Taking a bioprocess from the laboratory to commercial manufacturing involves far more than increasing vessel size. At larger scales, mixing, oxygen transfer, heat removal, feed distribution, reactor geometry and control systems can create conditions that are difficult to reproduce at development scale. Understanding these scale-dependent effects is critical to achieving consistent process performance, product quality and manufacturing robustness.
In this conversation, Mr Prashant Chawla, Sr. General Manager, Manufacturing at Biological E. Limited, shares practical insights from his experience in bioprocess scale-up, technology transfer, process validation and greenfield and brownfield projects. The discussion explores scale-defining parameters, scale-down models, process and equipment fit, troubleshooting unexpected scale-up behaviour, and the emerging role of PAT, AI/ML and predictive modelling in commercial bioprocessing.
1. At commercial scale, how much of a bioprocess’s behavior can realistically be predicted from laboratory and pilot-scale data? Which variables or scale-dependent phenomena become inherently less predictable as scale increases, and how should development teams establish the limits of their scale-up confidence before entering commercial manufacturing?
Honestly, less than most of us would like. Lab data is a great starting point, but I’d never treat it as proof that a process will behave the same way in a commercial vessel. The moment you scale up, the fluid environment stops being uniform. Mixing slows down, oxygen transfer gets harder, heat removal behaves differently, and the cells start experiencing local pockets of varying pH, DO, nutrients, CO₂, temperature and shear.
A handful of variables shift noticeably with scale:
- Mixing time. The time it takes to distribute feed, acid, base or gas through the vessel generally grows as the vessel gets bigger.
- Oxygen transfer. Your kLa, the volumetric oxygen transfer coefficient, can change because of differences in impeller design, gas flow, pressure, bubble size and power input.
- Power per volume. Holding power per unit volume constant often becomes impractical, whether because of equipment limitations, shear concerns or simply excessive energy consumption.
- Heat transfer. Temperature control tends to get sluggish because the surface-area-to-volume ratio drops as the vessel grows.
- Feed distribution. A concentrated feed can create local zones of high substrate concentration before it’s fully mixed in.
- Sensors and control. A single probe may not represent the whole vessel, especially once gradients show up.
OTR deserves special mention. At lab scale it’s easy to maintain, but a commercial vessel can be quietly running close to its oxygen transfer limit even while the controller settings look perfectly normal.
Here’s a picture I often use. Say you’re running a cell culture with a DO setpoint of 30%. In a lab bioreactor, everything mixes in a few seconds. In a large fermenter, though, a cell may be travelling from low-oxygen zones into well-aerated regions and back again. Your probe happily reads 30%, but at the level of the individual cell, those repeated swings can affect growth, metabolism and ultimately product yield.
So how do you know how far to trust your scale-up? Not by leaning on any single criterion. I’d build confidence on several fronts:
- Clearly identify your critical process parameters (CPPs) and critical quality attributes (CQAs).
- Where possible, run at an intermediate scale before going commercial. That’s where problems like pH correction issues or poor feed distribution tend to surface.
- Don’t aim just to match impeller speed or DO. Aim to reproduce the actual cell growth profile and the pH and temperature fluctuations the process experiences.
- Challenge the process at worst-case operating conditions: maximum feed rate, reduced aeration, maximum working volume and worst-case raw material lots.
2. Scale-up is often built around parameters such as oxygen transfer, mixing time, power input, agitation or gas-flow characteristics. How should engineers determine which parameters are truly scale-defining for a particular bioprocess, and how should those criteria influence equipment selection and the acceptable operating envelope at commercial scale?
Let me start with the uncomfortable truth: there’s no universal scale-up parameter. Anyone who tells you “just keep tip speed constant” or “just match kLa” is oversimplifying. The right answer comes from a risk assessment that asks one question: what is most likely to break this particular process?
Your answer to that question points you to what matters:
- If oxygen transfer is the main risk, prioritise oxygen transfer capacity, gas dispersion, pressure and oxygen uptake rate.
- If CO₂ accumulation worries you, look at stripping efficiency, pressure, gas flow and liquid height.
- If heat generation is the concern, assess cooling capacity and temperature recovery time.
- If the organism is sensitive to substrate spikes, focus on reproducing the feed concentration profile.
