Serum metabolites as diagnostic biomarkers for preterm labor: a metabolomics-based study

Abstract

Background

Preterm labor is a common high-risk condition during pregnancy, but current diagnostic approaches, such as cervical length measurement and uterine contraction monitoring, lack sufficient specificity and sensitivity. This study aims to explore potential biomarkers for threatened preterm labor using untargeted metabolomics, providing novel indicators to improve clinical diagnosis.

Methods

A total of 46 pregnant women from Jiangxi Maternal and Child Health Hospital were retrospectively enrolled in case–control study and divided into a preterm birth group (n = 23) and a control group (n = 23) based on gestational age. Serum metabolic profiles were analyzed using untargeted metabolomics. Differential metabolites were identified via Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) and P-value screening. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was used to explore their biological relevance. Logistic regression and Receiver Operating Characteristic (ROC) curve analysis were applied to evaluate potential biomarkers for threatened preterm labor.

Results

Using Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), key metabolites associated with preterm birth were identified, including cis-9-palmitoleic acid (AUC = 0.830), 2-amino-1-phenylethanol (AUC = 0.718), and phenylalanine (AUC = 0.708), as shown by ROC curves. These metabolites were significantly elevated in the preterm group and showed good diagnostic potential (AUC > 0.5).

Conclusions

Several serum metabolites were identified as potential biomarkers for threatened preterm labor, defined by regular uterine contractions, cervical changes, and abnormal vaginal discharge before 37 weeks. These findings may aid in the early diagnosis of preterm birth.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-025-07732-8.

Background

Preterm labor signs are a common and high-risk obstetric condition characterized by uterine contractions, cervical effacement and dilation, and changes in vaginal discharge occurring before 37 weeks of gestation, which may lead to or indicate imminent preterm birth (PTB) [1–3]. According to the World Health Organization statistics, approximately 15% of babies globally are born prematurely each year, which not only increases the risk of neonatal mortality but may also have long-term implications for children’s health and development [4–6]. Therefore, effective diagnosis and management of preterm labor signs are crucial for reducing the rate of PTB and its associated complications [7, 8]. In clinical practice, the diagnosis and intervention of preterm labor signs are particularly important as they directly impact the health of both the pregnant woman and the fetus.

The diagnosis of preterm labor signs mainly relies on clinical symptoms, cervical examination, and uterine contraction monitoring [4, 9, 10]. However, these conventional methods have notable limitations. For example, cervical length measurement and contraction assessment lack globally standardized protocols, leading to variability across medical institutions [11–13]. Moreover, the accuracy of conventional diagnostic methods in predicting the risk and timing of preterm birth remains limited [14], often failing to provide an adequate window for early prevention and intervention. Therefore, developing novel and more precise biomarkers is crucial for improving the screening and prevention of preterm labor signs.

Metabolomics is a comprehensive science that studies all metabolites within an organism, reflecting the physiological and pathological status of the organism by analyzing small molecule metabolites in biological samples [15–17]. In researching pregnancy-related diseases, metabolomics methods have shown powerful potential [18, 19]. Untargeted metabolomics enables an unbiased comparison of serum metabolic profiles between women with preterm labor signs and those with full-term pregnancies, allowing the identification of specific metabolic alterations associated with PTB. These metabolites not only help us understand the biological basis of preterm labor but may also serve as novel biomarkers for diagnosing and predicting preterm labor signs.

In this study, 46 pregnant women from Jiangxi Maternal and Child Health Hospital were enrolled and divided into a PTB group and a control group based on gestational age at delivery. Serum samples were analyzed using untargeted metabolomics based on high-performance liquid chromatography-mass spectrometry (LC–MS). The aim was to identify potential serum biomarkers associated with preterm labor signs, providing more accurate and timely tools for clinical screening. By detecting key metabolites in the serum of women with PTB, we hope to develop novel diagnostic markers that not only improve the accuracy of early diagnosis but also enable personalized and timely interventions. Ultimately, this research aims to offer new scientific evidence for preventing and managing PTB, reducing its associated health risks and improving outcomes for mothers and newborns.

