Discover how histone modifications unlock secrets of human aging, rivaling the precision of DNA methylation in decoding biological age.
Study: Histone mark age of human tissues and cell types. Image Credit: peterschreiber.media / Shutterstock
In a recent study published in the journal Science Advances, a group of researchers investigated the role of histone modifications in human aging by developing and evaluating histone-specific age prediction models across tissues and cell types.
Background
Aging involves complex cellular and molecular changes, including epigenetic modifications like Deoxyribonucleic acid (DNA) methylation and histone marks. Age predictors, or "clocks," have been developed using DNA methylation, transcriptomics, and blood chemistry data, with methylation-based models achieving a median absolute error of ~4 years.
While histone marks offer an interpretable framework based on the histone code, their potential for constructing accurate age predictors remains underexplored. Research has shown age-related shifts in histone modifications, suggesting their utility in modeling aging. However, the study emphasizes that sample size plays a critical role in determining the accuracy and reliability of such predictors.
Further research is needed to fully understand their role and to establish histone-based clocks comparable to existing methylation-based predictors.
About the Study
Researchers collected 1,814 human tissue chromatin immunoprecipitation sequencing (ChIP-seq) samples from the Encyclopedia of DNA Elements (ENCODE) project in bigWig format to generate and interpret histone-based age predictors.
The samples included seven histone modifications: histone H3 lysine 4 trimethylation (H3K4me3), histone H3 lysine 27 acetylation (H3K27ac), histone H3 lysine 27 trimethylation (H3K27me3), histone H3 lysine 4 monomethylation (H3K4me1), histone H3 lysine 36 trimethylation (H3K36me3), histone H3 lysine 9 trimethylation (H3K9me3), and histone H3 lysine 9 acetylation (H3K9ac).
Data processing involved averaging the negative base-10 logarithm of P-values’ signals across gene bodies using Homo sapiens annotations from Ensembl release 105. Samples with substantial missing features were discarded, and missing values were encoded as zero.
Various genomic regions were analyzed, including intergenic regions and Cytosine-phosphate-Guanine (CpG) dinucleotides. Embryonic samples were encoded with gestational week adjustments, while anonymized samples over 90 were assigned an age of 90.
To test in vitro performance, 568 additional samples spanning 12 histone marks were collected. Imputed data from ENCODE’s Avocado dataset added 1,379 samples, enhancing the training dataset. Age predictors employed Elastic Net regularization-based feature selection, principal component analysis (PCA) with truncated support vector decomposition, and automatic relevance determination regression, all implemented in Python. Performance evaluation used 10-fold nested cross-validation to prevent artificially inflated accuracy metrics, excluding cancer samples.
Histone-based predictors were compared to DNA methylation-based predictors, with the study noting the impact of differences in sample size and dataset distributions on the comparison. Predictor interpretation involved gene set enrichment analysis (GSEA), selecting genes significantly contributing to age prediction accuracy. Statistical analyses employed Python packages, ensuring validation.
Study Results
Focusing on seven histone marks (H3K4me3, H3K27ac, H3K9ac, H3K9me3, H3K27me3, H3K36me3, and H3K4me1), researchers used data from 1,814 human tissue samples spanning 82 tissues and age groups ranging from embryonic stages to 90-plus years. The samples represented diverse biological contexts and were processed using standardized methods to ensure consistency and reliability.
To create age predictors, researchers reduced the dimensionality of the data by averaging negative log-transformed P-values for each histone modification across gene bodies. These values were then transformed to stabilize the variance. Uniform manifold approximation and projection (UMAP) and PCA revealed distinct clustering based on histone type, with some age-related trends emerging, particularly for samples over 70 years old.
Histone marks showed significant correlations with age, particularly the repressive marks H3K9me3 and H3K27me3, which decreased with age, and the activating mark H3K4me3, which increased. Notably, the study observed that signal variance for all histone marks increased with age, highlighting epigenetic drift as a key factor in declining regulation. These observations informed the development of multivariate age predictors using ElasticNet for feature selection, principal component analysis to reduce noise, and automatic relevance determination regression for age estimation.
The histone-specific age predictors demonstrated robust performance, with H3K4me3 achieving the highest accuracy (Pearson’s r = 0.94, median absolute error = 4.31 years). Comparisons with DNA methylation-based predictors indicated comparable accuracy, particularly for activating histone marks, though the paper notes that DNA methylation predictors often have a younger skew in sample age distributions, which can affect performance comparisons. Additional experiments with imputed and primary cell data confirmed the reliability and accuracy of the histone mark predictors.
GSEA and pathway analyses highlighted developmental processes, transcriptional regulation, and ribonucleic acid (RNA)-related pathways as key contributors to age prediction. Histone-coding genes and age-related genes such as Homeobox D8 (HOXD8), Thioredoxin Interacting Protein (TXNIP), and Period Circadian Regulator 1 (PER1) were strongly associated with histone mark changes.
The study also introduced a pan-histone, pan-tissue age predictor, which leverages shared age-related trends across histone marks. This model not only performed comparably to histone-specific predictors but also emphasized the shared epigenetic patterns across the genome that underpin aging.
Conclusions
Since the development of DNA methylation-based age predictors, biohorology has rapidly expanded, offering biomarkers like telomere length, transcriptomics, and proteomics. While DNA methylation clocks are accurate, interpreting them is often challenging due to ambiguous gene associations. In contrast, histone mark-based predictors reveal genes linked to development, circadian regulation, and aging. Using ChIP-seq data, researchers created age predictors from seven histone marks.
Crucially, the study demonstrated that models trained on one histone mark could predict age using another, underscoring the shared epigenetic information across histone modifications. This research highlights the interpretability and potential of histone mark-based predictors as a robust tool for understanding epigenetic aging and developing age estimation models.