Introduction

Gestational diabetes mellitus (GDM) is mainly occurs and diagnosed in the second or third trimester of pregnancy. The documented prevalence of GDM varies widely, ranging from 1% to over 30%, owing to a lack of consensus and uniformity in the screening standards and diagnostic criteria for GDM across countries and regions [1,2,3]. Women with GDM may experience damage to one or more organ systems, including the liver, kidneys, placenta, and other organs. These women face significantly increased risks of delivering large for gestational age (LGA) infants, undergoing cesarean sections, experiencing postpartum hemorrhage, preterm delivery, and developing preeclampsia (PE) [4]. Additionally, they are at an increased risk of developing circulatory and metabolic diseases in later life [5, 6]. Neonates born to mothers diagnosed with GDM are at an increased risk for macrosomia, congenital malformations, asphyxia, respiratory distress, and neonatal hypoglycemia [4,5,6]. Emerging evidence suggests that early exposure to maternal intrauterine hyperglycemia has long-term adverse effects on offspring health, including an increased risk of developing adolescent hypertension and type 2 diabetes mellitus [6, 7]. Unfortunately, even prompt and effective insulin therapy following a GDM diagnosis cannot fully mitigate the risk of metabolic disorders in adulthood for these offspring [8]. Thus, early prediction and intervention for GDM are imperative, enabling preemptive measures ahead of formal diagnosis in later pregnancy to improve maternal and neonatal health outcomes.

With the increased use of assisted reproductive technology (ART) worldwide, obstetric and perinatal complications after ART are receiving increased amounts of attention. It has been consistently reported that compared to women who conceive spontaneously, women who conceive through in vitro fertilization (IVF) are at a higher risk of GDM [9, 10]. However, the early prediction of GDM among women undergoing IVF treatments remains largely unexplored. A prospective study by Coussa et al. identified the follicle-stimulating hormone (FSH)/luteinizing hormone (LH) ratio as a potential predictor for subsequent development of GDM in women who underwent IVF; however, the validity of the study is limited by the fact that over one-third of the participants were diagnosed with polycystic ovary syndrome (PCOS), a condition characterized by elevated LH levels and glucose metabolic disorders [11]. Therefore, it is imperative to develop a robust and innovative predictive model for identifying the high-risk GDM population during the first trimester of pregnancy, especially for women undergoing IVF treatment.

Despite extensive research, the pathogenesis of GDM is not fully understood [12]. During a normal pregnancy, maternal insulin resistance progressively increases due to elevated levels of placental hormones, which ensures sufficient energy availability for the growing fetus [6]. It has been suggested that GDM arises from inadequate pancreatic β-cell insulin secretion to compensate for insulin resistance, possibly combined with β-cell dysfunction [13, 14]. Additionally, the complex pathogenesis of GDM likely involves immune dysfunction and low-grade systemic inflammation. This involves a complex interplay of immune cells (such as B lymphocytes, T lymphocytes, macrophages, and neutrophils) and signaling pathways [15, 16]. Hence, there has been a growing focus on the shifts in pro-inflammatory and anti-inflammatory cytokines in women with GDM [17,18,19]. Among these studies, the role of interleukins (ILs) in GDM has received considerable attention [20, 21]; however, comprehensive analyses that combine IL family members with other inflammatory factors to examine their potential as predictive biomarkers in GDM are scarce.

In addition to inflammatory factors, adipocytokines such as leptin and adiponectin have been implicated in the inflammatory processes associated with GDM development, influencing glucose metabolism and insulin sensitivity [12, 22, 23]. However, their utility as independent predictive biomarkers for GDM is limited [24], partly due to the confounding effects of body mass index (BMI) and gestational weight gain. Moreover, emerging evidence suggests that serum growth differentiation factor-15 (GDF-15) serves as a biomarker for GDM during late pregnancy [25]. However, the predictive value of GDF-15 in early pregnancy remains underexplored. Although Thadhani et al. revealed that follistatin-like-3 (FSTL-3, also named FLRG) levels were lower in women who developed GDM compared to healthy women [26], other studies did not find altered first-trimester FSTL-3 levels in women with GDM [27, 28]. To date, the cytokine profiles of these biomarkers during the first trimester of pregnancy, particularly in IVF patients at high risk of GDM, have not been well characterized.