Methods

Study participants and research design

This retrospective case–control study included 46 pregnant women with signs of preterm labor who underwent regular prenatal care at Jiangxi Maternal and Child Health Hospital between January 1, 2021, and December 31, 2022. Participants were divided into a preterm birth group (less than 37 weeks of gestation) and a control group (37 weeks or more) based on gestational age at delivery. Inclusion criteria were singleton pregnancies between 28 and 37 weeks of gestation with either regular (at least four contractions in 20 min or eight in 60 min with cervical changes) or irregular uterine contractions, as irregular contractions may progress to preterm birth. Exclusion criteria included multiple pregnancies, gestational hypertension, chronic hypertension, gestational diabetes, fetal growth restriction, autoimmune diseases, and incomplete medical records.

Peripheral venous blood (5 mL) was collected before any clinical intervention, centrifuged at 3000 rpm for 10 min, and the upper serum layer was stored at − 80℃. The study was conducted by the Declaration of Helsinki and was approved by the Institutional Review Board of Jiangxi Maternal and Child Health Hospital (Ethics Approval No. EC-KT-202207). Written informed consent was obtained from all participants. Obstetric diagnoses were independently reviewed and confirmed by two senior obstetricians.

Clinical data collected included maternal age, body mass index, gravidity, parity, gestational age at sampling, interval since last delivery, and cervical length measured via transvaginal ultrasound. Laboratory data included complete blood count (white blood cells, red blood cells, platelets, neutrophil count and percentage, lymphocyte count and percentage, and red cell distribution width standard deviation), liver function tests (alanine aminotransferase, aspartate aminotransferase, total protein, albumin, alkaline phosphatase, and lactate dehydrogenase), and renal function indicators (creatinine and urea).

Reproductive tract health was evaluated through vaginal discharge tests and vaginal microbiota analysis. These assessments are important indicators of female reproductive health, reflecting the microbial balance of the vaginal environment, including bacterial species, quantity, pH, and cellular composition. Vaginal cleanliness is classified into four grades, which are directly related to the risk of gynecological diseases and are valuable for clinical diagnosis and treatment [20].

Blood samples were analyzed using the Sysmex-XN-2000 automated hematology analyzer (Sysmex Europe, Germany). Liver and kidney function were assessed by radioimmunoassay on the AU5800 automated biochemical analyzer (Beckman Coulter, USA) to exclude the influence of systemic disease on metabolic outcomes. Vaginal tests were conducted using the LTS-V400 automated vaginal infection analyzer (Guokang, Shandong, China) with Swiss staining and combined morphological and dry chemical methods.

Neonatal outcomes, including birth weight, Apgar score, and gestational age, were recorded by two experienced neonatologists (Table 1). The study design and grouping are shown in Fig. 1.

Table 1.

