Gain-Scheduled Optogenetic Feedback for Disturbance Rejection in Bacterial Batch Cultures

This paper presents a multiscale model-guided control framework that employs gain-scheduled PID and feedback-feedforward strategies to achieve robust disturbance rejection in batch E. coli cultures by adapting to growth-phase-dependent dynamics that limit fixed-gain controllers.

The Gist

Imagine you are trying to keep the temperature in a room perfectly steady at 72°F. If the room is empty and the windows are closed, it's easy: you just turn the heater on or off based on a simple thermostat.

But now, imagine that room is a living, breathing organism that is constantly changing size, growing, and shifting its internal needs. This is what scientists face when they try to control gene expression (turning genes on and off) inside a batch of bacteria growing in a lab.

Here is a simple breakdown of what this paper is about, using everyday analogies.

The Problem: The "Moving Target"

In a standard lab experiment (a "batch culture"), bacteria are put in a cup with food. They eat, grow, and multiply until they run out of food.

• The Goal: Scientists want the bacteria to produce a specific protein (like a glowing green light) at a constant level.
• The Catch: As the bacteria grow from a few cells to a crowded crowd, their internal "engine" changes.
  • Early stage: They are hungry and growing fast.
  • Late stage: They are full, tired, and slowing down.

The researchers found that a standard controller (like a fixed thermostat) fails here. It's like trying to drive a car using the same steering sensitivity whether you are driving a tiny go-kart or a massive semi-truck.

• If you use the "go-kart" settings on the "semi-truck," you crash (the signal goes wild).
• If you use the "semi-truck" settings on the "go-kart," you barely move (the signal doesn't react fast enough).

When the bacteria grow, the "rules of the road" change. A fixed controller can't handle this, especially if something disturbs the system (like suddenly adding more water to the cup, which dilutes the bacteria).

The Solution: The "Smart Driver"

The team developed a new way to control the bacteria using Optogenetics. Think of this as a remote control that uses light to tell the bacteria what to do.

• Green light = "Turn the gene ON."
• Red light = "Turn the gene OFF."

They created a computer system that watches the bacteria and adjusts the light instantly. But instead of using a "dumb" controller, they built a "Gain-Scheduled" controller.

The Analogy: The Adaptive Cruise Control
Imagine driving a car with Adaptive Cruise Control.

1. Fixed-Gain Controller (The Old Way): You set the car to always accelerate at the same rate when you hit the gas. If you are on a steep hill, you go too slow. If you are on a flat road, you fly off the handle.
2. Gain-Scheduled Controller (The New Way): The car has a sensor that knows if you are going uphill, downhill, or on a curve. It automatically adjusts how sensitive the gas pedal is.
   • When the bacteria are growing fast (like a steep hill), the controller becomes gentler so it doesn't overshoot.
   • When the bacteria are slowing down (like a flat road), the controller becomes sharper so it reacts quickly to mistakes.

The "Double-Action" Strategy

The researchers didn't stop there. They realized that sometimes the bacteria get hit with a massive shock (like a sudden 80% drop in population). A reactive controller (waiting to see the mistake and then fixing it) is too slow.

So, they added a Feedforward component.

• The Analogy: Imagine you are walking down a hallway and you see a puddle ahead.
  • Feedback (PID): You wait until you step in the puddle, feel the water, and then jump back.
  • Feedforward: You see the puddle before you step in, so you jump over it immediately.

By combining the "Smart Driver" (Gain-Scheduled) with the "Puddle Spotter" (Feedforward), they created a system that can handle both small bumps and massive shocks without losing control.

The Results: Three Different "Driving Modes"

The paper tested these controllers under different "disturbances" (shocks to the system) and found three distinct scenarios:

1. Small Disturbances (The Breeze): If the bacteria are just slightly disturbed, any controller works fine. It's like a gentle breeze; you don't need a special system to stay on the road.
2. Medium Disturbances (The Bump): If the bacteria are shaken a bit, the Gain-Scheduled Controller is the winner. It knows exactly how to adjust its sensitivity to prevent the bacteria from "overshooting" (getting too excited and producing too much protein).
3. Massive Disturbances (The Earthquake): If the bacteria are hit with a huge shock, the Combined Controller (Gain-Scheduled + Feedforward) is the hero. It acts instantly to recover the situation before the bacteria even realize they've been knocked off course.

Why This Matters

This research is a big step forward for Synthetic Biology.

• Before: Scientists could only control bacteria in very specific, steady environments (like a continuous flow of water), which is expensive and hard to do.
• Now: They can control bacteria in simple, cheap "batch" cups (like a standard test tube), even as the bacteria grow and change.

This opens the door to using living cells as reliable factories for making medicines, biofuels, and other products, ensuring they work perfectly even when conditions change.

In short: The paper teaches us how to drive a living, growing biological system by giving the computer a "smart steering wheel" that changes its sensitivity based on how the bacteria are feeling at that exact moment.

Technical Summary

1. Problem Statement

Precise regulation of gene expression in batch bacterial cultures is fundamentally challenging due to the time-varying nature of cellular physiology. Unlike continuous-flow systems (chemostats) where cells remain in a steady state, batch cultures undergo distinct growth phases (lag, exponential, stationary).

• The Core Issue: These growth-phase transitions alter the underlying input-output dynamics of gene expression through multiscale interactions (e.g., growth-dependent dilution, resource allocation changes).
• Limitation of Current Methods: Fixed-gain controllers (like standard PID) designed under constant-growth assumptions fail to provide robust disturbance rejection in batch cultures. A controller tuned for one physiological state becomes either overly aggressive (causing overshoot) or insufficiently responsive (causing slow recovery) as the culture transitions between growth phases.
• Specific Challenge: The paper focuses on rejecting population-scale disturbances (simulated as sudden dilution events) in E. coli cultures regulated by an optogenetic system.