Both GDM and hypertensive disorders of pregnancy, such as pregnancy-induced hypertension (PIH) and PE, are classified as cardiometabolic disorders that occur during pregnancy, and these complications share common pathogenetic mechanisms, including inflammation, oxidative stress, and vascular endothelial dysfunction. Our previous research into autoimmune antibodies revealed that first-trimester serum levels of antiphospholipid antibodies (aPSs) are altered in women diagnosed with PIH and PE, suggesting their potential as candidate biomarkers for hypertensive disorder of pregnancy [29, 30]. Furthermore, considering the pivotal role of immune and inflammatory processes in the pathogenesis of GDM, it is yet to be determined whether the levels of aPSs are dysregulated in first-trimester serum of women with GDM and whether they could serve as predictive biomarkers for this condition.

To explore the potential of cytokines and autoimmune antibodies as predictive biomarkers for GDM, here we examined and compared the first-trimester profiles of 48 cytokines, autoimmune antibodies, and several previously identified GDM biomarkers between the GDM group and the control group of women who underwent IVF treatment, aiming to identify novel biomarkers and provide predictive models for screening high-risk GDM patients.

Materials and methods

The detailed methods are provided in the online-only Supplementary Material.

Patients

We included 38 women aged 20–40 years who underwent their first cycles of in vitro fertilization (IVF) with or without intracytoplasmic sperm injection and achieved singleton live births from January 2020 to December 2020 in both the GDM group and the control group. Ethical approval for the use and analysis of blood samples and data from patients included in our study was obtained from the Institutional Ethical Committee of the Center for Reproductive Medicine of Shandong University (Ethical Review No. 67, 2023). All participants provided informed written consent.

Initially, a total of 1584 women with tubal factor as the only indication for IVF were screened for eligibility (Fig. 1). All women achieved singleton pregnancy, and none of them experienced vanishing twins or reduction of twins. Women who were diagnosed with PCOS or who underwent preimplantation genetic testing cycles or frozen-thawed oocyte cycles were excluded from this study. Women with chronic autoimmune disease (such as systemic lupus erythematosus, thyroid autoimmunity, or antiphospholipid syndrome), pregestational diabetes mellitus or pregestational hypertension were also excluded. Additionally, women without available first-trimester serum samples were also excluded. To mitigate potential interference from other pregnancy complications, women who developed hypertensive disorders of pregnancy (HDP), placenta previa or placental abruption were also excluded. Blood samples from participants were collected at 11–13 gestational weeks after IVF treatment and subjected to serum preparation and storage in our large human biobank. The follow-up of pregnancy complications in women after IVF treatment was achieved based on our large-scale assisted reproductive cohort platform.

Fig. 1: Flow diagram of participant screening and enrollment.
figure 1

IVF in vitro fertilization, HDP hypertensive disorders of pregnancy, GDM gestational diabetes mellitus, PSM propensity score matching.

Full size image

The baseline characteristics of patients with available first-trimester serum samples were compared and matched using propensity score matching (PSM) approach to control for potential confounding bias. Maternal age, paternal age, duration of infertility, BMI, preconceptional fasting glucose, gravidity, parity, ovarian stimulation protocols, fertilization method, number of embryos transferred, stage of embryo transfer, endometrial thickness before embryo transfer, and embryo transfer regimen which were weighted equally. The control group included 38 healthy women who were matched in a 1:1 ratio to the GDM group based on the propensity score with a standard caliper width of 0.2. The study flow diagram is shown in Fig. 1.

Outcomes

The diagnostic criteria of GDM is “one-step” strategy 75-g oral glucose tolerance test (OGTT) proposed by the International Association of the Diabetes and Pregnancy Study Groups (IADPSG), with the cutoffs of ≥92 mg/dL (5.1 mmol/L) for fasting, of ≥180 mg/dL (10.0 mmol/L) for 1 h and of ≥153 mg/dL (8.5 mmol/L) for 2 h plasma glucose concentration, at 24–28 weeks of pregnancy [1]. Low birthweight was defined in terms of weight at birth of <2500 g, and macrosomia was defined as beyond an absolute birth weight exceeding historically 4000 g or 4500 g, regardless of gestational age. Small for gestational age (SGA) was defined as birthweight lower than the 10th percentile of referential birthweight. LGA was defined as birthweight higher than the 90th percentile of referential birthweight.

