Introduction

Gestational diabetes mellitus (GDM) is the most common pregnancy complication, affecting up to 25% of pregnancies worldwide.¹ Along with disrupted glucose regulation, GDM is characterized by dyslipidemia.² Despite extensive research into the role of routine lipid parameters, more comprehensive research into the lipidomic milieu in GDM offers promising opportunities for deeper insights into pathogenetic mechanisms and risk assessment.

Cholesterol homeostasis is a balance between cholesterol synthesis and absorption, and can be estimated by noncholesterol sterol (NCS) quantification. Desmosterol, 7‑dehydrocholesterol, and lathosterol reflect the cholesterol synthesis efficiency, while campesterol and β-sitosterol serve as surrogate markers for cholesterol absorption.³

Sphingolipids have received increasing attention in cardiometabolic complications of pregnancy, especially GDM.⁴ An increase in ceramides C18 and C24 in early pregnancy was associated with a higher risk of GDM.⁵ Sphingosine‑1‑phosphate (S1P) promotes cell survival, boosts glucose uptake, and is implicated in insulin receptor activation. Cellular S1P‑ceramide balance plays a crucial role in maintaining cellular health, and may also be involved in metabolic regulation.⁶

Several studies have focused on identifying biomarkers and multimarker models for predicting GDM and adverse pregnancy outcomes.^4,7,8 A lipid fingerprinting analysis revealed several lipid species associated with GDM,⁴ while metabolomics- and proteomics‑based models showed a significant potential for early GDM detection.^7,8

This study aimed at providing a detailed characterization of the lipid species used in GDM risk assessment toward the end of the first trimester, with special emphasis on lipoprotein distribution, NCSs, and sphingolipids.

Patients and methods

Results

Baseline demographic data, clinical features, and basic lipid status of the participants are presented in Supplementary material, Table S1. There were no significant differences regarding age, smoking status, percentage of first‑time pregnancies, TC, HDL‑C, and LDL‑C levels between the women with GDM and those without pregnancy complications. Pregestational BMI values, gestational weight gain, as well as TG levels were significantly higher in the women with GDM (P = 0.01, P = 0.001, and P = 0.009, respectively).

First, we conducted a univariable logistic regression analysis to assess the possible associations of the distribution and size of the lipoprotein subclasses with the development of GDM (Supplementary material, Table S2). Higher prevalence of large LDLI and HDL2a subclasses demonstrated a significant negative prediction of GDM (P = 0.04 and P = 0.04, respectively). In contrast, small‑size LDLIIIA subclass positively predicted GDM development (P = 0.02).

Next, we examined the predictive potential of NCSs for GDM development (Supplementary material, Table S2). To this end, absolute plasma NCS concentrations were determined (desmosterol, 7‑dehydrocholesterol, lathosterol, campesterol, and β-sitosterol), along with NCS concentrations in HDL (desmosterol_hdl, 7‑dehydrocholesterol_hdl, lathosterol_hdl, campesterol_hdl, and β-sitosterol_hdl) and cholesterol homeostasis indices (cumulative synthesis index [CSI], cumulative absorption index [CAI], and cholesterol synthesis‑to‑absorption ratio [CSI/CAI]). Low levels of 7‑dehydrocholesterol, campesterol, β-sitosterol, and CAI significantly predicted GDM development in the examined population. The level of CSI/CAI was a significant positive predictor of GDM (OR, 2.435; 95% CI, 1.066–5.561; P = 0.03). Concerning the NCS profile of HDL particles, only β-sitosterol_hdl showed a significant predictive potential for GDM.

Moreover, we examined the ability of sphingolipid profile to predict GDM (Supplementary material, Table S2). The strongest negative predictor of GDM was CerC16:0 level (P <⁠0.001), meaning that the likelihood of developing GDM was significantly higher in the patients with low levels of this ceramide. On the other hand, high levels of CerC24:0 and SAP were significant predictors of GDM (P = 0.01 and P = 0.04).

When the previously known risk factors for pregnancy complications were included in the multiple logistic regression model for GDM (age, smoking status, pregestational BMI, and weight gain) the LDLI and HDL2a subclasses, desmosterol, 7‑dehydrocholesterol, β-sitosterol levels, CAI, CerC16:0, and SM levels remained independent negative predictors of GDM, while campesterol, CSI/CAI, and CerC24:0 levels did not maintain their predictive capacity (Supplementary material, Table S2).

To assess the diagnostic accuracy of the proposed models for the early prediction of GDM, ROC analysis was conducted (Table 1). First, we developed 5 single‑marker models that included individual lipid biomarkers as independent predictors of GDM (model A, B, C, D, and E; Table 1). Next, we developed multimarker models combining commonly considered risk factors for GDM (maternal age, smoking status, pregestational BMI) and different panels of the analyzed lipid biomarkers (models G, H, I, J, and K; Table 1). In this regard, we gradually added different panels combining the lipid parameters that showed significant predictive properties in the univariable and multivariable analysis. Model G included the specified demographic parameters and traditional lipid biomarkers (TC, TG) and demonstrated the lowest AUC value, as compared with the other multimarker models. By adding LDL and HDL diameter values to model G, we developed model H (demographic markers, TC, TG, LDL and HDL diameters), with marginally higher AUC value, but without significant difference (P = 0.49). Model I consisted of the demographic markers, TC, TG, CSI, and CAI, and it yielded a higher AUC value than both model G (P = 0.01) and H (P = 0.03). Next, we developed model J as a combination of the demographic factors, TC, TG, and the analyzed sphingolipid concentrations (CerC16:0, SM). This model showed better prediction ability than both model G (P = 0.02) and H (P = 0.05), while there was no significant difference in comparison with model I (P = 0.94). The highest AUC value was observed for model K, which included a panel of the demographic markers, TC, TG, LDL and HDL diameters, NCS, and sphingolipid levels (AUC, 0.967; 95% CI, 0.92–1; P <⁠0.001). It showed higher AUC than models G (P = 0.001), H (P = 0.002), and I (P = 0.02), but not than model J (P = 0.12).

