Gene Regulatory Networks
From autoregulation to oscillatory circuits — the mathematics of gene expression
A Brief History
Modelling Gene Expression
The flow of genetic information from DNA to protein follows the central dogma of molecular biology: DNA is transcribed into messenger RNA (mRNA), which is then translated into protein. Understanding gene regulation requires mathematical models that capture the essential dynamics of this process: production, degradation, and feedback.
The simplest model describes mRNA and protein as chemical species. mRNA is synthesized at a constant rate $\alpha_m$ and degraded with first-order kinetics at rate $\delta_m$. Similarly, protein is produced proportional to mRNA concentration and degraded at rate $\delta_p$:
$$ \frac{dm}{dt} = \alpha_m - \delta_m m, \quad \frac{dp}{dt} = \alpha_p m - \delta_p p $$where $m$ is mRNA concentration, $p$ is protein concentration, and $\delta_m$ and $\delta_p$ are degradation rate constants. At steady state ($\frac{dm}{dt} = \frac{dp}{dt} = 0$):
$$ m^* = \frac{\alpha_m}{\delta_m}, \quad p^* = \frac{\alpha_p \, \alpha_m}{\delta_p \, \delta_m} $$A key insight is that mRNA and protein operate on vastly different timescales. mRNA has a half-life of minutes, while proteins often persist for hours or longer. This timescale separation motivates the quasi-steady-state approximation: assuming mRNA reaches equilibrium quickly (setting $dm/dt \approx 0$), the dynamics are dominated by protein evolution.
The rapid equilibration of mRNA ($\tau_m \sim \delta_m^{-1}$ = minutes) versus slower protein evolution ($\tau_p \sim \delta_p^{-1}$ = hours) means that in many models, mRNA can be treated as instantaneously following gene expression dynamics. This reduces the two-variable system to a single ODE for protein. Reference: Alon (2007) An Introduction to Systems Biology, Chapter 1.
Autoregulation
A gene can regulate its own expression through its protein product—either accelerating its own production (positive autoregulation) or slowing it down (negative autoregulation). These regulatory architectures are among the most fundamental network motifs in living cells.
Negative Autoregulation
In negative autoregulation, the protein product represses transcription of its own gene. A common model uses a Hill function to represent the nonlinear repression:
$$ f(p) = \frac{\alpha}{1 + (p/K)^n} $$where $\alpha$ is the maximal transcription rate, $K$ is the binding affinity (dissociation constant), and $n$ is the Hill coefficient. When $n > 1$, the response is ultrasensitive—a sharp transition around the threshold $p = K$. The protein dynamics become:
$$ \frac{dp}{dt} = \frac{\alpha}{1 + (p/K)^n} - \delta p $$Negative autoregulation has two important effects: it speeds up response time to environmental changes and reduces noise in protein levels. The system acts as a homeostatic "buffer"—if protein rises above steady state, the increased repression pushes it back down. This makes the circuit robust to fluctuations and perturbations. Becskei and Serrano (2000) demonstrated these principles experimentally using engineered circuits in bacteria.
Positive Autoregulation
In contrast, positive autoregulation allows a protein to enhance its own transcription. The Hill function now has exponent $n$ in the numerator:
$$ f(p) = \frac{\alpha p^n}{K^n + p^n} $$This creates slower response times and can generate bistability: two stable steady states with an unstable fixed point between them. Once induced to a high level, positive autoregulation locks the system in the "on" state. This is exploited in developmental and immune systems for decisions with long-term memory.
The Hill function $f(p) = \frac{\alpha p^n}{K^n + p^n}$ models transcription when $n$ cooperative copies of a protein must bind to a regulatory site. For $n=1$, the response is hyperbolic and gradual. For $n > 1$, the curve becomes sigmoidal, displaying a sharp switch-like response. The steepness is greatest when all binding sites have comparable affinity—a signature of cooperativity.
The Genetic Toggle Switch
One of the first synthetic gene circuits was the genetic toggle switch, constructed by Gardner, Cantor, and Collins (2000) in E. coli. It consists of two genes, each encoding a protein that represses the other. This mutual repression creates bistability, allowing the cell to "remember" its state—either gene 1 is on and gene 2 off, or vice versa.
