📄 cancer biology

How optimal control of cellular cost shapes population-level tumor growth dynamics

This paper employs a continuous-time Markov decision framework to demonstrate that the specific structure of objective functions—such as the type of cost penalties and reward constraints—fundamentally dictates whether a tumor population exhibits regulated, dormant-like dynamics or sustained, uncontrolled growth.

Original authors: Shrestha, P., George, J. T.

Published 2026-09-18

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Shrestha, P., George, J. T.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

In the crowded landscape of modern biology, cancer is often viewed as a runaway train, a chaotic collection of cells growing without restraint until they overwhelm the body. Yet, tumors are not merely passive victims of treatment or immune attacks; they are dynamic populations that constantly adapt to their surroundings. Just as a forest fire might slow down when it runs out of fuel or a bacterial colony might stop expanding when space runs out, cancer cells possess a hidden capacity to regulate their own numbers. They can sense stress, such as a lack of nutrients or the presence of immune cells, and adjust their behavior to survive. This ability to shift between aggressive growth and a quiet, dormant state is a key feature of how tumors persist. The question that has long puzzled scientists is not just whether tumors can adapt, but how they decide when to grow and when to stop. Is this regulation a simple reaction to external pressure, or does it stem from a complex internal calculation of costs and benefits?

A team of researchers at Texas A&M University and MD Anderson Cancer Center has explored this question by treating a tumor not as a chaotic mess, but as a population making strategic decisions. They built a mathematical model that imagines the tumor as a single, unified group of cells constantly weighing the benefits of multiplying against the risks of dying. In their view, every time a cell divides, it risks triggering a death signal, and every time it tries to hide from the immune system, it pays a metabolic price. The researchers used a framework known as a continuous-time Markov decision process, a method that allows them to simulate how a population might change its behavior over time to maximize its survival. They did not look at individual cells or specific genes; instead, they focused on the big picture, asking how the overall rules of the game—how much it costs to grow, how strong the external pressure is, and how the tumor values the future—shape the final outcome.

The study tested several different scenarios, or "rules," to see which ones would lead to a tumor that grows forever, one that settles into a stable size, and one that goes dormant. In the first scenario, the researchers imagined a tumor that faces a specific threshold: once it grows large enough to be easily seen by the immune system, the cost of continuing to grow spikes dramatically. In this case, the simulation showed the tumor behaving like a cautious traveler. It would grow quickly when small and hidden, but as it approached the size where it would be detected, it would deliberately slow down. The population would hover in a narrow band of size, neither growing out of control nor shrinking away, effectively entering a state of regulated dormancy. This suggests that if a tumor's environment makes large size particularly dangerous, the tumor will naturally evolve to stay small.

In a second scenario, the researchers removed that sharp threshold and instead assumed that the cost of growing increases smoothly and steadily as the tumor gets bigger, much like how a city becomes harder to manage as its population swells. Here, the results were similar but sharper. The tumor would grow until it reached a specific, optimal size where the cost of adding one more cell exactly balanced the benefit of having it. At this point, the population would stabilize, creating a self-regulating system that naturally resists both uncontrolled expansion and total collapse. This behavior mimics the logistic growth seen in many natural populations, but it arises not from a fixed rule imposed by nature, but from the tumor's own calculated response to increasing costs.

However, the researchers also tested a scenario where the reward for growing was simple and constant, with no extra penalty for getting large. In this case, the tumor did not learn to stop. Even when the cost of regulating its own death signals was high, the model showed the tumor continuing to grow or hovering near a neutral state without ever finding a stable, resting size. This finding is crucial because it rules out the idea that the mere presence of costs or constraints is enough to stop a tumor. The study suggests that for a tumor to naturally settle into a dormant or stable state, the pressure to stop must specifically increase as the tumor gets larger. If the cost of growing stays the same regardless of size, the tumor will not find a reason to stop.

