Characteristics of Placental Bacterial DNA Communities in Pregnant Women with Gestational Diabetes Mellitus and Association with Early Neonatal Gut Microbiota and Short-Term Perinatal Outcomes

Study population

As an observational prospective cohort study, it included 104 pregnant women who delivered at our hospital between March 2025 and August 2025. The study flow is shown in Fig. 1. Inclusion criteria were age 18–40 years, delivery at 37–42 weeks of gestation, singleton pregnancy with natural conception, establishment of prenatal medical records and regular antenatal examinations, completion of oral glucose tolerance test screening during pregnancy, and provision of written informed consent and voluntary participation. Exclusion criteria were multiple pregnancies or conception via assisted reproductive technologies such as in vitro fertilization, preexisting hyperglycemia or glucose metabolism abnormalities identified before 20 weeks of gestation, severe pregnancy complications during the current pregnancy (such as placenta previa or placenta accreta), chromosomal abnormalities in either parent or the fetus, antibiotic use within 3 months prior to delivery, premature rupture of membranes, presence of any of the following conditions: autoimmune diseases, congenital heart disease, hematological diseases, chronic hypertension, chronic kidney disease, insulin resistance, thyroid dys-function, or other severe medical or surgical disorders, or fewer than 500 valid 5-region 16S rRNA sequence reads. This study was approved by the Ethics Committee of Hangzhou Linping District Maternal & Child Health Care Hospital and was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent.

Fig. 1.

Sample size calculation

According to the literature, the operational taxonomic unit counts of 16S rRNA placental microbiota were 159.5 ± 34.57 in the GDM group and 176.25 ± 27.86 in the control group (Zheng et al. 2017). Using the two-sided t-test formula for comparing the means of two independent samples: n=2×(Z1-α/2+Z1-β)2σ2/δ2, with a significance level of α = 0.05, power 1 – β = 0.80, pooled standard deviation σ ≈ 31.40, and mean difference between groups δ = 16.75, the calculated sample size per group was approximately 56 cases, yielding a total sample size of 112. Considering an anticipated dropout or noncompliance rate of approximately 20%, the final sample size was set at 70 per group, for a total of 140 participants. The study initially planned to enroll 140 participants, but due to the limited number of subjects who met all inclusion and exclusion criteria and completed maternal-infant paired sample collection during the study period, 115 were preliminarily enrolled. After sequencing quality control (QC), 104 cases were finally analyzed.

Data collection

All clinical and laboratory data were collected from the medical records of all participants and reviewed by experienced obstetricians. Maternal demographic and anthropometric information included age, pre-pregnancy weight, pre-delivery weight, GWG, and body mass index (BMI). Other data collected included dietary habits, education level, and mode of delivery.

Early pregnancy biochemical markers included fasting blood glucose (FBG), postprandial blood glucose (PBG), glycated hemoglobin (HbA1c), total cholesterol (TC), triglycerides (TG), total bilirubin, and total bile acids. All pregnant women underwent a 75-g oral glucose tolerance test between 24 and 28 weeks of gestation. After at least 8 hours of overnight fasting, 5 mL of fasting venous blood was collected in the morning. Blood samples were centrifuged at 3,000 rpm for 15 minutes. Biochemical parameters were measured using a fully automated biochemical analyzer (AU5800, Beckman Coulter, USA) according to standard procedures. PBG levels were measured following the same protocol after ingestion of standardized glucose to ensure comparability of glycemic control.

Short-term neonatal outcomes included birth weight, neonatal Apgar scores, and neonatal hospitalization.

Disease and group definitions

The diagnosis of GDM was based on the 75-g oral glucose tolerance test performed at 24–28 weeks of gestation, with any of the following criteria confirming GDM: FPG ≥5.1 mmol/l, 1-hour postprandial glucose ≥10.0 mmol/l, or 2-hour postprandial glucose ≥8.5 mmol/l (Hod, Kapur 2015). GWG was calculated as the difference between maternal weight at delivery and pre-pregnancy weight to reflect weight changes during pregnancy (Jia, Bo 2025). GWG was categorized according to the 2009 Institute of Medicine guidelines for GWG, as shown in Supplementary Table I. Education level was classified into four categories based on the 2011 International Standard Classification of Education: lower secondary or below, upper secondary, post-secondary non-tertiary (college), and university degree.

