New artificial intelligence tool identifies aging patterns in hematopoietic stem cells

Aging progressively affects the functioning of our body and, among other things, deteriorates the ability of the hematopoietic system- the set of organs and tissues responsible for producing blood cells- to maintain adequate blood cell production. Understanding and measuring this process is especially relevant to study how blood stem cells age and to identify strategies to preserve or recover their function.

The study, led by Dr Maria Carolina Florian, a researcher at the program of Regenerative Medicine at Bellvitge Biomedical Research Institute (IDIBELL) and ICREA Research Professor, and Dr Paula Petrone, a researcher at the Barcelona Supercomputing Center – Centro Nacional de Supercomputación (BSC-CNS), and the Barcelona Institute for Global Health (ISGlobal), a centre supported by the “la Caixa” Foundation, presents ChromAgeNet, an artificial intelligence-based tool that identifies aging-associated patterns in microscopy images of hematopoietic stem cells by analyzing the 3D organization of chromatin, the material made up mainly of DNA and proteins found in the cell nucleus that packages DNA and regulates which genes are active, ultimately determining cell identity and function. The work has been a central part of the doctoral thesis of Pablo Iañez, researcher at ISGlobal, and brings together expertise in stem cell biology, aging, image analysis and artificial intelligence. The results are published in Aging Cell, an international journal of reference in the field of aging research.

To develop the model, the researchers analyzed three-dimensional images of mouse hematopoietic stem cell nuclei, stained with DAPI, a simple and widely used technique for visualizing DNA. Using a convolutional neural network -a type of artificial intelligence model designed to analyze images-, ChromAgeNet learned to distinguish young cells from aged cells.

Based on the appearance of the nucleus, the model demonstrated a 77% probability of correctly distinguishing cells into two groups, outperforming even a machine learning model based on chromatin features previously defined by the researchers.

Detect what the eye does not see

One of the relevant aspects of the work is that the differences associated with aging are not necessarily visible to the naked eye in microscopic images. Stem cell aging is a heterogeneous process and changes in the architecture of the nucleus can be very subtle. ChromAgeNet allows, however, to identify combinations of spatial chromatin features that contain information about the aging state of cells.

In addition, the researchers analyzed the performance of the model to determine which specific features of the images are the most helpful in distinguishing young cells from aging cells. This analysis identified, among other factors, chromatin entropy, heterochromatin located at the periphery of the nucleus, and certain chromatin condensates as predictive features of age-associated state.

Knowing which elements of the DNA organization the model uses to make its predictions allows us to go beyond a simple classification between young and aged cells and provides information on the characteristics of the nuclear architecture associated with aging. This approach can complement other age biomarkers, such as the so-called epigenetic clocks, which estimate biological age based on certain chemical changes in DNA, including those related to methylation.

A new way to look for rejuvenation strategies

The team also explored the potential of ChromAgeNet as a screening tool for treatments capable of modifying age-associated characteristics. As a proof of concept, the model was applied to aged hematopoietic stem cells treated with different epigenetic drugs to assess whether these treatments produced changes in chromatin organization compatible with a younger state.

These results do not demonstrate that the treatments have functionally rejuvenated cells, but they do show the potential of ChromAgeNet as a tool to detect age-associated changes in response to different interventions and explore potential rejuvenation strategies.

The use of DAPI, a low-cost stain that is easy to incorporate into microscopy protocols, together with the small number of parameters in the model, facilitates its possible application in high-throughput microscopy workflows, systems that allow large quantities of samples to be analyzed. This could contribute in the future to study a large number of compounds and accelerate the identification of candidates for cell rejuvenation strategies.

As part of the work, the researchers have also made available to the scientific community a dataset of three-dimensional images of hematopoietic stem cells, together with ChromAgeNet. This resource can facilitate the development and validation of new computational tools to study the aging of these cells, an area for which there are currently few publicly available imaging datasets.

Overall, the study demonstrates that the three-dimensional organization of DNA contains quantifiable information about the aging of blood stem cells and that artificial intelligence can help extract this information from microscopic images. ChromAgeNet thus provides a new tool to study cellular aging and explore possible strategies to preserve or recover hematopoietic stem cell functions in a faster and more scalable way.