The team also explored what happens when the tumor has a limited "budget" for regulating its own behavior, representing a finite amount of energy or signaling capacity available to fight off death signals. When the reward for growing was linear, a larger budget simply allowed the tumor to grow bigger and faster, but it still did not create a stable stopping point. But when the reward structure included a penalty for getting too big, the limited budget forced the tumor to be strategic. It would use its resources to grow until it reached a preferred size, then stop, effectively regulating itself around a specific population scale. This highlights that the structure of the trade-offs matters more than the sheer amount of resources available.

These simulations, while mathematical in nature, offer a new way to look at why some tumors remain dormant for years while others explode in size. The researchers found that the key to regulation lies in how the difficulty of growing changes as the tumor gets bigger. If the environment makes it increasingly hard to grow as the tumor expands—whether through immune detection, resource scarcity, or internal stress—the tumor will naturally find a way to stop. If, however, the difficulty remains constant, the tumor will not stop on its own. This insight shifts the focus from simply trying to kill every cancer cell to understanding the rules that govern their growth. It suggests that therapies which change the cost landscape—making it progressively more expensive for the tumor to grow larger—could be more effective at inducing long-term control than those that merely attack the cells directly. The study concludes that the behavior of a tumor is not just a reaction to the outside world, but a complex, calculated response to the specific shape of the challenges it faces.

Technical Summary: Optimal Control of Cellular Cost and Tumor Growth Dynamics

Problem Statement
Tumor progression is frequently modeled as a passive response to external therapies or immune pressure. However, biological evidence suggests that tumor populations possess the capacity for population-level regulation of proliferation and apoptosis through phenotypic plasticity. While optimal control theory is widely used in mathematical oncology, the prevailing paradigm typically treats the clinician as the active agent optimizing against a passive tumor. Conversely, evolutionary game theory and adaptive therapy have begun to model tumor adaptation, yet there remains a gap in frameworks that explicitly treat the tumor population itself as an adaptive agent optimizing its survival under sustained environmental and immune pressure. This paper addresses how the structure of trade-offs among growth incentives, regulatory costs, and extrinsic death pressures shapes the macroscopic behavior of tumor populations, specifically focusing on the emergence of regulated, dormant, or sustained growth dynamics.

Methodology
The authors develop a Continuous-Time Markov Decision Process (CTMDP) framework to model a tumor population as a controlled birth–death process.

State and Control: The state variable $N(t)$ represents the effective tumor population size. The control variable $a \in [0,1]$ represents a population-level regulatory response that modulates the balance between intrinsic proliferation and susceptibility to regulated apoptosis.
Dynamics: The per-cell birth rate is $\lambda(i,a) = ria$ , and the death rate is $\mu(i,a) = \delta i(1-a) + \delta_0 i$ . Here, $r$ is the intrinsic proliferation rate, $\delta$ is the regulated death rate, and $\delta_0$ represents extrinsic, uncontrolled death pressure (e.g., immune signaling or therapy). The control $a$ shifts the balance: higher $a$ favors proliferation and suppresses regulated death, while lower $a$ favors apoptosis.
Baseline Reference: A baseline signaling level $a^*$ is defined where per-capita birth and death are balanced ( $a^* = (\delta + \delta_0)/(r + \delta)$ ). Regulatory costs are defined as deviations from this baseline.
Objective Functions: The study evaluates five distinct objective structures to determine how cost scaling influences optimal policies:
1. Threshold-Dependent Cost ( $C_1$ ): Adds a cost that increases sharply when the population exceeds a recognition scale $L$ , modeling immune surveillance thresholds.
2. Quadratic Population Cost ( $C_2$ ): Incorporates superlinear costs ( $i^2$ ) representing crowding or metabolic burden, balanced against a linear growth reward.
3. Unbounded Linear Reward ( $C_3$ ): Balances a linear growth reward against the baseline regulatory cost, without additional state-dependent penalties.
4. Constrained Linear Reward: Maximizes linear growth subject to a global budget constraint ( $\beta$ ) on cumulative regulatory deviations.
5. Constrained Quadratic Penalty: Maximizes a concave reward (linear benefit minus quadratic penalty) subject to the same global budget constraint.
Numerical Approach: Optimal policies are computed using policy iteration and linear programming on a truncated state space. The resulting dynamics are analyzed via the induced normalized per-capita drift $g(i)$ and stochastic Gillespie simulations.

