Travelling Waves in Gene Expression: A Mathematical Model of Cell-State Dynamics in Melanoma

This paper presents a piecewise-linear mathematical model of a three-transcription-factor gene regulatory network to demonstrate how strong intercellular signaling drives travelling waves of gene expression, ultimately determining the dominant cell-state characteristic in melanoma populations.

The Gist

Imagine a melanoma tumor not as a solid lump of identical cells, but as a bustling, chaotic city where every citizen (cell) has a different job, a different mood, and a different way of reacting to the world. Some cells are "workers" (proliferating), some are "hiding in bunkers" (drug-tolerant), and some are "adventurers" (invasive).

This paper is like a mathematical weather forecast for that city. The authors are trying to figure out: Why do some parts of the tumor look like a peaceful village while others look like a war zone? And how does the whole city decide which "personality" to adopt?

Here is the breakdown of their discovery, using simple analogies:

1. The Three "Traffic Lights" Inside Every Cell

Inside every melanoma cell, there is a tiny control panel with three main switches (Transcription Factors):

• SOX10 (The Identity Card): Keeps the cell feeling like a skin cell.
• MITF (The Energy Switch): Tells the cell to grow and divide.
• ZEB1 (The Escape Artist): Tells the cell to stop being a skin cell, become tough, and start invading other areas.

These three switches talk to each other. If ZEB1 gets too loud, it tells SOX10 and MITF to shut up. If MITF is loud, it tells ZEB1 to quiet down. It's a constant tug-of-war inside the cell.

2. The "Light Switch" Analogy (The Math Part)

Usually, these switches are dimmer switches—they can be slightly on or slightly off. But the authors made a clever simplification: they treated them like light switches that are either fully ON or fully OFF.

By doing this, they could map out exactly what happens. They found that the cell can settle into specific "modes" or states, just like a car can be in "Park," "Drive," or "Reverse."

• State A (The Sleepy Cell): High MITF, Low ZEB1. It's differentiated and hard to kill with drugs, but it doesn't spread.
• State B (The Invader): Low MITF, High ZEB1. It's aggressive, moves around, and is hard to catch.
• The "Bistable" Trap: The most interesting finding is that for many settings, a cell can be either State A or State B, but it's unstable in the middle. It's like a ball sitting on a hilltop; it will eventually roll down to one side or the other, but it won't stay balanced in the middle.

3. The "Wave of Change" (The Big Discovery)

Here is where the paper gets really cool. The authors asked: What happens when you have a whole neighborhood of these cells?

They realized that cells talk to their neighbors. If a cell in State A (Sleepy) starts to feel the pressure from its neighbors who are in State B (Invader), it might switch sides.

Imagine a stadium wave. If a few people stand up, the person next to them might stand up too, and soon the whole stadium is standing.

• The Travelling Wave: The authors found that if the "communication" between cells is strong enough, a specific state (like the "Invader" state) can sweep through the tumor like a wave. It converts the "Sleepy" cells into "Invaders" as it moves.
• The Winner Takes All: Depending on the exact settings (the "parameters" of the tumor), one state will always win. If the "Invader" wave is strong, the whole tumor becomes aggressive. If the "Sleepy" wave is stronger, the whole tumor becomes dormant.

4. Why Does This Matter? (The Real-World Impact)

This explains why tumors are so tricky to treat.

• Heterogeneity: You might see a tumor that looks half "Sleepy" and half "Invader." The paper suggests this isn't random; it's because the "wave" of change got stuck or blocked.
• Therapy Resistance: If a drug kills the "Sleepy" cells, the "Invader" wave might sweep through the remaining cells, making the tumor come back even stronger.
• The "Blockage" Theory: The paper suggests that if we can disrupt the "communication" (the wave) between cells, we might be able to stop the tumor from switching into its dangerous, invasive mode. It's like cutting the power to the stadium speakers so the wave can't spread.

Summary

Think of the tumor as a city where the citizens are constantly arguing about whether to stay home or go out and explore.

• The Math is the rulebook that predicts who wins the argument.
• The Wave is the moment when the "Go Out" crowd convinces the "Stay Home" crowd to join them, sweeping through the city.
• The Goal is to understand the rules of this argument so doctors can intervene, perhaps by silencing the loudest voices (the signals) to stop the tumor from becoming an unstoppable invader.

The authors have built a map of this argument, showing exactly where the "tipping points" are, which could help design better treatments to keep the tumor in a "Sleepy" state forever.

Technical Summary

1. Problem Statement

Melanoma is characterized by high intratumour heterogeneity and phenotypic plasticity, where cells switch between distinct transcriptional states (e.g., proliferative, invasive, drug-tolerant). This plasticity drives therapy resistance and immune evasion. While the underlying gene regulatory networks (GRNs) involving transcription factors (TFs) like MITF, SOX10, and ZEB1 are known, the mechanisms governing how these states emerge locally and how they coordinate spatially across a tumor population remain unclear. Specifically, the paper addresses:

• How a minimal GRN generates multiple stable cell states.
• How intercellular communication influences the spatial distribution of these states.
• Under what conditions a specific cell state dominates a population via travelling waves of gene expression.

2. Methodology

The authors employ a multi-scale mathematical modeling approach combining dynamical systems theory and reaction-diffusion equations.

