Associations between arginine metabolism and gestational diabetes mellitus: an exploratory longitudinal study using widely targeted metabolomics

Abstract

Objective

To characterize longitudinal metabolic alterations associated with gestational diabetes mellitus (GDM) and to identify candidate metabolite signals for earlier risk assessment using a widely targeted metabolomics platform.

Methods

In this prospective cohort, 35 women who developed GDM and 35 matched healthy controls underwent fasting blood sampling in early pregnancy (6–13 weeks) and mid-pregnancy (24–28 weeks). Widely targeted metabolomics and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed. Multivariate analyses, logistic regression, receiver operating characteristic (ROC) analysis, internal cross-validation, and restricted cubic spline modeling were applied within an exploratory framework.

Results

In early pregnancy, 11 differential metabolites were identified, including 6 amino acid-related metabolites, but no robust signal remained after false discovery rate (FDR) correction. By mid-pregnancy, 35 differential metabolites were identified, including 13 amino acid-related metabolites. Among 19 amino acid-related metabolites examined in focused analyses, L-arginine was the only amino acid-related metabolite that remained significant after FDR correction (FDR = 0.043). Higher mid-pregnancy L-arginine was associated with increased odds of GDM (aOR = 2.840, 95% CI 1.481–5.446), was positively correlated with 1-h (r = 0.28) and 2-h (r = 0.26) glucose levels during OGTT, and showed modest discrimination (AUC = 0.754; mean cross-validated AUC = 0.754). Quartile analyses showed a more pronounced risk increase in the highest exposure group, and restricted cubic spline analysis suggested an overall positive association with a possible nonlinear trend at higher levels. KEGG analyses highlighted arginine- and amino acid-related pathways, including arginine and proline metabolism and arginine biosynthesis, while network analysis suggested a potential link to mTOR signaling.

Conclusion

In this exploratory longitudinal analysis, pregnancies complicated by GDM showed progressively increasing amino acid-related metabolic disturbances from early to mid-pregnancy. Mid-pregnancy L-arginine emerged as a candidate metabolic signal associated with GDM risk, post-load glycemia, and greater risk elevation at higher levels. These findings support further investigation of amino acid-focused metabolic profiling as a research-stage approach for earlier GDM risk identification.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-026-09414-5.

Introduction

Gestational diabetes mellitus (GDM) is one of the most common metabolic complications of pregnancy and is characterized by hyperglycemia and insulin resistance first recognized during gestation. Its global prevalence has been estimated to be as high as 14% [1]. GDM is associated with increased risks of adverse maternal and perinatal outcomes, including preeclampsia, excessive fetal growth, preterm birth, and long-term metabolic disturbances in both mothers and offspring [2]. Early identification of women at high risk for GDM may enable timely preventive interventions and thereby reduce maternal morbidity [3].

Accumulating evidence suggests that, in women who subsequently develop GDM, metabolic abnormalities may already be present in early pregnancy [4]. Longitudinal monitoring from early to mid-pregnancy may therefore help delineate the metabolic trajectory of disease development and improve understanding of the temporal characteristics of these alterations [5]. Such an approach may not only advance understanding of the pathophysiological evolution of GDM but also provide clues for earlier risk assessment.

Metabolomics enables systematic characterization of dynamic changes in small-molecule metabolites and thereby links biochemical perturbations to clinical phenotypes [6]. As such, it has become an important tool for identifying disease-related metabolic signatures and candidate biomarkers. Untargeted metabolomics has offered broad metabolic coverage but has been relatively limited in quantitative accuracy and downstream validation, whereas targeted metabolomics provides high quantitative precision but is generally restricted to predefined metabolite panels. Widely targeted metabolomics combines the strengths of both approaches by maintaining broad metabolite coverage while improving sensitivity, quantitative performance, and scalability, making it particularly suitable for characterizing the dynamic metabolic changes that occur during pregnancy [7].

Against this background, the present study aimed to characterize longitudinal metabolic alterations associated with GDM from early to mid-pregnancy, identify early metabolic disturbances related to subsequent disease development, and screen candidate metabolite biomarkers with potential translational relevance. To this end, we conducted a prospective cohort study and applied widely targeted metabolomics to samples collected at different gestational stages. We further integrated metabolomic features with clinical information to provide evidence supporting earlier risk identification for GDM.

Methods

Study design and participants

This study was designed as a prospective longitudinal cohort study. From May 2022 to December 2023, pregnant women receiving routine antenatal care at Peking Union Medical College Hospital (PUMCH) were consecutively recruited. Eligible participants were women aged 20–45 years with a singleton pregnancy, a gestational age of 6–13 weeks at enrollment, and the ability to complete follow-up. Exclusion criteria were a previous diagnosis of type 1 or type 2 diabetes mellitus, known hyperlipidemia or hyperbilirubinemia, abnormal liver function, renal insufficiency, or any severe systemic disease that might affect pregnancy outcomes or metabolite levels.