Comparison of clinical characteristics between two groups of study subjects

Clinical characteristics	PTB (n = 23)	C (n = 23)	P value
Basic information
Age (years)	27.96 ± 5.63	28.61 ± 3.53	0.640a
BMI (kg/m2)	24.19 ± 2.55	25.57 ± 3.51	0.135a
Number of pregnancies(times)	2 (1, 3)	2 (1, 3)	0.549a
Number of deliveries(times)	1 (0, 1)	0 (0, 1)	0.293b
Gestational age at sampling (weeks)	31.44 ± 2.72	32.26 ± 1.88	0.251a
Time interval since the last delivery (years)	2 (0, 4)	0 (0, 5)	0.499b
Cervical length (mm)	24.00 (9.00, 30.00)	30.00(27.00, 30.50)	0.049b
Complete blood count
White blood cells (× 109/L)	11.87 (9.15, 16.22)	9.63(8.03, 12.57)	0.048b
Red blood cells (× 109/L)	3.77 ± 0.33	3.70 ± 0.28	0.408a
Platelets (× 109/L)	222.22 ± 68.08	198.35 ± 58.66	0.209a
Neutrophil count (× 109/L)	9.15 (7.00, 13.53)	7.04(5.77, 9.69)	0.044b
Neutrophil percentage (%)	78.70(77.10, 89.00)	74.10(71.4, 79.3)	0.051b
Lymphocytes (× 109/L)	1.55 ± 0.59	1.62 ± 0.45	0.632a
Lymphocyte percentage (%)	13.29 ± 6.51	17.17 ± 6.70	0.053a
Monocytes (× 109/L)	0.69 ± 0.26	0.70 ± 0.16	0.883a
Monocyte percentage (%)	5.68 ± 2.40	7.20 ± 2.04	0.026a
Eosinophils (× 109/L)	0.04(0.03, 0.09)	0.06(0.03, 0.14)	0.120b
Basophils (× 109/L)	0.03(0.02,0.05)	0.03(0.02, 0.03)	0.535b
Basophil percentage (%)	0.20(0.20, 0.30)	0.20(020, 0.30)	0.668b
Red cell distribution width (fL)	43.64 ± 3.33	45.13 ± 3.31	0.137a
Liver function indicators
ALT (U/L)	10.13 ± 4.66	8.43 ± 3.46	0.169b
AST (U/L)	19.26 ± 5.39	16.70 ± 3.21	0.056a
Total protein (g/L)	66.34 ± 4.34	65.17 ± 5.21	0.411a
Albumin (g/L)	35.24 ± 2.79	34.42 ± 2.76	0.671a
Alkaline phosphatase (U/L)	127.70 ± 49.68	92.96 ± 29.00	0.006a
Lactate dehydrogenase (U/L)	191.76 ± 49.63	172.26 ± 44.58	0.201a
Renal function indicators
Creatinine (μmmol/L)	41.36 ± 5.03	39.65 ± 5.80	0.297a
Urea (mmol/L)	2.96 ± 0.78	3.12 ± 0.72	0.485a
Vaginal cleanliness
Abnormal vaginal discharge (%)	2 (8.70)	5 (21.74)	0.412c
Microbiota imbalance (%)	4 (17.39)	4 (17.39)	1.000c
Neonatal outcome
Gestational age at delivery (weeks)	34.10 (29.75, 35.65)	38.60(38.10, 40.10)	< 0.001b
Amount of bleeding during delivery (mL)	260 (200, 315)	260 (200, 340)	0.903b
Newborn weight (kg)	2.14 ± 0.57	3.24 ± 0.41	< 0.001a
Apgar score (1–10)	9.00 (8.50, 9.00)	10.00 (10.00, 10.00)	< 0.001b

Continuous variables are presented as mean ± standard deviation (a) or median [interquartile range] (b), depending on data distribution. Comparisons were made using independent-samples t-test (a) or Mann–Whitney U test (b). Categorical variables were compared using Chi-square or Fisher’s exact test (c). A p-value < 0.05 was considered statistically significant

Fig. 1.

Metabolomics Study Design of Serum Samples from Women with Preterm Labor Signs. Note: This study included a total of 23 samples from women with preterm labor signs and full-term delivery, and 23 samples from women with preterm labor signs and PTB. Metabolomics data processing was conducted using untargeted metabolomics based on HPLC-HRMS, with peak extraction and normalization; compounds were identified based on compound databases. Statistical analysis involved the identification of differential metabolites associated with preterm labor signs, KEGG pathway enrichment analysis, and ROC selection of potential biomarkers

Metabolite extraction and UHPLC-MS analysis

After slowly thawing the samples at 4 °C, an appropriate amount of sample was added to a pre-cooled mixture of methanol/acetonitrile/water (2:2:1, v/v), followed by vortex mixing, low-temperature sonication for 30 min, standing at −20 °C for 10 min, and centrifugation at 14000 g at 4 °C for 20 min. The supernatant was then vacuum-dried and reconstituted with 100 μL of acetonitrile–water solution (acetonitrile: water = 1:1, v/v) for mass spectrometry analysis. After vortexing and centrifugation at 14000 g at 4 °C for 15 min, the supernatant was used for injection analysis.

An Agilent 1290 Infinity LC ultra-high performance liquid chromatography system (UHPLC, Thermo Fisher Scientific, USA) with a HILIC column was used for metabolite separation. The column temperature was maintained at 25 °C, with a flow rate of 0.5 mL/min and an injection volume of 2 μL. The mobile phase composition included A: water + 25 mM ammonium acetate + 25 mM ammonium hydroxide and B: acetonitrile. The gradient elution program was as follows: 0–0.5 min, 95% B; 0.5–7 min, B linearly decreased from 95 to 65%; 7–8 min, B linearly decreased from 65 to 40%; 8–9 min, B was maintained at 40%; 9–9.1 min, B linearly increased from 40 to 95%; 9.1–12 min, B was maintained at 95%. Throughout the analysis, samples were kept at 4 °C in the automatic sampler to minimize instrument signal fluctuations. Samples were randomly analyzed in consecutive order to mitigate the impact of instrumental signal fluctuations. Quality control (QC) samples were inserted into the sample queue to monitor and assess system stability and experimental data reliability.

