Technical publication · Fermentation control

Fed-batch feed rate calculation for exponential feeding

Abstract. This method note derives an exponential fed-batch feed profile from the biomass inventory, target specific growth rate, true growth yield, maintenance demand and feed assay. A deterministic calculator then screens the planned fermentation against the calibrated feed-pump range, maximum working volume and a qualified whole-reactor oxygen-transfer limit. Equations, declared dimensionally consistent engineering units, a reproducible 1,000 L example, sensitivity cases and implementation checks make every result reviewable.

Acatian EngineeringPublished and reviewed 12 September 202625 min technical publication

Initial feed3.47 L/h Final feed14.63 L/h Final biomass33.77 kg Projected O₂ equality12.87 h

Synthetic carbon/energy-substrate screen at μ_set = 0.12 h⁻¹. No declared nominal upper limit is exceeded during the twelve-hour phase; oxygen-transfer equality is projected 0.87 h later only if the same profile continues. This is a calculation basis—not a fermentation recipe.

Scope and design question

Convert a target growth rate into an executable feed-pump profile.

Question answered

For a cell-free carbon/energy-limiting feed in a biomass-growth phase, what initial flow, exponential ramp and cumulative feed support a constant target specific growth rate—and when would each declared equipment or oxygen-transfer equality occur?

Model boundary

Set t = 0 at feed start after depletion and require residual limiting substrate S ≈ 0. The screen assumes no broth withdrawal, the main feed alone changes volume, and μ_set, Y^G_X/S, m_S, S_F, q_O2 and a conservative minimum OTR_cap remain constant. It omits induction and product-associated substrate demand, inhibition, by-products, decay, rheology, heat removal and gas-path limits.

Engineering use

Use the result as an initial open-loop schedule and nominal constraint screen for high-cell-density or precision fermentation. Large-scale feed zones, mixing and oxygen limitation can break the ideal well-mixed assumption [2]. The response to an exponential regime is organism and process specific; one recombinant Yarrowia lipolytica study evaluated epoxide-hydrolase production under this strategy [6].

Nomenclature

Declare mass, concentration and whole-reactor rates explicitly

On small screens, swipe the table horizontally to compare every column.

SymbolDefinitionUnit used here

M_XDry biomass inventory in the vessel; calculate in g and convert to kg only for displayg

μ, μ_setActual and targeted specific biomass growth rateh⁻¹

Y^G_X/SMaintenance-free true growth yield used with a separately estimated maintenance coefficient; not an observed overall yieldg_X g_S⁻¹

m_SMaintenance-associated substrate consumption coefficientg_S g_X⁻¹ h⁻¹

q_SSpecific limiting-substrate uptake used by the screeng_S g_X⁻¹ h⁻¹

S_FAvailable limiting-substrate concentration in feedg_S L⁻¹

F, V_FInstantaneous feed flow and cumulative feed volumeL h⁻¹; L

q_O2Specific oxygen uptake rate on dry biomassmmol O₂ g_X⁻¹ h⁻¹

OTR_capQualified whole-reactor oxygen-transfer capacity at the intended statemol O₂ h⁻¹

02 · Interactive engineering calculator

Calculate an exponential feed and screen the first modeled equality

Deterministic model · single limiting substrate · no withdrawal · constant parameter basis

Specific substrate uptake0.260 g/g/hq_S = μ/Y^G_X/S + m_S Initial feed flow3.47 L/hF₀ at feed start Final feed flow14.63 L/hAt 12.0 h Cumulative feed93.04 L55.83 kg substrate Final biomass33.77 kg30.89 g/L after dilution Final culture volume1,093.04 LBelow 1,600 L limit Final oxygen uptake270.12 mol/hAgainst 300 mol/h OTR capacity First modeled equality12.87 hProjected O₂ equality if continued · +0.87 h

Derivation and calculation sequence

Close biomass and substrate demand before programming the pump.

1 · Initial inventory

Convert concentration to a mass inventory at feed start: M_X,0 = X₀V₀. Keep dry-cell-weight, viable-cell or another biomass basis consistent across X, yield, maintenance and q_O2.

2 · Substrate demand

The Pirt maintenance relation separates growth-associated demand from substrate required to maintain biomass [1]. Use the maintenance-free true growth yield: q_S = μ_set/Y^G_X/S + m_S. An observed overall yield at the target μ instead satisfies 1/Y_obs = 1/Y^G + m_S/μ; adding maintenance to Y_obs would double-count it.