The study protocol was approved by the Ethics Committee of PUMCH (I-25PJ0478). Written informed consent was obtained from all participants before enrollment. All data were used exclusively for research purposes and were managed under strict confidentiality. The study was conducted in accordance with the principles of the Declaration of Helsinki.

Fasting venous blood samples were collected from all participants during both early and mid-pregnancy. After sterile processing, samples were stored at − 80 °C until analysis. All participants underwent a 75-g oral glucose tolerance test (OGTT) at 24–28 weeks of gestation. GDM was diagnosed according to the 2010 International Association of Diabetes and Pregnancy Study Groups criteria, defined as any one of the following plasma glucose values meeting or exceeding the corresponding thresholds: fasting glucose 5.1 mmol/L, 1-h post-load glucose 10.0 mmol/L, or 2-h post-load glucose 8.5 mmol/L [8].

Case and control selection and matching

A total of 267 pregnant women were enrolled. Among them, 180 participants provided blood samples at both early and mid-pregnancy and were therefore included in the longitudinal metabolomics analysis. The participant flow, sample collection, GDM diagnosis, and analytical workflow are shown in (Fig. 1).

Fig. 1.

Flowchart of participant enrollment, follow-up, and metabolomics analysis

Among the 180 women with paired early- and mid-pregnancy samples, 35 were diagnosed with GDM. To improve baseline comparability, healthy controls were selected from women who did not develop GDM during the same study period using a 1:1 frequency-matching strategy based primarily on maternal age and pre-pregnancy body mass index (BMI) [1]. Ultimately, 35 healthy pregnant women were included as controls. The matching results are presented in (Table 1).

Table 1.

Characteristics of the pregnant women involved in the study

Variable			GDM (n = 35)	Control (n = 35)	P-value
Age, y, mean (SD)			33.51 (4.41)	32.74 (3.76)	0.434
Pre-pregnancy BMI, kg/m2, mean (SD)			22.32 (3.14)	21.63 (2.57)	0.318
Systolic blood pressure, mmHg, mean (SD)			116.80 (11.18)	115.14 (10.72)	0.529
Diastolic blood pressure, mmHg, mean (SD)			69.17 (10.50)	65.66 (10.78)	0.172
Education level, n (%)		Master or above	18 (51.4%)	20 (57.1%)	0.379
		Bachelor degree	13 (37.2%)	14 (40.0%)
		College degree	4 (11.4%)	1 (2.9%)
Gravidity, n (%)		1	18 (51.4%)	27 (77.1%)	0.009
		2	11 (31.4%)	7 (20.0%)
		≥ 3	6 (17.2%)	1 (2.9%)
Parity, n (%)		0	26 (74.3%)	31 (88.6%)	0.363
		1	9 (25.7%)	4 (11.4%)
Family history, n (%)		Hypertension	11 (31.4%)	9 (25.7%)	0.597
		Diabetes	12 (34.3%)	5 (14.3%)	0.051
		PCOS	5 (14.3%)	0 (0.0%)	0.020
History of GDM, n (%)			4 (11.4%)	0 (0.0%)	0.039
FPG, mmol/L, mean (SD)			4.82 (0.39)	4.80 (0.36)	0.625
GA, %, mean (SD)			13.28 (1.30)	13.28 (1.09)	0.867
HGB, g/L, median (IQR)			130.50 (13.25)	132.00 (8.00)	0.888
WBC, ×109/L, median (IQR)			8.51 (2.19)	7.57 (3.26)	0.080
ALP, U/L, median (IQR)			53.00 (16.50)	51.00 (16.00)	0.295
GGT, U/L, median (IQR)			16.00 (10.26)	14.00 (7.00)	0.104
ALT, U/L, median (IQR)			17.00 (10.25)	13.00 (8.00)	0.002
AST, U/L, median (IQR)			18.00 (7.25)	18.00 (4.00)	0.571
CRP, median (IQR)			1.22 (2.03)	0.65 (0.72)	0.003
TC, mmol/L, median (IQR)			4.23 (1.24)	4.05 (0.66)	0.050
TG, mmol/L, median (IQR)			1.04 (0.67)	0.70 (0.41)	< 0.001
HDL, mmol/L, median (IQR)			1.53 (0.42)	1.49 (0.28)	0.245
LDL, mmol/L, median (IQR)			2.38 (0.95)	2.16 (0.73)	0.215
OGTT, mmol/L, median (IQR)	0 h		4.70 (0.60)	4.30 (0.30)	< 0.001
	1 h		9.70 (1.59)	6.83 (1.64)	< 0.001
	2 h		8.80 (1.80)	6.20 (1.80)	< 0.001