Discussion

In this study, we performed a comprehensive lipidomic analysis to gain a better insight into the lipid metabolism in GDM, and to evaluate the predictive abilities of specific lipidome patterns in GDM risk assessment.

Identifying reliable biomarkers linked to the prediction and progression of GDM could strengthen our understanding of its underlying mechanisms and offer new perspectives and targets for prevention and treatment.⁷

Previous research has showed marked differences in the lipid profile in women in healthy pregnancy and GDM‑complicated pregnancy.⁴ As opposed to conventional lipid parameters that have been extensively investigated with regard to GDM, only few studies focused on the role of cholesterol homeostasis markers in this population.^17,18 NCS are sensitive indicators of cholesterol metabolism changes, which makes them attractive parameters with a potential use in the prevention, prediction, and treatment of various cardiometabolic disturbances.¹⁹

In line with these findings, our results demonstrated that cholesterol homeostasis disturbances are linked to GDM development (Supplementary material, Table S2). Our results concerning cholesterol absorption markers are consistent with previous research.^17,18 Plasma concentrations of β-sitosterol and campesterol were strong negative predictors of GDM development. The negative correlation between insulin resistance and cholesterol absorption has been previously demonstrated; however, the exact mechanism behind this is not clear.¹⁷ An increase in the CSI/CAI ratio was associated with over a 2.4‑time higher risk of GDM (Table 1). However, after adjustment for common GDM risk factors, this association did not remain significant. On the other hand, CAI remained an independent negative predictor of GDM, supporting the strong negative relationship between cholesterol absorption and GDM (Table 1). Moreover, plasma concentration of β-sitosterol_hdl significantly predicted GDM (Table 1). An adequate mother‑to‑fetus cholesterol transport is partly dependent on the cholesterol absorption efficiency.²⁰ Reduced cholesterol absorption and the resulting impaired HDL maturation may lead to an insufficient cholesterol supply for the fetus in GDM. In support of this finding, the prevalence of HDL2a was negatively associated with GDM in this study, indicating an inverse relationship between large‑size HDL particles and GDM development (Supplementary material, Table S2). GDM is postulated to alter HDL composition and reduce its ability to facilitate cholesterol efflux.⁹ Accordingly, the prevalence of the largest and least atherogenic LDLI subclass negatively predicted, while the small‑size LDLIIIB particles positively predicted GDM in our study (Supplementary material, Table S2). This is consistent with previous reports on greater prevalence of small dense LDL subclasses in pregnancies with GDM and a positive correlation between fasting glucose level and LDLIVA and LDLIVB.¹⁹

Ceramides and S1P emerge as novel diagnostic markers and therapeutic targets implicated to have multiple roles in preeclampsia and GDM. We observed a positive predictive value of SAP and CerC24:0 concentration and a negative predictive value of CerC16:0 and SM levels for GDM development (Supplementary material, Table S2). SM species have been recognized to show protective effects in pregnancy, with lower levels observed in women with GDM. However, our results concerning the levels of CerC16:0 and CerC24:0 are contrary to the expected buildup of the short‑chain cytotoxic ceramide species (CerC16:0, CerC18:0), usually seen in GDM pregnancies, and low levels of long‑chain CerC24:0.⁵ A possible explanation for lower CerC16:0 content could be regarding it as a compensatory protective mechanism against its harmful effects.

Finally, we developed different predictive models to discriminate women with GDM from controls. They were single‑marker models and multimarker models using different combinations of lipid markers and established risk factors for GDM (maternal age, smoking, pregestational BMI; Table 1). Diverse marker combinations performed significantly better than individual variables. More importantly, the stepwise addition of the advanced lipid markers to the models consisting of routine lipid profile significantly improved their discriminatory potential for GDM. The best predictive performance was seen in model K that contained the most comprehensive combination of all parameter groups analyzed in this study, that is, age, smoking status, pregestational BMI, TC, TG, HDL diameter, LDL diameter, CSI, CAI, CerC16:0, and SM (Table 1).

Although dyslipidemia in patients with GDM has been reported previously,^2,17,18,20 this study is the first to provide a detailed lipid analysis by combining traditional lipid markers and lipoprotein subclass distribution with sphingolipid and NCS levels. However, this work has some limitations. Firstly, the small number of patients with GDM and the lack of glycemic control levels limit its strength. Due to the observational design of the study, we were unable to elucidate all aspects of the complex relationship between lipid metabolism and GDM, hence further research is needed to delve into this matter. Finally, future studies using machine learning models could better predict the development of GDM.