Sample collection and measurement of candidate biomarkers

First-trimester serum samples were collected from a total of 38 women diagnosed with GDM and 38 matched normoglycemia controls. The serum levels of 48 cytokines, total immunoglobulins (IgA, IgM, and IgG), aPS autoantibodies (including aPS IgA, aPS IgM, and aPS IgG), aPS immune complexes (including aPS-IgA IC, aPS-IgM IC, and aPS-IgG IC), and previously reported underlying GDM biomarkers such as leptin, adiponectin, GDF-15, and FSTL-3 were measured in these samples.

The measurement of candidate biomarkers was performed using the Bio-Plex Pro Human Cytokine Screening Panel, 48-plex (Bio-Rad, #12007283) for cytokine analysis, and enzyme-linked immunosorbent assays (ELISA) for specific biomarker quantification. Detailed information regarding the sample collection procedure and the specific methods employed for measuring the candidate biomarkers can be found in the online-only Supplementary Material.

Construction and evaluation of first-trimester serum predictive machine-learning models’ performance

First-trimester serum predictive models for GDM were developed using machine-learning methods such as K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), utilizing the “sklearn” and “xgboost” packages in Python. Model parameters were optimized using a grid search with fivefold cross-validation. The dataset of 76 samples was randomly divided into a training set (n = 53) and a testing set (n = 23) using a 7:3 ratio through the “train_test_split” function.

These five models were trained and tested using this split and ROC curves were generated to evaluate ehir performance. Additionally, SHapley Additive exPlanations (SHAP) module in Python v3.12.0 was also utilized to further interpret the operations of the XGBoost model. Evaluation of models’ performance by utilizing these metrics on both training and test datasets is crucial for validating the accuracy of predictions and ensuring a well-fitted performance.

Statistical analysis

Graphical use of histograms, Q–Q plots, and the Shapiro‒Wilk test was used to test the normality of continuous variables. While normally distributed data were presented as the means ± standard deviations and analyzed by the student’s test, nonnormally distributed data were distributed as medians with interquartile range and compared by the Mann–Whitney U test. Categorical variables were presented as counts (percentages) and were compared using either the chi-square analysis or Fisher’s exact test. P values < 0.05 were considered statistically significant, and cytokines with P values < 0.05 were included in the Receiver Operating Characteristic (ROC) analysis. ROC curves were drawn to analyze the independent and combined values of specific cytokines and aPS antibodies in predicting GDM.

All statistical analysis was performed using SPSS 26.0 (IBM, Armonk, NY, USA), GraphPad Prism 9.0 (GraphPad Software, San Diego, CA, USA), and Python 3.12.0 (New Castle, DE, USA).

Results

Baseline characteristics and pregnancy outcomes of the participants

In this study, we included 38 women who developed GDM following IVF treatment and 38 matched normoglycemia controls. The baseline characteristics, including maternal age, paternal age, duration of infertility, BMI, preconceptional fasting glucose levels, pregnancy history, ovarian stimulation protocols, characteristics of the ovarian stimulation cycle, and embryo culture and transfer methods, were comparable between the two groups (Table 1). Due to the relatively limited sample size, we did not observe significant differences in pregnancy outcomes, including delivery mode, gestational weeks at delivery, the risks of infants being low birthweight, macrosomia, small for gestational age (SGA), or large for gestational age (LGA), between these two groups. All 76 women included in our study delivered at full term (Supplementary Table 1).

Table 1 Baseline characteristics of the participants.
Full size table

Profiling of first-trimester serum cytokines reveals revealed dysregulation of ten cytokines in GDM development

