Fermentation Scale-Up: Translating Microwell Hits to Bench Bioreactor Performance

Microwell screening is fast, high-throughput, and inexpensive. A 96-well microfermentation run can evaluate 24 or more construct variants in the time it takes to set up a single bench bioreactor experiment. That throughput is why we use it — but it creates a specific problem: the productivity rankings you observe in microwell format frequently change when you scale to a stirred-tank bioreactor. The construct ranked first in the microwell may be third at bench scale, and the one you nearly discarded may be your production winner.

Understanding why this happens is the key to designing scale-up experiments that preserve the predictive value of your screening data and avoid costly surprises at CMO handoff.

What Changes Between Microwell and Bioreactor

Four physical and chemical variables change simultaneously when you move from a 96-well plate to a stirred-tank system. Each can independently affect productivity ranking.

Dissolved oxygen (DO) profile

In a 96-well deepwell plate on an orbital shaker, dissolved oxygen is supplied through the liquid surface. The oxygen transfer rate (OTR) is limited by the surface-to-volume ratio of the well and the shaker frequency. Typical OTRs in deep-well plate microfermentation are 5–20 mmol O₂/L/h under standard conditions. A 5L stirred-tank bioreactor with sparging and agitation routinely achieves 50–200 mmol O₂/L/h and can maintain DO at a defined setpoint throughout the run.

Constructs that perform poorly under DO fluctuation — which is exactly what happens in microwell format during high-growth phases — may be strongly DO-sensitive. You won't know from the microwell data alone. This is why DO-controlled shake flask experiments (using baffled flasks and DO monitoring) are a useful intermediate validation step before committing to bioreactor runs with your top candidates.

pH control (or the lack of it)

In microwell format, there is no pH control. pH drifts as the culture acidifies (from CO₂ accumulation, organic acid overflow, or both) or alkalinizes (from ammonia release). Depending on the carbon source, initial pH, and culture density, the pH excursion can be 1–2 units over a 24-hour run. Some promoters are pH-sensitive. Some product pathways have pH-dependent flux. And some chassis strains are simply more pH-tolerant than others.

When a construct underperforms at bench bioreactor with tightly controlled pH, it's sometimes because the microwell screening inadvertently selected for a strain that performs well under the pH regime of the screening condition — not the production condition. Running a small pH-sensitivity screen (microwell plates at buffered vs. unbuffered conditions) before bioreactor scale-up can flag this confound early.

Feeding strategy and nutrient availability

Microwell screening typically uses a batch format with a fixed initial media composition. Bench bioreactors commonly use fed-batch strategies — controlled glucose feeding to maintain a target concentration, reducing overflow metabolite accumulation and extending the production phase. The productivity advantage of fed-batch over batch fermentation in E. coli is typically 3–8 fold in volumetric titer for high-yield constructs.

A construct that looks moderate in batch microwell screening may dramatically outperform a "better" microwell hit under fed-batch conditions if it has higher metabolic efficiency at low glucose concentrations — exactly the condition fed-batch creates. We've seen two cases in our partner programs where the microwell rank-2 construct became the clear production winner after fed-batch optimization, while the rank-1 construct hit an overflow limit at the glucose concentrations typical of fed-batch operation.

Agitation and shear

Mammalian cells are sensitive to shear stress in a way that bacterial and yeast cells generally are not. In CHO fed-batch culture, agitation and sparging introduce mechanical shear that can damage cells at the impeller tip or at the bubble surface. The Kolmogorov length scale — the smallest turbulent eddy in the bioreactor — should be compared to the cell diameter when setting agitation parameters. For CHO cells (10–20 µm diameter), eddy sizes below 50–80 µm start to cause damage. This isn't relevant in microwell format, where shear forces are much lower, which means CHO constructs that perform consistently in microwell format may show viability differences between scale-up conditions depending on impeller design and agitation speed.

The Scale-Up Ladder: A Structured Approach

Rather than jumping directly from microwell to a 5L bioreactor run, a four-stage ladder reduces the probability of expensive surprises.

1. Microwell screening (0.5–2 mL): Identify the top 4–6 construct candidates. This stage screens for expression level, solubility, and basic construct robustness, not for final productivity under process conditions.
2. Deep-well fed-batch mimicry (2–8 mL): Commercially available deep-well fermentation systems (BioLector, m2p-labs; Ambr 15 for mammalian) allow pH and DO monitoring and simplified feeding. This intermediate format catches pH and nutrient sensitivity issues without using bioreactor time.
3. Shake flask with DO/pH monitoring (100–500 mL): Baffled shake flasks with dissolved oxygen sensors confirm the rank order under aerobic conditions. This stage is particularly important for constructs that showed DO sensitivity signals at the deep-well stage.
4. Bench bioreactor (1–5L): Full process development with controlled pH, DO, temperature, and feeding strategy. Use the top 2 candidates from the shake flask stage. Define the process parameters that will be transferred to the CMO.

Scale-Independent Parameters to Establish Early

Some parameters translate directly from small scale to large scale when properly defined. Getting these right in the bench bioreactor prevents rework at the CMO.

• Specific oxygen uptake rate (sOUR) at induction. Defines the aeration requirement per unit biomass — used by the CMO to set initial sparging strategy.
• kLa target for DO maintenance. The mass transfer coefficient required to maintain DO above your setpoint at peak biomass. Scale-up engineers use this to specify impeller type, speed, and sparging rate in the large-scale vessel.
• Induction temperature and timing. Temperature shifts at induction (e.g., from 37°C to 25°C for E. coli, or from 36°C to 31°C for CHO fed-batch) should be confirmed at bench scale and included in the technical transfer dossier.
• Feeding profile. Glucose set point, feed concentration, and feeding algorithm (exponential feed, DO-stat, or pH-stat) should be locked in the bench bioreactor before CMO handoff.

What We Transfer to CMO Partners

The goal of the bench bioreactor phase is to generate a technical transfer package that a CMO process development team can execute without extensive re-optimization. In our experience, CMO ramp-up timelines are shortest when the transfer package includes: process definition (vessel geometry, impeller, sparging, setpoints), seed train protocol (inoculum concentration, passage number), feeding recipe with acceptable deviation ranges, and acceptance criteria for each run (minimum viable titer, viability threshold for CHO, and induction-phase productivity window).

CMOs are not passive recipients of cell banks. They will adapt the process to their equipment. What they need from the development team is a well-characterized process with defined setpoints and known acceptable variation ranges — not a protocol written for a specific bioreactor model that may not be available at their facility.

A Note on Data Quality at Each Stage

Scale-up decisions are only as good as the data underlying them. Microwell runs in duplicate are not sufficient replication for construct ranking decisions. We use triplicate minimum at the microwell stage and run each construct in at least two independent screening batches before advancing. That sounds excessive for a high-throughput screen, but a 10–15% coefficient of variation in microwell titer is normal, and ranking decisions based on duplicate data from a single batch carry substantial noise.

At the bench bioreactor stage, two independent runs of the top construct are the minimum basis for a process definition that gets transferred to a CMO. One run is a data point. Two runs with consistent performance is a basis for a specification.