In practice, the best strategy is usually a hybrid criterion. Two contrasting cases make this clear. Take a microbial fermentation with high oxygen demand: keeping tip speed constant alone may not cut it, because a large fermenter can have perfectly acceptable impeller velocity yet still fall short on oxygen transfer. Now flip it around. For a mammalian cell culture, chasing a very high kLa could force you into aggressive aeration and agitation that actually damages the cells.
So the scale-defining parameters, and the equipment and operating envelope that follow from them, have to be chosen based on three things together: the biology, what the equipment can realistically deliver, and the risk to product quality.
3. A commercial bioreactor can experience spatial and temporal variations in pH, dissolved oxygen, nutrient concentration and CO₂ that may not exist to the same extent at laboratory scale. To what extent should reactor geometry, impeller configuration, gas-distribution strategy and mixing characteristics be considered when assessing whether a process developed at smaller scale will remain robust commercially?
To a very large extent, and it helps to think about it from the cell’s point of view.
Picture yourself as a single cell in a large bioreactor. You’re experiencing two kinds of variation.
The first is spatial variation, meaning differences from one physical location to another. Different regions of the vessel can hold different levels of DO, CO₂, pH, glucose and other nutrients, temperature, and even cell concentration. The area right near a concentrated feed inlet, for instance, might briefly carry a high substrate concentration while the liquid around it is much leaner.
The second is temporal variation, meaning how your surroundings change over time. As you circulate, you repeatedly pass through a nutrient-rich feed zone, then a low-oxygen region, then a high-CO₂ region, and around again. You’re not living in a steady environment at all; you’re living through a fluctuating exposure pattern. That’s why the bulk average value can hide what the cells are genuinely experiencing.
And these fluctuations have real consequences. They can shift specific growth or production rates, change cell viability, and push up by-product formation.
This is exactly why reactor geometry, impeller configuration, gas distribution and mixing characteristics belong at the centre of any robustness assessment. A sound risk assessment for heterogeneity should include multiple-point measurement of pH, DO and temperature; scale-down reactors that reproduce the gradients you’ve identified; CFD to understand the fluid dynamics inside the vessel; and a comparison of cell physiology and product quality after gradient exposure.
4. What role should a well-designed scale-down model play in predicting commercial-scale performance? Beyond matching operating parameters, what evidence is needed to demonstrate that the model reproduces the relevant stresses, gradients and process history experienced by the commercial system?
A scale-down model is essentially a way of bringing the commercial vessel’s “bad days” into the lab. It’s an experimental system designed to reproduce the environmental stresses cells encounter at commercial scale, so you can study them cheaply and deliberately.
A good one should help you answer five questions. Does the organism tolerate commercial-scale gradients? Do growth and productivity stay stable? Does metabolism drift toward undesirable by-products? Is the proposed operating range genuinely robust? And what should the CPP ranges be for commercial manufacturing?
But matching operating parameters isn’t enough to call a model valid. I’d look for evidence on three levels:
- Physical comparability: mixing time, oxygen transfer behaviour and gas residence time should line up with the commercial system.
- Biological comparability: similar growth rate, oxygen uptake, CO₂ evolution and viability.
- Product quality comparability: attributes such as potency, purity, host cell protein and residual DNA should match.
If the model checks out on all three, you can start trusting what it tells you about commercial performance.
5. When a process that performs well at development scale shows unexpected behavior after scale-up, how do you distinguish a genuine scale-dependent effect from equipment differences, control-system behavior, raw-material variability or gaps in process understanding? Could you share an example where the initial assumption about the root cause proved incorrect?
The first thing I’d say is: resist the urge to blame scale straight away. When a process behaves differently after scale-up, your job is to work out whether the cause is the equipment, weak process control, gaps in process understanding or variation in raw materials. I approach it a bit like a detective, eliminating suspects one at a time.
Step one: make sure the discrepancy is real. Rule out poor sampling, calculation or unit-conversion mistakes, probe calibration errors and analytical method performance issues. You’d be surprised how often the “problem” disappears here.
Step two: compare the engineering data. Put development and production data side by side for OTR, feed addition rate, agitation, mixing time and glucose or feed consumption rate.
Step three: compare the procedures and inputs. Look at media preparation, raw material composition, possible microbial contamination, differences in operators or timing, and the age or physiological state of the inoculum.
