Epigenetic quantification of immunosenescent CD8+ TEMRA cells in human blood

Abstract

Age-related changes in human T-cell populations are important contributors to immunosenescence. In particular, terminally differentiated CD8⁺ effector memory CD45RA⁺ TEMRA cells and their subsets have characteristics of cellular senescence, accumulate in older individuals, and are increased in age-related chronic inflammatory diseases. In a detailed T-cell profiling among individuals over 65 years of age, we found a high interindividual variation among CD8⁺ TEMRA populations. CD8⁺ TEMRA proportions correlated positively with cytomegalovirus (CMV) antibody levels, however, not with the chronological age. In the analysis of over 90 inflammation proteins, we identified plasma TRANCE/RANKL levels to associate with several differentiated T-cell populations, including CD8⁺ TEMRA and its CD28^- subsets. Given the strong potential of CD8⁺ TEMRA cells as a biomarker for immunosenescence, we used deep-amplicon bisulfite sequencing to match their frequencies in flow cytometry with CpG site methylation levels and developed a computational model to predict CD8⁺ TEMRA cell proportions from whole blood genomic DNA. Our findings confirm the association of CD8⁺ TEMRA and its subsets with CMV infection and provide a novel tool for their high throughput epigenetic quantification as a biomarker of immunosenescence.

Keywords: CD8+ T-cells; CMV; biomarkers; epigenetics; human aging; inflammation.

FIGURE 1

Increased proportions and high interindividual variability of CD4⁺ and CD8⁺ T‐cell subsets. (a) Schematic picture of studied CD4⁺ and CD8⁺ T‐cell populations. (b–e) Relative sizes of CD4⁺ and CD8⁺ T‐cell subsets among CD4⁺ (b) and CD8⁺ (c) compartments and among whole blood cells (WBC) (d for CD4⁺ subsets, e for CD8⁺ subsets). Red point shows the mean and adjacent line a standard deviation. The color bar shows signal‐to‐noise ratio (SNR) calculated as mean/SD with brighter color denoting higher SNR value (in brackets). In addition, mean and standard deviation are written next to each measurement. The two heatmaps (f and g) are based on correlation matrices that contain pairwise Pearson's correlation coefficients in CD4⁺ and CD8⁺ T‐cell subsets, respectively

FIGURE 2

T‐cell subset dynamics in old individuals. The dynamics of CD4⁺ (a) and CD8⁺ (b) T‐cell subset sizes in old age (≥65 years), respectively, via moving average. (c) Scatterplots of CD4⁺ and CD8⁺ T‐cell subset changes in old age. Scatterplots report Pearson's correlation coefficient and an adjusted p‐value

FIGURE 3

CMV‐specific antibody level correlations with T‐cell subsets in old individuals. (a) Age‐gender distribution of CMV positive and negative individuals. (b) The levels of anti‐p150d1 and p50d2 antibodies in CMV positive and negative individuals are shown as luminescence units (LU) of luciferase enzyme activity as boxplots. (c) ROC curve for the p150d1 and p150d2 fragments’ LIPS analysis shows the classification performance by dividing individuals into CMV positives and negatives. (d) Correlation between antibody levels to p150d1 and p150d2 fragments in LIPS measurements. (e) Correlation between p150d1 specific LIPS results and T‐cell subset proportions in flow cytometry shown together with and without age‐adjusted Pearson's correlation coefficient and adjusted p‐values

FIGURE 4

Plasma inflammation markers correlations with T‐cell subsets in old individuals. (a) Correlation between CD4⁺ and CD8⁺ T‐cell subset proportions and plasma inflammation markers measured by proximity extension profiling and shown as a clustered heatmap. (b) Top 10 correlations of TRANCE with the proportions of CD4⁺ and CD8⁺ T‐cell subsets. The inflammatory protein levels are shown as normalized protein expression (NPX) values, a metric that is on a log₂ scale and where a higher value indicates a higher protein level. The Pearson's correlation coefficient and adjusted p‐value for each correlation are shown

FIGURE 5

CD8⁺ TEMRA associations with methylations levels of selected CpG sites. (a) The correlations between methylation levels of CpG sites that were incorporated into the prediction model and CD8⁺ TEMRA cell proportions in WBC. (b) PCA calculated on the methylation levels of those 7 CpG sites in (a) and colored according to the level of percentages of CD8⁺ TEMRA/WBC. (c) The accuracy of the final model in red together with predictions of models that were built on resampled training dataset using linear (light gray) and ridge (dark gray) regression models. Those illustrate the variability caused by selecting different training and test set