Dissolved carbon dioxide is the most frequently underestimated scale-up challenge in mammalian cell culture. A CHO fed-batch process that runs comfortably at 40–60 mmHg pCO₂ in a 2 L bioreactor can reach 150–220 mmHg at manufacturing scale, silently eroding growth rate, product titer, and glycosylation quality. Unlike dissolved oxygen, which triggers an alarm the moment DO drops below the setpoint, dissolved CO₂ accumulation is gradual, often invisible without dedicated sensors, and only apparent in hindsight when the batch underperforms.
This article covers the scale-up engineering of dissolved CO₂ control: why pCO₂ rises with vessel size, the thermodynamics behind the accumulation, sensor technologies for real-time measurement, four stripping strategies with quantitative guidance, and a worked example from 2 L to 2,000 L. For a broader overview of pCO₂ process control in cell culture, see the companion guide on dissolved CO₂ (pCO₂) control in mammalian cell culture.
Why Dissolved CO₂ Accumulates at Large Scale
Dissolved CO₂ accumulates at manufacturing scale because the physics of gas–liquid mass transfer shift against CO₂ removal as vessel dimensions increase. Four mechanisms drive this accumulation, and all four worsen simultaneously with scale.
1. Decreased Surface-Area-to-Volume Ratio
The headspace–liquid interface is the primary path for CO₂ removal at bench scale. A 2 L vessel with a 10 cm diameter has a surface-area-to-volume (SA/V) ratio of roughly 40 m−1. At 2,000 L (about 1.3 m diameter), SA/V drops to approximately 4 m−1. This 10-fold reduction means that surface gas exchange, which dominated CO₂ stripping at small scale, becomes negligible at large scale.
2. Hydrostatic Pressure Increases CO₂ Solubility
Henry’s law dictates that gas solubility increases with partial pressure. A tall 2,000 L bioreactor has a liquid height of roughly 1.5–2.0 m, adding 0.15–0.20 bar of hydrostatic pressure at the sparger. This raises the equilibrium dissolved CO₂ concentration by 15–20% at the bottom of the vessel relative to the surface. At 15,000 L or above, liquid heights of 3–4 m increase bottom-of-vessel pCO₂ by 30–40%.
3. Gas Residence Time and Bubble Saturation
As vessel height increases, bubbles spend more time rising through CO₂-rich medium. In a 2 L vessel, a bubble traverses the liquid in 1–2 seconds. In a 2,000 L vessel, the path is 5–8 seconds. During this transit, bubbles absorb CO₂ from the surrounding liquid and approach saturation. Once saturated, a bubble no longer strips CO₂. At manufacturing scale, Xing et al. (2017) showed that air sparge bubbles can reach 80–90% CO₂ saturation before exiting the liquid, dramatically reducing their stripping capacity.
4. Reduced Mixing Efficiency
Large-scale bioreactors develop spatial heterogeneity in dissolved gas concentrations. CO₂-rich zones form near regions of high cell density or near the bottom of the vessel where hydrostatic pressure is highest. Mixing times increase from 5–10 seconds at 2 L to 30–60 seconds at 2,000 L and 60–120 seconds at 15,000 L, meaning cells in poorly mixed regions are exposed to elevated pCO₂ for extended periods between mixing events.
Henry’s Law and the CO₂ Equilibrium System
The dissolved CO₂ concentration in cell culture medium is governed by Henry’s law and the carbonate buffer equilibrium. Understanding these two relationships is essential for predicting pCO₂ at any scale and designing effective stripping strategies.
Henry’s Law for CO₂
Henry’s law states that the equilibrium concentration of a dissolved gas is proportional to its partial pressure above the liquid:
CCO₂ = HCO₂ × pCO₂
At 37 °C in aqueous media, the Henry’s law constant for CO₂ is HCO₂ ≈ 0.034 mol/L/atm (or equivalently 2.3 × 10−4 mol/L/mmHg). This means CO₂ is roughly 30 times more soluble than O₂ (HO₂ ≈ 0.0013 mol/L/atm at 37 °C), which is why dissolved CO₂ accumulation is a bigger problem than oxygen supply at large scale.
