Analytics & PAT
Automated feed systems for bioreactors
The metabolic reason fed-batch feeding exists, the open-loop problem created by the absence of a reliable in-line glucose sensor, and the physical limits — osmolality, tubing drift and mixing time — that bound what a feed strategy can do.
A feed system exists because of a metabolic fact rather than a logistical one. Give a mammalian or microbial culture all the glucose it could use at the start, and it will not use it efficiently. Above a threshold uptake rate, glycolytic flux exceeds what the cell’s oxidative capacity can accept, and the overflow leaves as a partially oxidised by-product — lactate in CHO culture, acetate in E. coli, ethanol in Saccharomyces under the Crabtree effect. Carbon is wasted and the by-product itself inhibits growth and shifts product quality. Fed-batch is therefore a deliberate starvation: keep the residual sugar low so that specific uptake stays under the overflow threshold, and the same total carbon goes into biomass and product instead of into lactate.
Feeding blind
That control objective is clear. The problem is that the variable it targets — the concentration in the tank right now — is usually not measured. There is no equivalent of a pH electrode for glucose that survives weeks of sterile culture without drift or fouling, so most feed control is not feedback control at all. It is a pre-computed profile, an exponential ramp derived from a target specific growth rate and an assumed yield coefficient, executed open-loop and corrected from daily off-line samples.
The alternatives all substitute an indirect signal for the one that matters. Off-gas analysis gives oxygen uptake and carbon dioxide evolution rates, from which consumption can be inferred if the stoichiometry holds. Capacitance probes read the dielectric response of intact, membrane-bounded cells and so track viable biomass, letting feed be scaled per cell rather than per litre. DO-stat and pH-stat schemes exploit the fact that a starving culture stops respiring or stops acidifying, and feed on that transient. Each is a proxy, and each fails where its assumption fails — off-gas when metabolism shifts, capacitance when cell size or morphology changes mid-run.
What the hardware actually limits
Peristaltic pumping is chosen because the fluid path stays inside a closed disposable tube, which preserves sterility and containment. Its weakness is metrological: delivery per revolution depends on tubing that fatigues under repeated occlusion, so calibrated flow drifts over a long campaign and the delivered volume no longer matches the commanded one. The honest correction is gravimetric — the feed vessel sits on a load cell and mass loss, not pump revolutions, defines the dose.
Two further ceilings are chemical. Feeds must be concentrated, because vessel headspace is finite and dilution would undo the enrichment; but concentrating glucose and amino acids raises osmolality, and above roughly 350–400 mOsm/kg mammalian cells respond with reduced growth and altered glycosylation. Solubility forces a second line: tyrosine and cysteine will not stay dissolved at neutral pH in a concentrated bolus, so they are fed separately at high pH.
The last limit is spatial. A feed enters at one point, and in a large vessel the blend time is on the order of a minute or more. Near the addition port the culture briefly sees a concentration far above the setpoint — the exact condition the strategy was designed to avoid — which is why feed point placement and dilution of the feed stream are process decisions, not plumbing details.