A case that taught me this lesson: during the scale-up of a polysaccharide from 20 L to 600 L, we saw low yields. Because we’d recently changed raw material suppliers, everyone’s first instinct was raw material variation. It turned out the process was fine. The real culprit was an error in the analytical calculation, caused by improper sample handling by the analyst. It’s a good reminder that the most obvious suspect isn’t always the guilty one.
6. When transferring a process into an existing brownfield manufacturing environment, how do differences in equipment geometry, vessel configuration, utilities, automation and validated operating ranges affect process-equipment fit and the ability to reproduce development-scale performance? Where should engineers adapt the process, and where is it more appropriate to challenge the equipment configuration?
Brownfield transfers are really a negotiation between the process and the facility. You’re fitting a new process into equipment and systems that already exist, so the first step is a hard look at the bioreactor design, utility availability, automation and controls, and the validated operating limits for pH, pressure and temperature.
Vessel size is only the beginning. Geometry matters just as much, including impellers, spargers, feed locations, mixing, cooling, oxygen transfer and available working volume, because all of these shape process performance. And beyond the hardware, you need the right control systems, sensors, alarms and data recording, along with trained people who genuinely understand the process.
So where do you bend, and where do you push back?
Adapt the process when the existing equipment has a reasonable, scientifically justified limitation and product quality isn’t affected. A classic example: the existing vessel can’t reach the development-scale agitation speed, but it can deliver equivalent oxygen transfer capacity or acceptable mixing through a different impeller or gassing strategy. In that case, adjusting the process is perfectly sensible.
Challenge the equipment when it causes poor mixing, insufficient oxygen transfer, inadequate cooling, unstable control, operation outside validated limits, or inconsistent product quality. Those aren’t things you should engineer around; they’re signs the equipment configuration itself needs to be reconsidered.
7. As predictive modelling, PAT, AI/ML and digital-twin approaches become more capable, where do you see their greatest practical value in commercial bioprocessing? What would need to be demonstrated around model validation, data integrity, uncertainty, change control and ongoing performance before such predictions could be relied upon for a GMP manufacturing or quality decision?
I’m genuinely optimistic here, with a healthy dose of caution. The practical value is real, and I see it in four areas:
- Early warning. Spotting abnormal trajectories well before they turn into an OOS.
- Predictive control. Forecasting where the process is heading and recommending adjustments before it drifts out of its acceptable range.
- Real-time release. When analytical measurements, models and process understanding are all mature, PAT can support much faster decisions.
- Scale-up and tech transfer. Comparing fermenters, predicting oxygen transfer limitations and flagging risky operating regions before you commit to them.
Now the caution. These tools are only as good as the data, sensors and science underneath them. I’d trust a model for a manufacturing or quality decision only once it has performed consistently across different batches, equipment, raw materials and operating conditions.
Before a model goes anywhere near a GMP decision, it needs to be tested independently against real production data, with its accuracy and reliability properly documented. It should then be reviewed for accuracy by the people who own that data. And there must always be a fallback, such as conventional laboratory testing, in case the model fails.
If I had to sum it up in one line: AI and PAT should strengthen process understanding, never stand in for it.
About the Guest
Mr Prashant Chawla
Sr. General Manager, Manufacturing, Biological E. Limited
Mr Prashant Chawla is a Postgraduate in Bioprocess Engineering with experience leading upstream and downstream operations across multiproduct manufacturing facilities.
His expertise includes scale-up, technology transfer and process validation of bacterial and recombinant vaccines, along with process engineering and CQV activities for greenfield and brownfield projects.
He also brings strong people management and project leadership capabilities, with experience leading multiple projects and large teams. He is a Lean Six Sigma Green Belt certified professional, with knowledge of ISO 14001:2015, OHSAS 18001 and ISO 50001:2018.
Disclaimer
The views and opinions expressed in this editorial are those of the interviewee and are based on his professional experience across pharmaceutical supply chains, networked manufacturing, regulatory strategy and technology-enabled operations. They do not necessarily reflect the official views, policies, or positions of Hello Pharma, its management, or its affiliates.
Hello Pharma does not endorse or take responsibility for any specific technical, commercial, or regulatory interpretations presented in this article. Readers are encouraged to independently evaluate the information shared, review applicable regulatory guidance, and rely on their own experience, expertise, and professional judgement before making decisions related to manufacturing strategy, partner selection, technology transfer, or regulatory compliance.