The Carbonate Buffer Equilibrium
Once dissolved, CO₂ enters a rapid equilibrium with carbonic acid and bicarbonate:
CO₂(aq) + H₂O ↔ H₂CO₃ ↔ HCO₃⁻ + H⁺
The first dissociation constant is pKa1 = 6.35 at 37 °C. At the typical cell culture pH of 7.0–7.4, the Henderson–Hasselbalch equation shows that approximately 95% of total dissolved inorganic carbon exists as bicarbonate (HCO₃⁻), with only about 5% as dissolved CO₂. This ratio has a critical implication: only the dissolved CO₂ fraction participates in gas–liquid transfer. The bicarbonate pool acts as a reservoir that continuously replenishes dissolved CO₂ as stripping removes it, making CO₂ removal inherently slower than O₂ supply.
Most cell culture media contain 20–26 mM sodium bicarbonate (NaHCO₃) as pH buffer. This deliberately maintains a large dissolved carbon pool. Goudar et al. (2007) demonstrated that eliminating bicarbonate addition reduced pCO₂ accumulation by 30–50% in high-cell-density CHO cultures, though this requires alternative pH control strategies such as HEPES or increased base addition.
Worked Example: Equilibrium pCO₂ from Bicarbonate
Given: Cell culture medium at pH 7.2 with 22 mM NaHCO₃, 37 °C.
Step 1: Calculate [CO₂] from Henderson–Hasselbalch:
pH = pKa1 + log([HCO₃⁻] / [CO₂])
7.2 = 6.35 + log(22 mM / [CO₂])
log(22 / [CO₂]) = 0.85 → 22 / [CO₂] = 7.08
[CO₂] = 3.11 mM = 3.11 × 10⁻³ mol/L
Step 2: Calculate equilibrium pCO₂ from Henry’s law:
pCO₂ = [CO₂] / HCO₂
pCO₂ = 3.11 × 10⁻³ / 0.034 = 0.091 atm
pCO₂ = 0.091 × 760 = 69 mmHg
Result: At pH 7.2 with 22 mM bicarbonate, the equilibrium pCO₂ is approximately 69 mmHg. This is the baseline pCO₂ from the buffer alone, before cells add any metabolic CO₂. At peak cell density (15–20 × 10₆ cells/mL), metabolic CO₂ production raises steady-state pCO₂ to 100–180 mmHg depending on the stripping capacity of the system.
How to Measure pCO₂ in Bioreactors
Real-time pCO₂ monitoring is the prerequisite for effective CO₂ control. Without it, pCO₂ accumulation is only discovered post-run from off-line blood gas data, when it is too late to adjust. Four measurement approaches are available, each with distinct trade-offs between accuracy, response time, and single-use compatibility.
| Technology | Example | Range | Response (t90) | Accuracy | Sterilizable | Single-use |
|---|---|---|---|---|---|---|
| Severinghaus (electrochemical) | Mettler InPro 5000i | 1–400 mmHg | 3–5 min | ±5 mmHg | Autoclavable | No |
| Optical fluorescence | PreSens SP-CD1 | 10–250 hPa | 2–4 min | ±5% of reading | Gamma-stable | Yes (patch) |
| Off-gas derived | Calculated from exhaust CO₂ | Indirect | Depends on gas flow | ±10–20% | N/A | N/A |
| Blood gas analyzer (off-line) | Radiometer ABL / Nova BioProfile | 5–800 mmHg | 60 s (measurement) | ±2 mmHg | N/A | N/A |
Severinghaus sensors use a CO₂-permeable membrane over an internal electrolyte coupled to a pH electrode. As dissolved CO₂ diffuses through the membrane, it acidifies the electrolyte, and the pH change is converted to pCO₂. The Mettler Toledo InPro 5000i is the most widely deployed in-situ pCO₂ probe in bioprocessing. Its main limitations are a 3–5 minute t90 response time and periodic recalibration every 1–2 weeks due to electrolyte drift.