After separation, mass spectrometry was analyzed using a Triple TOF 6600 mass spectrometer (SCIEX, USA) in both positive and negative electrospray ionization (ESI) modes. The ESI source settings included the following parameters: Gas1: 60, Gas2: 60, Curtain Gas (CUR): 30 psi, Ion Source Temperature: 600 °C, Ion Spray Voltage (ISVF): ± 5500 V (for both positive and negative modes). The mass range for the first mass spectrometry scan was 60–1000 Da, and for the second, it was 25–1000 Da. The accumulation time for the first mass spectra scan was 0.20 s/spectrum; for the second, it was 0.05 s/spectrum. The second mass spectra scan was performed using data-dependent acquisition (IDA) mode with peak intensity value filtering, with declustering potential (DP) set at ± 60 V, collision energy at 35 ± 15 eV, and IDA settings included a dynamic exclusion range of 4 Da, acquiring 10 fragment spectra per scan.

Data processing and analysis

Raw data were converted to.mzXML format using ProteoWizard and processed with XCMS software for peak detection, retention time alignment, and peak area extraction. Statistical analysis was performed after metabolite identification, data preprocessing, and quality assessment.

To ensure the stability and reproducibility of LC–MS analysis, all serum samples were pooled in equal volumes to generate quality control (QC) samples. One QC sample was injected after every ten test samples during the LC–MS run. Blank samples were also used to assess background noise. Instrumental stability was monitored by evaluating the consistency of retention times and peak areas of representative ions in the QC samples. Metabolic features with a relative standard deviation (RSD) greater than 30% in QC samples were excluded. Signal drift was corrected using a QC-based robust LOESS signal correction algorithm (QC-RLSC) to improve data reliability.

Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) was conducted using SIMCA software version 14.1 (Umetrics, Sweden), and 100 permutation tests validated model reliability. Variable Importance in Projection (VIP) scores were calculated to identify influential variables. Differences in metabolite intensities were assessed using the Mann–Whitney U test, and linear regression was applied to analyze associations between pairs of compounds, both conducted in IBM SPSS Statistics (version 21, IBM Corp., USA).

Differential metabolites were selected based on VIP > 1.0 and p < 0.05. Correlation and clustering analysis were used to evaluate metabolite relationships and grouping. Identified differential metabolites were subjected to the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. Finally, ROC curve analysis was performed to assess the diagnostic value of these metabolites in distinguishing between the preterm birth and full-term groups.

Results

Baseline characteristics of the participant

To identify potential biomarkers associated with preterm labor signs, a cohort of 46 pregnant women presenting with such signs was analyzed. Based on clinical outcomes, participants were categorized into a preterm birth group (n = 23) and a full-term birth group (n = 23). Baseline clinical characteristics are summarized in Table 1. Significant differences were observed between the two groups in cervical length, white blood cell count, neutrophil count, monocyte percentage, alkaline phosphatase level, gestational age at delivery, neonatal birth weight, and Apgar scores. No significant differences were found in the remaining baseline variables.

Selection of differential metabolites related to preterm labor signs

Metabolomic profiling was performed on QC samples, the preterm labor signs with full-term delivery group, and the preterm labor signs with preterm birth group. Carboxylic acids and their derivatives, glycerophospholipids, and fatty acyl metabolites were the most abundant compound classes detected. Principal component analysis (PCA) demonstrated tight clustering of QC samples in both positive and negative ion modes, indicating high experimental reproducibility (Figures S1). OPLS-DA was used to compare serum metabolic profiles between the two groups. Clear separation was observed in both ion modes, suggesting significant metabolic differences (Fig. 2).

Fig. 2.

OPLS-DA Score Plot and Permutation Test of Non-Polar Metabolites. Note: (A) OPLS-DA score plot of metabolites in positive ion mode; (B) Permutation test validation of OPLS-DA model in positive ion mode; (C) OPLS-DA score plot of metabolites in negative ion mode; (D) Permutation test validation of OPLS-DA model in negative ion mode

Based on the criteria of VIP > 1, p < 0.05, and fold change > 1.5 or < 0.67, a total of 10 significantly different metabolites were identified in positive ion mode and 24 in negative ion mode (Figs. 3A-D, Table 2). Among them, cis-9-palmitoleic acid, 2-amino-1-phenylethanol, phenylalanine, and DL-phenylalanine were significantly upregulated in the preterm birth group, with fold changes of 1.684, 1.752, 1.528, and 1.663, respectively. The corresponding p-values were 6.079E-05, 0.004, 0.003, and 0.008. The median serum levels of these metabolites in the preterm birth group were 3,372,246.056, 1,101,747.347, 1,984,155.972, and 593,909.247, respectively, compared with 1,795,308.802, 567,446.048, 1,159,624.650, and 314,838.601 in the full-term group. In contrast, indole-3-butyric acid was significantly decreased in the preterm group, with a fold change of 0.601.