The toggle switch is governed by the system:
$$ \frac{du}{dt} = \frac{\alpha_1}{1 + v^{\beta}} - u, \quad \frac{dv}{dt} = \frac{\alpha_2}{1 + u^{\gamma}} - v $$where $u$ and $v$ are the (normalized) levels of the two proteins, $\alpha_1$ and $\alpha_2$ control maximal expression, and $\beta$ and $\gamma$ are Hill coefficients representing the strength of repression. The degradation rates are normalized to 1.
For bistability to exist, the parameters must satisfy specific conditions. The key requirement is that each gene's self-intersection curve (the nullcline where $du/dt = 0$ or $dv/dt = 0$) crosses the other's multiple times, creating multiple fixed points. With sufficiently strong repression ($\beta, \gamma \gg 1$) and sufficient expression ($\alpha_1, \alpha_2$ large enough), the circuit exhibits a characteristic S-shaped phase portrait:
- Two stable fixed points: $(u^*, 0)$ and $(0, v^*)$ (high-low and low-high states)
- One unstable saddle point at an intermediate state
- Trajectories collapse to the stable states from either basin of attraction
Bistability requires that each repressor be strong enough to fully suppress its target. Mathematically, the nullclines must intersect with sufficient negative slope to create three equilibria. The synthetic toggle switch demonstrated that bistable genetic circuits could be engineered, opening a new field of synthetic biology. This principle is now used in engineered biosensors and biological memory devices. Reference: Gardner, T. S., Cantor, C. R., & Collins, J. J. (2000). Nature, 403, 339–342.
The Repressilator
While the toggle switch harnesses bistability, Elowitz and Leibler (2000) asked: what if you closed a feedback loop with three genes, each repressing the next? The result is the repressilator, a synthetic oscillator that produces periodic protein fluctuations.
The repressilator consists of genes encoding proteins that form a cycle: Gene 1 protein represses Gene 2, Gene 2 protein represses Gene 3, and Gene 3 protein represses Gene 1. Because there are an odd number of negative interactions, the circuit cannot reach a stable fixed point—instead, it oscillates.
The dynamics are:
$$ \frac{dp_i}{dt} = \frac{\alpha}{1 + p_{i-1}^n} - p_i, \quad i = 1, 2, 3 \pmod{3} $$where $p_i$ are the protein levels, $\alpha$ is the production rate, and $n$ is the Hill coefficient. Oscillations require:
- Sufficient repression strength ($n$ moderately large, $n \geq 2$)
- Sufficient production rate ($\alpha$ large relative to degradation)
- Timescale separation: if proteins degrade too quickly relative to mRNA, oscillations are damped
The repressilator can be viewed as a generalization of earlier oscillator models. Goodwin (1965) first showed mathematically that negative feedback with sufficient time delays or cooperativity could produce oscillations—a prediction confirmed by Elowitz and Leibler's synthetic circuit. Their measurements showed sustained periodic fluorescence with a period of roughly two hours, demonstrating that temporal dynamics could be programmed into living cells.
The biological importance of oscillators extends far beyond synthetic circuits. The circadian clock (Vilar et al., 2002), the segmentation clock in vertebrate embryos, and calcium signaling oscillations are all examples of how cells use oscillatory gene circuits to regulate crucial processes.
Explore: Gene Circuits
Feedback Loops and Network Motifs
Real gene regulatory networks are composed of smaller building blocks—recurring patterns of interactions known as network motifs. These motifs are overrepresented in biological networks compared to random networks and appear to be selected for by evolution because they perform specific computational or regulatory functions.
Negative Feedback Loops
A single negative feedback loop (as in negative autoregulation) tends to stabilize the system and reduce sensitivity to perturbations. However, with appropriate time delays or cooperativity, negative feedback can generate oscillations—as demonstrated by the repressilator. The key is that information takes time to propagate around the loop: the protein must be synthesized, and the repression must take effect, during which time the system overshoots, creating periodic behavior.
