Gestational diabetes DNA methylation markers confirmed in small cohort, fail larger validation

A routine blood draw taken in early pregnancy could one day reveal which women are heading toward gestational diabetes, if a new epigenetic study from South Africa has anything to say about it. Researchers at a primary healthcare clinic in Johannesburg set out to confirm whether three DNA methylation signatures, first spotted in the blood of women with gestational diabetes mellitus, hold up under rigorous testing. The answer is a fascinating split decision: the molecular signals were confirmed when the same samples were retested with a gold-standard technique, but they vanished when the researchers turned to a larger, independent group of pregnant women. Published in Molecular Genetics and Genomics, the study is a candid look at both the promise and the pitfalls of translating genome-wide methylation discoveries into clinically useful biomarkers, and it arrives at a moment when the global burden of gestational diabetes is climbing steadily.

Gestational diabetes mellitus, defined as glucose intolerance first diagnosed during pregnancy, affects approximately 16 percent of pregnancies worldwide, with recent estimates reporting a prevalence of 10.4 percent across Africa. The condition is driven by factors including obesity, advanced maternal age, excessive weight gain during pregnancy, previous gestational diabetes, and a family history of diabetes. Without timely diagnosis and appropriate glucose management, the consequences ripple outward for both mother and child. In the short term, women face elevated risks of preeclampsia, cesarean delivery, macrosomia, hypoglycemia in the newborn, and birth trauma. The long-term picture is even more sobering: women with prior gestational diabetes carry an eight to ten-fold increased risk of developing type 2 diabetes and a two-fold higher risk of cardiovascular disease compared to women without the condition, while their infants face heightened risks of future obesity and metabolic disorders. Identifying affected women early allows management strategies that can prevent or improve maternal and infant outcomes, which is precisely why the hunt for reliable, accessible biomarkers has intensified.

The molecular target of the new study is DNA methylation, an epigenetic mechanism that has become a central focus of metabolic disease research. Methylation involves the addition of a methyl group to the fifth carbon of cytosine dinucleotides, creating 5-methylcytosine, a reversible chemical modification known to regulate gene expression. When these modifications are dysregulated in genes central to glucose homeostasis, insulin signaling, and beta-cell function, they may contribute to metabolic disease. High-throughput genome-wide studies using Illumina BeadChip technologies in maternal peripheral blood, placenta, and cord blood have identified differentially methylated regions and loci associated with genes and pathways relevant to gestational diabetes pathophysiology. Importantly, physiological and methylation changes during pregnancy are reflected in whole blood, raising interest in screening maternal blood for biomarkers. Because profiling methylation in peripheral blood offers a feasible, minimally invasive approach to monitor high-risk individuals, it has attracted considerable attention as a potential early-warning system for the condition.

The Johannesburg team built on their own previous genome-wide methylation study, conducted in the peripheral blood of just 12 South African women with gestational diabetes and 12 without it. That BeadChip array analysis identified 1046 differentially methylated CpG sites mapped to 939 genes enriched in gestational diabetes-related metabolic pathways. Of these, 148 were hypermethylated and 898 were hypomethylated in women with gestational diabetes compared to controls. From this molecular haystack, the researchers pulled three of the most significant candidate CpG sites for validation. The first, cg23643951, sits within intron 4 of CAMTA1, a calcium/calmodulin-dependent transcription factor implicated in cell-cycle regulation and tumour suppression, located on chromosome 1 roughly 342.5 kilobases downstream from the gene’s transcription start site. The second, cg21910650, overlaps a shared GC-rich region linking KLHDC3 and MEA1 in a head-to-head orientation on chromosome 6, meaning methylation there could influence transcription of both genes simultaneously. The third, cg16306629, is an unannotated CpG in an intergenic region of chromosome 8, with COLECT10, a complement-lectin family protein gene, as the nearest regulatory region.

Validation proceeded in two carefully staged phases using bisulfite pyrosequencing, widely regarded as the gold standard for methylation quantification. In the technical validation phase, the researchers returned to the same 24 women from the original genome-wide study, matched as far as possible for age, body mass index, and gestational age, to confirm concordance between array-based and pyrosequencing-based measurements. In the biological validation phase, they recruited an independent sample set of 46 women with gestational diabetes and 135 without it, none of whom had been included in the original discovery study. Participants were recruited at less than 28 weeks of gestation from a prospective cohort at the Johannesburg clinic, with singleton pregnancies, self-reported black ethnicity, ages between 18 and 40 years, and HIV-negative status. Women with random glucose above 11.1 mmol/L or HbA1c above 6.5 percent were excluded. Gestational diabetes was diagnosed between 24 and 28 weeks using the 75-gram two-hour oral glucose tolerance test according to International Association of the Diabetes and Pregnancy Study Groups criteria. DNA was extracted from peripheral blood collected before diagnosis, bisulfite converted, and quantified by pyrosequencing, which measures percent methylation at each individual CpG site from the ratio of cytosine to total cytosine and thymine peak heights.