Based on GDM diagnosis, study participants were divided into GDM (n = 48) and normal control (n = 56) groups. To better understand how GWG influences pregnant women with GDM, we stratified the GDM group into inadequate (n = 17), appropriate (n = 16), and excessive (n = 15) GWG subgroups.

Sample collection

Placental samples were collected within 30 minutes after delivery. A region approximately 3 cm from the placental edge was selected. From the maternal side, a rectangular tissue piece measuring approximately 0.5 cm in width, 3 cm in length, and 1-1.5 cm in depth was excised, avoiding penetration through the fetal membrane surface. The superficial layer of approximately 0.25–0.5 cm from the maternal side was then removed. Nine cubic tissue blocks, each with an edge length of approximately 0.5 cm, were collected as study samples. Neonatal first-pass meconium samples were collected within a few hours after birth, approximately 200 mg per sample, ensuring that sterilized diapers were used by the infant to capture the initial gut microbial colonization as accurately as possible. All samples were placed in sterile, labeled cryotubes, rapidly frozen in liquid nitrogen, and stored at -80°C for DNA extraction and microbiota analysis.

5-Region 16S rRNA sequencing

The study employed 5-region 16S rRNA sequencing, targeting multiple variable regions (V2, V3, V5, V6, and V8) of the bacterial 16S rRNA gene to characterize bacterial communities in placental and neonatal meconium samples. Compared with single-region 16S sequencing, the multi-region strategy enhances taxonomic coverage and resolution, making it particularly suitable for exploratory detection of bacterial DNA signals in low-biomass samples such as the placenta (Fuks et al. 2018, Jiang et al. 2025). All negative controls were processed following the identical workflow. To account for potential contamination from the hospital and laboratory environment, as well as during different stages of sample processing, various types of controls were included, including sampling controls, DNA extraction controls, and no-template PCR amplification controls.

DNA was extracted from frozen samples using the CTAB method with the DP302-02 kit (TianGen, China). DNA concentration was measured using a Qubit 3.0 fluorometer (Invitrogen, USA) (Li et al. 2025). PCR amplification of the 16S rRNA gene was performed targeting the V2, V3, V5, V6, and V8 variable regions (Wang et al. 2025). The PCR reaction conditions were as follows: initial denaturation at 95 °C for 3 min; 35 cycles of 95 °C denaturation for 30 seconds, 55°C annealing for 30 seconds, and 72°C extension for 30 seconds; a final extension at 72°C for 5 minutes; and storage at 4°C.

The PCR products were quantified using a Qubit 3.0 fluorometer and pooled at equimolar concentrations. The pooled amplicons were purified with AMPure XP magnetic beads (Beckman Coulter, USA) to remove residual primers and impurities. Library concentrations were measured using the KAPA Library Quantification Kit (KAPA Biosystems) on a Roche LightCycler 480 real-time PCR system (Roche, Switzerland). The final libraries were sequenced on the NovaSeq 6000 platform (LC Biotechnology Co., Ltd., Hangzhou, China) using paired-end sequencing (2 × 150 bp).

Microbiome data analysis

Raw sequence data were processed in QIIME2. Adapter and primer sequences were first removed using the “cutadapt” plugin. The Short Multiple Regions Framework was applied to integrate reads from the five amplified regions into a unified microbial community profile, to enhance the taxonomic resolution of multi-region 16S rRNA sequencing data. QC and amplicon sequence variant identification were performed using the “DADA2” plugin.