GDM Gestational diabetes mellitus, FPG Fasting plasma glucose, GA Glycated albumin, HGB Hemoglobin, WBC White blood cell, ALP Alkaline phosphatase, GGT Gamma- glutamyl transferase, ALT Alanine aminotransferase, AST Aspartate aminotransferase, CRP C-reactive protein, TC Total cholesterol, TG Triglyceride, HDL High-density lipoprotein, LDL Low-density lipoprotein, OGTT Oral glucose tolerance test

In this study, matching was primarily used to improve baseline comparability between cases and controls. Given the possibility of residual confounding in an observational design, subsequent association analyses additionally adjusted for relevant covariates in regression models. Considering the sample size and the need to avoid model overfitting, the main adjusted models included only the core covariates of maternal age and pre-pregnancy BMI [9].

Sample processing and widely targeted metabolomics analysis

Serum sample pretreatment

Serum samples were thawed on ice and vortex-mixed for 10 s. A 50-µL aliquot of serum was then mixed with 300 µL of methanol, vortexed for 3 min, and centrifuged at 12,000 rpm for 10 min at 4 °C. The supernatant was collected and centrifuged again at 12,000 rpm for 5 min at 4 °C. The resulting supernatant was then kept at − 20 °C for 30 min, followed by a further centrifugation at 12,000 rpm for 3 min at 4 °C. Finally, 150 µL of supernatant was transferred into the insert of an autosampler vial for instrumental analysis.

Chemical reagents

HPLC-grade methanol and acetonitrile were purchased from Merck (Germany). HPLC-grade ammonium formate, ammonia solution, and formic acid were obtained from Shanghai Aladdin Biochemical Technology Co., Ltd. Commercial reference standards and isotope-labeled internal standards were purchased from Shanghai ZZBIO Co., Ltd.

Widely targeted metabolomics platform

Serum metabolite profiling was performed using an ultra-performance liquid chromatography–electrospray ionization tandem mass spectrometry (UPLC-ESI-MS/MS) platform. Chromatographic separation was carried out on a SCIEX ExionLC AD system equipped with an HSS T3 column (1.8 μm, 2.1 mm × 100 mm; Waters). The mobile phases consisted of water containing 0.1% formic acid and acetonitrile containing 0.1% formic acid. The column temperature was maintained at 40 °C, the flow rate was 0.4 mL/min, and the injection volume was 2 µL. The gradient elution program was as follows: 95:5 at 0 min, 10:90 at 11.0 min, maintained at 10:90 until 12.0 min, returned to 95:5 at 12.1 min, and maintained at 95:5 until 14.0 min.

Mass spectrometric detection was performed using a QTRAP^® LC-MS/MS system and a TripleTOF 6600 system equipped with an ESI Turbo Ion-Spray interface. These systems were used for multiple reaction monitoring (MRM)-based quantitative scanning and information-dependent acquisition (IDA)-based MS/MS spectral acquisition, respectively.

Quality control

To assess LC-MS instrument stability and ensure data reliability, pooled quality control (QC) samples were prepared by combining 10 µL of supernatant from each sample. Throughout the analytical run, one blank sample and one QC sample were injected at both the beginning and the end of the sample sequence, and one QC sample was inserted after every 10 study samples. In principal component analysis (PCA), QC samples clustered tightly, indicating stable instrument performance and acceptable within-batch data quality.

Metabolomics data preprocessing and quality control

Metabolites of different molecular weights were first separated by liquid chromatography. Characteristic ions for each metabolite were then monitored using the multiple reaction monitoring (MRM) mode of the triple quadrupole mass spectrometer, and the signal intensity of the corresponding ions was recorded as counts per second (CPS). Peak extraction and correction were performed using MultiQuant 3.0.3 software (AB SCIEX) based on mass-to-charge ratio (m/z) and retention time (RT). Only peak area data meeting the criteria of a signal-to-noise ratio (S/N) > 3 and an RT shift of no more than 0.2 min were retained for further analysis. For widely targeted and quantitative metabolomics analyses, metabolites with a coefficient of variation > 50% in QC samples or a missing rate > 20% within either group were excluded. For the remaining metabolites, missing values were imputed as one-half of the minimum observed value of the corresponding metabolite within the same group.

Statistical analysis

All statistical analyses were performed using SPSS version 27, R version 4.5.3, and SIMCA-P version 14.1. Continuous variables are presented as mean ± standard deviation, mean ± standard error, or median (interquartile range), as appropriate according to their distribution. Categorical variables are presented as counts (percentages). Baseline characteristics were compared between groups using the chi-square test, Student’s t test, or the Wilcoxon rank-sum test, as appropriate. All tests were two-sided, and P < 0.05 was considered statistically significant.

