🛡️ immunology

A computational investigation of pathways for ex vivo differentiation and dysfunction of healthy and patient T cells

This study employs a large-scale computational framework to analyze thousands of potential T cell differentiation pathways, revealing that cancer likely alters differentiation kinetics rather than pathway structure and that T cells dynamically switch between functional phenotypes during ex vivo expansion, thereby offering new insights to optimize adoptive cell therapy.

Original authors: Lacy, M. S., Binenbaum, Y., Buenzli, P. R., Jenner, A. L.

Published 2026-10-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lacy, M. S., Binenbaum, Y., Buenzli, P. R., Jenner, A. L.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The human immune system relies on a specialized army of white blood cells called T cells to hunt down and destroy cancer. In a promising medical approach known as adoptive cell therapy, doctors take these cells from a patient, grow them in vast numbers outside the body, and then reinfuse them to fight the disease. For this treatment to work, scientists must carefully manage how these cells grow in the lab. They need the cells to multiply into powerful attackers, but they must avoid a trap: if the cells are stimulated too much or for too long, they can become exhausted. An exhausted T cell is like a soldier who has fought so hard it has given up; it loses its ability to kill cancer and can no longer reproduce. The challenge for researchers has been understanding exactly how T cells decide their fate during this growth process. Do they follow a strict, one-way path from a young, initial state to a powerful attacker, and then to exhaustion? Or do they move back and forth between states? The answer is crucial because if we know the rules of this journey, we might be able to steer the cells toward becoming effective cancer killers rather than exhausted failures.

A team of researchers in Australia and the United States decided to solve this puzzle not by running more biological experiments, but by building a massive computational investigation. They recognized that the existing scientific literature was full of conflicting ideas about how T cells change. Some studies suggested a linear path, while others proposed branching routes or the ability to switch directions. To cut through this confusion, the team created a digital framework to test thousands of different possibilities at once. They started by gathering every known rule about T cell behavior from biology papers. These rules covered everything from how cells divide to how they might become exhausted. By mixing and matching these rules, they constructed nearly 8,000 unique mathematical models, each representing a different potential story of how T cells might behave in a culture dish.

The researchers then put these 8,000 stories to the test against real-world data. They used experimental results from a previous study where T cells from both healthy people and cancer patients were grown in the lab for eight days under different levels of stimulation. In these experiments, the cells were exposed to varying amounts of activating molecules attached to tiny rods. The team fed their computer models the starting conditions of these experiments and asked the models to predict what the cell populations would look like after eight days. They compared the computer predictions to the actual counts of cells observed in the lab. Most of the 8,000 models failed to match the reality of the experiment, producing results that were wildly off. However, a small group of models, fewer than 600, managed to accurately recreate the growth patterns seen in both healthy and patient cells.

The analysis of these successful models revealed a surprising truth about how T cells behave in the lab. The most accurate models did not show the cells moving in a single, straight line from one type to another. Instead, the data strongly suggested that T cells are highly flexible, constantly switching back and forth between being memory-like cells, which are long-lived and ready to reproduce, and effector-like cells, which are the immediate attackers. This bidirectional movement appears to be a key feature of how the population expands. The models also indicated that the cells that become exhausted likely do so because the effector-like cells turn on specific inhibitory receptors, essentially hitting the brakes on their own function. Interestingly, the best-fitting models did not require the effector-like cells to be the ones doing the most multiplying; instead, the memory-like cells seemed to be the primary engine driving the population growth, replenishing the other types as needed.

When the researchers compared the models fitted to healthy cells against those fitted to cancer patient cells, they found that the underlying structure of the journey remained the same. The path the cells took—switching back and forth between states—did not change based on whether the cells came from a healthy person or someone with cancer. What did change was the speed and sensitivity of the process. The models suggested that T cells from healthy individuals tend to multiply faster and require a different amount of stimulation to change their state compared to cells from patients. This implies that the disease might alter the pace of the cells' development rather than rewriting the rules of the road entirely. The study also found that the cells' decision to become exhausted was linked to the level of stimulation they received, with higher levels of activation making it more likely for them to upregulate the inhibitory markers that signal dysfunction.

This work provides a new way to think about the complex biology of immune cells. By testing thousands of potential pathways against real data, the researchers were able to rule out many long-held assumptions, such as the idea that T cells follow a rigid, one-way path to exhaustion. They have highlighted that the relationship between memory and effector cells is dynamic and fluid, not static. While the study relies on computer simulations and does not yet prove these mechanisms in a living body, it offers a set of clear, testable hypotheses for future scientists. It suggests that to improve cancer therapies, researchers should focus on maintaining this flexibility in the cells and understanding how to control the stimulation levels to keep the cells in a productive state. The findings offer a roadmap for optimizing how we grow these life-saving cells, potentially leading to better treatments that keep the immune system's soldiers sharp and ready for battle.

Technical Summary: Computational Investigation of T Cell Differentiation Pathways

Problem Statement
Adoptive T cell therapy relies on the ex vivo expansion of functional T cells to treat cancer. A critical challenge in this process is preventing the differentiation of these cells into exhausted (dysfunctional) phenotypes while maintaining memory-like and effector-like populations. However, the biological literature presents significant uncertainty and conflicting hypotheses regarding the structure of T cell differentiation pathways. It remains unclear whether T cells follow linear, branching, or bidirectional trajectories between naive, memory-like, effector-like, and dysfunctional states, and whether these pathways differ between healthy donors and cancer patients. This lack of a unified framework hinders the optimization of culture conditions for therapeutic production.

