Aging on a schedule: a mouse cell atlas argues for windows, not doses

Junyue Cao, who heads the Laboratory of Single-Cell Genomics and Population Dynamics at Rockefeller University, told Quanta Magazine that aging is not the slow failure of a repair system but a stepwise, programmed and orderly process, with discrete stages resembling those of embryonic development and defined by changes in molecular signals and specific cell populations ^[1]. That distinction has an operational edge: a sequence of stages implies windows, while wear and tear implies a dose you start early and never stop.

The standard account, which Cao says is inaccurate, holds that aging results from the degradation of molecules, organelles, cells and organs, from outside assault or inexorable breakdown, and becomes visible when repair cannot keep pace ^[2]. "The destruction of the system is programmed at a very early stage," Cao said ^[3].

The evidence he points to is an atlas. Cao and his team processed 21 million cells sampled from 14 tissues or organs in about 50 male and female mice at five life stages, and built a gene expression data set for each cell ^[4]. Each stage, he says, was marked by a dramatic decline in or expansion of specific cell types ^[5]. Two papers from the lab, published in Science in 2025 and 2026, report a redistribution of the cells that make up the body as mammals age and describe some of the epigenomic instructions guiding that process ^[6]. "There are molecular changes and maybe some other changes in aging," Cao said, "but they all converge in the remodeling of the cell society" ^[7].

The cell count is the headline number and the least interesting one. Twenty-one million cells across roughly 50 animals is about 420,000 cells per mouse ^[8], but the unit of biological replication is the mouse, not the cell, and 50 animals spread across five life stages and two sexes leaves on the order of five animals per sex per stage ^[9]. Stage boundaries drawn at that scale are hypotheses about when transitions happen, not calibrated dates.

Cao's route to this position ran through the alternative. As an undergraduate at Peking University he joined a computational drug design lab building peptides against molecules related to aging, and found the obstacle was not the chemistry but not knowing which molecules to target ^[10]. He then worked at the Jackson Laboratory as a research assistant, studying molecular pathways associated with aging in mice with the aim of finding a drug target ^[11]. Working on proteins after college, he found a daunting number of them associated with aging, with effects that depended on the cell type in which they operated ^[12]. He built high-throughput single-cell technology as a graduate student to quantify embryonic development, and turned it on aging when he opened his own lab in 2020 ^[13].

The claim with the most consequence outside a mouse room is that in humans the process likely begins before age 30 ^[14]. Read as staging rather than decay, that is not a warning about early damage; it says the first transition is finished before most people would ever be prescribed anything.

Quanta frames the interview around evidence for the programmed theory, what happens at each stage, and why aging mammals resemble trees shedding leaves ^[15]. The timing consequence follows from the model, not from any protocol Cao is quoted here as proposing. Three things to watch: whether independent labs recover the same stage boundaries in the same tissues, at animal counts that can support dates; whether the epigenomic instructions in the Science papers behave like instructions when perturbed rather than like correlates; and whether anyone runs the experiment the model demands, a single intervention delivered at one stage boundary, withheld at another, and scored against continuous dosing. Until that comparison exists, "programmed" is one investigator's reading of an atlas, and the atlas is the durable part.