candidate for targeted confirmation and integration with DBP, whereas lipidomics characterized the accompanying lipid remodeling and its relationships with ALC and DBP. Thus, the lipidomic findings provided complementary biological context for the metabolic and hemodynamic alterations associated with GDMHDP, while the primary prediction analysis remained focused on the ALC–DBP model, for which ALC had been confirmed by targeted LC–MS/MS in a separate validation cohort.

Beyond its role as a candidate biomarker, the association of reduced ALC with altered lipid metabolism raises the possibility that carnitine-related metabolic pathways may also have therapeutic relevance. Accumulating evidence indicates that ALC supplementation can influence metabolic and vascular function in several metabolic disorders, including diabetes. For instance, phase I/II clinical trials have reported beneficial effects of ALC treatment in diabetic peripheral neuropathy [39]. In cardiovascular and endothelial models, L-carnitine and its derivatives have also been associated with reduced oxidative stress and improved endothelial function [40, 41]. In the context of our findings, early ALC supplementation in women with GDM and low circulating ALC may have the potential to modulate mitochondrial lipid handling, limit the accumulation of lipotoxic intermediates, and influence hemodynamic stress. Although these observations do not establish the efficacy of ALC supplementation for preventing HDP, they provide a rationale for further investigation of carnitine-related metabolic pathways in this setting. Future prospective studies are warranted to determine whether ALC supplementation can safely modify the metabolic alterations associated with reduced ALC and reduce the subsequent risk of HDP. Notably, the established use of low-dose aspirin for preeclampsia prevention in high-risk pregnancies illustrates the broader potential value of early preventive intervention [42]. From this perspective, ALC supplementation may represent a metabolically targeted strategy worthy of further investigation in women with GDM and low circulating ALC levels.

Although this study analyzed maternal serum metabolites, the observed metabolic signatures may partly reflect the pathophysiological status of the placenta, the central organ in the pathogenesis of HDP. The placenta is highly enriched in mitochondria, and mitochondrial dysfunction is widely recognized as a key mechanistic contributor to the development of preeclampsia [43, 44]. In placental trophoblasts and decidual cells from patients with preeclampsia, markers of mitochondrial dysfunction and oxidative stress have been consistently reported to be markedly elevated. In this context, the reduced serum ALC observed in pregnant women may reflect increased placental demand for fatty acid metabolism under metabolic stress. For instance, reduced placental expression of key fatty-acid oxidation enzymes and the high-affinity carnitine transporter OCTN2 has been reported in preeclampsia, suggesting impaired placental fatty-acid oxidation and carnitine handling [45]. Consistently, the alterations in selected ether-linked lipids and ceramide species detected in our study may, at least in part, reflect lipid remodeling in placental or maternal vascular tissues [36]. Thus, reduced circulating ALC may partly reflect placental mitochondrial and fatty-acid metabolic alterations, although the tissue origin of the circulating signal cannot be determined from the present study.

The strengths of this study include its prospective design, separate discovery and validation cohorts, and targeted LC–MS/MS confirmation of the candidate metabolite identified by untargeted metabolomics. Several limitations should be acknowledged. The modest number of GDMHDP cases and the single-center internal validation may limit the generalizability of the findings. The use of the three-way intersection may have excluded metabolites that differed between GDM and GDMHDP but not in comparisons involving the Normal group. Although a pregnancy-specific improved triglyceride–glucose index (TyGIS) has been developed as a surrogate measure of insulin sensitivity, its calculation requires fasting insulin, which was unavailable in our cohort [46]. Larger, multicenter cohorts are needed to further validate the generalizability of our findings. The inclusion of fasting insulin in future studies would also enable a more comprehensive assessment of pregnancy-specific insulin sensitivity. Despite these limitations, the consistent reduction in ALC observed in both the discovery and validation cohorts supports the robustness of its association with subsequently diagnosed HDP in women with GDM.

Conclusion

In summary, reduced maternal serum ALC at 24–28 weeks of gestation was consistently associated with GDM complicated by subsequently diagnosed HDP and was reproduced by targeted LC–MS/MS in a separate validation cohort. Combining ALC with DBP improved predictive discrimination compared with either marker alone, supporting the complementary metabolic and hemodynamic information provided by these variables. The accompanying lipidomic patterns were consistent with altered mitochondrial fatty-acid handling, oxidative lipid stress, and lipid remodeling. These findings support serum ALC as a candidate metabolic marker for mid-gestational prediction of HDP in women with GDM, although larger multicenter prospective studies are needed to validate its generalizability and clinical utility.

Abbreviations

GDM

Gestational diabetes mellitus

HDP

Hypertensive disorders of pregnancy

GDMHDP

Gestational diabetes mellitus complicated by hypertensive disorders of pregnancy

ROC

Receiver operating characteristic

ALC

Acetyl-L-carnitine

DBP

Diastolic blood pressure

SBP

Systolic blood pressure

IDF

International Diabetes Federation

LC–MS

Liquid chromatography-mass spectrometry

OGTT

Oral glucose tolerance test

SGA

Small for gestational age

FGR

Fetal growth restriction

LGA

Large for gestational age

NICU

Neonatal intensive care unit

QC

Quality control

AJS

Agilent jet stream

ESI

Electrospray ionization

MRM

Multiple reaction monitoring

BK

Blank

FDR

False discovery rate

HB

Hemoglobin

TIC

Total ion chromatograms

PCA

Principal component analysis

OPLS-DA

Orthogonal partial least squares discriminant analysis

LASSO

Least absolute shrinkage and selection operator

NRI

Net reclassification improvement

IDI

Integrated discrimination improvement

Author contributions

Conceptualisation, XJ, QW, HL and WZ; Methodology, XJ, QW, HL, JG, XZ, HY and XH; Formal Analysis and Data Curation, JX and HL; Writing-Original Draft Preparation, XJ; Writing-Review and Editing, XJ, QW, XZ, ZL, DC, WZ and LM; Visualisation, XJ; Supervision, DC, WZ, and LM; All authors read and approved the final manuscript.

Funding

This work was supported by the National Key Research and Development Program of China [2021YFC2401000]; Zhejiang Province Traditional Chinese Medicine Key Laboratory Project [GZY-ZJ-SY-2303]; the Medical and Health Technology Program of Zhejiang Province [WKJ-ZJ-2324]; the 4+X Clinical Research of Women's Hospital, School of Medicine, Zhejiang University [ZDFY2022-4XB101]; the Zhejiang Provincial Natural Science Foundation of China [ZCLQN25H0402]; and Huadong Medicine Joint Fund of the Zhejiang Provincial Natural Science Foundation of China [LHDMD25H040001].

Data availability

Data beyond those included in the manuscript are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of the Obstetrics and Gynecology Hospital Affiliated to Zhejiang University School of Medicine (Approval No. IRB-20210112-R), and all participants provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.