Fig. 3.

Fold Change and Cluster Analysis of Differential Metabolites. Note: (A) Fold change of metabolites in positive ion mode; (B) Fold change of metabolites in negative ion mode; (C) Cluster analysis of metabolites in positive ion mode between preterm labor signs with PTB group and preterm labor signs with full-term delivery group; (D) Cluster analysis of metabolites in negative ion mode between preterm labor signs with PTB group and preterm labor signs with full-term delivery group

Table 2.

Significant differences in metabolites were observed in the serum of two groups of study subjects (TOP 5)

Name	Average		Median		IQR		Fold change	p-value	VIP
	PTB (n = 23)	C (n = 23)	PTB (n = 23)	C (n = 23)	PTB (n = 23)	C (n = 23)
Cis-9-palmitoleic acid	3390218.078	2013702.929	3372246.056	1795308.802	1381958.297	1125201.239	1.684	6.079E-05	10.246
2-amino-1-phenylethanol	1073742.710	612766.854	1101747.347	567446.048	923406.086	292166.700	1.752	0.004	8.845
Indole-3-butyric acid	600377.974	998169.5948	513432.111	691618.516	166079.214	1017416.548	0.601	0.005	6.015
Phenylalanine	1662827.000	1088368.252	1984155.972	1159624.650	1209849.444	417502.387	1.528	0.003	5.921
DL-phenylalanine	566209.907	340573.1826	593,909.247	314838.601	513797.121	205429.220	1.663	0.008	5.769

Pathway analysis of significantly different metabolites in serum.

Correlation analyses were performed separately for those identified in positive and negative ion modes to explore the functional relevance of the differential metabolites. Strong correlations were observed among 2-amino-1-phenylethanol, phenylalanine, and DL-phenylalanine (Figs. 4A and B). KEGG enrichment analysis was conducted to identify associated biological pathways. The statistical significance of enrichment was evaluated using Fisher’s exact test. The top five significantly enriched pathways were ABC transporters, protein digestion and absorption, aminoacyl-tRNA biosynthesis, biosynthesis of amino acids, and mineral absorption (Figs. 4C and D).

Fig. 4.

Pathway Analysis of Significant Differential Metabolites in Serum. Note: (A) Correlation analysis of differential metabolites in positive ion mode; (B) Correlation analysis of differential metabolites in negative ion mode; (C) KEGG enrichment pathway map of differential metabolites; (D) Enrichment pathway score plot of differential metabolites

Predicting the outcome of preterm labor signs leading to preterm delivery

To identify potential predictors of PTB among individuals with symptoms of threatened preterm labor, differential metabolites and clinical indicators were included in a logistic regression model. ROC curve analysis was conducted to assess their diagnostic performance. The seven variables with the highest predictive value were cis-9-palmitoleic acid (0.830), alkaline phosphatase (0.744), 2-amino-1-phenylethanol (0.718), phenylalanine (0.708), DL-phenylalanine (0.687), neutrophil count (0.673), and white blood cell count (0.670). Cis-9-palmitoleic acid showed the strongest predictive ability for PTB when used alone (Fig. 5, Table 3).

Fig. 5.

ROC Curve Analysis of Seven Biomarkers. Note: ROC curves of five metabolic markers, including cis-9-palmitoleic acid, alkaline phosphatase, DL-phenylalanine, 2-amino-1-phenylethanol, and phenylalanine, as well as neutrophil count and white blood cell count. The area under the curve (AUC) for each ROC curve is greater than 0.5, demonstrating their potential diagnostic value

Table 3.

