Mini bioreactors and DoE screening

A small vessel is not a small version of a large one. Scale-down starts by choosing a criterion to hold constant — power per unit volume, volumetric mass transfer coefficient, tip speed, mixing time, or gas flow per unit volume — and, as set out in high cell density culture systems, matching any one of these forces the others to differ. A mini bioreactor matched on kLa will not have the production tank’s shear field; matched on tip speed it will not have its oxygen transfer. The honest description of a screening platform is therefore not “predictive” but “predictive of the things that follow from the matched parameter”.

What small scale systematically gets wrong

The structural mismatch is homogeneity. Mixing time in a few millilitres is on the order of a second or less, so the culture experiences one uniform environment. A cubic-metre tank has circulation times long enough that a cell passes repeatedly through a well-fed, well-oxygenated zone near the impeller and a starved zone away from it, and many organisms respond to that oscillation with metabolic shifts — overflow metabolism, by-product formation, stress responses — that simply cannot appear in a homogeneous millilitre. This is why deliberately inhomogeneous two-compartment scale-down systems exist: they reintroduce the gradient the small vessel removed.

Geometry adds its own distortions. Surface-to-volume ratio is high, so evaporation concentrates the broth over a long run and headspace exchange contributes a share of gas transfer that is negligible at scale. Carbon dioxide is stripped efficiently for the same reason, so the acidification that troubles large vessels is absent, and a pH control strategy tuned at millilitre scale may be tuned against a problem that will not exist and blind to one that will.

Instrumentation has to change too. There is no room for glass electrodes and no volume to spare for sampling, so dissolved oxygen and pH are read optically from immobilised fluorescent sensor spots on the vessel wall, with response and drift characteristics of their own. Shaken formats — deepwell plates and shake flasks with optical readout — go further and dispense with impellers entirely, transferring oxygen through the deformed liquid surface, which makes their kLa a function of shaking frequency, orbit and fill volume rather than of stirrer power.

Why parallelism is the point

Change one factor at a time and a medium with a dozen candidate variables is unaffordable to optimise, and interactions between factors remain invisible by construction. Design of experiments treats the factors jointly: fractional factorial screens identify which few of many variables actually move the response, and a response-surface design then locates an optimum inside that reduced space, including curvature and interaction terms that one-factor-at-a-time work cannot detect. The catch is that these designs need many runs executed under identical conditions, and that is precisely what a bank of twenty-four or forty-eight independently controlled small vessels supplies. Parallel mini bioreactors did not create DoE; they made it affordable in a discipline where each experiment previously occupied a whole fermenter for a week.