Fed-Batch vs Perfusion Carbon Footprint: Which Process Emits Less per Kilogram?

Key differences at a glance

• The headline fed-batch vs perfusion carbon footprint is 3,980 against 1,624 kg CO₂e per kg. Same factors, same 400 g CO₂e/kWh grid, same cradle-to-gate boundary. Perfusion is 2.45 times lower.
• Most of that gap is the room. The perfusion preset assumes a 45 kW suite against 80 kW. Give both the same suite and perfusion is 2,722, which is 1.46 times lower. The size of the room explains 1.68x of the 2.45x.
• The real mechanism is product per suite-hour. Cleanroom HVAC is 93.5% of the fed-batch electricity, and it scales with time, not with litres. Perfusion makes 0.306 kg per day against 0.216.
• The published studies point the other way, for a reason. Bunnak et al. (2016) found standard perfusion emitting 17% more CO₂, driven by 35% more water. Those burdens scale with volume. Take the cleanroom out of our model, load perfusion with the literature's water and plastic penalties, and fed-batch wins by 7.1%.
• Run failure is what can actually erase it. On a matched suite perfusion still matches fed-batch at a 58.1% run success rate. Media, extra buffer and grid choice move the answer by single-digit percentages at most.

Fed-batch vs perfusion: side-by-side comparison

Both columns are calculator presets at a 400 g CO₂e/kWh grid with open emission factors (US EPA eGRID, UK Government conversion factors, eLCI). Both are single-use and incinerate 100% of their consumables.

How a fed-batch process spends its carbon

A fed-batch run inoculates the production bioreactor, grows the culture for around two weeks with concentrated feeds, and harvests the whole vessel once. The modelled case is the reference batch for this whole cluster: 2,000 L single-use bioreactor at 80% working volume, 3 g/L, 70% downstream yield, 90% of batches released, giving 3.024 kg of purified antibody. The equipment it runs on is the industry mainstream, from Sartorius Biostat STR bags to the single-use systems sold by Thermo Fisher and Cytiva.

Its carbon footprint is almost all electricity, and its electricity is almost all air. Of the 28,759 kWh the batch uses, 26,880 kWh is cleanroom HVAC running for 336 hours. Agitation and aeration together are under 80 kWh. That is why the CO2e benchmark table for biologics puts this batch at 9,510 kWh per kg and why every lever on the fed-batch side of this comparison is really a lever on hours of occupancy.

When fed-batch wins

On everything that scales with volume rather than time. A fed-batch run buys one working volume of medium, one set of bags and filters, and one harvest to purify. Per run it emits 19% less than the perfusion run (12,036 against 14,909 kg CO₂e), and in studies that count water and consumables but not the room, that is decisive. It also wins on robustness: a two-week run has less time to fail than a month-long one, and failed runs carry their full footprint with no product to divide it by.

How a perfusion process spends its carbon

A perfusion bioreactor holds cells at high density for weeks while fresh medium flows in and product-containing permeate flows out through a cell retention device, most often an alternating tangential flow filter such as the Repligen XCell ATF or a tangential flow depth filter such as KrosFlo TFDF. Our ATF vs TFF cell retention comparison covers the hardware. The modelled case is a 500 L vessel run for 30 days at one vessel volume per day, 1.2 g/L in the permeate, 75% downstream yield and 85% of runs released: 9.180 kg of product per run.

Each run uses more of everything that scales with volume: 30 working volumes of medium, 157 kg of single-use polymer against 82, 93,200 L of process water. It uses 21% more electricity per run too, because it occupies its suite for 30 days. But it makes three times the product, so every one of those burdens is divided by a larger number. Per kilogram, energy falls from 9,510 to 3,805 kWh and the carbon from plastic falls from 147.6 to 92.7 kg CO₂e.