Among the 48 cytokines, levels of ten cytokines including IL-1α, IL-1β, IL-3, IL-5, IL-7, IL-10, IL-12 (p40), IL-15, IL-18, and TNF-α were lower in the GDM group than in the control group. Among these ten cytokines, IL-1α, IL-3, IL-7, IL-10, IL-12 (p40), IL-15, and IL-18 have been identified as novel predictors of GDM during the first trimester in previous studies [21, 31]. These cytokines demonstrated moderate predictive values, with respective AUC scores: IL‑1α (0.715), IL-3 (0.663), IL-7 (0.735), IL-10 (0.715), IL-12 (p40) (0.670), IL-15 (0.718), and IL-18 (0.632). Previous research indicated that IL-5, IL-1β, and TNF-α serve as biomarkers for GDM in the second and third trimesters [32,33,34,35]. Corroboration of these findings, our study showed that first-trimester serum levels of IL-1β (AUC: 0.656, 95% CI: 0.532-0.779, P = 0.020), IL-15 (AUC: 0.718, 95% CI: 0.602-0.834, P = 0.001), and TNF-α (AUC: 0.641, 95% CI: 0.516-0.765, P = 0.035) also serve as candidate biomarkers for GDM. The detailed serum levels of 48 factors in the GDM and control groups are summarized in Table 2, with violin plots for these ten biomarkers presented in Fig. 2A.

Table 2 Cytokine profile in first-trimester serum of the GDM group and the normoglycemia control group.
Full size table
Fig. 2: Violin plots illustrating significantly different first-trimester serum biomarkers between the GDM group and the normoglycemia control group.
figure 2

A Violin plots of ten first-trimester serum cytokines, including IL-1α, IL-1β, IL-3, IL-5, IL-7, IL-10, IL-12 (p40), IL-15, IL-18, and TNF-α, in the GDM group (n = 38) and normoglycemia control group (n = 38). B Serum levels of five first-trimester serum antibodies (total IgM, total IgG, aPS IgA, and aPS-IgG IC) in the GDM group (n = 38) and normoglycemia control group (n = 38). Data are presented as the mean ± standard deviation or median (interquartile range). *P < 0.05; **P < 0.01; ***P < 0.001. aPS anti‑phosphatidylserine, GDM gestational diabetes mellitus, IC immune complex, Ig immunoglobulin, IL-1α interleukin-1α, IL-1β interleukin-1β, IL-3 interleukin-3, IL-5 interleukin-5, IL-7 interleukin-7, IL-10 interleukin-10, IL-12 (p40) interleukin-12 (p40), IL-15 interleukin-15, IL-18 interleukin-18, TNF-α tumor necrosis factor‑α.

Full size image

The ROC curves for the ten differentially expressed cytokines between the control and GDM groups are shown in Fig. 3A.

Fig. 3: ROC curve evaluation of first‑trimester serum biomarkers for GDM.
figure 3

A ROC curves for the following ten first-trimester serum cytokines: IL-1α (AUC: 0.715, 95% CI: 0.598-0.833, P = 0.001), IL-1β (AUC: 0.656, 95% CI: 0.532-0.779, P = 0.020), IL-3 (AUC: 0.663, 95% CI: 0.540-0.786, P = 0.014), IL-5 (AUC: 0.673, 95% CI: 0.550-0.795, P = 0.010), IL-7 (AUC: 0.735, 95% CI: 0.612-0.858, P < 0.001), IL-10 (AUC: 0.715, 95% CI: 0.595-0.834, P = 0.002), IL-12 (p40) (AUC: 0.670, 95% CI: 0.545-0.794, P = 0.011), IL-15 (AUC: 0.718, 95% CI: 0.602-0.834, P = 0.001), IL-18 (AUC: 0.632, 95% CI: 0.505-0.758, P = 0.048), and TNF-α (AUC: 0.641, 95% CI: 0.516-0.765, P = 0.035). B ROC curves for four first-trimester serum antibodies: total IgM (AUC: 0.774, 95% CI: 0.663-0.884, P < 0.001), total IgG (AUC: 0.855, 95% CI: 0.774-0.936, P < 0.001); aPS IgA (AUC: 0.665, 95% CI: 0.543-0.786, P = 0.014), and aPS-IgG IC (AUC: 0.721, 95% CI: 0.608-0.835, P < 0.001). C The ROC curve for the combined analysis of the top three first-trimester serum biomarkers (including total IgG, total IgM, and IL-7). D The ROC curve for the combined analysis of the top four first-trimester serum biomarkers (including total IgG, total IgM, IL-7, and aPS-IgG IC). E The ROC curve for the combined analysis of the top five first-trimester serum biomarkers (including total IgG, total IgM, IL-7, aPS-IgG IC, and IL-15) achieved a superior predictive value [AUC and 95% CI 0.906 (0.840-0.971), with a sensitivity of 75% and a specificity of 94.7%] for the development of GDM. aPS anti-phosphatidylserine, AUC area under the curve, CI confidence interval, IC immune complex, Ig immunoglobulin, IL-1α interleukin-1α, IL-1β interleukin-1β, IL-3 interleukin-3, IL-5 interleukin-5, IL-7 interleukin-7, IL-10 interleukin-10, IL-12 (p40) interleukin-12 (p40), IL-15 interleukin-15, IL-18 interleukin-18, ROC receiver operating characteristic, TNF-α tumor necrosis factor-α.