• Gene Regulatory Network (GRN) Simplification:

  • The model focuses on a core network of three TFs: SOX10 (S), MITF (M), and ZEB1 (Z).
  • Interactions include mutual inhibition (Z inhibits S and M; S inhibits M) and activation (S activates M).
  • The authors refine a previous 5-TF model by removing SOX9 and JUN (absorbing their effects into basal rates) and adding an inhibitory link from ZEB1 to SOX10.

• Mathematical Formulation:

  • Local Dynamics: The system is modeled using Ordinary Differential Equations (ODEs) based on mass-action kinetics and Hill-type sigmoidal functions to represent cooperative binding of TFs to promoter regions.
  • Piecewise-Linear Approximation: A key simplification assumes a large number of TFs ($n \to \infty$), allowing the Hill functions to be approximated by Heaviside step functions. This renders the system piecewise-linear and non-smooth, enabling analytical determination of all steady states and their stability.
  • Spatial Extension: To model tissue-level dynamics, the authors use a Voronoi tessellation to represent disordered cell packing. They introduce reaction-diffusion equations where diffusion terms represent the effect of diffusing signaling molecules (mediated by secondary signals) that couple TF levels between neighboring cells.

• Analysis Techniques:

  • Steady-State Analysis: Solving the non-smooth algebraic system to identify 19 possible uniform states (16 stable, 3 saddle points).
  • Bifurcation Analysis: Mapping parameter space (specifically threshold values $\hat{S}_M^0, \hat{M}_Z^0, \hat{Z}_S^0, \hat{Z}_M^0$) to identify regions of monostability and bistability.
  • Travelling-Wave Analysis: In the 1D limit, the authors derive conditions for travelling-wave solutions connecting two stable states. They construct piecewise-exponential solutions and derive a condition for the wave speed ($c$) to determine which state is spatiotemporally stable.

3. Key Contributions

• Analytical Classification of States: The paper provides a complete analytical classification of 16 stable steady states and 3 saddle points in the limit of large $n$, categorizing them into biologically relevant phenotypes (e.g., Hyper-differentiated, Mesenchymal-like, Neural-crest-like, Hybrid).
• Spatiotemporal Stability Threshold: The authors derive a specific analytical condition (Equation 20/22) that partitions the parameter space. This condition predicts which of two coexisting local stable states will dominate a cell population when intercellular communication is present.
• Mechanism of Heterogeneity vs. Homogeneity: The study demonstrates that while local dynamics may be bistable, strong intercellular coupling (diffusion) disrupts this bistability at the population level, forcing the system into a single spatiotemporally stable state. Conversely, weak coupling allows for heterogeneous patterns.
• Butterfly Catastrophe Discovery: The authors identify a "butterfly catastrophe" near $n=5$ where the system exhibits tristability, highlighting the sensitivity of the model to the cooperative binding assumption, though the large-$n$ limit remains robust for $n \ge 7$.

4. Key Results

• State Landscape: The model successfully reproduces known melanoma phenotypes. For instance, high MITF/low ZEB1 corresponds to a hyper-differentiated (drug-tolerant) state, while low MITF/high ZEB1 corresponds to a mesenchymal (invasive) state.
• Bistability and Bifurcations: The parameter space contains regions where two distinct stable states (e.g., Hyper-differentiated and Mesenchymal-like) coexist, separated by a saddle point. Transitions between these regions occur via saddle-node bifurcations.
• Travelling Waves and Dominance:
  • Simulations in 1D and 2D show that when two states are introduced in a population, a travelling wave propagates, and one state invades the other.
  • The direction of the wave (and thus the dominant state) is determined by the derived stability threshold. If the threshold condition is met, the wave speed $c > 0$ (State A dominates); otherwise, $c < 0$ (State B dominates).
  • Diffusion Dependence: High diffusion coefficients lead to a single homogeneous state across the tumor. Low diffusion coefficients result in the wave stalling, leaving persistent spatial heterogeneity (clusters of different states), which mimics the subpopulations observed in experimental data (Figure 1).
• Validation: The analytical travelling-wave solutions (piecewise-exponential) match numerical simulations with high accuracy, confirming the validity of the space-continuous approximation for the discrete cell system.

5. Significance

• Theoretical Insight: The paper bridges the gap between local gene regulatory dynamics and global tumor phenotypes. It provides a mathematical framework to understand how cell-cell communication can synchronize cell states, potentially explaining the formation of "encapsulated" homogeneous regions observed in tumors after therapy.
• Therapeutic Implications: The model suggests that therapeutic strategies could target the parameters governing the stability threshold (e.g., altering TF production/degradation ratios or blocking intercellular signaling). By manipulating these parameters, it might be possible to force the tumor into a uniform, drug-sensitive state or prevent the invasion of aggressive mesenchymal states.
• Experimental Guidance: The derived stability conditions offer testable hypotheses. For example, the model predicts that changing the threshold for ZEB1 inhibition of SOX10/MITF could switch the dominant phenotype. It also suggests that measuring diffusion rates of signaling molecules could predict whether a tumor will exhibit homogeneous or heterogeneous drug responses.
• Methodological Advance: The use of piecewise-linear approximations for complex GRNs allows for the exact enumeration of steady states and rigorous stability analysis, a feat difficult with standard non-linear ODE solvers alone.

In summary, this work establishes that melanoma cell-state plasticity is not merely a local property but a collective phenomenon governed by the interplay between intrinsic gene network bistability and extrinsic intercellular communication, with travelling waves serving as the mechanism for state dominance.