the microbe thrives in the flask, dies in the tank

Problem statement

Biomanufacturing — using engineered microorganisms, cell cultures, or enzymatic processes to produce chemicals, materials, and therapeutics — consistently fails during scale-up from laboratory flasks (milliliters) to production bioreactors (thousands of liters). Organisms engineered to produce target molecules at high yields in well-mixed, well-aerated shake flasks experience fundamentally different conditions at scale: heterogeneous dissolved oxygen, pH and nutrient gradients, shear stress from impellers, and metabolic byproduct accumulation. No generalizable framework exists to predict which laboratory-validated bioprocesses will fail at scale, or why, forcing companies into expensive and time-consuming empirical scale-up campaigns that fail 70-90% of the time.

Why this matters

The US bioeconomy is valued at over $1 trillion, and the 2022 Executive Order on Advancing Biotechnology and Biomanufacturing Innovation calls for expanding domestic biomanufacturing capacity across health, energy, agriculture, and industrial sectors. Biomanufactured products — from biofuels to bioplastics to cell-cultured meat — could displace petroleum-derived chemicals and reduce carbon emissions. However, the scale-up failure rate means that most promising laboratory strains never become production organisms. A single scale-up campaign for a biopharmaceutical can cost $50-200 million and take 3-5 years, with no guarantee of success.

What’s been tried and why it hasn’t worked

Traditional chemical engineering scale-up relies on dimensionless numbers (Reynolds, Damkohler) to maintain similarity across scales, but biological systems violate the assumptions behind these correlations — cells are not passive reactants but adaptive organisms that change their gene expression, metabolism, and growth behavior in response to environmental shifts. Scale-down models that replicate large-scale heterogeneity in laboratory bioreactors capture some failure modes but miss others because the temporal dynamics of gradient exposure differ. Computational fluid dynamics (CFD) coupled with metabolic models can simulate bioreactor conditions but require organism-specific kinetic parameters that are expensive to measure and often unreliable. High-throughput minibioreactors (ambr, BioLector) enable parallel screening of conditions but don't replicate the spatial heterogeneity that causes scale-up failures. Genome-scale metabolic models predict steady-state yields but not dynamic responses to the fluctuating conditions cells experience in large bioreactors.

What would unlock progress

A predictive framework integrating computational fluid dynamics, genome-scale metabolic modeling, and dynamic gene regulation models that can simulate organism behavior under the heterogeneous, time-varying conditions of industrial bioreactors. Machine learning models trained on paired small-scale/large-scale process data could identify early warning signatures of scale-up failure. Standardized protocols for characterizing organism responses to controlled environmental perturbations (oxygen shifts, pH pulses, shear steps) would generate the training data such models require.

Entry points for student teams

The computational door needs a laptop and no organism: couple open-source CFD (OpenFOAM, https://openfoam.org) to a published genome-scale metabolic model — E. coli iML1515 downloads free from BiGG Models (http://bigg.ucsd.edu/models/iML1515) and runs in COBRApy (https://opencobra.github.io/cobrapy/) — simulate the dissolved-oxygen and substrate gradients of a stirred tank, then ask how much predicted productivity is lost when cells are carried along those concentration trajectories instead of held at flask conditions. The wet-lab door should be deliberately narrow: impose oscillating dissolved oxygen on E. coli or S. cerevisiae in a benchtop fermenter and track growth, oxygen uptake, and acetate or ethanol accumulation — one perturbation, one readout — rather than attempting transcriptomics and metabolomics in the same term, which is three projects in one. That version takes a benchtop bioreactor and BSL-1 culture space, access most chemical- or bio-engineering departments that teach a bioprocess course already hold; the omics layers are the follow-on project once the perturbation protocol is stable. Relevant disciplines include bioengineering, chemical engineering, computational biology, and data science.