Black women in the United States are diagnosed with endometrial cancer at lower rates than White women but die from it at nearly twice the rate. Understanding the biological basis of this disparity is a critical research priority, but most genomic studies have historically enrolled predominantly White patients.
This study analyzed data from 393 endometrial tumor samples in The Cancer Genome Atlas (TCGA), a large publicly available database of cancer genomic data. The patients included 294 self-reported White individuals and 99 self-reported Black individuals.
Rather than relying solely on self-reported race (which reflects social identity), the researchers used genetic ancestry proportions derived from the patients' own DNA to estimate the percentage of African genetic ancestry in each tumor sample. Comparing self-reported race to genetically estimated ancestry allowed the team to ask which measure more precisely predicts molecular differences in tumors.
DNA methylation is a chemical modification where a small molecule (a methyl group) attaches to specific locations along the DNA strand, typically at sites called CpGs. Methylation can turn genes on or off without changing the underlying DNA sequence, a phenomenon studied in the field of epigenetics.
The researchers compared methylation patterns between patients with high African ancestry proportions versus those with low African ancestry proportions, identifying 471 differentially methylated CpG sites (DMCs) that showed statistically significant differences between the two groups.
Many of these differences appeared as hypomethylation (reduced methylation) in genes involved in drug metabolism pathways in the high African ancestry group. This is potentially important because altered drug metabolism can affect how effectively cancer treatments work, providing a molecular-level explanation for some of the observed differences in treatment outcomes.
With hundreds of differentially methylated sites identified, the challenge was to determine which ones were most biologically meaningful and least redundant. For this, the researchers applied Recursive Feature Elimination (RFE) combined with logistic regression, a machine learning technique that iteratively removes the least informative variables until only the most predictive set remains.
This process reduced 471 candidate DMCs to 38 Epigenetic Signature Genes (ESGs) that collectively distinguished high from low African ancestry tumors. A further step identified a core set of just 4 genes: TRPC5, APOBEC1, PLEKHG5, and WHSC1, which the authors call Core-ESGs.
TRPC5 is a calcium channel gene. APOBEC1 is part of a family of enzymes that can edit DNA and RNA and has been implicated in cancer mutation patterns. PLEKHG5 is involved in cell signaling. WHSC1 (also known as NSD2) is a chromatin-modifying enzyme linked to cancer progression. Each of these has distinct functions that may contribute to differences in tumor behavior across ancestry groups.
To determine whether the epigenetic differences have clinical consequences, the researchers examined whether the 38 ESG methylation patterns were associated with patient survival outcomes using Kaplan-Meier survival analysis, a standard statistical method for comparing survival curves between groups.
Two genes stood out as having statistically significant associations with worse survival specifically in patients with high African ancestry: APOBEC1 and PLEKHG5 (p=0.006). This finding suggests that the methylation alterations in these genes are not merely statistical curiosities but may have real consequences for patient outcomes.
The researchers also noted that genetic ancestry proportions were a stronger predictor of these epigenetic differences than self-reported race in several analyses, suggesting that future genomic studies examining racial health disparities may benefit from including genetic ancestry estimates alongside traditional demographic categories.
A key translational finding from this study is that 9 of the 38 ESGs have known therapeutic relevance: they encode proteins that are already targeted by existing drugs or are actively being studied as drug targets in cancer treatment. This overlap between ancestry-associated epigenetic changes and druggable targets opens a path toward ancestry-informed treatment strategies.
The broader implication is that the biology of endometrial cancer may differ in ways that are systematically linked to genetic ancestry, and that these differences could partly explain why standard treatment approaches do not perform equally well across all patient populations. Recognizing this is a step toward more equitable, personalized cancer care.
The authors emphasize that this is an exploratory study requiring validation in larger and more diverse cohorts. However, the combination of machine learning for feature selection and survival analysis for clinical validation provides a rigorous methodological foundation for follow-up research targeting these disparity-linked epigenetic differences.