Differentiation hierarchy in adult B cell acute lymphoblastic leukemia at clonal resolution

Alec Geßner, Adrien Jolly, Tjeerd Sijmonsma, Yves Matthess, Manuel Kaulich, Stefan Günther, Fabian Lang, Florian Buettner, Michael A. RiegerbioRxiv 2025 · DOI

TL;DR

Researchers have finally mapped the "family tree" of adult B-cell acute lymphoblastic leukemia (B-ALL), proving that the disease follows a strict, one-way hierarchy rather than being a chaotic collection of plastic cells. By combining single-cell barcoding with mathematical modeling, the team identified a specific "source" population of leukemia-initiating cells and demonstrated that as these cells mature, they lose their ability to drive the cancer.

What Problem Does This Paper Address?

In the world of blood cancers, Acute Myeloid Leukemia (AML) is the textbook example of a hierarchy: it has "queen bee" stem cells at the top that produce all other cancer cells. However, for B-cell Acute Lymphoblastic Leukemia (B-ALL), the most common leukemia in children and a high-risk disease in adults, the architecture has been a subject of intense debate.

Some researchers argued B-ALL also has a hierarchy, while others suggested the cells are "plastic," meaning any cancer cell could potentially revert to a stem-like state and restart the disease. This isn't just an academic disagreement—it dictates how we treat the disease. If the cancer is plastic, you have to kill every single cell; if it’s hierarchical, you must prioritize the "source" cells at the apex to prevent relapse.

Methods

The authors employed a sophisticated "track and model" approach:

  • Lentiviral Barcoding: They used the ClonTracer library to tag individual B-ALL cells with unique genetic "barcodes." When a cell divides, its progeny inherits the same barcode, allowing researchers to track "clones" (families of cells) across different environments.
  • Multiome Single-Cell Analysis: They combined single-cell RNA sequencing (scRNA-seq) with AbSeq, which uses oligonucleotide-labeled antibodies to measure 49 different surface proteins simultaneously. This provided a high-resolution map of both the internal "state" and external "markers" of the cells.
  • Serial Xenotransplantation: Barcoded patient cells were injected into immune-compromised (NSG) mice. Cells were then harvested and re-injected into secondary mice to test their long-term "leukemogenic" (cancer-starting) potential.
  • Mathematical Modeling: Since you can't watch a cell differentiate in real-time inside a mouse, the team used Approximate Bayesian Computation (ABC) to test six different mathematical models of cell growth. They compared simulated data against their experimental results to see which model (one-way hierarchy vs. two-way plasticity) actually fit the reality.

Datasets

The study utilized primary B-ALL cells derived from two adult patients (referred to as CR and BV). These cells were expanded in patient-derived xenograft (PDX) mouse models. The resulting single-cell dataset included over 35,000 leukemic cells.

Key Insights and Ideas

  • The Unidirectional Hierarchy: The most significant finding is that B-ALL differentiation is a one-way street. The mathematical models for "one-way" differentiation were over 95% more likely to be correct than models involving plasticity.
  • The "Source" Cluster: The researchers identified "Cluster 1" (characterized by high MHC-class I expression) as the apex of the hierarchy. These cells are less differentiated and act as the hub that produces more "mature" progeny.
  • Clonal Behavior is "Baked In": Interestingly, identical clones (cells with the same barcode) showed nearly identical differentiation patterns even when transplanted into different mice. This suggests that a cell’s "fate"—whether it stays a stem cell or matures—is controlled by internal programs rather than just random external signals.
  • Predicting the Future: Using an XGBoost machine learning model, the authors found they could predict the future fate of a "source" cell's progeny just by looking at the gene expression of the parent cell.

Key Results

  • Functional Validation: When researchers sorted cells by the marker CD44, they found a massive difference in lethality. Mice injected with "mature" CD44+ cells had a median survival of 154 days, and 40% didn't develop leukemia at all. In contrast, mice injected with the more "primitive" CD44- cells had a median survival of only 75 days.
  • Growth vs. Initiation: While mature cells were much worse at starting the leukemia, once the disease was established, they grew at roughly the same rate as the primitive cells. This proves that "leukemia initiation" and "cell proliferation" are two distinct biological processes.
  • Marker Identification: The study identified specific markers like MSI2 (associated with self-renewal) in the source cells and EBF1 (a driver of B-cell maturation) in the differentiated progeny.

Limitations and Future Work

The authors acknowledge a few key hurdles:

  1. Patient Diversity: The study focused deeply on two patient samples. Given that B-ALL is a highly diverse disease, these findings need to be validated across more genetic subtypes.
  2. The CD44 Caveat: While CD44 worked as a marker for these specific samples, the authors caution that it might not be a universal "discrimination marker" for all B-ALL cases.
  3. The "Why": While the study proves the hierarchy exists and is cell-intrinsic, it doesn't fully explain what (epigenetic marks or specific mutations) causes one clone to be "self-renewing" while another is "differentiation-biased."

This research moves the needle for B-ALL therapy, suggesting that future treatments should focus on the molecular signatures of the "source" cells to ensure that the roots of the cancer are truly eradicated.

Summary updated

Abstract

Abstract While a differentiation hierarchy with leukemia-initiating stem cells (LICs) at the apex is well documented for acute myeloid leukemia, the existence of LICs and their trajectories in B cell acute lymphoblastic leukemia (B-ALL) are debated. B-ALL is a malignant disease displaying considerable phenotypical and functional heterogeneity, yet the underlying cellular organization remains largely elusive. This study aims to investigate the hierarchical landscape of B-ALL by combining barcoding and multiome single cell data to unveil the differentiation architecture and temporal dynamics of leukemic differentiation at clonal and single cell resolution in vivo . Single-cell transcriptome and AbSeq analysis of lentiviral barcoded and transplanted patient-derived B-ALL cells revealed interconnected subpopulations, which could be prospectively isolated via surface markers and exhibited distinct leukemogenicity. Barcode tracking demonstrated the ability of cells to differentiate from a source cluster into differentiated progeny. Using machine learning we identified expression patterns predicting their differentiation potential of each barcoded cell within the immature cell compartment. To determine the dynamical properties of clones, we have designed a mathematical model that simulates the development of these subpopulations from a clonally diverse stem cell compartment. The model confirmed the most likely cluster of origin, supporting our functional data, and revealed the variability in dynamical properties between clones, while largely excluding plasticity. Hence, our work demonstrates a unidirectional differentiation in B-ALL with an immature population exhibiting a high leukemogenic potential. Key points A differentiation hierarchy with distinct leukemogenesis potential in B-ALL is demonstrated via clonal tracing and mathematical modeling. Clonal differentiation behavior, while heterogeneous, is cell intrinsically controlled and predictable, yet independent from expansion.
BibTeX
@misc{gener2025differentiation,
  author = {Alec Geßner and Adrien Jolly and Tjeerd Sijmonsma and Yves Matthess and Manuel Kaulich and Stefan Günther and Fabian Lang and Florian Buettner and Michael A. Rieger},
  title = {{Differentiation hierarchy in adult B cell acute lymphoblastic leukemia at clonal resolution}},
  year = {2025},
  doi = {10.1101/2025.09.10.674954},
  howpublished = {bioRxiv},
}
Figure