Model-guided low-temperature fermentation boosts enzyme output in yeast

Engineers trying to squeeze more therapeutic and industrial proteins out of yeast cells often face a frustrating trade-off: conditions that make cells secrete more protein also slow their growth, and vice versa. A new study used a dynamic mathematical model to predict the temperature and dilution rate at which a hybrid yeast strain would maximize secreted protein productivity per liter per hour, and chemostat experiments supported the predicted trend. The work, published in Bioresources and Bioprocessing, combines bioreactor experiments at three temperatures with a dynamic model originally developed for microalgae, then validates its predictions in continuous chemostat cultures.

The yeast in question, Komagataella phaffii, formerly known as Pichia pastoris, is one of the workhorses of industrial biotechnology. It grows rapidly to high cell densities in cheap media, performs the post-translational modifications that many human proteins need to function, and can devote up to roughly 80 percent of its total secreted protein output to a single recombinant product. Its GRAS status from the US Food and Drug Administration and QPS status from the European Food Safety Authority further widen its use for proteins destined for food, feed and pharmaceutical applications.

The research team, led by Eduardo Rocco-Martínez and colleagues, worked with an unusual host: a hybrid strain called S467, created by crossing the standard laboratory strain CBS7435 with the wild isolate NRRL Y-17741.

The central experimental variable was temperature. Batch cultures in 2-liter bioreactors were run in duplicate at 20, 25 and 30 degrees Celsius, with glucose as the carbon source. Lowering the temperature to 20 degrees cut the maximum specific growth rate by about half compared with 30 degrees, while 25 degrees behaved much like the standard condition. Yet the cold came with a striking benefit: total secreted protein concentration rose fourfold at 20 degrees. Because K. phaffii secretes very few native proteins, the secreted fraction largely reflects the recombinant product itself. Despite this boost in secretion, the highest volumetric β-glucosidase activity in the culture supernatant was still achieved at 30 degrees, because the reduced biomass formation at low temperature outweighed the per-cell secretion advantage.

To describe these dynamics quantitatively, the team fitted a Droop-based dynamic model to the batch data. Classical Monod-type models assume that the specific growth rate depends directly on the substrate concentration in the medium, but the experimental data showed a decoupling between substrate uptake and growth: cells kept growing after the external glucose signal would have predicted a slowdown, because they draw on nutrients stored inside the cell. The Droop framework captures this by introducing a cell quota, the amount of intracellularly accumulated substrate per unit of biomass, and the authors added a delayed coupling between substrate uptake and biomass synthesis. The model tracks four state variables: biomass, secreted protein, substrate (glucose) concentration and the cell quota.

The model fit the batch data well, with coefficients of determination ranging from 0.921 to 0.995 across all variables and temperatures. It also outperformed a conventional Monod-based model fitted to the same data. Parameter estimates showed low to moderate relative uncertainty, typically below 20 percent for the key growth and production parameters, and multi-start optimization confirmed a unique global optimum. The team also verified that fixing the half-saturation constant and the initial cell quota did not bias the results: sensitivity analyses changed the fitted parameters by less than 3 percent across wide ranges of the fixed values.

With the model fitted to the batch data, the researchers simulated continuous chemostat operation across the range of admissible dilution rates at each temperature. The simulations predicted that continuous cultivation at 20 degrees Celsius with a dilution rate of 0.045 per hour would maximize the volumetric productivity of total secreted protein, computed as the dilution rate multiplied by the steady-state protein concentration. In the simulations, this condition gave the lowest steady-state biomass among the temperatures evaluated, yet the highest predicted protein productivity, consistent with the larger maximum specific production rate fitted at 20 degrees. The predicted optimum proved robust: varying the half-saturation constant over two orders of magnitude shifted the optimal dilution rate by less than 2 percent.

The team then evaluated the prediction experimentally, running chemostats at 20 degrees with dilution rates of 0.025, 0.045 and 0.060 per hour, sampling after two complete volume changes and checking steady-state stability across residence times. The measured trends matched the simulations. Volumetric productivity peaked at 0.045 per hour among the tested conditions, exactly as predicted. The model, however, underestimated the absolute values: experimental biomass at the optimum dilution rate was 23 percent higher than predicted, and measured productivity ran about 36 percent above the simulated value. The authors attribute this to parameters derived from transient batch cultures not fully capturing cell physiology under sustained glucose limitation, and they note that shifting between cultivation regimes affects process variables disproportionately, with protein production more sensitive than biomass growth.

One of the most intriguing findings emerged when the researchers looked at enzyme quality rather than quantity. Specific β-glucosidase activity per unit biomass increased steadily with dilution rate, reaching its highest level at 0.060 per hour, even though bgl1 transcript levels and total secreted protein were lower there. Measurements of ER stress markers, including HAC1, KAR2, PDI and ERO1, showed no significant transcriptional changes across the steady states, pointing away from a transcriptional explanation. The authors hypothesize that faster-growing cells fold and process the enzyme more efficiently, face less severe nutrient limitation, or benefit from faster washout of extracellular proteases, and that glycosylation differences could also contribute. Whatever the mechanism, the implication is that cultivation conditions shape not just how much enzyme is secreted but how active it is.

For the broader bioprocessing community, the study offers a template: fit a dynamic model to inexpensive batch data, use it to shrink the experimental design space, then validate only the most promising operating points in continuous culture. The authors caution that additional dilution rates near the simulated optimum would be needed to pin down the true experimental optimum with greater precision, and they frame the approach as a framework to be evaluated for other proteins, strain backgrounds and cultivation conditions rather than a finished recipe. Still, the demonstration that a microalgae-derived model can guide yeast fermentation from batch characterization to a validated chemostat operating point gives process engineers a quantitative handle on a trade-off they have long managed by trial and error.

Subject of Research: Dynamic modeling and continuous cultivation to optimize recombinant β-glucosidase production in Komagataella phaffii

Article Title: Integrating dynamic modeling and continuous cultivation to enhance recombinant β-glucosidase production in Komagataella phaffii

Article References: Rocco-Martínez, E., Quezada, J., Berrios, J., Wolfe, K. H., Díaz-Barrera, A., Martínez, C., & Braun-Galleani, S. (2026). Integrating dynamic modeling and continuous cultivation to enhance recombinant β-glucosidase production in Komagataella phaffii. Bioresources and Bioprocessing, 13(1), Article 151. https://doi.org/10.1186/s40643-026-01147-9

Image Credits: AI Generated

DOI: 10.1186/s40643-026-01147-9

Keywords: Komagataella phaffii, β-glucosidase, Droop model, chemostat, hypothermia, bioprocess optimization, recombinant protein, continuous cultivation, mathematical modeling, biotechnology, Integrating, dynamic