Optical fluorescence sensors such as the PreSens SP-CD1 use a pH-sensitive fluorescent dye behind a gas-permeable membrane. CO₂ diffusing through the membrane changes the dye’s fluorescence, which is read optically through the vessel wall. These sensors are non-invasive, require no recalibration during a run, and are compatible with single-use bioreactors. Their main limitation is a narrower measurement range (typically up to 250 hPa ≈ 188 mmHg), which may not cover the full pCO₂ range in poorly controlled large-scale cultures.
Off-gas analysis provides an indirect estimate by measuring CO₂ concentration in the exhaust gas and back-calculating dissolved CO₂ using Henry’s law. This approach is simple and non-invasive but assumes the headspace and liquid are in equilibrium, which is often not the case at high cell densities where metabolic CO₂ production exceeds stripping capacity. Accuracy degrades to ±20% or worse under these conditions.
Off-line blood gas analyzers (Radiometer ABL series, Nova BioProfile) provide the most accurate pCO₂ measurement but require manual sampling, introducing a 5–15 minute lag and making real-time control impossible. They serve as the reference for calibrating in-situ sensors.
CO₂ Stripping Strategies: Engineering the Driving Force
CO₂ stripping works by maximizing the concentration gradient between dissolved CO₂ in the liquid and CO₂ in the gas phase. Four engineering approaches are available, often used in combination. The fundamental equation for CO₂ mass transfer rate is:
dCO₂/dt = kLaCO₂ × (CCO₂ − C*CO₂)
where kLaCO₂ is the volumetric CO₂ mass transfer coefficient, CCO₂ is the dissolved CO₂ concentration, and C*CO₂ is the equilibrium concentration determined by the CO₂ partial pressure in the gas phase. Reducing C*CO₂ (diluting gas-phase CO₂) or increasing kLaCO₂ (more bubbles, better mixing) both accelerate stripping.
| Strategy | Mechanism | Typical rate | pCO₂ reduction | Key limitation |
|---|---|---|---|---|
| Overlay gas sweep | N₂ or air across headspace reduces gas-phase pCO₂ | 0.05–0.2 VVM headspace | 15–30% | Limited by SA/V ratio at large scale |
| Macrosparging | Large bubbles (3–5 mm) absorb CO₂ during rise | 0.01–0.03 VVM | 30–50% | Foam generation; may strip DO |
| Increased agitation | Higher impeller speed increases surface renewal rate | Tip speed ≤1.5 m/s | 10–20% | Shear damage above 1.5 m/s tip speed |
| Low-bicarbonate medium | Reduces dissolved carbon pool and equilibrium pCO₂ | 0–12 mM NaHCO₃ | 30–50% of baseline | Requires alternative pH buffer (HEPES, increased base) |
Overlay Gas Sweep
Flowing N₂ or air through the headspace at 0.05–0.2 VVM continuously removes CO₂ from the gas above the liquid surface. This reduces the headspace pCO₂ and increases the driving force for CO₂ transfer from liquid to gas at the surface. At bench scale, overlay alone is often sufficient because the high SA/V ratio provides adequate surface area. At 2,000 L and above, the diminished SA/V ratio limits the effectiveness of overlay to a 15–30% pCO₂ reduction, making it a necessary but insufficient strategy on its own.
Macrosparging for CO₂ Stripping
Macrosparging with bubbles of 3–5 mm diameter is the most effective submerged CO₂ stripping method for mammalian cultures. Large bubbles rise faster (15–25 cm/s terminal velocity) than microsparger bubbles (0.5–2 mm, 5–15 cm/s), giving them less time to reach CO₂ saturation before exiting the liquid. This means each large bubble removes more CO₂ per unit volume of gas than a small bubble that has equilibrated. The trade-off is that large bubbles provide less oxygen transfer per unit gas volume than microsparger bubbles, so macrosparging is typically used alongside microsparging: the microsparger provides O₂, and the macrosparger strips CO₂.
The relationship between kLa for CO₂ and O₂ follows from the penetration theory:
kLaCO₂ / kLaO₂ = √(DCO₂ / DO₂) ≈ √(1.6 / 2.1) ≈ 0.87
Experimental values typically range from 0.80 to 0.90. This means that the same bubble population and aeration conditions strip CO₂ about 10–20% less effectively than they deliver O₂.
