Cybergenetic Gene Expression Control

• Cybergenetic gene expression control is a technology that integrates computer-based feedback with synthetic and natural circuits to dynamically regulate gene expression at single-cell and population levels.
• It employs both in silico and intracellular feedback architectures, using methods like optogenetics and chemical actuation to adjust transcription and metabolic processes.
• The approach manages intrinsic molecular noise through moment-based and predictive control strategies, enhancing applications in bioproduction, host-aware regulation, and cell-state engineering.

Cybergenetic gene expression control integrates computer-based feedback with synthetic and natural biological circuits to provide dynamic, quantitative control over gene expression at the single-cell and population levels. By leveraging real-time measurements and actuation (chemical, optical, or electrical), cybergenetic systems implement rigorous feedback and optimal control strategies to achieve not only setpoint tracking of protein means but also manipulation of entire population distributions, variances, and cellular phenotypes under intrinsic molecular noise. Applications span synthetic biology, bioproduction, metabolic engineering, and host-aware control, with architectures ranging from purely genetic intracellular circuits to model-based and data-enabled computer-in-the-loop platforms.

1. Feedback Control Architectures: In Silico and Intracellular

Early cybergenetic control paradigms are distinguished by their physical locus and information flow. In silico control employs a computer-mediated feedback loop, where sensing (e.g., via fluorescence or microscopy) is processed by an algorithm that determines actuator inputs (e.g., media composition, light intensity), which are imposed via microfluidics or optogenetics. Classical control laws (PI, MPC, optimal control) are common, with actuation directly modulating transcriptional or translational parameters.

In contrast, purely intracellular architectures implement synthetic feedback controllers entirely within the cellular circuitry. These can comprise modular genetic comparators, relay switches, and toggle modules. For example, the Discrete Comparator and Bistable Comparator circuits implement digital and analog feedback regulation, using synthetic modules to compare the process output to a reference transcription factor and adjust the activation/repression of the gene of interest accordingly. This can achieve robust setpoint tracking with rapid convergence, low overshoot, and negligible steady-state error, without the need for any external computation or actuation (Kazemi et al., 2016).

2. Stochasticity and Moment-Based Feedback Control

Intrinsic noise is an inherent property of gene expression. Cybergenetic strategies account for this molecular noise by focusing on control of statistical moments (mean, variance, covariance) of protein distributions.

Proportional-integral (PI) controllers can achieve robust, global setpoint tracking of the mean protein level in simple gene networks by manipulating the transcription rate, provided explicit sufficient conditions on controller gains are enforced. Multivariable PI control, involving an additional actuated input (e.g., mRNA degradation rate), enables simultaneous local robust tracking of both the mean and variance. The reference setpoints must satisfy required admissibility conditions (e.g., positivity, coefficient-of-variation constraints). The framework extends to more complex motifs (dimerization, nonlinear moment systems) using pure integral control, with explicit gain bounds guaranteeing stability and performance, circumventing the need for moment closure (Briat et al., 2012, Briat et al., 2018).

3. Model-Based Optimal and Predictive Control under Molecular Noise

Computational frameworks have expanded cybergenetic control to manipulate full population probability distributions, not just low-order moments. In the regime of fast mRNA degradation (“bursting”), the discrete-state Chemical Master Equation (CME) can be approximated with Partial Integro-Differential Equations (PIDEs) for the continuous protein-level PDF.

Within this setting, optimal control and Model Predictive Control (MPC) are formulated to minimize divergence between the controlled and desired protein PDFs, subject to PIDE constraints and actuator limits. The adjoint-based approach enables efficient gradient computation for large-scale trajectory optimization, and the method readily supports time-varying and arbitrarily shaped target distributions. In practice, this allows induction of complex multimodal distributions (e.g., controlled bimodality, moving peaks) in cell populations, using real-time adjustments of stimuli such as chemical inducers or light (Faquir et al., 2024).

