028_ Molecular mechanisms linking psychiatric medications to biological age
Research Question and Aims
Many psychiatric medications including mood stabilizers, antipsychotics, and antidepressants are also known to extend lifespan or alter aging in model organisms (McColl et al. 2007, Evason et. al 2008, Ye et al. 2014). These drugs influence DNA methylation and chromatin remodeling (Asai et al. 2013, Swathy and Banerjee 2017), but it is unknown whether they affect DNA methylation-based epigenetic clocks which are currently the most validated biomarkers of human aging (Horvath and Raj, 2018). We recently reported that clozapine is associated with a 7-year reduction in epigenetic age in men (Higgins-Chen et al. 2020). New preliminary data suggests many psychiatric medications affect epigenetic age, by influencing specific CpG modules within the clocks. Each CpG module may be linked to different hallmarks of aging (e.g. altered mitochondrial function, intercellular signaling, nutrient sensing, and proteostasis). This proposal aims to confirm these associations between medications, CpG modules, and specific biological processes utilizing PsyCourse data as an independent data set. While it is plausible that medications slow or alter the aging process, we will also investigate alternative explanations for the association between biological age and medications. Medications may exert acute effects on methylation, and thus longitudinal methylation data will be used to investigate if medications affect the rate of aging in CpG modules. Those with younger biological age may be less prone to medication side effects and thus more likely to be maintained on a medication, so we will assess the relationship between adverse events and CpG modules. We also aim to determine whether other factors previously reported to influence epigenetic aging, such as adiposity, stress, smoking, and alcohol do so through CpG modules distinct from those affected by medications.
Hypothesis: Each mood stabilizer, antipsychotic, or antidepressant influences the specific aging methylation modules and gene expression profiles associated with hallmarks of aging.
Analytic methods: The minfi R package will be used for standard quality control, normalization, and calculation of beta values from EPIC methylation array data. Individual epigenetic clock and CpG module scores, along with age residuals, will be calculated as described in Higgins-Chen et al. 2020 and Liu et al. (under revision). Smoking, alcohol, and BMI epigenetic predictors will also be calculated and correlated with phenotypic smoking, alcohol, and BMI variables.
Analyses for visit 1 and visit 3 will be conducted separately. Multiple regression analysis using the glm R package will be performed for every CpG module age residual, using two models: (1) Module age residual ~ Plate + Sex + med1 + med2 + ... + medN (2) Module age residual ~ Plate + Sex + med1 + med2 + ... + medN + smoking DNAm + alcohol DNAm + BMI DNAm + LEQneg + CTS Hierarchical clustering will be applied to t-values from the multiple regression for every independent variable and every CpG module age residual, with results displayed as a heatmap for both t-values and estimates. Results will be compared to our other two data sets.
CpG modules with strong and consistent relationships with specific medications will be further analyzed to determine if medication dose, duration of treatment, and medication adherence predict degree of age acceleration. Difference in age acceleration between visit 1 and visit 3 will be calculated for those modules, and regression analysis will be performed to determine if medications predict the degree of age change. Weighted gene co-expression network analysis (WGCNA) will be performed on proteomics data sets separately, using all individuals for which proteomics is available, including those without methylation data. Then, for those individuals with both proteomics and methylation data available, the degree of CpG module age acceleration will be correlated with eigengenes and Mahalanobis distance values for proteomic modules, and those modules will be analyzed for enriched GO and KEGG terms using the clusterProfiler or EnrichR R packages. Top hub genes (assessed by kME values) from proteomic modules will be integrated into a larger network analysis performed in Cytoscape where medications, CpG modules, and genes will be linked using pharmacogenomic databases.
Statistical power: Analysis of CpG modules influenced by medications using our other two data sets indicates effect size in our multiple regression models is f2 = 0.2 or greater. Assuming up to 6 different medications, a significance level of 0.05 and power of 0.8, power analysis indicates 87 samples will be needed.
Biological data (EPIC methylation, Lexogene mRNA, proteomics, and protein profiling analysis)
Note: Methylation data for all individuals for which it is available will needed for the study (including those without proteomics data). Proteomics data for all individuals for which it is available will be needed (including those without methylation data) in order to properly establish the proteomics modules via WGCNA. Then, multiomics analysis will be performed on the overlap between methylation and proteomics.