All serum samples are labeled with a recognizable study number. We will collect whole blood in a tube using a vacutainer. After collection, the blood is allowed to clot without hemolysis by leaving it undisturbed at room temperature for up to 30 minutes. After the clot is removed by centrifugation at 1500 rpm for 10 minutes, the resulting supernatant will be designated as serum. Using a pipette, the serum is then transferred into small aliquots. The samples are being stored in a deep freezer at –80 °C until analysis. The samples will then be processed in batches, along with appropriate internal standards (ISs) and blanks, for metabolomics in accordance with the protocols.

A major challenge in placental sample collection is the unpredictable timing of delivery during the third trimester and the mode of delivery, except in elective cesarean sections. Obtaining samples becomes difficult during nighttime deliveries or emergency cesarean sections. Placentas from morning deliveries will be used for metabolomics and histopathology, while those from nighttime deliveries will undergo histopathological examination only.

The placental samples are collected from 9 representative individuals, as described previously [12,13]. The sampling procedure is as follows. We collect the placenta within 30 minutes of delivery without placing it in saline or formalin. A photograph of the placenta is taken before sample collection. The samples are collected at equidistant points from the cord, avoiding the edges, fibrin deposits, infarcts, necrotic areas, and large blood vessels, with each piece measuring 1 to 2 cm in diameter. No additives are added to the placental tissue pieces. A total of 9 tissue pieces (3 fetal, 3 central, and 3 maternal) are placed into empty tubes and snap-frozen in liquid nitrogen. The tubes remain in liquid nitrogen for approximately 2 to 10 minutes. Afterward, the tubes are removed from liquid nitrogen and stored at –196 °C in a liquid nitrogen dewar until analysis. Photographs of the entire placenta are taken again, and the distance from the cord insertion site is measured to confirm the site of biopsy. The photographs are discarded after confirmation [8,14].

Cord blood is collected immediately after the delivery of the placenta once the cord is clamped and cut from the placenta. The sample processing and storage procedures are the same as those followed during antenatal blood collection.

All placentas will undergo comprehensive macroscopic and microscopic structural assessments [15]. Placental tissue will be assessed using the Amsterdam Placental Workshop Group (APWG) consensus classification, with maternal vascular malperfusion, fetal vascular malperfusion, and inflammatory lesions scored on a 0 to 3 scale for severity. GDM-specific features, including villous hypermaturity, trophoblastic hyperplasia, villous edema, and fibrinoid deposition, will also be assessed [16]. A comprehensive template for the macroscopic and microscopic evaluation of the placenta is provided in Table 1.

All electronically documented study information will be stored in a dedicated, secure, and restricted-access REDCap (Vanderbilt University) database. Only study investigators and key study staff will have access to the database.

All data entry and cleaning will be conducted by data entry staff using a double-entry verification process. All analyses will be performed using R (version 26; R Project for Statistical Computing). Descriptive statistics will summarize participant characteristics. All continuous variables will be checked for normality before analysis. All clinical characteristics and demographic data will be presented as either categorical data (percentages) or continuous data (means or medians, as appropriate). We will include all the data for final analysis even if not all the samples were obtained, following the intention-to-treat analysis.

All serum (antenatal, cord blood, and postnatal serums) and placental samples will undergo untargeted metabolomic profiling to identify hundreds of metabolites spanning carbohydrate, lipid, amino acid, and energy metabolism pathways. Metabolite profiling will be performed using liquid chromatography tandem mass spectrometry (LC-MS).

The acquired LC-MS or raw mass spectrometry (MS) data will initially be processed using Agilent MassHunter (Agilent Technologies Inc) software for data quality assessment, chromatographic inspection, and peak detection. The resulting data files will then be processed using MS-DIAL (version 5.5; RIKEN Center for Sustainable Resource Science; CompMS; MS-DIAL software program for untargeted metabolomics) for peak detection, noise filtering, chromatographic peak alignment, deconvolution of MS or MS spectra, and metabolite annotation. During MS-DIAL processing, appropriate parameters will be applied for mass detection, peak height threshold, minimum peak width, MS1 or MS2 tolerance, retention time tolerance, and chromatographic peak alignment based on the LC-MS or MS acquisition conditions. For metabolite annotation, accurate precursor mass and MS or MS spectral matching will be used against available reference libraries, with isotope pattern and retention time information considered where available.