4. Constraint-Based and Hybrid Physics-Informed Metabolic Cybergenetic Control

Metabolic cybergenetics targets not only gene expression but also overall metabolic fluxes and product yields in microbial systems. Dynamic constraint-based models (e.g., dynamic enzyme Flux Balance Analysis, deFBA) integrate gene expression, enzyme allocation, metabolism, and extracellular process conditions for the closed-loop optimal control of dynamic bioprocesses.

The resulting bilevel optimal control problems are computationally intensive. Recent advances employ machine learning surrogates, trained on grid-based FBA data, to embed the physics of the metabolic network into lower-dimensional, macro-kinetic ODE models coupled with expression. This “hybrid physics-informed” approach preserves critical constraints and trade-offs while enabling tractable single-level MPC and estimation even in the presence of unmeasured states. Case studies demonstrate that this approach can optimize optogenetically driven production yields (e.g., itaconate in E. coli), maintain real-time capability, and enhance robustness to plant-model mismatch (Espinel-Ríos et al., 2023, Espinel-Ríos et al., 2024).

5. Data-Enabled, Model-Free Predictive Control

Where explicit modeling is challenging or systems are highly variable, data-enabled predictive control (DeePC) provides a sample-efficient, model-free cybergenetic alternative. DeePC constructs Hankel matrices from I/O data and reformulates control as a regularized quadratic program that predicts output evolution based purely on past measurements. Incorporation of basis functions for known input nonlinearities (e.g., Hill, Michaelis–Menten) allows convex QP formulations with built-in integral action and zero steady-state offset.

DeePC demonstrates robust multi-input, multi-output (MIMO) control performance for gene expression and host growth, requiring only $\mathcal{O}(10^2)$ samples—orders of magnitude less than deep reinforcement learning or neural-network-based controllers. It is robust to both parametric errors and substantial measurement noise and is competitive with optimal-model-based MPC while forgoing explicit system identification (Perreault et al., 4 Jan 2026).

6. Applications: Dynamic Bioproduction, Host-Aware Regulation, Cell-State Engineering

Cybergenetic gene expression control underpins advanced applications in bioproduction, synthetic biology, and host-aware regulation. In dynamic fed-batch fermentation scenarios, such as optogenetically controlled ATPase expression for enforced ATP wasting, cybergenetic MPC achieves significant improvements in product (lactate) yield, productivity, and robustness to process drift. Physics-informed and DeePC-based controllers similarly enable dynamic adjustment of intracellular protein expression and host growth rates, often under significant plant–model mismatch or in data-scarce regimes.

Population-level distributional control enables not only mean regulation but also masterminding cellular fate distributions—key for cell-fate programming, noise utilization, and differentiation protocols. Simulation and experimental platforms have demonstrated reliable tracking, shaping of distributional features (e.g., bimodality), and rapid recovery from disturbances at both the single-cell and population levels (Faquir et al., 2024, Espinel-Ríos et al., 2023, Perreault et al., 4 Jan 2026).

7. Limitations and Perspectives

The principal limitations of current cybergenetic approaches relate to model accuracy (especially for constraint-based and dynamic FBA schemes), identification of dose–response relationships, and the need for informative real-time measurements. Intracellular circuit-based controllers are constrained by circuit burden, retroactivity, and compositional context effects; in silico controllers depend on measurement/actuator latency and process observability. Machine learning surrogates, while tractable, require careful validation against plant data.

Future directions emphasize: scaling to multi-enzyme or even multicellular consortia control, integrating machine learning for system identification, expanding control to higher-dimensional phenotypic targets, and implementing robust, adaptive schemes to address unmodeled disturbances and biological variability (Espinel-Ríos et al., 2024, Espinel-Ríos et al., 2023, Faquir et al., 2024, Perreault et al., 4 Jan 2026, Briat et al., 2018, Kazemi et al., 2016, Briat et al., 2012).