The technical validation delivered a clean sweep. Pyrosequencing confirmed decreased methylation at cg23643951 within CAMTA1, showing a 1.53-fold decrease with a p-value of 0.040, and decreased methylation at cg21910650 within KLHDC3/MEA1, showing a 1.57-fold decrease with a p-value of 0.004, in women with gestational diabetes. Meanwhile, methylation at cg16306629 near COLECT10 was increased 4.5-fold with a p-value of 0.003. All three signals were consistent with the original array findings, demonstrating that the targeted pyrosequencing assays faithfully reproduced the genome-wide measurements at these specific loci. The assays were optimised with methylation standards ranging from 0 to 100 percent to assess conversion efficiency and performance, and samples failing quality control were omitted from analysis. Within the technical validation group, women with gestational diabetes also showed higher fasting and one-hour glucose tolerance test concentrations than women without the condition, as expected.

Then came the twist. When the researchers applied the same pyrosequencing assays to the larger, independent biological validation group, no significant differences were observed at any of the three CpG sites between women with and without gestational diabetes. The methylation signals that had survived retesting of the original samples simply failed to reproduce in a broader cohort. This discordance is a common and frustrating pattern in epigenetic biomarker research, and the authors dissected it with unusual transparency. The genome-wide methylation differences at individual CpG sites were likely modest, meaning genuine methylation differences could fall below the threshold of statistical detection in a more diverse validation sample. The original discovery phase had applied relaxed selection criteria, identifying candidate CpG sites using an unadjusted p-value threshold of less than 0.01 in the absence of false discovery rate correction, a choice that increases the odds of selecting signals that cannot survive independent testing.

Sample selection discordance between the discovery and replication cohorts likely compounded the problem. The technical validation group was tightly matched for age, body mass index, and gestational age, and comprised the very samples used in the original array study, thereby minimising confounders known to influence methylation profiles and replicating the original patterns. In contrast, the biological validation group was broader and more heterogeneous, with greater interindividual variability, particularly in gestational age, which ranged from 10 to 27 weeks. This mismatch between discovery and replication cohorts is a recognised methodological challenge, explaining why methylation signals detectable under tightly controlled conditions often fail to reach significance in more representative settings. Platform-related technical differences between the Illumina EPIC BeadChip array and bisulfite pyrosequencing may also have contributed, since array platforms are subject to probe-level biases, background noise correction, and normalisation assumptions that differ fundamentally from quantitative pyrosequencing measures, and discordance can arise at individual loci where methylation differences are small. Additional considerations include the heterogeneous cell-type composition of peripheral blood, which was statistically adjusted for in the original array study but not in the biological validation group, and unmeasured lifestyle factors such as diet, physical activity, smoking, and alcohol consumption, all of which can influence DNA methylation and leave residual confounding that cannot be excluded.

Yet the study was not without encouraging findings. Within the technical validation group, methylation at cg21910650 within KLHDC3/MEA1 was inversely correlated with fasting plasma glucose, showing a Spearman rho of -0.625 with a p-value of 0.015, and with one-hour glucose tolerance test levels, showing a rho of -0.555 with a p-value of 0.034. Methylation at cg16306629 near COLECT10 was positively correlated with fasting plasma glucose, showing a rho of 0.709 with a p-value of 0.006. These associations align with previous research showing that lower placental gene-specific methylation is associated with higher maternal glucose concentrations in women with gestational diabetes, suggesting that differential methylation at particular loci may be linked to glucose tolerance during pregnancy. Notably, a correlation analysis between gestational age at collection and methylation at the three candidate CpG sites in the biological validation sample showed no significant association, suggesting gestational age is unlikely to explain the absence of methylation differences there. To the authors’ knowledge, this is the first validation study investigating CpG-specific methylation changes within CAMTA1, KLHDC3/MEA1, and COLECT10 in women with and without gestational diabetes in a South African population, addressing the significant under-representation of African populations in gestational diabetes epigenetic research. The biological validation cohort was larger than those used in previous studies elsewhere, suggesting the work was sufficiently powered and that the lack of replication could reflect population variability. The researchers propose that future longitudinal follow-up studies of mothers with gestational diabetes and their offspring are warranted to determine whether altered methylation at these loci is associated with glucose dysregulation and long-term metabolic health, and they call for standardised protocols covering sample collection timing, processing, and normalisation to enable more direct comparisons across populations and methodologies. For now, the three CpG sites remain candidates rather than biomarkers, but the associations with glucose concentrations provide a foundation for future research to validate their clinical utility as early detectors of gestational diabetes.

Subject of Research: Candidate DNA methylation markers for gestational diabetes: findings from technical and biological validation in a South African population

Article Title: Candidate DNA methylation markers for gestational diabetes: findings from technical and biological validation in a South African population

Article References: Dias, S., Masete, M., Adam, S., Mutavhatsindi, H., & Pheiffer, C. (2026). Candidate DNA methylation markers for gestational diabetes: findings from technical and biological validation in a South African population. Molecular Genetics and Genomics, 301(1), Article 205. https://doi.org/10.1007/s00438-026-02528-7

Image Credits: AI Generated

DOI: 10.1007/s00438-026-02528-7

Keywords: Candidate, methylation, markers, gestational, diabetes, findings, technical, biological, validation, South, African, population