Positive Feedback Loops
Positive feedback (where a protein enhances its own production or another protein that enhances the first) creates bistability and switches. Once the system crosses a threshold, it locks into a high state and resists returning to the low state. This is exploited in biological decision-making: phage lambda's lysis-lysogeny switch, mammalian Wnt signaling, and immune system activation all use positive feedback to create hysteresis.
Feed-Forward Loops
A coherent feed-forward loop (FFL) consists of three genes: gene A regulates gene B, gene A also regulates gene C, and gene B regulates gene C. There are two types depending on whether the paths are activating or repressing.
An incoherent feed-forward loop occurs when the direct path (A→C) has opposite sign to the indirect path (A→B→C). FFLs perform temporal filtering: they respond to sustained signals but ignore brief pulses. This is useful for distinguishing transient noise from genuine changes in the environment. Alon and colleagues (2007) showed that these motifs appear frequently in bacterial chemotaxis and other sensory systems.
Time Delays and Oscillations
A fundamental principle is that negative feedback with sufficient time delay can generate oscillations, while positive feedback with delay can lead to bistability. The delay comes from protein synthesis time, mRNA degradation, or from multi-step regulatory cascades. Even linear systems (without cooperativity) can oscillate if delays are present—though the repressilator and circadian clock rely on both nonlinearity and time delays to achieve robust oscillations.
Stochastic Gene Expression
Gene expression is fundamentally a stochastic process. Transcription and translation are discrete events; at any moment, only a handful of mRNA or protein molecules may be present. This inevitably leads to noise in gene expression levels—fluctuations that can be observed experimentally and must be accounted for in understanding cell behavior.
Elowitz et al. (2002) distinguished two sources of noise in their landmark study using fluorescent proteins:
- Intrinsic noise: arises from the stochastic nature of individual biochemical reactions (transcription, translation, binding events). It scales inversely with molecule number and is most significant when protein or mRNA levels are low.
- Extrinsic noise: arises from fluctuations in the cellular environment (ribosome availability, polymerase concentration, growth rate variations). It affects all genes similarly and can be large even for high-concentration species.
A simple model of gene expression noise uses the Gillespie algorithm, which simulates the stochastic reactions:
- Transcription (mRNA produced at rate $\nu_m$)
- mRNA degradation (each mRNA destroyed at rate $\delta_m$)
- Translation (each mRNA produces proteins at rate $\nu_p$)
- Protein degradation (each protein destroyed at rate $\delta_p$)
At each time step, the algorithm determines which reaction occurs and advances time by an exponentially distributed interval. This correctly captures the full statistics of fluctuations around the deterministic mean trajectory.
An important measure is the Fano factor $F = (\langle p^2 \rangle - \langle p \rangle^2) / \langle p \rangle$, the ratio of variance to mean. For Poisson processes (pure intrinsic noise), $F = 1$. For a more tightly controlled gene (negative autoregulation), $F < 1$. For a gene subject to large extrinsic fluctuations, $F \gg 1$.
Swain et al. (2002) showed that the noise in protein levels depends on the relative timescales of mRNA and protein lifetime. If mRNA is short-lived (rapid averaging), noise is dampened; if mRNA is long-lived, fluctuations in transcription propagate to protein noise. This has profound implications for how cells tolerate (or suppress) noise in critical genes like those encoding developmental regulators.
Gene expression noise can be decomposed into intrinsic (due to random biochemical events) and extrinsic (due to cell-to-cell variation in environment) components. Negative autoregulation suppresses intrinsic noise but not extrinsic noise, whereas positive feedback and long mRNA lifetimes amplify both. References: Elowitz, M. B., et al. (2002) Science, 297, 1183–1186; Swain, P. S., et al. (2002) Proc. Natl. Acad. Sci. USA, 99, 12795–12800.
Exercises
Conceptual Questions
- Explain the difference between a negative feedback loop (e.g., inhibition of a transcription factor by its own product) and positive feedback. What dynamical signatures does each produce?
- What is a Hill function and why is the Hill coefficient $n > 1$ necessary for bistability? Sketch the nullclines for a toggle switch and identify stable and unstable equilibria.
- In gene oscillators (like the synthetic repressilator), explain why a delay in feedback is essential for sustained oscillations. What happens if the feedback is instantaneous?