Increased Agitation
Raising impeller speed improves CO₂ stripping by increasing the surface renewal rate at the gas–liquid interface and reducing boundary layer thickness. However, mammalian cells are shear-sensitive, and tip speeds above 1.5 m/s risk cell damage and lysis. In practice, agitation can contribute a 10–20% reduction in pCO₂ within the safe shear window, making it a useful supplement to sparging but not a primary stripping strategy.
Low-Bicarbonate Medium
Reducing or eliminating sodium bicarbonate from the medium directly lowers the total dissolved inorganic carbon pool, reducing the equilibrium pCO₂. Goudar et al. (2007) demonstrated that replacing 24 mM NaHCO₃ with HEPES buffer in a perfusion CHO culture reduced steady-state pCO₂ from 120 mmHg to 60 mmHg at the same cell density. The trade-off is that HEPES is expensive at manufacturing scale ($50–100/kg vs $0.50–1.00/kg for NaHCO₃), and pH control must rely more heavily on NaOH/HCl addition, which raises osmolality over the course of a fed-batch.
A Worked Scale-Up: From 2 L to 2,000 L
The following worked example predicts steady-state pCO₂ at manufacturing scale and demonstrates the effect of stripping strategies. The numbers are representative of a typical CHO mAb fed-batch process.
Worked Example: Predicting pCO₂ at 2,000 L
Given parameters:
- Cell density at peak: 15 × 10₆ cells/mL
- Specific CO₂ production rate: qCO₂ = 3.5 × 10⁻⁹ mmol/cell/h
- Medium bicarbonate: 22 mM NaHCO₃, pH 7.2
- Temperature: 37 °C (HCO₂ = 0.034 mol/L/atm)
- Working volume: 1,500 L (in a 2,000 L vessel)
Step 1: CO₂ production rate
QCO₂ = qCO₂ × X × V
QCO₂ = 3.5 × 10⁻⁹ mmol/cell/h × 15 × 10⁹ cells/L × 1,500 L
QCO₂ = 3.5 × 15 × 1,500 × 10⁻⁹⁺⁹ mmol/h
QCO₂ = 78,750 mmol/h = 78.8 mol/h
Step 2: CO₂ stripping rate without active measures
At 2,000 L with standard air microsparging at 0.01 VVM for DO control only:
kLaCO₂ (microsparger only) ≈ 2–4 h⁻¹
Assume kLaCO₂ = 3 h⁻¹
At steady state: QCO₂ = kLaCO₂ × V × (CCO₂ − C*)
C* ≈ 0 (if headspace pCO₂ kept low by overlay)
CCO₂ = QCO₂ / (kLaCO₂ × V)
CCO₂ = 78.8 mol/h / (3 h⁻¹ × 1,500 L)
CCO₂ = 0.0175 mol/L
pCO₂ = CCO₂ / HCO₂ = 0.0175 / 0.034 = 0.515 atm
pCO₂ ≈ 391 mmHg
This is clearly too high. In practice, the headspace CO₂ would build up, reducing the driving force and reaching a lower steady state of approximately 180–220 mmHg (since C* is not zero but increases with headspace CO₂). The exact value depends on the headspace ventilation rate, but without active stripping, pCO₂ far exceeds the 150 mmHg action limit.
Step 3: With macrosparging + overlay
Add macrosparger at 0.02 VVM air (30 L/min)
Add N₂ overlay at 0.1 VVM headspace (50 L/min)
Combined kLaCO₂ ≈ 8–12 h⁻¹
Assume kLaCO₂ = 10 h⁻¹
CCO₂ = 78.8 / (10 × 1,500) = 0.00525 mol/L
pCO₂ = 0.00525 / 0.034 = 0.154 atm
pCO₂ ≈ 117 mmHg
With moderate overlay, this reduces further to approximately 80–100 mmHg, well within the acceptable range for CHO cell culture.