Logistic regression analysis of potential serum biomarkers and clinical indicators associated with threatened preterm labor

Variable	β Coefficient	SE	Wald χ2	p-value	OR (Exp(β))	95% CI for OR	AUC	95% CI for AUC
Cis-9-palmitoleic acid	0.0	0.0	10.5335	0.0012	1.0	1.00–1.00	0.830	0.696–0.934
Alkaline phosphatase	0.0256	0.0105	5.9393	0.0148	1.0259	1.01–1.05	0.744	0.590–0.881
2-amino-1-phenylethanol	0.0	0.0	6.4912	0.0108	1.0	1.00–1.00	0.718	0.530–0.866
Phenylalanine	0.0	0.0	6.8649	0.0088	1.0	1.00–1.00	0.708	0.535–0.860
DL-phenylalanine	0.0	0.0	5.9195	0.015	1.0	1.00–1.00	0.687	0.526–0.870
Neutrophil count	0.1696	0.0862	3.8745	0.049	1.1848	1.00–1.40	0.673	0.497–0.823
White blood cell	0.1584	0.0815	3.7812	0.0518	1.1716	1.00–1.37	0.670	0.502–0.817

SE Standard Error, OR Odds Ratio, CI Confidence Interval. p-value < 0.05 was considered statistically significant

Discussion

Preterm labor signs are a significant concern in perinatal medicine due to the potential long-term adverse effects on maternal and infant health, making effective prediction and diagnosis a research focus [21–23]. While various methods and indicators are currently used to assess the risk of PTB, they often have limitations, such as the inability of methods like cervical length measurement and uterine contraction monitoring to accurately predict all cases of PTB [24–26]. Therefore, developing new biomarkers to predict preterm labor signs earlier and more accurately holds significant clinical value. This study used metabolomics methods to explore serum biomarkers associated with PTB, to provide a new screening tool for clinical practice.

In contrast to previous targeted metabolomics or proteomics approaches, this study employed untargeted metabolomics technology, allowing for unbiased exploration of a wide range of metabolites. This approach not only increases the possibility of discovering unknown biomarkers related to preterm labor signs but also enhances coverage and sensitivity, revealing more comprehensive metabolic changes [27, 28]. Through this method, we can observe more subtle metabolic changes relevant to the disease, which are often overlooked in traditional methods.

The enriched metabolic pathways identified in this study may be involved in the pathogenesis of preterm birth. ABC transporters are essential for transmembrane transport of nutrients and metabolites, and dysfunction in this pathway may impair placental nutrient delivery [29]. Disruption in protein digestion and absorption may affect protein availability, critical for hormone synthesis and pregnancy maintenance [30]. Altered amino acid biosynthesis and aminoacyl-tRNA biosynthesis could lead to amino acid imbalance, affecting fetal development and placental function [31]. Impaired mineral absorption, particularly of calcium and iron, may influence uterine contractility and fetal growth [32]. These findings suggest possible mechanisms linking metabolic disturbances to increased risk of preterm birth.

Several metabolites closely associated with preterm labor signs were identified through untargeted metabolomics analysis, including cis-9-palmitoleic acid, 2-amino-1-phenylethanol, and phenylalanine, which have rarely been reported in previous studies. The seven predictive indicators identified in this study may contribute to the onset of preterm birth through distinct biological mechanisms. Cis-9-palmitoleic acid, an unsaturated fatty acid, is involved in membrane structure and cellular signaling. Abnormal elevation of this metabolite may impair placental cell function and disrupt the intrauterine environment, thereby increasing the risk of preterm birth [27]. 2-amino-1-phenylethanol, phenylalanine, and DL-phenylalanine are related to amino acid metabolism. Amino acid imbalance may interfere with protein synthesis and neurotransmitter regulation, potentially disrupting maternal–fetal physiological balance and triggering preterm labor [3]. Neutrophil and white blood cell counts reflect the maternal immune status during pregnancy, and inflammatory responses are recognized as key contributors to preterm birth [6]. Notably, alkaline phosphatase (ALP), which is easily measurable in clinical practice, is involved in multiple metabolic processes and is closely related to placental development and function. Elevated ALP levels may indicate placental dysfunction, impair fetal nutrient supply, and contribute to preterm birth [4]. These hypotheses require further experimental validation. Nonetheless, the newly identified metabolites in this study not only deepen our understanding of the metabolic mechanisms underlying preterm birth but also provide promising diagnostic targets, which are crucial for elucidating the biochemical basis of the disease.

