The work sits off on its own: there is almost nobody nearby, the closest neighbor at a distance of 0.59. Around it lie studies of aging through methylation, cellular landscapes, and loss of coordination in gene networks. Most neighbors examine either the composition of cell populations or coordinated shifts in expression, but they do not try to separate a loss of precision in executing the program from a change in the program itself. This paper fills that niche: it offers a quantitative measure of disarray specifically, not of state change.
For each cell, the authors assess how far its expression of every gene deviates from a predicted value, which is computed from the rest of the transcriptome using a regularized model. Technical sequencing noise is accounted for through a UMI sampling model from the SAVER method. A cell's dyscoordination is the mean squared deviation across genes, normalized to the expected variation. The method requires no clustering or predefined markers. Validation was performed on simulations with an added coordinated program or independent noise, on allele-specific data from hybrid mice, on cells after irradiation and induced aging, and on atlases of bone marrow, T cells, rat and human kidney, and mouse liver.
In simulations, coordinated changes in expression barely increased dyscoordination, whereas independent noise did. Allele-specific analysis revealed a positive correlation between dyscoordination and the classical measure of intrinsic transcriptional noise, and the link grew stronger as the genetic difference between haplotypes increased. Irradiating cells with a dose of 12 Gy raised dyscoordination, while clearing senescent cells with a senolytic lowered it. In mouse and human bone marrow, dyscoordination rose monotonically with age in hematopoietic stem and progenitor cells (ρ = 0.397, p < 2.2×10⁻¹⁶), but not in terminally differentiated cells. In T cells, dyscoordination was higher in large clones, in cytotoxic states, and in high-grade tumors, and it fell in persistent clones after immunotherapy. In rat and human kidney, dyscoordination increased with age, more strongly in the S2 and S3 segments of proximal tubules in rats, and across all tubular epithelia in humans. In mouse liver, dyscoordination rose in hepatocytes, especially in zones 2 and 3, and correlated with mitotic age according to EpiTrace (r = 0.31, p < 2.2×10⁻¹⁶). Genes that lose expression precision are enriched for genome maintenance and proteostasis pathways, while genes activated in highly dyscoordinated cells are enriched for inflammatory and fibrotic programs.
Unlike neighboring studies that rely on aging markers or clustering, this paper introduces a marker-free measure that applies to any cell type without retraining. It is the first to systematically separate loss of coordination from coordinated reprogramming, and it shows that these processes can move in opposite directions. The work fits into the broader current of cellular aging research, but it offers a new tool that can be layered onto existing atlases.
The method could be extended to other models of technical noise and other types of omics data, such as spatial transcriptomics or multi-omic profiles. It would be interesting to test whether dyscoordination can serve as an early marker of dysfunction in preclinical studies or as a biomarker of therapy response. It also remains to be determined whether elevated dyscoordination is a cause or a consequence of declining regenerative potential.
The measure is applicable in gerontology, immunology, oncology, and regenerative medicine for mapping vulnerable cell populations. It could be used to assess the effect of anti-aging interventions and to stratify cells in atlases.
Check reproducibility across a broader set of tissues and species. Investigate the causal relationship between dyscoordination and functional decline using genetic or pharmacological manipulations.
In aging biology, it remains an open question how molecular damage turns into functional decline of tissues. This work links the accumulation of stochastic expression disturbances to loss of cellular identity and activation of inflammatory programs, which speaks directly to the problem of aging heterogeneity and the search for reversible targets.
In mouse hematopoietic stem cells, dyscoordination rises with age almost linearly: in 30-month-old mice it is on average 40% higher than in 1-month-old mice. Yet terminally differentiated monocytes show no such increase, even though they age too.
In Greg Egan's novel Permutation City, people digitize their consciousness but run into the problem of copy degradation from accumulated random errors. A similar idea underlies this work: cellular aging can be seen as a gradual loss of fidelity in reproducing the program, not just as a change in the program itself.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.