Beyond setpoints: why bioreactor control strategies matter for reproducibility
Reproducibility starts where setpoints end: in the control strategy that defines how your process actually runs
Dirk Hebel
Global Head of Product Management, Infors HT
A process that worked once is repeated months later. Same cell line, same medium, same recorded setpoints. The result is different. Biology and materials explain many of these cases. But another candidate tends to go unexamined because it was never written down: how the bioreactor was configured to reach those setpoints.
A setpoint describes a target. It says nothing about the route. A dissolved oxygen target of 50 percent does not tell you whether the controller raised agitation first, opened the gas flow, or enriched with oxygen, nor in what proportion.
PID gains determine how hard the controller reacts and whether it overshoots. A deadband decides when it reacts. Calibration history determines what the measurement meant. All shape the conditions the cells experienced, and none appear in a protocol that lists targets.
How much does it matter? Harcum and colleagues published a CHO fed-batch study in 2022 that puts a number on one case. Their preliminary run reached about 1.3 g/L; by the third tuning experiment they were at about 2.0. They had adjusted PID gains and actuator operating ranges, so this is not a clean before-and-after on tuning alone. Their conclusion: these settings belong in the reported process description.
The practical cost shows up at a handoff. A process moves to another operator, vessel, or site. If the protocol carries only targets, the receiving team rebuilds the configuration from memory, email and whatever the last person wrote down. That is where mismatches enter. It also makes the inevitable question hard to answer: when the second site gets a different result, was it the biology, the scale, or were the systems never set up the same way?
Documentation discipline alone does not fix this, because the thing to document is large and changes often. It becomes tractable when configuration is part of the run rather than a description written afterwards: controller settings saved as reusable presets, tied to the batch strategy, stored with the data instead of beside it.
A recent perfusion cultivation run with FHNW using the INFORS HT Multifors 3 bench-top bioreactor and eve® bioprocess software shows what that looks like. NIST CHO cells, 1.7 L working volume, a tangential flow filtration loop, 23 days with 15 at steady state. The control strategy was not only DO and pH. Volume and feed were controlled gravimetrically using balances, and cell bleeding was automated from an inline capacitance signal. Those elements often sit outside the bioreactor record, handled by a spreadsheet or operator decision. Here they were part of one defined strategy.
The run also illustrated the original point. Capacitance held its setpoint while offline cell count gradually declined because the cells were growing larger. The controlled variable was stable; the thing people assume it represents was not. Knowing which variable you are holding, and what it does and does not tell you, is part of the strategy too. One cultivation cannot demonstrate reproducibility, and it is not offered as that. What it demonstrates is scope: variables shaping a process can sit inside a single defined, recorded strategy. Three things worth taking into the next experiment. Define how the process is controlled, not only where it is held. Keep that definition together with the data. And when evaluating any platform, ask what it captures, not only what it controls.
This is a guest blog. Its author/s is/are responsible for content within it, which does not necessarily reflect the opinions or positions of BIA.
