Longitudinal analysis strategies for modelling epigenetic trajectories.

Area of Science:

• Epigenetics
• Genomics
• Biostatistics

Background:

• DNA methylation levels change dynamically over time.
• Modeling these DNA methylation trajectories is key to understanding their biological significance.
• Computational costs limit traditional multilevel modeling to subsets of CpG sites identified via epigenome-wide association studies (EWAS).

Purpose of the Study:

• To propose and evaluate a computationally efficient linear regression method for longitudinal DNA methylation analysis.
• To compare this method against traditional multilevel modeling and EWAS-based approaches.
• To identify epigenetic changes related to exposure using longitudinal DNA methylation data.

Main Methods:

• Proposed linear regression with cluster-robust standard errors (sandwich estimator) as a computationally efficient alternative to multilevel modeling.
• Compared linear regression and multilevel modeling with three EWAS approaches (baseline, any time-point, all time-points).
• Applied methods to blood DNA methylation data from the Accessible Resource for Integrated Epigenomics Studies (ARIES) to assess prenatal smoking exposure.

Main Results:

• Linear regression with cluster-robust standard errors was 74 times more efficient than multilevel models, yielding similar effect estimates.
• EWAS restricted to baseline identified fewer associations than time-point specific EWAS or longitudinal modeling.
• Both longitudinal methods identified comparable CpG sites associated with prenatal smoking exposure (>70% agreement).

Conclusions:

• Linear regression with cluster-robust standard errors is an effective and efficient method for longitudinal DNA methylation data analysis.
• This method provides a practical alternative for large-scale epigenomic studies.
• The findings support the use of this approach for identifying dynamic epigenetic changes related to environmental exposures.