When perfusion wins

Whenever time-proportional burden dominates, which in a classified mammalian suite it always does. The comparison that matters is not litres or runs but kilograms per suite-day. Madabhushi, Pinto and Lin (2022) made the same point from the materials side: the process mass intensity of their continuous process was comparable to batch, but because its output per unit time was several-fold higher, its energy per gram of drug substance was likely lower.

The fed-batch vs perfusion carbon footprint, stage by stage

Both presets run through the open-factor LCA calculator with nothing changed. Figures are kg CO₂e per kg of purified product.

Perfusion is lower on every stage per kilogram, and higher on almost every stage per run. That pattern is the whole story in miniature: the process does more, for longer, and gets more out. Electricity is still 93.7% of the perfusion carbon footprint, and 32,400 of its 34,926 kWh is cleanroom air.

One row deserves a flag. The calculator charges media per litre of working volume, which suits fed-batch. For perfusion the run feeds 30 working volumes, so the honest input is the medium summed over the run. Entering it that way (glucose and salts multiplied by 30) adds 198 kg CO₂e to the run and moves the result from 1,624.0 to 1,645.6, a 1.33% change. Medium is a large mass flow and a small carbon flow, as the process mass intensity guide found for the fed-batch case too.

Where the 2.45x actually comes from

The two presets differ in eleven inputs at once, so the headline ratio is a bundle. The useful way to unbundle it is through the term that dominates both carbon footprints. Cleanroom HVAC energy per kilogram is:

HVAC kWh per kg = suite HVAC power (kW) × 24 ÷ product per day (kg/day)

Fed-batch: 80 × 24 ÷ 0.216 = 8,889 kWh/kg. Perfusion: 45 × 24 ÷ 0.306 = 3,529 kWh/kg. The ratio, 2.52, is the product of two separate things: an 80 to 45 kW suite (1.78x) and 0.306 to 0.216 kg per day (1.42x). Only the second is a property of perfusion. The first is an assumption that a 500 L train fits in a smaller classified room than a 2,000 L one, which is reasonable, but it is a facility decision and not a process advantage.

Running the full model with the suite held equal separates them cleanly:

kg CO₂e per kg product, 400 g CO₂e/kWh.

On a matched suite, intensification is worth 1.46x to 1.49x. That is still a large, real reduction, nearly a third off the fed-batch figure, and it comes purely from more product per hour of conditioned air. The rest of the headline, 1.64x to 1.68x depending on which suite you hold fixed, is the smaller room. On a log scale that is 55% to 58% room and the rest process.

The comparison can also be turned around. If a perfusion train needed a larger suite, because of the retention skid, media hold tanks and the extra harvest handling, how large could it get before the advantage vanished? Holding everything else at the preset, perfusion matches fed-batch at 120.1 kW of suite HVAC, 1.5 times the fed-batch room. Any perfusion facility that comes in below that wins on carbon.

Equal output is a third lens. To match the fed-batch train's 0.216 kg per day, a perfusion vessel at these settings needs only 353 L. That is the sense in which intensification is sometimes described as a footprint reduction in the literal, square-metre meaning, and the carbon follows the square metres.

Why the published studies say fed-batch is greener

If the model ended at the table above, this would be an easy page. It does not, because the two best-known studies that compared these modes directly reached the opposite conclusion.

• Bunnak et al. (2016) built a combined LCA and cost-of-goods framework for mAb manufacture and found that a standard perfusion process, pooling every 4 days, had cost of goods similar to fed-batch but consumed 35% more water, demanded 17% more energy and emitted 17% more CO₂. Water was their most important category, because the energy in their model was largely the energy of making purified water and water for injection. Extending pooling to 8 days made perfusion the greener option.
• Pollock, Ho and Farid (2013) compared fed-batch with spin-filter and ATF perfusion under uncertainty. ATF perfusion offered up to 20% lower cost of goods, but fed-batch performed better on environmental sustainability, with lower water and consumable use, and was the more robust process once failure rates were simulated.