Widely targeted metabolomic analysis

Unsupervised PCA was performed in SIMCA-P 14.1 (Umetrics AB, Umeå, Sweden) to evaluate the overall distribution of the metabolomic data. Orthogonal partial least-squares discriminant analysis (OPLS-DA) was used to identify metabolic patterns related to between-group discrimination. Model performance was assessed using R²X, R²Y, and Q². To evaluate the risk of overfitting, 200-permutation testing was performed. Q² values were obtained from the internal cross-validation procedure embedded in the modeling workflow.

Between-group differential metabolite analysis was conducted using Student’s t test, with P < 0.05 used as the nominal threshold for screening differential metabolites. Fold change (FC) was used to describe the magnitude of between-group differences, and variable importance in projection (VIP) values derived from the OPLS-DA model were used to evaluate the contribution of individual metabolites to group separation. Differential metabolites were therefore defined at the nominal significance level (P < 0.05) [10], while VIP values and false discovery rate (FDR)-adjusted P values were additionally reported to indicate variable importance and result robustness, respectively. The P values, VIP values, FDR-adjusted P values, and FCs for differential metabolites in early and mid-pregnancy are provided in Tables S1 and S2.

Based on metabolite annotation, metabolites classified at the first-level category as “Amino acid and its metabolites” among the differential metabolites identified in early and mid-pregnancy were combined, yielding a panel of 19 amino acid-related metabolites for subsequent focused analyses. For data normalization, total area peak normalization (TAPN) was first applied to correct for overall differences in signal intensity across samples. Z-score transformation was then performed on the TAPN-normalized data to standardize differences in scale across metabolites. The resulting standardized Z-scores were used in subsequent analyses of associations between candidate metabolites and clinical phenotypes, as well as in dose-response analyses. TAPN- and Z-score–processed results for 19 amino acid-related metabolites in early and mid-pregnancy are provided in Supplementary Material 1.

Association analyses

For the selected candidate metabolites, univariable logistic regression was used in Model 1 to evaluate their associations with GDM risk, with odds ratios (ORs) and 95% confidence intervals (95% CIs) estimated. Model 2 was adjusted for maternal age and pre-pregnancy BMI to assess whether the observed associations were independent of these core clinical factors. Model 3 was further adjusted for gravidity and family history of diabetes. Spearman correlation analysis was used to assess the correlations between candidate metabolites and OGTT glucose levels. In sensitivity analyses, given the sparse-data structure of certain baseline variables with significant between-group imbalance, Firth penalized logistic regression was additionally performed by adding previous GDM history or PCOS history separately to Model 2.

Receiver operating characteristic (ROC) curve analysis was performed to evaluate the discriminatory ability of candidate metabolites for GDM, and the area under the curve (AUC) was calculated. Internal validation was performed using repeated stratified 5-fold cross-validation with 20 repeats, and the mean, standard deviation, median, and 2.5th–97.5th percentiles of the cross-validated AUCs were reported [11, 12].

To explore potential metabolic heterogeneity within GDM, exploratory subgroup analyses were performed according to OGTT abnormality patterns [13]. Based on the results of the 75-g OGTT at 24–28 weeks of gestation, women with GDM were classified into an impaired fasting glucose subgroup (IFG; isolated OGTT-0 h ≥ 5.1 mmol/L), an impaired glucose tolerance subgroup (IGT; OGTT-1 h ≥ 10.0 mmol/L and/or OGTT-2 h ≥ 8.5 mmol/L with fasting glucose below the diagnostic threshold), and a mixed subgroup. Because the IFG and mixed subgroups were both small, these participants were further combined into a fasting-involved group (FIG) for supplementary analyses. Associations between L-arginine and different phenotypic subgroups were then evaluated using the same regression framework as in the main analysis.

To assess the dose-response relationship between L-arginine and GDM risk, participants were categorized into quartiles according to mid-pregnancy L-arginine levels. Using the lowest quartile (Q1) as the reference group, ORs for the remaining quartiles were estimated using logistic regression. Restricted cubic spline (RCS) regression was further used to model the potential non-linear relationship between mid-pregnancy L-arginine and GDM risk.

Results

Participant characteristics and baseline features

Among the 180 pregnant women included in the longitudinal analysis, 35 (19.44%) were ultimately diagnosed with GDM, which was generally consistent with the prevalence reported in previous studies [14]. According to the predefined matching strategy, 35 healthy pregnant women who did not develop GDM during the same study period were selected as controls. No missing values were present in the clinical baseline variables included in the analysis. Compared with the healthy control group, women in the GDM group had higher gravidity (P = 0.009), a higher prevalence of polycystic ovary syndrome (PCOS) history (P = 0.020), and a higher frequency of previous GDM (P = 0.039). Laboratory testing further showed that the GDM group had significantly higher levels of C-reactive protein (CRP, P = 0.003), alanine aminotransferase (ALT, P = 0.002), total cholesterol (TC, P = 0.050), and triglycerides (TG, P < 0.001) than the healthy control group (Table 1).

