Abstract
Sex differences in methylation status have been observed in specific gene-disease studies and healthy methylation variation studies, but little work has been done to study the impact of sex on methylation at the genome-wide locus-to-locus level. This study investigates the genomic sex effect on saliva DNA methylation of 197 subjects using 20,493 CpG sites. Using T-tests, PCA, and ICA, the researchers identified sex influences on both the X chromosome and autosomes. A novel correction method based on Independent Component Analysis (ICA) is proposed to remove these confounding effects in association studies.
Introduction
In Epigenome-Wide Association Studies (EWAS), identifying chemical "tags" or methylation patterns that correlate with diseases is a primary goal. However, biological sex represents a significant source of variation that can confound these results. While X-chromosome inactivation is a known driver of methylation differences, the extent of sex influence on the 22 pairs of autosomes is often overlooked. If a methylation site is influenced by both sex and a target phenotype (e.g., age, smoking, or cancer), sex becomes a confounding variable that can lead to false discoveries.
This study utilizes a primary dataset of saliva DNA from 197 subjects (54 females, 143 males) and a validation dataset of peripheral blood cell DNA (1,298 CpG sites) to map these differences and develop a robust correction tool.
Methodology
The researchers analyzed over 20,000 CpG sites using the Illumina Infinium Methylation Assay. Three statistical approaches were compared:
- Two-sample T-tests for direct associations between sex and specific sites.
- Principal Component Analysis (PCA) to identify broad variance factors.
- Independent Component Analysis (ICA) to isolate statistically independent sources.
The study proposes an ICA-based correction algorithm. By identifying the specific "Independent Component" most strongly correlated with sex, researchers can mathematically subtract its contribution from the dataset, effectively "de-noising" the signal while preserving other phenotypic data.
Results and Discussion
The analysis revealed that sex differences are not confined to the X chromosome. While 678 sites on the X chromosome showed significant associations, 12 autosomal sites also passed stringent Bonferroni correction. In total, 580 autosomal sites showed strong differences between males and females, clustering into functional groups such as DNA transcription and RNA splicing.
Interestingly, the overlap of sex-affected genes between saliva and blood was high for the X chromosome (81%) but low for autosomes (8%), suggesting that autosomal sex-based regulation is highly tissue-specific.
The ICA-based correction proved superior to PCA. While PCA distributed the sex signal across multiple components, ICA isolated it into a single factor. After applying the ICA correction, the number of sites significantly associated with sex dropped to zero, while associations with other phenotypes, such as marijuana use, became clearer (increasing from 8 to 11 identified sites).
Conclusion
This research highlights the necessity of accounting for biological sex in genomic association studies. The proposed ICA-based correction method provides a data-driven way to disentangle sex influences from other phenotypic signals. Future work should expand this mapping to a wider variety of human tissues to create a comprehensive "sex-effect atlas" and further explore the biological mechanisms—such as hormonal influences—driving these epigenetic differences.
Summary updated