Full size image

Analysis of previously identified serum biomarkers for GDM development in first-trimester serum

We conducted ELISA assays on the first-trimester serum samples to investigate the potential differences in the levels of previously identified GDM biomarkers, including leptin, adiponectin, GDF-15, and FSTL-3, between GDM patients and control groups after IVF treatment. However, no significant differences were observed in the serum levels of these biomarkers during the first trimester, indicating their limited utility in predicting GDM (Supplementary Table 2).

Screening of first-trimester serum antibodies identified novel biomarkers for early prediction of GDM

To assess the potential utility of first-trimester serum autoimmune antibodies in the early prediction of GDM, the optical density (OD) values at 450 nm for total IgA, IgM, and IgG, as well as antiphospholipid (aPS) antibodies and aPS antibody immune complexes (ICs), were compared between the GDM group and the control group, as shown in Table 3. The OD values at 450 nm for total IgM [2.06 (1.97, 2.19) vs. 2.24 (2.16, 2.31), P < 0.001], total IgG [0.50 (0.39, 0.59) vs. 0.70 (0.58, 0.82), P < 0.001], aPS IgA [1.05 (0.79, 1.25) vs. 1.22 (0.99, 1.46), P = 0.014], and aPS-IgG IC [1.41 (1.09, 1.54) vs. 1.60 (1.39, 1.80), P < 0.001] were lower in the GDM group than in the control group.

Table 3 Antibody profile in first-trimester serum of the GDM group and the normoglycemia control group.
Full size table

Violin plots and ROC curves were generated for the serum levels of total IgM, total IgG, aPS IgA, and aPS-IgG IC, as shown in Fig. 2B and Fig. 3B, respectively. ROC curve analysis revealed favorable predictive values for total IgM (AUC: 0.774, 95% CI: 0.663–0.884, P < 0.001), total IgG (AUC: 0.855, 95% CI: 0.774–0.936, P < 0.001), aPS IgA (AUC: 0.665, 95% CI: 0.543–0.786, P = 0.014), and aPS-IgG IC (AUC: 0.721, 95% CI: 0.608–0.835, P < 0.001), offering novel insights into the early prediction of GDM.

Development of first-trimester serum predictive models for GDM

Based on the predictive efficacy of selected cytokines, previously identified GDM biomarkers, and autoimmune antibodies, we identified the top five serum biomarkers (total IgG, total IgM, IL-7, aPS-IgG IC, and IL-15) according to the AUC area under the ROC curve and established combined predictive models incorporating these biomarkers. The combinations of the first three, first four, and all five biomarkers yielded superior predictive accuracy [AUC and 95% CI: 0.904 (0.837–0.971), 0.904 (0.836–0.971), 0.906 (0.840-0.971), respectively; sensitivities and specificities: 77.8% and 92.1%, 77.8% and 92.1%, 75% and 94.7%, respectively] for GDM development, surpassing the predictive value of any single biomarker (Fig. 3C, D, E and Supplementary Table 3). These findings highlight the efficacy of combining serum autoimmune antibodies and cytokines for the early prediction of GDM.

Moreover, we evaluated five diverse GDM predictive models, including KNN, LR, SVM, RF, and XGBoost, using the top five serum biomarkers: total IgG, total IgM, IL-7, aPS-IgG IC, and IL-15 (Fig. 4A, B). The XGBoost model exhibited robust predictive capability for GDM in early pregnancy, achieving AUCs of 0.995 (95% CI: 0.995–1.000, P < 0.001) for the training set and 0.867 (95% CI: 0.789–0.952, P < 0.001) for the test set. The SHAP analysis, visualized as a bee swarm plot, highlighted the individualized impact of each biomarker, identifying total IgG as the most critical predictor, followed by total IgM, IL-7, aPS-IgG IC, and IL-15. A beeswarm plot further illustrated the association between the levels of these five biomarkers and GDM risk, thereby enhancing the interpretability of their impact within the predictive model (Fig. 4C).