Metabolic profile characteristics and multivariate pattern recognition

PCA was first performed to assess the overall distribution of the metabolomic data. A modest between-group separation trend was observed at the global metabolic level, although substantial overlap remained. In early pregnancy, the first two principal components explained 10.80% and 6.37% of the total variance, respectively, whereas in mid-pregnancy they explained 9.89% and 6.80% of the total variance, respectively (Fig. 2A-B).

Fig. 2.

PCA score plots, OPLS-DA score plots, and permutation tests of metabolomic profiles between GDM and healthy control groups in early and mid-pregnancy. A–B PCA score plots; (C–D) OPLS-DA score plots; (E–F) 200-time permutation tests. A, C, and E represent early pregnancy; B, D, and F represent mid-pregnancy

OPLS-DA was subsequently conducted to explore discriminatory metabolic patterns between groups. In early pregnancy, the OPLS-DA model yielded an R²X of 0.128, an R²Y of 0.636, and a Q² of − 0.385; in 200-permutation testing, the permutation P values for R²Y and Q² were 0.855 and 0.640, respectively (Fig. 2C and E). In mid-pregnancy, the corresponding R²X, R²Y, and Q² values were 0.129, 0.670, and − 0.094, with permutation P values of 0.840 for R²Y and 0.145 for Q² (Fig. 2D and F).

Within the supervised analytical framework, a modest separation trend was observed between groups, although model robustness was limited. Overall, global metabolic differences between groups were modest, and the sample distribution may still have been influenced by other sources of variation.

Differential metabolites in early and mid-pregnancy

Characteristics of differential metabolites in early pregnancy

A total of 11 differential metabolites were identified between the two groups in early pregnancy, including 8 upregulated and 3 downregulated metabolites (Figure S1, Table S1). Detailed legends for supplementary figures are provided in Supplementary Material 2. In terms of metabolite class, amino acid and its metabolites accounted for the largest proportion, comprising 6 metabolites, all of which were upregulated. This pattern suggests that women who later developed GDM had already shown a tendency toward more active amino acid metabolism in early pregnancy. Although these early-pregnancy differential metabolites reached nominal statistical significance, none remained robust after FDR correction, suggesting that metabolic differences between groups were still limited at this stage.

3.3.2 Characteristics of differential metabolites in mid-pregnancy

A total of 35 differential metabolites were identified, including 31 upregulated and 4 downregulated metabolites, indicating that GDM-related metabolic disturbances became more pronounced with gestational progression (Fig. 3A, Table S2). The spectrum of differential metabolites was also more diverse in mid-pregnancy, encompassing free fatty acids, oxidized lipids, amino acids and their derivatives, organic acids and their derivatives, and hormone-related compounds (Table S2). Amino acid and its metabolites remained the predominant metabolite class, comprising 13 metabolites, of which 12 were upregulated and 1 was downregulated. After FDR correction, only L-arginine (FDR = 0.043) and Phosphatidylethanolamine lyso alkenyl 18:2 (FDR = 0.039) remained significant, among which L-arginine was the only amino acid-related metabolite. N6-Acetyl-L-Lysine reached nominal statistical significance (P = 0.049), but was considered only a borderline differential signal for its VIP value (VIP = 0.911).

Fig. 3.

Identification of L-arginine as a key candidate metabolite and its association with GDM risk and glycemic phenotypes. A Volcano plot of differential metabolites in mid-pregnancy. The x-axis represents log2 fold change and the y-axis represents -log10 nominal P values derived from Student’s t-test. Points were color-coded according to nominal statistical significance and direction of change for visualization. B ROC curve analysis for the diagnostic performance of L-arginine. C Longitudinal changes of L-arginine levels from early to mid-pregnancy. D–F Spearman correlation between mid-pregnancy L-arginine and OGTT. G Forest plot of GDM risk according to L-arginine quartiles. H RCS analysis between L-arginine and GDM risk

Identification of the key candidate metabolite and its associations with GDM risk and glycemic phenotypes

Identification of L-arginine

Because amino acid and its metabolites were the predominant differential metabolite class in both early and mid-pregnancy, we performed a focused analysis. Among the 19 amino acid-related metabolites included in this analysis, L-arginine detected in positive ion mode in mid-pregnancy emerged as one of the most distinct metabolic features differentiating GDM pregnancies from healthy pregnancies and remained statistically significant after FDR correction (P < 0.001, VIP = 1.612, FDR = 0.043, FC = 1.300) (Fig. 3A, Table S3).