Neither result contradicts the model. They are measuring a different mix of burdens. Every input in a bioprocess inventory is either time-proportional (cleanroom HVAC, utilities that run while the suite is occupied) or volume-proportional (medium, buffer, water, bags, filters, waste). Perfusion is better on the first and worse on the second. Which mode wins depends on which group dominates the boundary.

That can be tested directly. Take the perfusion preset and load it with the literature's direction of travel: water per kilogram 1.35 times the fed-batch figure (245,893 L per run instead of 93,200) and single-use polymer per kilogram 1.5 times the fed-batch figure (373 kg per run instead of 157). These are stress-test assumptions chosen to match the published direction, not measurements. Then vary the cleanroom.

kg CO₂e per kg product, 400 g CO₂e/kWh. Fed-batch is the unchanged preset in every row.

With the cleanroom out and the literature's penalties in, fed-batch wins, which is the published result reproduced. Put any cleanroom back in and the verdict flips: the break-even is 2.3 kW of suite HVAC, 3% of the 80 kW in the fed-batch preset. At that point HVAC is still only 29% of the fed-batch electricity. A classified suite running continuously for two to four weeks is far above it.

So the disagreement is a boundary question, the same kind our gate-to-gate vs cradle-to-gate comparison found reversing the single-use verdict. A study that models the process train and its utilities but not the building around it will tend to favour fed-batch. A study that charges the room will favour perfusion, and by a margin that dwarfs the water and plastic penalties. Both are internally correct. Before quoting either, check which one you are reading.

There is one more caveat on our side. The preset gives perfusion half the water per kilogram of fed-batch, while Bunnak et al. found more. If downstream buffer scales with grams purified rather than with litres harvested, perfusion needs 181,200 L per run instead of 93,200, roughly fed-batch's water intensity. That raises its carbon footprint by 4.10%, to 1,690.7. Water intensity is genuinely uncertain for perfusion. Its carbon consequence is not large.

What would erase the perfusion advantage?

Each input below was moved on its own until perfusion matched the fed-batch 3,980 kg CO₂e/kg, first in the preset 45 kW suite and then in a matched 80 kW suite.

Run failure is the realistic threat. A failed perfusion run consumes a month of suite time and every bag and litre of medium, and yields nothing. On a matched suite the advantage holds down to a 58.1% success rate, so a process that loses four runs in ten is no better than fed-batch on carbon. The sensitivity is steep: at 75% success the matched-suite figure is 3,085, at 65% it is 3,560. This is the environmental face of the robustness concern Pollock et al. raised, and it is why a perfusion carbon claim should always state the success rate assumed. Our contamination investigation guide covers the failure mode that ends most long runs.

Titer and water are not realistic threats. Perfusion would need to fall below 0.82 g/L in the permeate on a matched suite before it lost, and water would have to rise 35-fold. Water costs about 6.9 kg CO₂e per cubic metre here once distillation energy is counted, which is too little to move a carbon footprint dominated by the room.

Run length helps, with diminishing returns. Holding one vessel volume per day and stretching the run, the per-kilogram figure goes 1,865 (14 days), 1,730 (20), 1,624 (30), 1,554 (45), 1,519 (60). Longer runs spread the fixed set of consumables over more product. This is the same lever as Bunnak's longer pooling interval, and it only pays if the success rate holds as the run lengthens, which the model does not assume for you.

Does the electricity grid change the fed-batch vs perfusion answer?

Not the verdict. Carbon per kilogram is exactly linear in grid intensity for both processes, kg CO₂e/kg = a × grid + b, with fed-batch at a = 9.5104, b = 176.00 and perfusion at a = 3.8046, b = 102.18. The ratio therefore moves only between two limits.

Across all 26 US eGRID subregions the preset ratio runs from 2.370 to 2.473, and the matched-suite ratio from 1.457 to 1.480. Perfusion is lower in every one. The absolute gap is what moves: 840 kg CO₂e/kg in upstate New York, 4,420 in MRO East.

