Work

Epigenetic clocks and inflammaging: pitfalls caused by ignoring cell-type heterogeneity

Year: 2025
Type: article
Abstract: The epigenome-wide association study (EWAS) field is now over a decade old [ 1 ]. The aim of these EWASs is to identify differentially methylated cytosines (DMCs) associated with a wide range of exposures and phenotypes, including aging. Overwhelmingly, the tissue of choice has been whole blood (WB) or peripheral blood mononuclear cells (PBMC). These are complex tissues composed of many different immune cell types, each one characterized by a unique DNA methylation (DNAm) profile [ 2 , 3 , 5 ]. Some of the earliest EWASs already noted the importance of taking cell-type heterogeneity (CTH) into account when identifying differentially methylated cytosines (DMCs) [ 6 , 7 ]. For instance, one of the earliest EWAS by Liu et al. studied rheumatoid arthritis (RA), an inflammatory autoimmune disease, demonstrating how the overwhelming majority of RA-associated DMCs merely reflect the increase in the neutrophil to lymphocyte ratio that is associated with a condition like RA [ 6 ] (Fig. 1 a, b). Indeed, most of these DMCs disappear upon adjustment for this shift in the neutrophil to lymphocyte ratio (Fig. 1 b). In other words, most of the original RA-DMCs do not reflect DNAm changes that happen in specific immune cell types of RA cases compared to controls. Thus, it is surprising that a recent study by Skinner and Conboy [ 8 ] does not adjust for this underlying CTH when analyzing the exact same EWAS by Liu et al. Indeed, numerous articles have, over the last decade, continued to emphasize the critical need to estimate immune cell–type fractions in blood and to adjust for this variation [ 9 , 10 , 11 , 12 ]. The estimation of these immune cell–type fractions is performed with a cell-type deconvolution algorithm that is in effect a multivariate least squares regression where the weights representing the cell-type proportions in the sample are constrained to lie between 0 and 1, with their summation also constrained to equal 1 (or a number less or equal than 1) [ 13 , 14 ]. The covariates in this constrained regression problem are DNAm reference profiles of the underlying cell types defined over cell-type specific marker CpGs, thus defining a DNAm reference panel [ 9 ]. Such DNAm reference panels are amply available for tissues like adult blood and are currently defined at the resolution of 12 immune cell types, including memory and naïve T-cell subsets [ 4 , 15 ]. Recent articles have also highlighted the importance of considering CTH when building and interpreting epigenetic clocks [ 16 , 17 , 18 , 19 ]. Hence, it is especially surprising how Skinner and Conboy do not acknowledge cell-type heterogeneity even once in their whole manuscript, i.e., no biological or statistical justification is given for ignoring CTH. Thus, here we aim to discuss their recent findings taking CTH into account and to demonstrate that conclusions can be dramatically altered when adjustment for CTH is made.
Source: GeroScience
Language: English
Open Access status: hybrid
Estimated APC paid: $2,890
OpenAlex ID: W4409920911
FWCI: 2.804
Cites: 71
Cited by: 14
Related to: 9