Methodology
The authors developed a large-scale computational hypothesis-testing framework to evaluate thousands of potential differentiation pathways against experimental data.

Data Source: The study utilized ex vivo expansion data from Zhang et al. [30], involving CAR T cells from healthy individuals and patients with acute lymphoblastic leukemia and chronic lymphocytic lymphoma. The data included cell counts for naive, memory-like, effector-like, and dysfunction-associated (PD-1/TIM-3 co-expressing) subpopulations after 8 days of culture under varying densities of polyclonal stimulation (anti-CD3/anti-CD28).
Model Construction: The authors constructed 7,840 unique mathematical models based on a literature review of 20+ differentiation features (e.g., linear vs. branching differentiation, bidirectional switching, stimulus dependence, proliferation rates). These features were encoded as binary selections (on/off) or parameter variations within a system of linear ordinary differential equations (ODEs).
- The model compartments included Naive ( $N$ ), Memory-like ( $M$ ), Effector-like ( $E$ ), and Dysfunction-associated ( $D$ ) cells.
- Differentiation rates were modeled as functions of stimulus density ( $s$ ), utilizing Michaelis-Menten kinetics where appropriate.
Model Fitting and Selection: Each of the 7,840 models was fitted to the experimental data using the Broyden–Fletcher–Goldfarb–Shanno (BFGS) optimization algorithm to minimize a loss function (sum of squared errors) for both healthy and patient datasets.
Acceptance Criteria: Models were "accepted" if their predictions for Day 8 cell counts fell within one standard deviation of the experimental mean for all subpopulations (excluding naive cells) and all stimulus densities.
Statistical Analysis: To identify dominant biological features, the authors calculated Yule's Q coefficient of association. This metric quantified the likelihood of a model being accepted given the presence or absence of a specific feature, allowing the authors to distinguish between features that are strongly supported by the data versus those that are neutral or unsupported.

Key Contributions and Results

Pathway Structure vs. Kinetics: The study found a strong correlation (Pearson $r = 0.896$ ) between the model fits for healthy and patient data. This suggests that cancer alters the kinetics (rates of differentiation and stimulus sensitivity) of T cell differentiation rather than the fundamental structure of the pathway. Healthy and patient-derived T cells appear to follow similar topological pathways, differing primarily in parameter values (e.g., proliferation rates and stimulus thresholds).
Bidirectional Differentiation: The most accurate models (lowest loss) overwhelmingly included bidirectional differentiation between memory-like and effector-like compartments (Feature 1c). This challenges the traditional view of a strictly unidirectional linear progression. The authors hypothesize that T cells dynamically switch between these functional phenotypes ex vivo, allowing memory-like cells to replenish effector pools.
Effector-Like Dynamics:
- Proliferation: The best-fitting models did not require explicit proliferation of effector-like cells (Feature 3b off) to match the data, suggesting their expansion is driven by transitions from other compartments.
- Dysfunction: Models indicated that effector-like cells are the primary source of dysfunction-associated cells (Feature 2c on), upregulating inhibitory receptors.
- Reversibility: The data did not support the back-differentiation of dysfunction-associated cells into effector-like states (Feature 1g off), suggesting that once cells co-express PD-1 and TIM-3 in this context, they do not revert.
Stimulus Dependence: The analysis suggested that differentiation to effector-like states is largely stimulus-independent (once initiated), whereas differentiation to memory-like states and the upregulation of inhibitory receptors are stimulus-dependent.
Parameter Differences: Healthy donor T cells were modeled to proliferate more rapidly than patient-derived cells. Furthermore, healthy cells appeared to require higher stimulus levels to differentiate into memory-like states but lower levels to begin upregulating inhibitory receptors compared to patient cells.

Significance and Claims
The paper claims to provide a systematic, data-driven method to resolve conflicting biological hypotheses by testing thousands of pathway variations simultaneously. Its primary significance lies in:

Generating Testable Hypotheses: The framework identifies specific biological mechanisms (e.g., bidirectional switching, lack of effector proliferation) that are most consistent with current ex vivo expansion data, offering clear targets for future experimental validation via lineage tracing or longitudinal phenotyping.
Informing Therapeutic Strategies: By suggesting that cancer alters differentiation kinetics rather than pathway structure, the study implies that optimizing culture conditions (stimulus density, duration) could potentially normalize patient T cell behavior to match healthy donors.
Prioritizing Research: The analysis highlights specific areas of uncertainty (e.g., the role of asymmetric division, the exact nature of branching from naive cells) where future experimental data would most effectively reduce model ambiguity.

The authors modestly note that their conclusions are limited by the lack of time-series data (only Day 0 and Day 8) and the absence of dead cell counts, which prevents distinguishing between differentiation, differential proliferation, and survival. They emphasize that their findings are specific to short-term ex vivo expansion under polyclonal stimulation and IL-2 supplementation.

Technical Summary: Computational Investigation of T Cell Differentiation Pathways

More like this