Identification of hematopoietic stem cells gene expression signature using systems biology and machine learning approaches

Abstract

Hematopoietic stem cells with long-term engraftment potential (long-term hematopoietic stem cells - LT-HSC) are the only cells capable of reconstituting the bone marrow, generating all the blood cell lineages, therefore having the major role in a bone marrow transplant context. LT-HSCs are rare cells and, given that the success of a transplant depends on the amount of LT-HSC transplanted, their expansion in vitro has always been a goal in Medicine. However, reproducing conditions that allow these cells to expand to desirable levels without losing their potential remains a challenge. In attempts to optimize the expansion protocols, it is therefore necessary to evaluate both the number of LT-HSCs and their functionality, another important challenge. Although some surface markers of cultured LT-HSC have been recently reported, since they are molecules susceptible to environmental variations, they may lose their validity depending on the cultivation condition, making it necessary to evaluate the maintenance of the functionality of these cells by xenotransplantation assays at each change in the cultivation protocol. Thus, the hypothesis of this work is that the analysis of gene co-expression networks, a systems biology approach, could contribute to the identification of a set of more robust LT-HSC markers, not variable with the cultivation or manipulation situation. Our objective is to use this approach, coupled with machine learning analyses, to verify if the expression of hub genes of co-expression modules associated with LT-HSC would reflect the relative amounts of these cells under different conditions. For this, we will analyze publicly available LT-HSC transcriptome data, as well as transcriptome and proteomics data generated from fresh and cultured cells in our laboratory to select a set of core genes of the modules found as associated with LT-HSC. The expression levels of these genes will be used to train machine learning algorithms in order to detect the minimum set of genes and the algorithm that will generate a score with the best predictive accuracy of enrichment of the cells of interest. This model will then be validated by in silico analysis of public data, by gene expression analysis (by quantitative PCR) of samples grown in conditions known to have higher or lower abundancy of LT-HSC and of selected test conditions and, finally, by xenotransplantation assays. Additionally to the contribution to the understanding of the cell biology of LT-HSC, our results may allow the design of a quantification strategy for the enrichment of these cells that would have important application in clinical contexts. (AU)