Associations of L-arginine with GDM risk and glycemic phenotypes

Univariable logistic regression analysis showed that higher mid-pregnancy L-arginine levels were associated with an increased risk of GDM (OR = 2.842, 95% CI: 1.498–5.390, P = 0.001). This association remained after adjustment for maternal age and pre-pregnancy BMI (aOR = 2.840, 95% CI: 1.481–5.446, P = 0.002) and was materially unchanged in Model 3 and subsequent sensitivity analyses (Table 2).

Table 2.

Association between L-arginine and GDM under primary and sensitivity models

Model	P	OR (95%CI)
Model 1	0.001	2.842 (1.498–5.390)
Model 2	0.002	2.840 (1.481–5.446)
Model 3	0.002	2.929 (1.461–5.869)
Model2 + previous GDM	< 0.001	2.555 (1.452–5.096)
Model2 + PCOS history	< 0.001	2.668 (1.460–5.692)

Note: Model 1 was unadjusted. Model 2 was adjusted for maternal age and pre-pregnancy BMI. Model 3 was additionally adjusted for gravidity and family history of diabetes. Because previous GDM history and PCOS history were sparse variables, additional sensitivity analyses were performed by adding each variable separately to Model 2 using Firth penalized logistic regression

ROC curve analysis indicated that L-arginine had modest discriminatory ability for GDM (AUC = 0.754) (Fig. 3B). Internal validation further showed a mean AUC of 0.754 (SD = 0.117), a median AUC of 0.755, and a 2.5th–97.5th percentile range of 0.673–0.847, indicating overall moderate discriminatory performance with some variability across repeated data splits (Figure S2, Table S4).

Longitudinal analysis revealed divergent trajectories of L-arginine between the two groups. From early to mid-pregnancy, L-arginine levels increased in the GDM group but decreased in healthy controls (Fig. 3C). Moreover, mid-pregnancy L-arginine levels were positively correlated with 1-h (r = 0.28) and 2-h (r = 0.26) plasma glucose concentrations during the 75-g OGTT (Fig. 3D–F).

Dose-response relationship of L-arginine

When participants were categorized into quartiles according to mid-pregnancy L-arginine levels, a dose-dependent pattern in the association between L-arginine and GDM risk was observed. Using the lowest quartile (Q1) as the reference group, the risk of GDM increased across L-arginine quartiles. This pattern was consistent in both the unadjusted model and the model adjusted for maternal age and pre-pregnancy BMI. Specifically, no clear increase in risk was observed in Q2, the risk tended to increase in Q3, and a significant increase was observed in Q4 (Fig. 3G). RCS analysis further showed an overall association between L-arginine and GDM risk (P for overall = 0.005), with a more apparent rise in risk at higher L-arginine levels (Fig. 3H). The test for nonlinearity was of borderline significance (P for nonlinear = 0.094), suggesting a possible nonlinear trend.

Exploratory subgroup analyses according to OGTT abnormality patterns

Exploratory subgroup analyses were performed according to patterns of OGTT abnormalities (Table S5). The association between L-arginine and GDM was more evident in the IGT and remained stable after adjustment for maternal age and pre-pregnancy BMI (aOR = 3.038, 95% CI: 1.410–6.548, P = 0.005). No statistically significant association was observed in the IFG, whereas the association became stronger after women with mixed abnormalities were combined with those with isolated fasting abnormalities into the FIG. Overall, these findings suggest that the association between L-arginine and GDM may be more pronounced in phenotypes characterized by post-load hyperglycemia.

Pathway enrichment and network analysis

KEGG pathway enrichment analysis was performed to further explore the biological context. In early pregnancy, the differential metabolites were enriched in four metabolic pathways, among which arginine and proline metabolism was the only amino acid-related pathway. By mid-pregnancy, the number of enriched pathways increased to 74, including 10 pathways related to amino acid metabolism. L-arginine was involved in 14 enriched pathways, with arginine and proline metabolism remaining enriched and arginine biosynthesis additionally identified (Table 3). KEGG network analysis further suggested that L-arginine related to the mTOR signaling pathway and nitric oxide synthesis-related reaction/enzyme nodes (Figure S3). All enriched metabolic pathways identified in early and mid-pregnancy are listed in Table S6.

Table 3.