Bacterial taxonomic annotation was performed based on the Greengenes 2 database (2024) constructed by Nejman et al. to identify the most likely 16S sequences and calculate relative abundances. To reduce the impact of low-abundance noise on subsequent analyses, sample read counts were normalized, samples with fewer than 500 total valid reads (including negative controls) were excluded, and bacterial features with relative abundance <10⁻⁴ were filtered out. Using a stringent contaminant filtering workflow (Li et al. 2025), with a 30% prevalence threshold, potential experimental and environmental contaminants were identified and removed. Remaining signals were considered to originate from genuine tissue-resident microbial communities. We set a threshold of <500 reads for QC to exclude samples with insufficient sequencing depth, where community composition estimates are unstable and prone to random noise.

Given that placental tissue is a low-biomass sample prone to bacterial DNA contamination from reagents, environments, and experimental procedures, we included sequencing reads from the negative control samples in the QC and contaminant-identification pipeline, along with the study samples. Controls were not used for biological comparisons between the GDM and Normal groups, but to assess background bacterial DNA signals and identify contaminants. Following contaminant filtering strategies from previous low-biomass microbiome research (Li et al. 2025), we adopted a prevalence-based identification method using negative controls: a bacterial feature detected in ≥30% of negative control samples was designated as a potential contaminant and removed prior to downstream analysis. After low-abundance filtering and contaminant removal, the remaining bacterial features were used to characterize bacterial DNA communities in placental and meconium samples. Of note, while this pipeline helps reduce the impact of potential contaminant signals, it cannot completely rule out residual contamination in low-biomass placental samples.

To assess the composition and distribution of dominant bacteria across taxonomic levels and study groups, we calculated relative abundances of each taxon and visualized them using stacked bar plots. α-diversity, evaluated using Shannon and Simpson indices, reflected within-sample richness and evenness of microbial community. β-diversity was used to assess differences in microbial community composition between samples, calculated using the Bray-Curtis distance, which reflects dissimilarity in both composition and relative abundance. Differences in β-diversity between groups were tested using permutational multivariate analysis of variance (PERMANOVA), which is typically used to determine whether statistically significant differences existed in overall microbial community structure between groups, with results visualized through principal coordinates analysis (PCoA). To identify differentially abundant microbial taxa between groups, linear discriminant analysis effect size (LEfSe) was applied, which assesses both the statistical significance of differential microbiota and their contribution to intergroup differences, with thresholds set at linear discriminant analysis score > 2 and P < 0.05.

Assessment of microbial sharing between placenta and meconium

To evaluate the degree of microbial sharing between the placenta and meconium, the microbial source tracking algorithm FEAST (v0.1.0) was used (Shenhav et al. 2019). FEAST is a source tracking model that estimates the relative contribution of candidate source samples to a target sample based on similarities in microbial community composition (Gao et al. 2023). Of note, FEAST primarily estimates potential source contributions based on compositional similarity between samples; the results reflect relative source contribution patterns and are not equivalent to actual biological transmission strength. To ensure reliability, only samples with a sequencing depth ≥5,000 reads were included. FEAST was run with 1,000 expectation-maximization iterations. The microbial source contributions were quantified per sample, averaged, and normalized so that they sum to 100%. A shared taxon was defined as a bacterial species detected in both a mother’s placental sample and her infant’s meconium sample. The proportion of shared taxa was calculated as the percentage of participants with at least one shared species among the total number of participants evaluated at that time point.

Visualization and statistical analysis

All downstream data processing and analyses were conducted in R (v4.3.1). Data are presented as mean ± standard deviation or median (interquartile range) for continuous variables, compared using t-tests or nonparametric tests, and as n (%) for categorical variables, compared using χ² or Fisher’s exact test. Statistical tests were conducted using the “rstatix” package (v0.7.2) (Kassambara 2023). Visualization analyses were performed using the “ggplot2” (v3.4.4) and “ComplexHeatmap” (v2.15.4) packages (Gu 2022). All figures were generated from normalized data to ensure comparability and reproducibility.

Given that delivery mode may affect neonatal meconium microbiota composition and source tracking results, we performed subgroup analyses by delivery mode to assess the robustness of GDM-associated microbial alterations and placental source similarity-based contribution estimates. Due to the limited number of vaginal deliveries, particularly in the GDM group, these stratified analyses were considered exploratory.