Math models expose myeloid bias mechanisms in hematopoiesis

In this issue of Blood, by integrating mathematical modeling and experimental approaches, Singh et al show how myeloid bias arises from a combination of hematopoietic stem cell (HSC) differentiation and progenitor proliferation biases.¹ Through an elegant synthesis of mouse data at both the cellular and expression levels with mechanistic mathematical models, the authors quantify and compare specific regulatory processes that drive shifts in lineage contributions during hematopoiesis, particularly in the context of inflammation or aging.

Blood homeostasis and stress response originate from the tight regulation of multipotent HSCs, which include both long-term and short-term HSCs. Although long-term HSCs sustain self-renewal and are capable of full hematopoietic reconstitution, short-term HSCs preferentially differentiate into 3 multipotent progenitors (MPPs) subtypes with distinct lineage potential, namely erythroid/megakaryocyte, myeloid, and lymphoid MPPs.² Inflammatory conditions and aging are associated with alterations in hematopoiesis, particularly a “myeloid bias,” characterized by increased myeloid and reduced lymphoid cells.^3,4 Although MPPs are downstream of HSCs, they play a critical role in maintaining hematopoiesis both under normal conditions and during stress.⁵ However, the quantitative and mechanistic details of their contribution to myeloid bias remain to be fully elucidated.

In this study, the authors investigated whether changes in the differentiation rates of short-term HSCs alone can account for myeloid bias in MPPs and explored additional mechanisms that might contribute. To this end the authors used a chronic inflammation mouse model (IκB^–), which is characterized by increased myeloid-biased MPPs. To interpret quantitative changes in short-term HSC and MPP numbers between IκB^– mice and wild-type (WT) controls, the authors developed a mathematical model of hematopoiesis, which contained a set of 5 ordinary differential equations (ODEs) inspired by previous models.^6,7 The model explicitly distinguished between distinct differentiation processes toward MPP subtypes and their specific net proliferation. Their findings suggest that individual parameters between the MPP subtypes are necessary to describe the observed myeloid bias in IκB^– mice. Specifically, the authors conclude that the myeloid bias arises from a combination of 2 factors: an increased differentiation rate from short-term HSCs to myeloid-biased MPPs and increased difference between net proliferation and differentiation of these MPPs. Although the specific contribution of each factor cannot be conclusively determined, their model demonstrated that the combination of both factors best explains the transition from WT to IκB^– parameters.

To further investigate these mechanisms driving myeloid bias in the IκB^– phenotype, Singh et al used single-cell RNA sequencing (scRNA-seq) of hematopoietic stem and progenitor cells from both WT and IκB^– mice. By employing the pseudotime algorithm Slingshot, they reconstructed continuous developmental pathways and quantified HSC and progenitor densities along the differentiation continuum. Building upon their initial ODE model, they advanced to a partial differential equation (PDE) framework, which was calibrated against the scRNA-seq data and produced a detailed description of cell densities across the differentiation landscape.^8,9 The model suggests that myeloid bias in IκB^– mice is driven by increased net proliferation of early myeloid-primed progenitors, although decreased differentiation cannot be ruled out at this point. Subsequently, the authors showed that increased differential gene expression was found for proliferative markers in early myeloid progenitors in IκB^– mice, thereby confirming the major role of increased net proliferation in myeloid bias.

To substantiate these findings in a broader context, the authors applied their mathematical models to additional datasets from aged mice and patients with myeloid neoplasms, both of which exhibited elevated NF-κB activity. By integrating these models with experimental data, they quantified changes in differentiation flux and proliferation dynamics, confirming enhanced proliferation in myeloid-biased progenitors. These results support the notion of a conserved mechanism where inflammatory signaling and elevated NF-κB activity drive myeloid bias through dysregulated HSC dynamics and increased expansion of myeloid progenitors.

From a methodological perspective, the article exemplifies the rigorous, iterative, data-driven modeling approach needed to quantify experimental results. Although the models can incorporate different, potentially competing hypotheses, the experiments were used to distinguish between them. The ODE and PDE models not only offer a robust quantitative framework and a theoretical basis for analyzing dynamical data on hematopoiesis under various experimental conditions, but also highlight the potential of mathematical and computational models to provide a more conceptual and mechanistic interpretation of the underlying biology. In particular, the PDE model used to untangle and quantify differentiation on the scRNA level is a fine example of how mathematical models can even integrate high-dimensional datasets, making them accessible for a mechanistic interpretation. The further acceptance of such studies relies on transparent communication of model limitations and the robustness of the methodology used for adapting the model to experimental data. Clearly explaining these aspects ensures that readers can critically assess the model’s reliability and applicability to the particular biological questions.

Twenty years ago, Cohen suggested that “mathematics is biology’s next microscope, only better.”¹⁰ In the age of big data and artificial intelligence, one might question the relevance of mechanistic mathematical modeling. Is mathematics still a microscope for biology? The work of Singh et al provides an emphatic yes, demonstrating how mathematical models not only deliver critical biological insights but also structure data into interpretable frameworks and uncover underlying mechanisms. In the rapidly evolving world of biology, the question is not whether mathematics is still useful but rather which mathematics will be needed to address the challenges of the future. For now, this microscope is far from obsolete and will gain even more importance.

Conflict-of-interest disclosure: The authors declare no competing financial interests.