Key Contributions and Results
The study identifies that the emergence of restoring dynamics (where growth transitions to suppression) is not guaranteed by adaptive control alone but depends critically on how the objective function scales with population size.

Restoring Dynamics via Non-Linear Penalties:
- Threshold and Quadratic Models: Both formulations generate restoring dynamics. The threshold model produces a transition from growth to suppression near a recognition scale $L$ , creating a broad region of near-zero drift. The quadratic model produces a sharper, more localized transition around an interior population scale. In both cases, the optimal policy shifts from high proliferation at small sizes to suppression (or even $a < a^*$ ) at larger sizes, stabilizing the population.
- Mechanism: Restoring behavior arises when the marginal cost of proliferation increases with population size, making continued expansion progressively unfavorable.
Absence of Restoring Dynamics in Linear Models:
- Unbounded Linear Reward: When growth rewards and regulatory penalties scale proportionally with population size, the model fails to generate a restoring regime. The optimal policy remains above $a^*$ , resulting in sustained or near-neutral growth without a finite growth horizon. The system may approach zero drift under strong regularization but does not cross into negative drift.
- Constrained Linear Reward: Even with a global budget constraint, a linear reward structure merely expands the region of positive drift as the budget increases. It does not create an intrinsic finite population scale or a restoring transition.
Role of Constraints and Reward Structure:
- Constrained Quadratic Model: When a global budget is combined with a reward structure that includes an intrinsic finite scale (quadratic penalty), the model generates restoring dynamics. The budget determines the strength and duration of deviations from baseline, while the quadratic penalty ensures that continued expansion eventually becomes unfavorable, leading to regulation around an interior scale.
Emergent Logistic-like Behavior:
- The paper demonstrates that logistic-like growth dynamics (positive drift below a carrying capacity, negative drift above) can emerge from the optimization of state-dependent policies without imposing logistic density dependence directly. This occurs when the optimized policy produces a decreasing per-capita drift with an interior zero crossing.

Significance and Claims
The paper claims that regulated tumor dynamics, including dormancy and population stabilization, are fundamentally dependent on the scaling of growth incentives, regulatory costs, and constraints with population size.

Theoretical Insight: The work challenges the notion that adaptive control alone is sufficient to generate restoring dynamics. Instead, it posits that such behavior emerges only when the effective trade-off between growth and regulatory burden changes with the population state (e.g., via superlinear costs or threshold effects).
Phenomenological Utility: The objective functions are presented not as mechanistic descriptions of specific molecular pathways, but as phenomenological representations of biological trade-offs (e.g., immune recognition, metabolic burden, resource limitation).
Interpretation of Tumor Behavior: The results suggest that observed tumor behaviors (sustained growth vs. regulated dormancy) may reflect the underlying structure of these trade-offs. For instance, a transition to near-dormant behavior may indicate that the tumor has entered a regime where continued proliferation is increasingly costly relative to the baseline signaling state.
Therapeutic Implication: The authors suggest that therapeutic interventions could be viewed as modifying the effective cost landscape (e.g., increasing immune pressure or metabolic constraints) to shift the tumor into a regime where continued growth is increasingly unfavorable, rather than solely focusing on reducing tumor burden.

The study remains modest in its scope, acknowledging that the model is a minimal representation of a single effective population and that parameters are phenomenological rather than direct biological observables. The primary contribution is the identification of the classes of objective structures capable of producing specific macroscopic tumor dynamics.

Technical Summary: Optimal Control of Cellular Cost and Tumor Growth Dynamics

More like this