Cell Division History Determines Hematopoietic Stem Cell Potency

ABSTRACT Changes in the ability of individual stem cells to self-renew based on their developmental history may underpin aging. Here, we developed a high-throughput paired daughter cell assay that combines single cell profiling with machine learning and mathematical modeling to probe in detail how multivariate gene expression patterns change through hematopoietic stem cell (HSC) divisions. We first collated a library of single cell gene expression patterns from various different cell types in the hematopoietic hierarchy taken from young, adult and aged mice. We then trained an artificial neural network (ANN) to accurately predict both cellular identity and developmental age directly from gene expression profiles. Once this classifier had been trained we used it to investigate HSC division patterns. To do so, we isolated highly purified long term repopulating HSCs, allowed them to divide in various different culture conditions, separated daughter cells using a micro-manipulator, immediately profiled their gene expression patterns, and used our trained ANN to compare daughter cells identities. We found that the propensity of individual HSCs to divide symmetrically or asymmetrically changed dramatically with age, with self-renewal ability starting high but sharply declining with age. Furthermore, we observed that while HSC cell divisions were essentially deterministic in young and old age and are primarily regulated by cell-intrinsic factors, they become highly stochastic and sensitive to niche instruction in mid-life. Analysis of evolving division patterns indicated that the propensity of an individual HSC to self-renew depends upon the number of times it has divided previously, with self renewal ability lost after four divisions. To explain these results we propose a mathematical model of proliferation that naturally gives rise to an evolving, age-structured stem cell population that is able to support life-long hematopoiesis without proliferative exhaustion. This model accurately predicts how stem cell numbers increase with age, and explains why regenerative potency simultaneously declines. By accounting for both intrinsic stochasticity and niche direction, it also reconciles the instructive and stochastic views of stem cell fate.