- Distinguish between intrinsic (demographic) and extrinsic (cell-to-cell variability in gene expression) noise. Which type of noise can be suppressed by negative feedback?
- Why might a feedforward loop (where one TF activates another and also directly activates the target) provide different dynamics than simple sequential regulation?
Computer Problems
- Hill Function and Bistability. Implement the toggle switch model: $\frac{dx}{dt} = \frac{1}{1+(y/K)^n} - \delta x$ and $\frac{dy}{dt} = \frac{1}{1+(x/K)^n} - \delta y$ with $K = 1$, $\delta = 1$, and vary $n = 1, 2, 3, 5$. Plot the phase plane for each and identify the bifurcation point where bistability emerges. Compute stable and unstable equilibria.
- Synthetic Repressilator Oscillations. Implement the repressilator with three genes: $\frac{dP_i}{dt} = \alpha_0 + \frac{\alpha}{1+(P_{i-1})^n} - \beta P_i$. Use $\alpha_0 = 0.5$, $\alpha = 100$, $n = 2$, $\beta = 1$. Simulate over 500 time units and plot all three protein concentrations. Verify periodicity and compute the oscillation period.
- Negative Feedback Noise Suppression. Implement stochastic simulations (Gillespie algorithm) for: (a) constitutive production with Poisson noise, and (b) negative feedback regulation. Use binding rate $k_{\text{on}} = 1$, unbinding $k_{\text{off}} = 0.1$, production rate $r = 10$, degradation $\gamma = 1$. Compare noise (coefficient of variation) between the two systems.
- Feedforward Loop Dynamics. Implement a feedforward motif where gene A activates gene B (with delay $\tau$) and also directly activates target gene Z, while B also activates Z. Compare with a serial cascade (A→B→Z). Show how the feedforward circuit creates pulse-like responses with sharper kinetics.
- Cooperative Binding and Switch Sharpness. For a promoter with multiple binding sites, implement the Hill function with variable Hill coefficients ($n = 1, 2, 4, 8$) and plot the response curve (TF concentration vs transcription rate). Quantify the steepness using the dynamic range and threshold response width. Show how cooperativity creates switch-like behaviour.
References
- Alon, U. (2007). An Introduction to Systems Biology: Design Principles of Biological Circuits. Chapman & Hall/CRC.
- Becskei, A. & Serrano, L. (2000). Engineering stability in gene networks by autoregulation. Nature, 405, 590–593.
- Gardner, T. S., Cantor, C. R., & Collins, J. J. (2000). Construction of a genetic toggle switch in Escherichia coli. Nature, 403, 339–342.
- Elowitz, M. B. & Leibler, S. (2000). A synthetic oscillatory network of transcriptional regulators. Nature, 403, 335–338.
- Goodwin, B. C. (1965). Oscillatory behavior in enzymatic control processes. Adv. Enzyme Regul., 3, 425–438.
- Vilar, J. M., Kueh, H. Y., Barkai, N., & Leibler, S. (2002). Mechanisms of noise-resistance in genetic oscillators. Proc. Natl. Acad. Sci. USA, 99, 5988–5992.
- Elowitz, M. B., Levine, A. J., Siggia, E. D., & Swain, P. S. (2002). Stochastic gene expression in a single cell. Science, 297, 1183–1186.
- Swain, P. S., Elowitz, M. B., & Siggia, E. D. (2002). Intrinsic and extrinsic contributions to stochasticity in gene expression. Proc. Natl. Acad. Sci. USA, 99, 12795–12800.
- Gillespie, D. T. (1977). Exact stochastic simulation of coupled chemical reactions. J. Phys. Chem., 81, 2340–2361.
- Savageau, M. A. (1974). Comparison of classical and autogenous systems of regulation in inducible operons. Nature, 252, 546–549.
- Ferrell, J. E., Jr. & Machleder, E. M. (1998). The biochemical basis of an all-or-none cell fate switch in Xenopus oocytes. Science, 280, 895–898.
- Tyson, J. J., Chen, K. C., & Novak, B. (2003). Sniffers, buzzers, toggles and blingers: dynamics of regulatory and signaling pathways in the cell. Curr. Opin. Cell Biol., 15, 221–231.