Transcriptional drift in aging cells: A global decontroller

Abstract

As cells age, they undergo a remarkable global change: In transcriptional drift, hundreds of genes become overexpressed while hundreds of others become underexpressed. Using archetype modeling and Gene Ontology analysis on data from aging Caenorhabditis elegans worms, we find that the up-regulated genes code for sensory proteins upstream of stress responses and down-regulated genes are growth- and metabolism-related. We observe similar trends within human fibroblasts, suggesting that this process is conserved in higher organisms. We propose a simple mechanistic model for how such global coordination of multiprotein expression levels may be achieved by the binding of a single factor that concentrates with age in C. elegans. A key implication is that a cell's own responses are part of its aging process, so unlike wear-and-tear processes, intervention might be able to modulate these effects.

Keywords: aging; archetype analysis; transcriptional drift.

Fig. 1.

The relative changes of the two archetypes of genes with age. Using normalized nonnegative matrix factorization, we identified two key archetypes: one that increases with age (red) and one that decreases (green).

Fig. 2.

Much of the transcriptome changes with age. (A) Pearson correlation coefficient of all genes with the increasing archetype. Genes with high positive R² are strongly monotonically increasing whereas genes with high negative R² are monotonically decreasing. Genes selected as archetype centers are to the Left and Right of the purple lines. (B) Rescaled expression data for genes with correlation coefficients ≥0.9 (Top) and ≤−0.9 (Bottom). The black lines represent the mean trajectory to better illustrate the shape of the curves.

Fig. 3.

The cumulative factor model captures the linear increase in up-regulated expressions and the Michaelis–Menten decrease in down-regulated expressions. Best-fit regressions to the data using the equations in the text. The average value of ${\text{PC}}_{\text{exp}}$ across all genes in each subset is plotted in blue. (A) models the signaling genes with an R² of 0.9999 and (B) models the growth genes with an R² of 0.9964.

Fig. 4.

Monotonically increasing and decreasing archetypes trajectories are present in human fibroblasts and can distinguish between increased aging mutations. HC cells are normal human fibroblasts, whereas SURF1 cells are fibroblasts with an accelerated aging mutation. A1 and A2 represent the two archetypes for each cell line. Our model was trained on technical replicates HC1 and HC2, then validated on technical replicate HC3 to show that similar cells show archetypes with similar trajectories. We then applied our model to SURF1 technical replicates 1 and 2 which produced archetypes that have lower/higher values at any given time point for the decreasing/increasing trajectories, suggesting that these cells have accelerated aging.

Fig. 5.

The transcriptome of human fibroblasts follow similar patterns to C. elegans with age. (A) Pearson correlation coefficient of all genes vs. the increasing archetypes. Genes with high positive R² are strongly monotonically increasing whereas genes with high negative R² are monotonically decreasing. Genes selected as archetype centers are to the Left and Right of the purple lines. (B) Rescaled expression data for genes with correlation coefficients ≥0.9 (Top) and ≤−0.9 (Bottom). Black lines show mean trajectory to better illustrate the shape of the curves.