Quality control (QC) samples will be prepared by pooling representative samples and analyzed after every 10 study samples throughout the analytical sequence to monitor instrument performance, signal stability, analytical reproducibility, and potential signal drift. Blank samples will also be included to identify background signals and contaminants originating from solvents, sample preparation, or the analytical system. Features with substantial contributions from blank samples will be removed during blank filtering prior to downstream analysis.

ISs or spiking controls will be added to the sample extraction protocol to monitor extraction efficiency and analytical variability. Following initial QC and filtering, metabolite intensities will be subject to IS-based normalization to correct for extraction-related and instrumental variability. Appropriate ISs will be used to normalize the corresponding metabolite signals. Following IS normalization, the data will be further processed using median-based normalization to minimize the remaining sample-to-sample variation.

Metabolite annotation will be performed based on a combination of accurate precursor mass, retention time, isotope pattern, and MS or MS spectral matching against available reference spectral libraries. The resulting quality-controlled and normalized metabolite dataset will subsequently be used for statistical analysis and biological interpretation, including pathway analysis using appropriate metabolomics bioinformatics tools.

The distribution of metabolite abundances will subsequently be evaluated using medians and IQRs, particularly because metabolomics data may not follow a normal distribution and can contain extreme values or outliers. The normalized metabolomics data will then be log-transformed to reduce right-skewness and stabilize variance.

For the statistical analyses, given the high dimensionality of metabolomics data and the large number of metabolites evaluated simultaneously, correction for multiple hypothesis testing is essential to reduce the risk of false-positive findings. We will further expand the statistical analysis section of the manuscript to explicitly describe our multiple-testing strategy.

For 2-group comparisons, Student t test or the Mann-Whitney U test will be used as appropriate, while ANOVA or Kruskal-Wallis tests will be used for comparisons involving more than 2 groups. The resulting metabolite-wise nominal P values will be subjected to Benjamini-Hochberg false discovery rate (FDR) correction across the corresponding set of simultaneously tested metabolites. The resulting FDR-adjusted P values will be used to determine statistical significance, with an adjusted P value of <.05 considered significant.

In addition to statistical significance, metabolite-level effect sizes will be considered to assess the magnitude of biological changes. Fold changes and/or log₂ fold changes will be calculated and reported alongside nominal P values and FDR-adjusted q-values. Thus, the interpretation of differential metabolites will be based on both the statistical evidence and the magnitude of the observed metabolic alteration.

For correlation analyses involving multiple metabolites and clinical or biochemical variables, Spearman correlation coefficients will be calculated, and the resulting P values will similarly be corrected using the Benjamini-Hochberg FDR procedure. This will ensure that the increased probability of false-positive findings arising from multiple correlation tests is appropriately controlled.

Annotated metabolites will subsequently be used for biological interpretation and pathway analysis. KEGG (Kyoto Encyclopedia of Genes and Genomes) will be used for metabolic pathway mapping, while MetaboAnalyst (version 6) will be used for pathway analysis and visualization [16,17,18].

The metabolites and their pathways that are significant at different time points (antenatal, placental, cord blood, and postnatal time points) will be tabulated and correlated to determine whether any metabolites or their pathways differ at a specific time point. Moreover, metabolites and their pathways identified in the placenta will be correlated with microscopic architecture scores from the APWG and with GDM-specific changes in the placenta using Spearman rank correlation. Univariate and multivariate logistic regression analyses will also be performed to assess the association between the microscopic architecture score and the identified metabolites. To compare the 2 groups, the chi-square test or an independent-samples t test will be used, with P<.05 considered significant.