The Four Candidate Causes of a pH Excursion
pH in a mammalian cell culture bioreactor is controlled by CO₂ addition to lower pH and base (typically sodium bicarbonate or Na₂CO₃) addition to raise pH. The setpoint for most CHO-based mAb processes is in the 6.9–7.2 range; acceptable deviation bands are typically ±0.1–0.15 pH units depending on the cell line. When the measured pH drifts outside the control band, the deviation investigation needs to distinguish between four candidate cause categories:
- Control loop fault — the controller is not responding correctly to the deviation signal (PID tuning, actuator issue, controller hardware)
- Probe calibration error — the measured pH value does not reflect the true culture pH (probe fouling, electrode drift, calibration offset)
- Feed strategy effect — the pH control system is responding correctly, but a feed addition event changed the buffering capacity or acid/base loading faster than the controller could compensate
- Metabolic cause — the culture's acid production rate (lactate, CO₂) or base consumption rate changed due to a metabolic state shift
These four candidates have meaningfully different CAPA implications. A control loop fault requires an engineering CAPA targeting the control system. A probe error requires a sensor management CAPA (calibration intervals, probe qualification protocol). A feed strategy effect may require a process development CAPA (revised feed addition rate or timing). A metabolic cause may require a process development CAPA targeting the upstream process (seed train, media, inoculum density). Misclassifying any of these results in a CAPA that does not address the actual mechanism.
Step 1: Controller Output Before Probe Calibration
The first investigation step — and the step most frequently skipped in informal investigations — is to plot the controller output signal alongside the measured pH. The controller output is the CO₂ valve position (0–100%) and the base pump activation signal. If pH is drifting low (acidic excursion) and the CO₂ valve is already at 0% (fully closed), the controller is responding correctly to the deviation — it has closed the CO₂ supply and the pH is still falling. This tells you the controller is not the fault; the acid loading from the culture exceeds the controller's ability to compensate by reducing CO₂ alone. The investigation moves to feed strategy or metabolic causes.
If pH is drifting low and CO₂ remains partially or fully open when it should be closing, the controller is not responding — this is a control loop fault, and the investigation focuses on the control system: is the PID control loop configured correctly? Did the controller receive the process variable (PV) signal? Is the CO₂ valve responding to the commanded position?
For batch CHOP-2025-062, a 500L CHO fed-batch process at a Phase 2 vaccine candidate program, pH began drifting from 7.10 downward at hour 55. The CO₂ valve position tag showed 0% throughout the drift — the controller had closed CO₂ completely. The base pump activation record showed intermittent activation at 15-minute intervals with volumetric additions of 2–3 mL each. The controller was functioning and responding; the acid load was exceeding the controller's capacity. This moved the investigation to the feed event at hour 52.
Step 2: Probe Calibration as a Parallel Check
Probe calibration should be evaluated in parallel with the controller output check, not sequentially. The key data points are: the last in-process calibration timestamp (available in the batch record), the two-point calibration offset at the last calibration, and whether offline pH measurements taken during the excursion window agree with the probe reading.
If an offline pH measurement taken during the excursion (e.g., from a Mettler Toledo handheld pH meter on a daily sample) agrees with the in-line probe reading within 0.05 pH units, the probe is provisionally validated and the investigation can proceed to controller and process causes. If the offline measurement shows a 0.10 pH unit or greater disagreement with the in-line probe, probe accuracy is compromised and any pH-based control action during the deviation window is suspect.
A probe calibration error that biases the measured pH low will cause the controller to add base continuously, potentially driving the culture to a higher-than-intended pH. A probe calibration error that biases high will allow the culture pH to fall without triggering control action. Both patterns produce investigation confusion if the probe condition is not assessed early.
We are not saying probe calibration is the most common cause of pH excursions — it is not. The most common cause in a well-maintained process is feed-related pH loading or culture metabolic state changes. The reason probe calibration should be evaluated early is that it is the fastest hypothesis to close: one offline pH measurement either confirms or refutes the probe as a factor. Closing it quickly narrows the remaining investigation scope.
