From: DNA methylation aging clocks: challenges and recommendations
Significant issue | Current problem | Potential solutions/advances |
---|---|---|
Biological age measures. No single measure or “gold standard” of biological age is likely to be possible | Simultaneously measuring multiple contributing biological processes that are changing with age | Focused analysis on specific components of aging biology and/or specific age-related diseases |
Prediction versus mechanistic insight | Predictors are by design not optimal for mechanistic insight but are nevertheless used | Separate prediction (using sparse CpG sets) from mechanistic studies (based on whole DNA methylome/integrated epigenome) |
Age-associated changes in non-dividing cells | Uncertainty over mechanism and current ability to dissect apart intrinsic (intracellular) from extrinsic (whole tissue) changes | Analyze aging in single cells. Also, determine whether construction of single-cell clocks is possible |
Confusion between epigenetic correlations of aging and the aging process itself | Aging is a complex multi-systemic process. Specific evidence is required that the epigenetic changes observed in DNA methylation clocks are driving the aging process itself | To reduce confusion for those outside the epigenomics field, publications need to be clear that epigenetic observations usually only represent a biomarker of aging |
Bias of DNA methylation clock training sets | Uncertainty of the contribution of deviation between predicted and actual age to biological aging or prediction error. Clocks trained on small samples are prone to confounding by cell composition | Larger studies, as well as increasingly focused tissue/disease-relevant clocks and cell type-specific information |
Contribution to DNA methylation clock signals of cell type proportions and rare cells and/or clonality | Uncertainty whether cell type deconvolution increases or decreases biomarker power for different diseases | Refined single-cell analysis to separate tissue cell proportional changes from intrinsic cell changes for specific diseases. Also, purified cell type analyses and further refinement of cell type deconvolution methodology |
Pan-tissue aging changes | Separation of true pan-tissue changes from cell proportion changes | Single-cell analysis to identify cell proportion changes from individual cell changes. Also, purified cell type and statistical cell type deconvolution analyses |
Aging-related increased variability in DNA methylation versus directional changes | Difficult to deconvolute intrinsic from extrinsic changes in heterogeneous cell type-derived DNA, as well as to delineate technical from biological variation | Single-cell analysis to differentiate cell proportion from individual cell changes. Use of multiple technical and statistical methodologies to dissect origin of variability, including deep-targeted BS-seq and third-generation sequencing |
Construction of a clock at the single-cell level | Currently technically challenging, particularly due to missing data in each individual cell | Imputation may be helpful, but ultimately improvements in single-cell analysis will be required |
Identification of disease-related changes | Uncertainty whether capturing the most disease relevant changes | Improved methodology: latest array increased enhancer CpGs focus—improved high-throughput power to identify tissue-specific and disease-specific loci. Also, increased deep BS sequencing |
Disease mechanism is unknown | Role of aging-related epigenomic deterioration contributing to age-related disease pathology is not appreciated | Discovery of disease-related mechanisms through disease-relevant cell and tissue-type epigenomic analysis of aging-related changes |
Regulatory role of DNA methylation is more complex than classical models | Complex interplay between transcription factors and epigenomic factors impacts on outcome within different functional loci (promoters, enhancers, insulators, transcribed regions, etc.) | Detailed experimental evidence within specific genetic loci and in disease-relevant cell types, including appropriate disease stressors, to infer potential repressive and/or activating roles |
Differentiation between DNA methylation loss due to reduction in active processes required for maintenance, or active enzymatic-driven loss | Firm evidence required in appropriate cell types of decay without cell division. May be more prevalent at dynamic enhancer regions. Neuronal cells have high post-mitotic expression of DNMTs and TETs plus high 5-hydroxymethylcytosine (5hmC) | Detailed models studying DNMT and TET expression in disease-appropriate cell types. Assaying the specific products of TET activity, such as 5hmC |
Functionality of DNA methylation changes is often assumed | Crossing statistical significance thresholds does not infer function. Statistical differences between quantitative and categorial measures | Acknowledged functional evidence deficiency in results and that further integrated disease-relevant tissue genomic/epigenomic/transcriptomic analyses in appropriate models are required |
Low reproducibility of DNA methylation clocks in model organisms reduces the utility of published clocks | Technical issue because of the reliance on low-depth sequencing due to the lack of available commercial DNA methylation arrays in non-human species | Higher-depth base-resolution sequencing studies are required to improve portability of DNA methylation clocks between experiments. Also, new methodology robust to stochastic missing data |
Aging DNA methylation sites are only partially conserved among different mammalian species | Reduced insights to be made from comparative studies | Integrative whole epigenome analyses to identify common mechanistic processes |
Role of DNA methylation and rare modifications, such as 5-hydroxymethylation (5hmC) in specific functional loci, such as enhancers | Large-scale base-resolution analyses currently performed using bisulfite conversion. This does not differentiate between 5hmC and 5mC | Oxidative bisulfite sequencing and new methodologies, such as a non-destructive DNA deaminase, and third-generation direct modification analysis |
Interconnected role of DNA modifications and chromatin modifications | Unknown directionality and causative effects of cross talk between these different epigenomic modifications | In vitro, organoid, and model organism evaluation of epigenetic machinery with age. Integration of DNA modifications, histone post-translational marks, and transcriptional data into a single integrated aging model |
Population variation in DNA methylation clock measures | Genetic variation may be influencing clock measures directly, or impacting on relevant causative factors, such as inflammation and immunological aging | Integrating genetic effectors into clock and age-related measures, including haplotypic information. This will also lead to insights into causal or mechanistic pathways |
Many different DNA methylation clock models | Many available clocks and ad hoc application and interpretation of results can result in suboptimal robustness of findings | Systematic evaluation of methods with a priori assumptions about the meaning of associations of various measures |
Forensic use of DNA methylation clocks to determine legal age | Robustness of DNA methylation clocks across populations, tissues, and environments is unknown. Furthermore, the impact of acute and chronic inflammatory processes needs assessment | Assess variability by the analysis of large, diverse, and well-powered datasets in the range of tissues likely to be employed (whole blood, buccal cells, etc.) |
DNA methylation clocks as a de facto measure of an individual’s “health” | Associations with biological aging are cross-sectional and epidemiological. Accuracy within an individual and in other populations the clock is not derived from is unknown | Longitudinal studies required to assess clock changes within an individual over time. Requires appropriately powered studies across diverse populations. Re-commercialization—public must be protected by provision of accurate data regarding estimate/error rates |