Fig. 4: Machine-learning models using first-trimester serum biomarkers for GDM prediction.
figure 4

A ROC curves for GDM prediction based on the top five serum biomarkers (total IgG, total IgM, IL-7, aPS-IgG IC and IL-15) across five machine-learning models, evaluated using the training set. Performance metrics for each model are as follows: KNN (AUC: 0.800, 95% CI: 0.711–0.830, P < 0.001), LR (AUC: 0.849, 95% CI: 0.793–0.909, P < 0.001), SVM (AUC: 0.875, 95% CI: 0.808–0.947, P < 0.001), RF (AUC: 0.957, 95% CI: 0.923–0.998, P < 0.001), XGBoost (AUC: 0.995, 95% CI: 0.995–1.000, P < 0.001). B ROC curves for GDM prediction based on the top five serum biomarkers (total IgG, total IgM, IL-7, aPS-IgG IC and IL-15) across five machine-learning models, evaluated using the test set. Performance metrics for each model are as follows: KNN (AUC: 0.749, 95% CI: 0.711–0.786, P < 0.001), LR (AUC: 0.778, 95% CI: 0.706-0.853, P < 0.001), SVM (AUC: 0.782, 95% CI: 0.731–0.836, P < 0.001), RF (AUC: 0.850, 95% CI: 0.800–0.907, P < 0.001), XGBoost (AUC: 0.867, 95% CI: 0.789–0.952, P < 0.001). C Beeswarm plot of the top five SHAP-based risk factors for GDM prediction. Each dot represents one pregnant individual and indicates their relative risk of developing GDM. The colors of the dots reflect the feature values on the bottom side (horizontal): blue indicates lower values, while red indicates higher values of each risk factor. aPS anti-phosphatidylserine, AUC area under the curve, CI confidence interval, GDM gestational diabetes mellitus, IC immune complex, Ig immunoglobulin, IL-7 interleukin 7, IL-15 interleukin 15, KNN K-Nearest Neighbors, LR logistic regression, RF random forest, ROC receiver operating characteristic, SHAP SHapley Additive exPlanations, SVM support vector machine, XGBoost eXtreme Gradient Boosting.

Full size image

Discussion

Despite growing research on potential predictors of GDM, the early prediction of GDM during the first trimester of pregnancy and among women who underwent IVF treatments is still significantly unexplored. In this study, significant differences were identified in the serum levels of certain cytokines and autoimmune antibodies between pregnant women who later developed GDM and healthy controls during the 11–13 weeks of gestation. The combined analysis incorporating five significantly altered cytokines and antibodies, including total IgG, total IgM, IL-7, aPS-IgG IC, and IL-15, demonstrated notable predictive efficacy for GDM.

Although previous studies have reported an association between inflammatory mediators and the development of GDM [36, 37], these studies mainly focused on the second and third trimesters of pregnancy. Consequently, the exploration of first-trimester inflammatory biomarkers and their role in early GDM prediction remains relatively unexplored, and the existing findings are inconsistent. In our study, we observed decreased expression of most pro-inflammatory cytokines, including IL-1α, IL-1β, IL-12 (p40), IL-18, and TNF-α, in the first-trimester serum levels of women who developed GDM in late pregnancy compared to non-GDM women. It has been suggested that pro-inflammatory cytokines contribute to the angiogenesis and remodeling of uterine arteries, facilitating embryo implantation in the early stage of pregnancy [38], which implies these down-regulated first-trimester pro-inflammatory cytokines may trigger placental vascular dysplasia in the development of GDM. In contrast, serum levels of pro-inflammatory cytokines such as IL-1β, IL-18, and TNF-α may increase in the second or third trimesters in women with GDM, possibly as a compensation mechanism [36].