3 · Feed profile

With dM_X/dt = μ_setM_X, biomass increases exponentially. The full limiting-substrate inventory balance is d(SV)/dt = FS_F − q_SM_X. Only when S ≈ 0 after depletion does S_FF = q_SM_X, giving the same exponential factor. Cell and substrate balances provide this basis for exponential feeding [3].

Biomass, uptake and instantaneous feedM_X(t) = M_X,0e^μt
q_S = μ/Y^G_X/S + m_S
F(t) = q_SM_X(t)/S_F = F₀e^μt

Here F₀ = q_SM_X,0/S_F. This is dimensionally consistent when M_X is in g, q_S in g_S g_X⁻¹ h⁻¹, and S_F in g_S L⁻¹; F is then L h⁻¹. The profile targets biomass growth, not a guaranteed measured μ.

Cumulative feed and volumeV_F(t) = F₀[e^μt − 1]/μ
V(t) = V₀ + V_F(t)

The integral assumes that the programmed feed is the only volume-changing stream and that evaporation, sampling, base, acid, antifoam and gas-associated losses are negligible. Include those streams in the volume balance when they are material. Gravimetrically calibrate the installed flow path across its intended range as an engineering qualification. In one microscale study, inferred delivery was about 30% below an earlier pump calibration, illustrating why delivered mass—not the controller setpoint alone—must be verified [7].

First-constraint screenOUR(t) = q_O2M_X(t)/1000
F_min ≤ F₀; t_O2 = ln(OTR_cap/OUR₀)/μ
t_pump,max = ln(F_max/F₀)/μ
t_volume = ln[1 + μ(V_max − V₀)/F₀]/μ

The factor 1000 converts mmol h⁻¹ to mol h⁻¹. The initial flow must lie inside the calibrated pump range; the first upper equality is min(t_O2, t_pump,max, t_volume). A time beyond the planned endpoint is an extrapolation that applies only if the same profile continues. Here OTR_cap is constant and must represent the conservative minimum whole-reactor capacity across the phase at the chosen DO floor, relevant volume, gas rate, pressure, composition, rheology and power. If capacity varies, solve the first root of OUR(t) = OTR_cap(t) numerically. Induction and production burden also require separate terms: Sandén et al. reported precursor limitation at high μ and carbon/energy limitation at low μ after induction in a recombinant E. coli process [4].

Reproducible worked example

Calculate a twelve-hour feed for a 1,000 L fermentation.

Declared basis

V₀ = 1,000 L; X₀ = 8 g/L; μ_set = 0.12 h⁻¹; Y^G_X/S = 0.50 g/g; m_S = 0.02 g/g/h; S_F = 600 g/L; q_O2 = 8 mmol/g/h; duration = 12 h. The equivalent observed yield at this μ is 0.4615 g/g.

Equipment limits

Calibrated pump range = 1–60 L/h; maximum working volume = 1,600 L; qualified whole-reactor OTR capacity = 300 mol/h. These are synthetic inputs for reproduction, not equipment recommendations or organism-specific operating ranges.

Initial calculation

M_X,0 = 8,000 g and q_S = 0.12/0.50 + 0.02 = 0.260 g/g/h. Therefore F₀ = (0.260 × 8,000)/600 = 3.4666667 L/h. Kilograms below are display conversions.

Calculation ledger

Worked result at the twelve-hour feed endpoint

On small screens, swipe the table horizontally to compare equation, substitution and result.

QuantitySubstitutionResult

Growth factore^0.12×124.22070

Final feed3.4666667 × 4.22069581714.6317455 L/h

Cumulative feed(3.4666667/0.12)(4.220695817 − 1)93.0423236 L

Substrate fed93.0423 × 600 / 100055.8254 kg

Final biomass inventory8.00 × 4.2207033.7656 kg

Final volume and X1,000 + 93.0423; 33,765.6 / 1,093.04231,093.0423 L; 30.8914 g/L

Final OUR8 × 33,765.6 / 1000270.1245 mol/h

Endpoint OTR headroom300 − 270.1245; 300 / 270.124529.8755 mol/h; OTR/OUR = 1.1106

Projected upper equalitymin(12.8742, 23.7596, 25.6708)Oxygen transfer at 12.8742 h if continued

05 · Target-growth-rate sensitivity

A modest μ increase can cross the oxygen-transfer boundary.