The zero-grid limit is the interesting one. With fully clean power the room costs nothing, so suite size drops out entirely and both bases converge on 1.72x. What remains is pure intensification of the volume-proportional burdens: perfusion's plastic, media and water per kilogram are lower because it makes more product per set of consumables. That makes the gap more robust, not less, as electricity decarbonises, but it also hands the decision back to the inventory items the literature worries about. On a clean grid, the plastic per kilogram is the number to watch, as the decarbonising biomanufacturing guide found for the single-use versus stainless choice.

Pros and cons on carbon

Which comparison should you trust for your decision?

Whatever the decision, state three things beside any fed-batch vs perfusion number: whether cleanroom HVAC is inside the boundary, the suite power assumed for each, and the run success rate. Without them the figure cannot be reproduced, and the direction of the result cannot be trusted.

Real-world use cases

Per run, per day and per year

Per-kilogram figures decide the fed-batch vs perfusion comparison, but sites report absolute emissions, so the basis matters. The intensity advantage carries through to absolute emissions here because perfusion makes the same product with less, unlike a titer increase in a fixed fed-batch schedule, which cuts intensity without cutting the annual total.

48.4 kg/yr is the annual output of the fed-batch train at 16 batches per year, from the life cycle costing case. Per-day figures exclude turnaround time between runs.

For the cost side of the same choice, which Pollock et al. found also tends to favour ATF perfusion, the fermentation economics calculator gives cost of goods per gram, and the perfusion calculator sizes the medium and cell-specific perfusion rate that set a perfusion run's volume-proportional burdens.

Equipment landscape

The carbon result depends far more on how a process is run than on whose hardware it runs on, but the equipment sets what is possible: retention device, bag format and how large a room the train needs.

Fed-batch production bioreactors

• Sartorius Biostat STR: single-use stirred tanks up to 2,000 L, the format of the modelled fed-batch case.
• Thermo Fisher HyPerforma: single-use bioreactors including the DynaDrive line for high-density culture.
• Cytiva Xcellerex: single-use stirred tanks used for both fed-batch and perfusion operation.
• Getinge Applikon: glass and stainless bioreactors common in process development.

Perfusion cell retention

• Repligen XCell ATF: alternating tangential flow, the most widely used retention device for CHO perfusion.
• Repligen KrosFlo TFDF: tangential flow depth filtration, also used for perfusion and clarification.
• Cytiva TFF cell retention: recirculating tangential flow approaches to perfusion.
• Levitronix: magnetically levitated pumps used in TFF perfusion loops.
• MilliporeSigma and Meissner: filters and single-use assemblies for harvest and media handling.

A recurring vendor claim is that perfusion cuts CO₂ by around half against fed-batch. That is consistent with the preset comparison here, but the figure should be read with this page's decomposition in mind: much of it can be a smaller room rather than a better process. For the process-mode comparison without the carbon, see batch vs fed-batch vs perfusion.

Related calculators and tools

Related articles

Resources and references

• Bunnak, Allmendinger, Ramasamy, Lettieri & Titchener-Hooker (2016), Biotechnology Progress 32(5):1324–1335. Life-cycle and cost of goods assessment of fed-batch and perfusion-based manufacturing processes for mAbs.
• Pollock, Ho & Farid (2013), Biotechnology and Bioengineering 110(1):206–219. Fed-batch and perfusion culture processes: economic, environmental, and operational feasibility under uncertainty.
• Madabhushi, Pinto & Lin (2022), New Biotechnology 72:122–127. Comparison of process mass intensity of continuous and batch manufacturing processes for biologics.
• Bielser, Wolf, Souquet, Broly & Morbidelli (2018), Biotechnology Advances 36(4):1328–1340. Perfusion mammalian cell culture for recombinant protein manufacturing: a critical review.

Further reading

• Walther et al. (2015), Journal of Biotechnology 213:3–12. The business impact of an integrated continuous biomanufacturing platform for recombinant protein production.
• US EPA eGRID. Public-domain source of the 26 subregion grid factors used for the regional ratios on this page.