For comparison, at 2 L bench scale with the same cell density: surface gas exchange dominates (SA/V ≈ 40 m⁻¹), effective kLaCO₂ ≈ 20–40 h⁻¹, giving a steady-state pCO₂ of 40–60 mmHg. The 3–4× difference between 2 L and 2,000 L is the scale-up gap this article addresses.
Gas Mixing Calculator
Calculate the N₂/air/O₂ blend for simultaneous CO₂ stripping and DO control at any bioreactor scale.
Biological Impact of Elevated pCO₂
Elevated dissolved CO₂ affects CHO cell growth, productivity, and product quality through a well-characterized mechanism. CO₂ freely diffuses across the cell membrane. Inside the cell, carbonic anhydrase converts CO₂ to carbonic acid, lowering intracellular pH (pHi). This pHi drop propagates to the Golgi apparatus, where it reduces the activity of glycosyltransferase enzymes responsible for galactosylation and sialylation. Brunner et al. (2017) showed that pH and pCO₂ interact significantly, making it essential to consider both parameters together during process characterization.
| Parameter | 40 mmHg | 80 mmHg | 120 mmHg | 160 mmHg | 200 mmHg |
|---|---|---|---|---|---|
| Specific growth rate μ (% of control) | 100 | 95 | 82 | 65 | 48 |
| Specific productivity qP (% of control) | 100 | 98 | 90 | 78 | 62 |
| G0F glycoform (%) | 45 | 48 | 55 | 62 | 68 |
| Galactosylation G1F+G2F (%) | 38 | 35 | 28 | 22 | 18 |
| Acidic charge variants (%) | 18 | 20 | 25 | 32 | 38 |
The data show a threshold effect: below 80 mmHg, performance is largely unaffected. Between 80 and 120 mmHg, growth and quality begin to shift measurably. Above 120 mmHg, the effects accelerate. The common industry action limit of 150 mmHg corresponds to the inflection point where quality impact becomes difficult to compensate for downstream.
Zhu et al. (2005) reported that elevated pCO₂ and osmolality interact synergistically. In fed-batch cultures, osmolality often rises to 350–400 mOsm/kg from nutrient additions and base titration. When high pCO₂ (>150 mmHg) coincides with high osmolality (>380 mOsm/kg), the combined growth inhibition exceeds the sum of individual effects. This synergy is another reason why CO₂ control at scale matters beyond the pCO₂ number alone.
Scale-Up Calculator
Translate your bench-scale parameters (P/V, kLa, tip speed, VVM) to production scale and identify CO₂ accumulation risk.
DOE for Optimizing CO₂ Stripping at Scale
Optimizing CO₂ stripping is a multivariate problem. Overlay flow rate, macrosparger VVM, agitation RPM, and bicarbonate concentration all interact, and each one affects not only pCO₂ but also DO, foam, and shear. A design of experiments (DOE) approach is far more efficient than one-factor-at-a-time testing for mapping these interactions.
A practical DOE for CO₂ stripping optimization would include:
- Factors: overlay gas flow (0–0.2 VVM), macrosparger VVM (0–0.03), agitation RPM (80–200), NaHCO₃ concentration (0–22 mM)
- Responses: steady-state pCO₂, DO setpoint maintenance, peak VCD, final titer, G0F%, galactosylation %, acidic variants %
- Design: A definitive screening design (DSD) with 4 factors requires only 13 runs and captures main effects, quadratics, and some two-factor interactions
The DOE results enable construction of a response surface model that predicts pCO₂ as a function of the four factors, allowing engineers to find the operating point that minimizes pCO₂ while maintaining DO control and staying within shear limits. For QbD filing, DOE Pro generates the design space boundary diagram showing the proven acceptable ranges for each parameter.
DOE Generator
Design an experiment to optimize CO₂ stripping parameters. The free tool handles screening and optimization designs; DOE Pro adds D-optimal, split-plot, Bayesian optimization, and the QbD design space report.
Frequently Asked Questions
What is an acceptable pCO₂ range for mammalian cell culture?
