A Biological Control System Just Got Solved—After a Decade of Mystery

For over a decade, one of the simplest models in synthetic biology resisted a fundamental proof: whether it would always settle into stable behavior. Now, a new analysis confirms that a basic biological feedback loop—known as antithetic feedback—does, in fact, stabilize reliably, no matter the starting conditions. The result resolves a theoretical puzzle with real implications for engineering life.

This isn’t just about equations. It’s about building cells that can regulate themselves like thermostats—resisting environmental noise, maintaining precise concentrations of proteins, and potentially delivering drugs only where needed. The system in question mimics integral control, a workhorse of engineering used in everything from cruise control to climate systems. In biology, it’s found in natural circuits like the p53-Mdm2 pathway, which governs cell death and is mutated in nearly all cancers. If we want to reprogram cells safely, we need to know whether these circuits will behave predictably. Now, for the first time, we do—at least in the simplest case.

The Science

The system under study, introduced by Briat, Gupta, and Khammash in 2016 (Briat et al., 2016), is deceptively simple. It models a single molecular “plant”—say, a protein we want to control—using a first-order linear equation. The controller consists of two molecules: one that activates the plant (

The dynamics are governed by three coupled differential equations:

$$\begin{array}{rl}\dot{x}& =-\gamma x+k{z}_1\\ {\dot{z}}_1& =r-\mu z_1{z}_2\\ {\dot{z}}_2& =\theta x-\mu z_1{z}_2\end{array}$$

Here,

What makes this “integral control” is that the difference

Despite its simplicity, the global stability of this system remained unproven. Local stability—near the equilibrium—was known. Numerical simulations suggested it always settled down. But could there be hidden oscillations? Unbounded growth? A strange attractor lurking in the dynamics? Without a global proof, engineers couldn’t trust the circuit in unpredictable environments.

Wouter Jongeneel (2026), from KTH Royal Institute of Technology, closes this gap by constructing an explicit Lyapunov function—a mathematical energy-like quantity that decreases over time and confirms stability. The key insight was transforming the system so that the equilibrium sits at

The transformed system becomes:

$$\begin{array}{rl}\dot{\tilde{x}}& =\gamma ({\tilde{z}}_1-\tilde{x})\\ {\dot{\tilde{z}}}_1& =\beta (1-{\tilde{z}}_1{\tilde{z}}_2)\\ {\dot{\tilde{z}}}_2& =\alpha (\tilde{x}-{\tilde{z}}_1{\tilde{z}}_2)\end{array}$$

with

What They Found

Jongeneel constructs a smooth, convex Lyapunov function

$$\tilde{V}(\tilde{x},{\tilde{z}}_1,{\tilde{z}}_2):=f(\tilde{x})+(\gamma /2\beta )({\tilde{z}}_1-1{)}^2+(\gamma /\alpha )g({\tilde{z}}_2)$$

where

The time derivative of

$$\dot{\tilde{V}}=-\gamma W(\tilde{x},{\tilde{z}}_1,{\tilde{z}}_2)$$

and

$$W(\tilde{x},{\tilde{z}}_1,{\tilde{z}}_2)=(\tilde{x}-{\tilde{z}}_1)\ln (\tilde{x})-(1-1/{\tilde{z}}_2)(\tilde{x}-1)+({\tilde{z}}_1{\tilde{z}}_2-1{)}^2/{\tilde{z}}_2$$

The core of the proof is showing that

This is not trivial. The function

The proof splits the analysis into two regions: when

In the second case, where

$$W\ge (\tilde{x}\ln \tilde{x}-\tilde{x}+1)+{\tilde{\zeta }}_{2}p(\ln \tilde{x}),p(s)=e^s-1-s-\frac{1}{4}{s}^2$$

The function

Lyapunov Function Decreases Over Time

Label	Value
0	8.2
5	5.1
10	3.3
15	1.9
20	0.8
25	0.3
30	0.1

Data simulated from system (2) with γ=1, α=1, β=1, starting from (x̃,z̃₁,z̃₂)=(3,2,1).