Selected KEGG pathways involving L-arginine in mid-pregnancy

Pathway	Total	Hit	P value
Arginine biosynthesis	7	3	0.093
Arginine and proline metabolism	8	3	0.133
D-Amino acid metabolism	8	4	0.027
Aminoacyl-tRNA biosynthesis	10	4	0.064
Metabolic pathways	93	17	0.329
Biosynthesis of amino acids	16	4	0.273
ABC transporters	12	4	0.120
Protein digestion and absorption	9	4	0.044
Amyotrophic lateral sclerosis	4	3	0.015
Pathways of neurodegeneration - multiple diseases	4	3	0.015
Central carbon metabolism in cancer	8	3	0.133
mTOR signaling pathway	2	2	0.028
Chagas disease	2	2	0.028
Amoebiasis	3	3	0.004

Total indicates the total number of annotated compounds in the corresponding KEGG pathway, and Hit indicates the number of differential metabolites mapped to that pathway. Only pathways involving L-arginine are shown. Multiple-testing correction was performed for pathway analysis, but no pathway remained significant after FDR adjustment (all adjusted P values > 0.050); therefore, pathway-level findings should be interpreted as exploratory

Discussion

In this prospective longitudinal cohort, we found that GDM-related metabolic differences became progressively more pronounced from early to mid-pregnancy. In early pregnancy, a modest trend toward between-group metabolic divergence was detectable, although no robust signal remained after FDR correction. By mid-pregnancy, metabolic abnormalities associated with GDM were more evident, as reflected by a marked increase in the number of differential metabolites. Amino acid and its metabolites constituted the predominant class of differential metabolites at both time points, prompting a focused analysis of amino acid-related metabolic alterations.

Among the 19 amino acid-related differential metabolites examined, L-arginine detected in positive ion mode during mid-pregnancy was one of the most strongly differentiated metabolites and the only amino acid-related metabolite that remained statistically significant after FDR correction (FDR = 0.043). L-arginine also showed relatively large effect sizes in both the univariable and adjusted models (aOR = 2.840, 95% CI 1.481–5.446). Moreover, its associations were directionally consistent across OGTT glucose levels, longitudinal trajectories, and quartile-based risk analyses, and it showed moderate discriminatory performance in internal validation. Further analyses across L-arginine quartiles suggested an overall positive association with GDM risk, with a more pronounced increase in risk among participants with higher L-arginine levels.

Previous studies examining the relationship between L-arginine and GDM have reported inconsistent findings [15, 16]. Burzynska-Pedziwiatr et al. reported that L-arginine levels were lower in GDM than in those with normal glucose tolerance and were inversely associated with GDM [17]. In contrast, more recent studies with larger sample sizes, together with meta-analytic evidence, have generally supported a positive association between elevated circulating arginine levels and GDM. In a meta-analysis of 14 case-control studies including 416 women with GDM and 736 controls, Xu et al. found that GDM was associated with higher maternal arginine concentrations (SMD = 0.46, 95% CI 0.11–0.81) [18]. Similarly, in a Chinese pregnant population, Hong et al. reported higher L-arginine levels in women with GDM than in controls (39.8 vs. 35.9 µmol/L, P < 0.001), and the association between L-arginine and increased GDM risk remained significant after FDR correction (aOR = 1.398, 95% CI 1.207–1.619) [19]. In the same study, L-arginine was positively correlated with 1-h glucose levels and was enriched in the arginine biosynthesis pathway. The discrepancies across studies may reflect differences in analytical platforms, timing of sample collection, fasting status, and other design-related factors. By using a prospective longitudinal design, the present study provides additional evidence that L-arginine-related metabolic disturbances associated with GDM may already be detectable in early pregnancy and become more pronounced by mid-pregnancy.

Against this background, consideration of the potential biological relevance of elevated L-arginine may help contextualize its associations with GDM risk and dysglycemia. L-arginine, a major substrate for nitric oxide (NO) synthesis [20], contributes to the maintenance of glucose–insulin homeostasis under physiological conditions [21]. In the present study, pathway enrichment analysis showed that differential metabolites identified in early pregnancy were already involved in arginine and proline metabolism, whereas by mid-pregnancy L-arginine was enriched in multiple pathways, including arginine biosynthesis, arginine and proline metabolism, and mTOR-related pathways. KEGG network analysis further linked L-arginine to mTOR signaling and NO synthesis-related reaction/enzyme nodes (Figure S3 and S4, Table S7). These findings suggest that altered L-arginine levels in GDM may reflect remodeling of arginine-related metabolic networks.