2. Methodology

The authors developed a comprehensive framework combining multiscale modeling, frequency-domain analysis, and advanced control theory.

A. System and Modeling (GEAGS Framework)

• Biological System: The study uses the CcaSR two-component optogenetic system in E. coli. Green light induces sfGFP expression, while red light represses it.
• Multiscale Model (GEAGS): The authors utilize the Gene Expression Across Growth Stages (GEAGS) framework. This model couples intracellular gene expression kinetics with population-level logistic growth.
  • It introduces Rate Modifier Functions (RMFs) that empirically link growth phase (defined by the ratio of current cell density to carrying capacity, $f_{RMF} = C/C_{max}$) to molecular rates (transcription, translation, degradation, and dilution).
  • This captures how translational resource availability and degradation rates shift dynamically as the culture moves from exponential to stationary phase.
• Simulation Platform: Simulations replicate the LEMOS (LED Embedded Microplate for Optogenetic Studies) hardware, using discrete 10-minute control intervals with specific light actuation windows.

B. Disturbance Formulation

• Disturbance Type: A "step perturbation" simulating a sudden dilution of the cell population ($C$) at a specific time $t_d$.
• Magnitude: Disturbances ($\rho$) ranged from 20% to 80% dilution, effectively resetting the culture to different growth phases (e.g., from stationary back to mid-log or early exponential).

C. Analysis and Controller Design

1. Frequency-Response Analysis: The authors performed open-loop frequency-response analysis (Bode plots) on the linearized GEAGS model at various operating points.
   • Finding: System gain and phase lag are highly dependent on the growth phase. Low-density (early growth) states exhibit high gain and low lag, while high-density (stationary) states exhibit low gain and high lag.
2. Controller Architectures:
   • Fixed-Gain PID: Served as the baseline for comparison.
   • Gain-Scheduled PID (PID-GS): A controller where gains ($K_P, K_D, K_I$) are dynamically scaled based on the physiological state variable $f_{RMF}$.
     • Logic: As the system gain increases (early growth), the controller gain is reduced to prevent overshoot. As the system becomes sluggish (stationary phase), the controller gain is increased to improve responsiveness.
   • Gain-Scheduled PID with Feedforward (PID-GS-FF): Combines the PID-GS with a feedforward term.
     • Logic: The feedforward term ($u_{FF}$) reacts immediately to the measured change in cell density ($f_{RMF}$) to compensate for the disturbance before it creates a tracking error, while the feedback loop corrects residual errors.

3. Key Contributions

1. Demonstration of Growth-Dependent Constraints: The paper provides quantitative evidence (via Bode analysis) that fixed-gain controllers cannot achieve robust disturbance rejection across the entire batch trajectory due to shifting plant dynamics.
2. Novel Controller Architectures:
   • Development of a PID-GS that adapts to cellular physiological states.
   • Development of a PID-GS-FF that explicitly compensates for growth perturbations using real-time population data.
3. Evaluation Framework: Introduction of a composite scalar performance metric ($M$) that penalizes both slow settling time and sustained tracking error (ITAE). This framework allows for systematic comparison across a grid of disturbance magnitudes and timing.

4. Results

The study compared the three controllers (Fixed-PID, PID-GS, PID-GS-FF) across varying disturbance magnitudes ($\rho$) and disturbance times ($t_d$).

• Fixed-Gain PID: Performed poorly across the board. It caused significant overshoot in mid-log phases (due to high system gain) and slow recovery in stationary phases (due to low system responsiveness).
• PID-GS (Feedback Only):
  • Performance: Significantly reduced overshoot in intermediate disturbances ($\rho = 40-60\%$) by lowering controller gains during high-resource availability phases.
  • Limitation: Under large disturbances ($\rho = 80\%$), the reduced gain led to slower recovery times compared to the fixed-gain controller.
• PID-GS-FF (Feedback + Feedforward):
  • Performance: Achieved the best overall performance, particularly for large disturbances ($\rho = 70-80\%$).
  • Mechanism: The feedforward term provided rapid, proactive signal recovery immediately after the disturbance, while the gain-scheduled feedback prevented the overshoot associated with the sudden resource surge.
• Operating Regimes: The analysis identified three distinct regimes:
  1. Low Disturbance: All controllers perform similarly.
  2. Intermediate Disturbance: PID-GS is optimal (suppresses overshoot).
  3. High Disturbance: PID-GS-FF is optimal (fast recovery + stability).

5. Significance and Conclusion

• Scientific Impact: This is the first study to design a growth-aware disturbance rejection controller for gene expression in bacterial batch cultures. It moves beyond the "steady-state" assumption common in synthetic biology control.
• Practical Utility: The results provide a clear guideline for controller selection in bioprocessing. For robust regulation in batch cultures, controllers must account for the shifting operating point of the biological plant.
• Future Directions: The authors propose experimental validation using the LEMOS platform, requiring real-time estimation of the scheduling variable (OD600) and automated logic to switch between control strategies based on disturbance magnitude.

In summary, the paper establishes that robust control of living systems with shifting operating conditions requires dynamic adaptation of controller parameters, specifically through gain scheduling and feedforward mechanisms, to overcome the inherent limitations of fixed-gain approaches in batch bioprocesses.