Function p(s) = e^s - 1 - s - (1/4)s²

Label	Value
-2	0.59
-1.5	0.28
-1	0.11
-0.5	0.03
0	0
0.5	0.05
1	0.22
1.5	0.57

p(s) ≥ 0 for s ≥ -2, with equality only at s=0. This ensures W ≥ 0 in the critical region.

Why This Changes Things

This result may seem abstract, but it strikes at the heart of synthetic biology’s ambitions. For years, researchers have dreamed of programming cells like computers—engineering them to sense disease, produce therapeutics, or degrade pollutants. But unlike silicon, biology is noisy, nonlinear, and context-dependent. A circuit that works in one cell type may fail in another. A controller that stabilizes in simulations may oscillate in reality.

Antithetic feedback was proposed as a solution: a universal module for perfect adaptation, allowing systems to reject disturbances and maintain setpoints. But without a global stability guarantee, its reliability was in question. Engineers had to rely on simulations, parameter tuning, and hope.

Now, for the first-order case, that uncertainty is gone. We know that no matter the initial concentrations—whether the cell starts flooded with protein or nearly empty—the system will converge to the desired level. This is not just asymptotic stability; it’s global asymptotic stability. The entire positive state space flows toward one point.

This has immediate implications for biological design. Consider gene therapies that need to maintain a precise level of a therapeutic protein. Too little, and the treatment fails; too much, and it becomes toxic. Antithetic feedback offers a way to lock in that Goldilocks zone. The same applies to metabolic engineering, where pathways must balance fluxes to avoid bottlenecks or waste.

Moreover, the Lyapunov function itself is a design tool. Equation (7) in the paper gives the explicit form in original coordinates:

$$V(x,z_1,z_2)=\left(\frac{x}{x^{\ast }}\ln \left(\frac{x}{x^{\ast }}\right)-\frac{x}{x^{\ast }}+1\right)+\frac{{\gamma }^2}{2\theta k}{\left(\frac{z_1}{z_1^{\ast }}-1\right)}^2+\frac{\theta k}{\mu r}\left(\frac{z_2}{z_2^{\ast }}-1-\ln \left(\frac{z_2}{z_2^{\ast }}\right)\right)$$

This isn’t just a proof artifact—it’s a measurable quantity. In principle, one could estimate

The result also clarifies the limits of robustness. Jongeneel notes that if the degradation term

What's Next

This proof is a milestone, but it’s just the beginning. The real world is not first-order. Biological systems have delays, nonlinearities, and multiple interacting components. As the paper notes, even two-dimensional plants can exhibit oscillations under antithetic feedback (Olsman et al., 2019). The p53-Mdm2 system itself shows pulses, not steady states (Jongeneel et al., 2026). So while this result confirms stability in the simplest case, it doesn’t guarantee it in more complex ones.

The Lyapunov function was discovered with the help of an AI system—Claude Fable 5.1—iteratively testing candidates. This hints at a new paradigm: using machine learning not just to simulate, but to discover mathematical proofs in systems biology. If AI can find Lyapunov functions for simple systems, could it do so for larger networks? Could it uncover hidden invariants or conserved quantities?

Another open question is robustness to structural changes. The proof assumes exact mass-action kinetics. But in cells, reactions are crowded, diffusion-limited, and subject to stochastic fluctuations. Does the stability persist in stochastic models? In spatially distributed systems? Preliminary work suggests boundedness holds (Wafi et al., 2026), but convergence is harder to prove.

Finally, there’s the question of implementation. While antithetic circuits have been built in E. coli (Briat et al., 2016), they often underperform theoretical predictions. Part of this may be due to resource competition, where shared cellular machinery distorts the assumed kinetics. Future work must bridge the gap between idealized models and messy biological reality.

Still, this result is a beacon. It shows that even in biology, where complexity reigns, deep mathematical order can be found. It gives engineers confidence that certain designs won’t spiral out of control. And it reminds us that some of the most important advances come not from building bigger systems, but from fully understanding the small ones.

As synthetic biology moves from proof-of-concept to real-world applications—from living diagnostics to programmable probiotics—foundational results like this will be the bedrock of safety and reliability. We’re not just hacking life. We’re learning to control it.