The OPLS-DA modeling and ROC curve analysis utilized in this study have provided an effective tool for assessing and validating potential biomarkers. The use of OPLS-DA models enabled us to distinguish metabolites significantly associated with PTB from complex datasets, while ROC curve analysis further confirmed the diagnostic potential of these metabolites. Compared to traditional statistical methods, these advanced techniques offer higher accuracy and reliability, making the research results more compelling.

The application of ROC curve analysis in this study confirmed the potential of the discovered metabolites as biomarkers for predicting preterm labor signs. By calculating the AUC values, we could evaluate the diagnostic performance of these biomarkers and predict their feasibility in practical clinical applications. These metabolites show good sensitivity and specificity, indicating promising prospects for their future clinical utility. However, prospective validation studies are needed to confirm their diagnostic utility and generalizability in broader populations.

This study, based on untargeted metabolomics analysis, successfully identified several novel serum metabolites associated with signs of preterm labor. These findings expand the pool of potential biomarkers for preterm birth and offer new directions for clinical prediction and diagnosis. However, several limitations should be acknowledged. First, the sample size was relatively small (n = 46), which may have reduced the statistical power and limited the generalizability of the results. Second, all samples were collected from a single medical center, introducing potential regional and population-related biases. In addition, the dataset was not split into training and validation sets, which may have led to model overfitting and overestimating the diagnostic performance of some biomarkers, such as cis-9-palmitoleic acid (AUC = 0.830). From a methodological perspective, the analysis relied on high-end LC–MS platforms, and the lack of standardized detection procedures and limited accessibility in primary care settings may hinder clinical translation. Moreover, the study did not explore the complementary value of the identified metabolites with established clinical indicators, such as cervical length or fetal fibronectin, leaving the independent or combined predictive value of these biomarkers unclear. Dynamic changes in metabolite levels over time (e.g., through serial sampling) were also not assessed. Finally, the absence of long-term follow-up data on maternal and neonatal outcomes limits our ability to evaluate the broader implications of metabolite changes following preterm birth.

Future studies should involve larger sample sizes across multiple centers and regions to enhance the representativeness and applicability of findings. Adopting a training-validation cohort design (e.g., a 70:30 split) and applying regularized modeling techniques such as LASSO regression to reduce the risk of overfitting is recommended. Additionally, developing rapid and accessible detection tools based on LC–MS/MS is warranted, along with efforts to integrate these biomarkers with routine clinical assessments such as uterine contraction monitoring and cervical ultrasound. Incorporating lifestyle, environmental, and other contextual variables into multi-factorial models may improve risk stratification. Prospective cohort studies with long-term follow-up are also needed to validate these biomarkers’ clinical utility and evaluate the long-term impact of preterm birth on maternal and neonatal health. These efforts will help bridge the gap between laboratory research and clinical application.

Conclusion

This study applied untargeted metabolomics to analyze serum samples from pregnant women with preterm labor signs and full-term pregnancies. Several key metabolites, including cis-9-palmitoleic acid, 2-amino-1-phenylethanol, and phenylalanine, were identified as significantly altered in the preterm group, suggesting their involvement in the pathophysiology of PTB. ROC analysis demonstrated their strong diagnostic potential in distinguishing PTB cases. These findings offer new insights into the metabolic mechanisms underlying PTB and provide a foundation for biomarker development and mechanistic research. Clinically, these metabolites, especially cis-9-palmitoleic acid, may serve as valuable tools for early diagnosis.

Authors’ contributions

Qiuhong Yi and Hua Lai drafted the manuscript. Jiusheng Zheng and Xiaoming Zeng conceived and designed the study. Bicheng Yang gave technical guidance and critically revised the manuscript. Qin Li and Chen Wang collected serum samples and the data. Xiao Zhou and Lijun Liao performed the experiments. Siming Xin and Feng Zhang analyzed the data. All authors reviewed and approved the final version of the manuscript.

Funding

This study was supported by Jiangxi Provincial Key R&D Program-General Project (20203BBGL73130), Jiangxi Provincial Applied Research Cultivation Program (20212BAG70006), Jiangxi Provincial Health Commission Science and Technology Program Project (202110093) and Funded by Jiangxi Provincial Key Clinical Specialty Construction Project.

Data availability

All data can be provided upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Helsinki Declaration and approved by the Ethics Committee of Jiangxi Maternal and Child Health Hospital (Ethical Approval Number EC-KT-202207). Informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.