Most CHO and HEK293 processes target 30–100 mmHg during the growth phase and tolerate gradual accumulation to 120–150 mmHg at peak cell density. Growth inhibition becomes measurable above 100–120 mmHg, with a 30–50% reduction in specific growth rate at 160–200 mmHg. Glycosylation shifts appear above 120 mmHg. An action limit of 150 mmHg is common, though cell-line-specific tolerance should be established during process characterization.
How does CO₂ stripping differ from O₂ transfer in bioreactors?
CO₂ stripping is harder for three reasons. First, CO₂ is about 30 times more soluble than O₂ at 37 °C. Second, dissolved CO₂ exists in equilibrium with bicarbonate; only the free CO₂ fraction (about 5% at pH 7.2) participates in gas–liquid transfer. Third, kLaCO₂ is approximately 0.85× kLaO₂ because CO₂ diffuses more slowly. These factors together mean CO₂ accumulates faster than O₂ depletes at scale.
Can you use pure N₂ for CO₂ stripping in bioreactors?
Yes, pure N₂ maximizes the driving force since incoming bubbles contain zero CO₂. However, N₂ simultaneously strips dissolved oxygen. In practice, N₂ is used as an overlay gas (headspace sweep) rather than a submerged sparge, or blended with air/O₂ so the oxygen-containing portion maintains DO while N₂ dilutes headspace CO₂. A gas mixing calculator helps balance the blend.
Why is kLaCO₂ lower than kLaO₂?
The volumetric mass transfer coefficient depends on liquid-phase diffusivity. CO₂ diffuses more slowly than O₂: DCO₂ ≈ 1.6 × 10⁻⁵ cm²/s versus DO₂ ≈ 2.1 × 10⁻⁵ cm²/s at 37 °C. By penetration theory, kLaCO₂/kLaO₂ = √(DCO₂/DO₂) ≈ 0.87. Experimental measurements in stirred bioreactors report 0.80–0.90.
Does elevated pCO₂ affect all cell lines equally?
No, pCO₂ sensitivity varies across cell lines and clones. CHO-K1 and CHO-DG44 generally tolerate up to 120 mmHg with modest effects. Some HEK293 lines show growth inhibition above 80 mmHg. NS0 cells tend to be more sensitive than CHO. Within CHO, high-producing clones may be more sensitive. Cell-line-specific tolerance should be established during clone selection using qualified scale-down models that mimic manufacturing-scale pCO₂ profiles.
Related Tools
- Gas Mixing Calculator — Calculate N₂/air/O₂ blend ratios for simultaneous CO₂ stripping and DO control
- Scale-Up Calculator — Translate P/V, kLa, tip speed, and VVM from bench to production scale
- OTR/kLa Estimator — Estimate oxygen transfer rate and kLa from bioreactor parameters
References
- Xing Z, Lewis AM, Borys MC, Li ZJ. A carbon dioxide stripping model for mammalian cell culture in manufacturing scale bioreactors. Biotechnol Bioeng. 2017;114(6):1184–1194. doi:10.1002/bit.26232
- Brunner M, Fricke J, Kroll P, Herwig C. Investigation of the interactions of critical scale-up parameters (pH, pO2 and pCO2) on CHO batch performance and critical quality attributes. Bioprocess Biosyst Eng. 2017;40(2):251–263. doi:10.1007/s00449-016-1693-7
- Goudar CT, Matanguihan R, Long E, Cruz C, Zhang C, Piret JM, Konstantinov KB. Decreased pCO2 accumulation by eliminating bicarbonate addition to high cell-density cultures. Biotechnol Bioeng. 2007;96(6):1107–1117. doi:10.1002/bit.21116
- Zhu MM, Goyal A, Rank DL, Gupta SK, Vanden Boom T, Lee SS. Effects of elevated pCO2 and osmolality on growth of CHO cells and production of antibody-fusion protein B1: a case study. Biotechnol Prog. 2005;21(1):70–77. doi:10.1021/bp049815s
- Mostafa SS, Gu X. Strategies for improved dCO2 removal in large-scale fed-batch cultures. Biotechnol Prog. 2003;19(1):45–51. doi:10.1021/bp0256263