Mining old transcriptomes to predict HSC age

In this issue of Blood, Svendsen et al propose to the research community a new and robust transcriptomic signature of aging and provide an online resource to help assess and understand the aging of hematopoietic stem cells (HSCs).¹ 

What causes our hematopoietic system to age? Is aging of the hematopoietic system due to intrinsic changes in HSCs? To answer these burning questions about the immunosenescence of the elderly, many groups have used transcriptome profiling of HSCs from mice of different ages, a strategy with great potential. After almost 2 decades, we have discovered a plethora of age-associated genes and almost as many potential mechanisms to explain HSC aging.^2,3  However, due to the quantity and divergence of results, probably because of differences in experimental parameters (ages, aging conditions, mouse strains, and platforms), we still lack a complete understanding of HSC aging.^2,3 

In an attempt to reconciliate these studies, Svendsen et al took advantage of computational biology to revisit 16 transcriptomic datasets of aged HSCs that used a variety of designs and platforms, ranging from the first microarray study⁴  to their latest RNA-sequencing study.⁵  To overcome the complexity of interexperimental variations, the authors developed tools for calculating and flattening transcriptomic data, without losing key information. The strength of this strategy lies in using different approaches to validate the robustness of their age-related gene list and smooth out the platform-dependent bias of each platform. They succeeded in providing a list of ∼100 genes whose expression indicates aged HSCs, thus defining a unique and robust aging signature (AS) in mouse HSCs.

The benefits of developing a consensual AS are multifaceted. The AS can serve as an “aging fingerprint” in searching for an aging phenotype in any mutant HSCs. To assist in this, they provide a user-friendly online resource (https://agingsignature.webhosting.rug.nl/) that allows any biologist to easily query their transcriptomic signature in relation to aging. As an extension, by testing the robustness of their AS on “rejuvenated HSCs” treated with β adrenaline, the authors opened an interesting application of their resource. They demonstrated the possibility of using it to test the efficacy of rejuvenation strategies, such as bone marrow niche change or pharmacological intervention, which are under development.⁶ 

A limitation of this type of signature is that it reflects the total cell population and fails to capture individual cellular differences. Single-cell approaches have highlighted the heterogeneity of the HSC population, and we can expect that most future transcriptomic analyses will be performed at single-cell resolution.⁷  The authors were aware of this and, using machine learning algorithms, managed to extract from their global AS a shortened list of 20 genes that can predict whether an individual HSC is young or old. This age predictor list is unique and will be very important in assessing the aging heterogeneity of an HSC population after different treatments or stresses. Considering that this predictor indicates the biological age of a cell, it may be useful as a biomarker when evaluating the age of HSCs in relation to the clonal evolution of age-related myeloid diseases.

Beyond being a useful predictor of age, the AS is a goldmine for shedding light on functional changes in aged HSCs. Looking at gene enrichment, Svendsen et al pointed out that 20% of the AS genes encode membrane proteins. This implies that age-related changes impact the way an old HSC perceives its microenvironment.⁸  The authors functionally validated one of their top genes, the Selp gene, encoding P-selectin, a platelet surface marker. They showed that re-expressing this aging marker on the surface of young HSCs blocked their erythroid differentiation potential. Altogether, these data demonstrate that the AS identifies genes that regulate aging-related processes. Their list certainly contains other nuggets, including the stress gene Nupr1, an interesting HSC aging candidate that has recently been described as an HSC quiescence regulator.⁹ 

By reanalyzing all these transcriptomic studies, Svendsen et al also made the interesting and surprising observation that the RNA content is higher and RNA polymerase II more active in aged HSCs. Do aged HSCs increase their transcriptional rate to compensate for altered proteolysis? Is it associated with an epigenetic drift toward an opening of the chromatin that occurs with aging?¹⁰  The authors present convincing data in this direction. By assessing the relaxed state of chromatin as a function of age, they demonstrate that aged HSCs have many more accessible and, thus, open sites than young cells. It is therefore tempting to look for a correlation between the high RNA content and the relaxed chromatin state of aged HSCs. Although this certainly needs to be studied further, it may suggest that older cells are more transcriptionally active.

In an effort to reconcile years of transcriptomics, the authors have developed a powerful resource that will be very useful to the hematology community for not only understanding the mechanisms of aging but also assessing the biological age of our stem cells.