Furthermore, there is a shift during normal pregnancy from Th1 cytokines (including IFN-γ, TNF-β, IL-2, and TNF-α), which are typically downregulated, to Th2 cytokines (including IL-4, IL-5, IL-6, IL-10, and IL-13), which are upregulated. In our study, the observed decrease in serum IL-5 and IL-10 levels in the GDM group compared to the control group suggests a potential disruption in this normal Th1 to Th2 cytokine shift, which may contribute to the development of GDM. We also observed lower serum levels of first-trimester biomarkers in the GDM group that received little attention in previous studies, including IL-3, IL-7, and IL-15. Infertility, associated with chronic low-grade inflammation may influence these levels [39]; this preexisting inflammatory state may persist, affecting inflammation-related signal transduction and altering cytokine profiles, even after achieving successful pregnancy via IVF treatment. Our study also explored the potential involvement of previously identified first-trimester GDM biomarkers including GDF-15 and FSTL-3 [25, 26, 40], but found no significant differences in their serum levels between the GDM and control groups. Further studies are warranted to validate the utility of these biomarkers for predicting GDM.

Adipokines, such as adiponectin and leptin, have been implicated in the dysregulation associated with GDM, contributing to the metabolic disorder characteristic of this pregnancy complication [41, 42]. Some studies have reported decreased adiponectin levels and increased leptin levels during the first and second trimesters in women who conceive spontaneously and develop GDM, compared to those without the condition [43, 44]. However, our findings indicate no significant differences in serum adiponectin and leptin levels between women who subsequently developed GDM and normoglycemic controls. This is consistent with the study by Coussa et al., which reported comparable first-trimester serum adiponectin levels between GDM women and non-GDM women after IVF treatment [11]. This suggests potential differences in GDM prediction between IVF patients and the naturally conceived population. It has been shown that adiponectin and leptin levels may be affected during the ovarian stimulation process [45,46,47], probably leading to the unique kinetics of adiponectin and leptin in women who underwent IVF treatment, thus weakening the disparities in serum levels between GDM and non-GDM women.

Given the important role of immune dysregulation in the development of GDM [48], here we investigated the serum levels of specific autoimmune antibodies as potential first-trimester biomarkers for GDM. We found decreased OD values at 450 nm for total IgM, total IgG, aPS IgA, and aPS-IgG IC in the GDM group than in the control group, indicating the decreased antibody responses in the early stage of pregnancy may lead to the development of GDM. This reduction in antibody levels suggests a potential B-cell defect in GDM patients, characterized by insulin resistance [49], where B cells are crucial for producing antibodies and cytokines that regulate immune responses [50]. Dysfunction of B cells may thus cause a decreased immune response and decreased autoimmune antibody levels. Moreover, the disturbance of inflammatory factors could interact with immune reactions, contributing to altered autoimmune antibody expression in the GDM pathophysiological process. Our findings support the notion that autoimmunity plays a role in GDM development, highlighting the importance of further investigating autoimmune responses in early pregnancy as a potential pathway to understanding GDM pathogenesis.

Our study was strengthened by strict inclusion and exclusion criteria, the use of propensity score matching (PSM) to ensure comparability of baseline characteristics, and the emphasis on high-risk populations who underwent IVF treatment. Although we excluded other complications such as HDP, 3 women (7.9%) in the GDM group and 2 women (5.3%) in the control group who delivered SGA infants were observed, which may have marginally affected our findings considering the similar inflammatory processes and dysregulation of immune of SGA with GDM. In addition, our study was limited by the small sample size and retrospective nature, and prospective studies with larger cohorts are needed to confirm our findings. However, owing to the limitations of the sample volume, we did not have available data regarding the fasting plasma glucose (FPG), serum insulin levels or insulin resistance during pregnancy. Correlating these clinical markers with available clinical measurements would have added value. Furthermore, our study cohort was restricted to women who conceived through IVF, and the applicability of our findings to women who conceive spontaneously warrants future investigation.

Conclusions

In conclusion, alterations of first-trimester serum cytokines and autoimmune antibodies were observed in women undergoing IVF treatment who subsequently developed GDM. A combined analysis of five significantly altered biomarkers, including total IgG, total IgM, IL-7, aPS-IgG IC, and IL-15 yielded pronounced predictive values prior to the typical screening period for GDM.