Step 3: Feed Event Correlation
For fed-batch processes, pH excursions that coincide with a feed addition window — typically within 0–6 hours of a bolus feed event — should be examined for feed-strategy causation before metabolic causes are evaluated. Bolus glucose additions can transiently lower pH by increasing the carbon loading and driving lactate production. The magnitude depends on the feed volume, the glucose concentration in the feed, and the culture's metabolic state at the time of addition.
The data sequence is: pull the feed addition log from the batch record (pump volume × concentration × timing), overlay the lactate trend if in-line or near-line lactate is available, and calculate whether the observed pH drop is consistent in magnitude and timing with the documented feed event. A 50 mL bolus glucose feed into a 500L bioreactor at 400 g/L concentration increases the culture glucose by approximately 0.4 g/L — enough to drive a measurable transient increase in OUR and corresponding CO₂ production.
For batch CHOP-2025-062, the hour-52 feed event was a 55 mL bolus glucose addition at the scheduled day-3 feed time. The lactate trend, from daily offline samples, showed an increase from 1.8 g/L at hour 48 to 3.1 g/L at hour 72. The timing and magnitude of the pH drop (7.10 to 6.91 over 8 hours) were consistent with the metabolic response to the glucose bolus combined with the active exponential growth phase, where OUR is highest and CO₂ evolution correspondingly high. The deviation was classified as a feed-strategy-driven pH excursion during an expected high-metabolic-demand window.
The CAPA Fork: Process Development vs. Engineering
The classification determines the CAPA fork. A feed-strategy cause creates two options: accept the current feed strategy with updated pH alert limits that account for the post-feed transient, or revise the feed strategy to reduce the bolus glucose volume and supplement with more frequent smaller additions to avoid the metabolic spike. The choice depends on whether the excursion impacted product quality attributes and on the clinical stage of the program.
An engineering CAPA for a control loop fault has a different structure: identify the fault component, replace or reconfigure, verify controller response in a non-GMP validation run or with a controlled process hold test, and document the qualification evidence. This CAPA can be closed faster than a feed-strategy CAPA, which may require a formal process change and potentially an additional development run to confirm the revised strategy before the next clinical batch.
The investigation record needs to document not only the classified cause but the reason the other candidate causes were ruled out. An inspector reviewing a deviation record for a pH excursion should see: the controller output trend confirming or ruling out loop fault, the probe calibration data confirming or flagging probe condition, the feed event correlation, and the metabolic data. The classification is the conclusion; the data is the evidence. Without the evidence, the classification is not defensible under 21 CFR 211.192.
For programs where pH excursions are recurring across multiple batches, the Fermentile analytics platform allows the MSAT team to compare pH profiles across the batch population, flagging whether the excursion pattern is isolated or systematic. A systematic pattern — pH drifting in the same direction at the same process hour across multiple batches — is a process characterization finding that typically belongs in a CPV-level trending review rather than an individual batch deviation. Making that distinction correctly determines whether the CAPA is batch-scoped or program-scoped, which has significant implications for validation status and change control under the Fermentile deviation management workflow.
Misclassification Patterns to Watch For
Three misclassification patterns are common in pH excursion investigations. The first is attributing a feed-strategy excursion to "culture metabolic shift" without linking the metabolic shift to its upstream cause — the feed event. "Culture metabolic shift" is not a controllable cause; "bolus glucose feed driving transient lactate spike exceeding CO₂ control capacity" is. The CAPA implications are different.
The second pattern is using the probe calibration check as a closing step rather than an early parallel check. When the probe calibration is found to be within specification at the end of the investigation, the record shows that calibration was verified — but it does not show that the verification happened early enough to inform the investigation direction. Regulators reviewing the investigation record want to see the verification sequence, not just the outcome.
The third pattern is closing the control loop branch without documenting the specific controller output data reviewed. "Controller reviewed, no fault found" is not an evidence statement. "CO₂ valve position tag BIO-02-CO2-POS reviewed from 14:00–22:00; valve at 0% from 15:30 onward consistent with correct controller response to low pH error signal" is an evidence statement. The second version survives regulatory scrutiny; the first does not.