The three cases change only μ_set; all other worked-example inputs remain fixed. At 0.15 h⁻¹, predicted OUR reaches 300 mol/h at 10.30 h, 1.70 h before the planned end; the listed twelve-hour biomass and OUR are unconstrained projections beyond that equality, not executable states. Kim et al. combined exponential feeding with fixed glucose additions that stopped after a defined dose and restarted on a pH depletion signal [5]. A comparison of four Pseudomonas putida PHA feeding strategies likewise reported oxygen limitation under some strategies [9]; feedback cannot remove a physical transfer limit.

On small screens, swipe horizontally. Download the rounded, machine-readable values and complete fixed-input basis below.

μ_set, h⁻¹F₀ → F_end, L/hFeed, LFinal biomass, kgFinal OUR, mol/hResult

0.102.93 → 9.7468.0626.56212.49No nominal exceedance; projected O₂ equality at 15.45 h if continued

0.123.47 → 14.6393.0433.77270.12No nominal exceedance; projected O₂ equality at 12.87 h if continued

0.154.27 → 25.81143.6348.40387.18Modeled O₂ equality at 10.30 h; endpoint values unconstrained

Download worked-example CSV

05B

From equation to controlled operation

Treat the open-loop profile as a testable hypothesis.

Calibrate delivery

Verify feed density, substrate assay, line prime, pump turndown, pulsation, backpressure and gravimetric delivery at representative head and temperature. Program mass flow when density varies. Record actual delivered feed, not only the controller setpoint.

Observe substrate limitation

Trend dissolved oxygen, oxygen uptake, carbon dioxide evolution, respiratory quotient, off-gas, base addition, residual substrate, by-products, biomass and feed mass. Hans et al. demonstrated recursive model refitting and feed-profile updates across 24 parallel mini-bioreactors [8]. A metabolic-model-based strategy has also been developed for methanol feeding in Pichia pastoris [10].

Define intervention rules

Specify holds, feed caps and abort criteria for DO, OUR/OTR margin, exhaust oxygen, carbon dioxide, pressure, temperature, foam, weight and pump faults. A feed controller should degrade to a safe state when a critical measurement becomes unavailable.

Limitations

What an exponential feed calculation does not prove

On small screens, swipe horizontally to compare failure mode, consequence and control.

Failure modeConsequenceRequired control

Residual substrate accumulatesFeed no longer maps directly to growth; overflow metabolism or inhibition may appear.Measure residual substrate/by-products and reduce or adapt feed.

Y^G_X/S or m_S driftsThe calculated substrate demand becomes biased as physiology changes.Estimate parameters over the intended phase and reconcile carbon and biomass balances.

q_O2 held constantOUR and the oxygen constraint time are wrong during induction, product formation or metabolic shifts.Use measured OUR trajectories and a state-dependent model where justified.

OTR capacity treated as constantRising volume, viscosity, antifoam or gas-path limits erode real transfer capacity.Qualify k_La, C*, gas handling and operating policy across the phase.

Well-mixed vessel assumedLarge-scale cells encounter transient substrate-rich and oxygen-poor zones.Combine mixing-time evidence, feed-location review, scale-down experiments and compartment/CFD work where risk warrants.

Nominal pump curve usedActual delivery deviates at low flow, high pressure or with concentrated feed.Calibrate the installed flow path and alarm on independent mass change.

Practical implementation checklist

Minimum evidence before approving a fed-batch feeding strategy

Biological basis

Biomass definition and assay; feed-start inventory; carbon/energy-limiting substrate; true growth yield Y^G_X/S; maintenance basis; μ_set rationale; residual-substrate and by-product data; q_O2 and q_CO2; induction or metabolic transitions; uncertainty and validity range.

Equipment and transfer

Feed assay and density; pump calibration, minimum stable flow and capacity; line volume and prime; load-cell accuracy; maximum working volume; evaporation and additions; mixing time and feed location; OTR/OUR margin; gas, pressure, foam and heat-removal limits.

Control and review

Open-loop schedule; online measurements; sample plan; feed cap; alarms and interlocks; sensor-loss state; deviation logic; mass and carbon closure; parameter owner; version and approval; scale-down challenge; post-batch reconciliation and predefined acceptance criteria.