In light of previous evidence regarding its role as an NO precursor and its involvement in nutrient sensing and metabolic regulation through mTOR signaling [22], elevated L-arginine may also be linked to dysregulated glucose metabolism during pregnancy. Under physiological conditions, appropriate levels of L-arginine have been shown to promote insulin sensitivity and GLUT4-mediated glucose uptake through the cNOS–sGC–cGMP–PKG pathway, thereby supporting glucose homeostasis [23]. However, under inflammatory or oxidative stress conditions, excessive activation of inducible nitric oxide synthase (iNOS) may promote increased NO production from elevated L-arginine via the iNOS pathway, potentially disrupting insulin signaling and impairing β-cell function. In parallel, high concentrations of L-arginine may activate mTORC1 through the CASTOR1/GATOR2 axis [24], thereby interfering with IRS-1 signaling [25] and suppressing autophagy [26], ultimately contributing to impaired glucose homeostasis [27] and reduced insulin sensitivity [28] (Fig. 4). These interpretations remain hypothetical, as enzyme activity, NO production, and mTOR signaling were not directly measured in the present study. Accordingly, our pathway enrichment and network analyses should be viewed as providing exploratory biological context rather than mechanistic proof.

Fig. 4.

Proposed L-arginine-related metabolic and signaling pathways potentially involved in GDM

The associations between L-arginine and different OGTT abnormality patterns may not be uniform, suggesting that the metabolic disturbances reflected by L-arginine may exhibit some degree of phenotypic specificity [13]. IFG is generally considered to be more closely related to impaired insulin secretion or the early stage of insulin resistance, whereas IGT is more strongly associated with defective post-load glucose disposal [29]. On this basis, we performed exploratory subgroup analyses according to OGTT abnormality patterns. The association between L-arginine and GDM appeared more pronounced in the IGT subgroup, whereas in the IFG subgroup the direction of association was consistent with that of the main analysis but the statistical evidence was weaker. This pattern was also in line with the positive correlations observed between L-arginine and 1-h and 2-h OGTT glucose levels in the main analysis, suggesting that L-arginine-related metabolic alterations may be more closely linked to post-load dysglycemia. However, this finding may also have been influenced by the small sample size and limited statistical power of the IFG subgroup. Taken together, these results suggest possible metabolic phenotypic heterogeneity in the association between L-arginine and GDM, which warrants further evaluation in larger studies [30].

This study has several strengths. First, its prospective longitudinal design spanned two key gestational windows, from early to mid-pregnancy, allowing dynamic assessment of metabolic alterations associated with GDM. Second, the use of a widely targeted metabolomics platform enhanced metabolite coverage and enabled broader characterization of disease-related metabolic features. Third, the analytical framework incorporated rigorous quality control, multiple-testing correction, and dose-response analyses, allowing a more nuanced interpretation of the findings. Finally, beyond the primary analyses, we included internal validation and exploratory subgroup analyses based on OGTT abnormality patterns, which strengthened the completeness and objectivity of the study from complementary perspectives.

Several limitations should also be acknowledged. First, as an observational study, this work does not allow causal inference. Second, the sample size was relatively modest, and some subgroup analyses were based on small numbers, which may have limited the precision of the effect estimates. Third, the OPLS-DA and ROC results should be regarded primarily as exploratory, and their practical value requires confirmation in larger samples and independent external cohorts. In addition, although matching and covariate adjustment were applied, residual confounding cannot be excluded. Finally, the pathway analyses mainly provided potential biological clues and require confirmation in dedicated functional studies; accordingly, any interpretation of the mechanistic role of arginine in the development and progression of GDM should remain cautious.

Conclusion

This study identified progressively increasing amino acid-related metabolic disturbances from early to mid-pregnancy. Among these, mid-pregnancy L-arginine emerged as a candidate metabolic signal associated with GDM risk, post-load glucose levels, and a possible nonlinear risk pattern. These findings support further investigation of amino acid-centered metabolic profiling in GDM and suggest that arginine-related metabolic alterations may provide useful clues for earlier risk identification and more refined risk stratification. Further validation in larger samples and independent cohorts, together with mechanistic studies, will be needed to clarify its potential role in risk assessment and intervention stratification.

Authors' contributions

Conceptualization, Y.S., W.Z., and Y.L.; methodology, Y.S. and W.Z.; software, J.Z.W.; validation, Y.S. and L.K.M.; formal analysis, Y.L. and Y.N.L; investigation, Y.N.L.; resources, S.H.Z.; data cu-ration, Y.L and J.Z.W.; writing—original draft preparation, Y.L.; writing—review and editing, Y.S.; visualization, Y.L.; supervision, Y.S. and L.K.M.; project administration, S.H.Z and W.Z.; funding acquisition, L.K.M and Y.S. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Beijing Municipal Commission of Science and Technology, grant number Z231100004623010; the National Key Research and Development Program of China, grant number 2022YFC2703304; the National Key Clinical Specialty Construction Project, grant number YS00078; the Peking Union Medical Foundation Pregnancy Health and Disease Metabolomics Project, grant number 01020007.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to participant confidentiality and institutional data protection regulations, but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Peking Union Medical College Hospital (I-25PJ0478). All participants were fully informed of the study objectives, procedures, potential risks, and benefits before enrollment. Written informed consent was obtained from each participant prior to participation in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.