Primary technical sources

References

Pirt, S. J. “The maintenance energy of bacteria in growing cultures.” Proceedings of the Royal Society B 163, 224–231 (1965). doi:10.1098/rspb.1965.0069.
Enfors, S.-O. et al. “Physiological responses to mixing in large scale bioreactors.” Journal of Biotechnology 85, 175–185 (2001). doi:10.1016/S0168-1656(00)00365-5.
Cheng, L.-C.; Wu, J.-Y.; Chen, T.-L. “A pseudo-exponential feeding method for control of specific growth rate in fed-batch cultures.” Biochemical Engineering Journal 10(3), 227–232 (2002). doi:10.1016/S1369-703X(02)00002-5.
Sandén, A. M. et al. “Limiting factors in Escherichia coli fed-batch production of recombinant proteins.” Biotechnology and Bioengineering 81(2), 158–166 (2003). doi:10.1002/bit.10457.
Kim, B. S. et al. “High cell density fed-batch cultivation of Escherichia coli using exponential feeding combined with pH-stat.” Bioprocess and Biosystems Engineering 26(3), 147–150 (2004). doi:10.1007/s00449-003-0347-8.
Maharajh, D.; Lalloo, R.; Görgens, J. “Effect of an exponential feeding regime on the production of Rhodotorula araucariae epoxide hydrolase in Yarrowia lipolytica.” Letters in Applied Microbiology 47, 520–525 (2008). doi:10.1111/j.1472-765X.2008.02425.x.
Funke, M. et al. “Bioprocess Control in Microscale: Scalable Fermentations in Disposable and User-Friendly Microfluidic Systems.” Microbial Cell Factories 9, 86 (2010). doi:10.1186/1475-2859-9-86.
Hans, S. et al. “Automated Conditional Screening of Multiple Escherichia coli Strains in Parallel Adaptive Fed-Batch Cultivations.” Bioengineering 7(4), 145 (2020). doi:10.3390/bioengineering7040145.
Guzik, M. et al. “A polyhydroxyalkanoates bioprocess improvement case study based on four fed-batch feeding strategies.” Microbial Biotechnology 15(3), 996–1006 (2022). doi:10.1111/1751-7915.13879.
Boojari, M. A. et al. “Developing a metabolic model-based fed-batch feeding strategy for Pichia pastoris fermentation through fine-tuning of the methanol utilization pathway.” Microbial Biotechnology 16(6), 1344–1359 (2023). doi:10.1111/1751-7915.14264.

FAQ

Technical FAQ

Questions engineers ask before programming exponential feeding

How do you calculate an exponential fed-batch feed rate?

Calculate initial biomass mass M_X,0 = X₀V₀, specific substrate demand q_S = μ_set/Y^G_X/S + m_S, and initial flow F₀ = q_SM_X,0/S_F. Then program F(t) = F₀e^μt while the depletion, growth-phase and physical-constraint assumptions remain valid.

Why include a maintenance coefficient in the feed equation?

True growth yield accounts for maintenance-free substrate conversion to new biomass; maintenance represents demand associated with sustaining existing biomass. An observed overall yield already includes maintenance and must not be combined with m_S without conversion. Every parameter must use the same biomass, substrate and physiological basis.

How should the target specific growth rate be chosen?

Select μ_set below the organism- and condition-specific growth capability and jointly screen product quality, overflow metabolism, induction burden, oxygen and heat transfer, mixing, gas handling, feed delivery and downstream consequences. It is a qualified process target, not a percentage of a universal maximum.

How are oxygen, pump and working-volume limits checked?

Confirm F₀ is at or above the pump’s calibrated minimum, then calculate when predicted OUR equals the conservative qualified OTR capacity, F(t) equals maximum calibrated pump flow and cumulative feed reaches available volume. A time after the planned endpoint is only a projection if the profile continued. Add heat, DO setpoint, exhaust gas, pressure, foam and substrate-mixing limits separately.

Is open-loop exponential feeding better than feedback control?

Neither is universally better. Open-loop feeding is simple and reproducible when the initial state and kinetics are reliable. Feedback can correct biological and delivery deviations but depends on informative, available signals and qualified logic. Many processes use a validated feed-forward profile with bounded feedback or supervisory overrides.