CO₂e Benchmarks for Biologics: Reference Figures and How to Read Them

Somebody has asked you for the carbon footprint of a kilogram of your drug substance, and somebody else has sent you a published figure to compare it against. The two numbers will not agree, and the reason is almost never the one people reach for.

This page does two things. It gives CO₂e benchmarks for six biologics processes calculated on a single, fully stated basis, so that they are comparable with each other. Then it shows how far a benchmark can move without the process changing at all, which is the part that decides whether any published comparison means anything. Every number here comes from the bioprocess LCA calculator, so all of it is reproducible rather than quoted.

A figure on its own is not a benchmark

A carbon figure for a biologic is a ratio. Emissions on top, product mass underneath, and both sides carry assumptions that a bare number hides completely.

Four items have to travel with the figure or it cannot be used by anyone else.

• The grid emission factor, in g CO₂e/kWh. Electricity is between 71% and 97% of the cradle-to-gate total in every process modelled below, so this single input dominates.
• Titer and downstream yield, because they set the denominator. A process that reports a low number by making more product is not the same as one that reports a low number by emitting less.
• The system boundary. Whether cleanroom HVAC, purified water generation, resin production and waste incineration are inside or outside the fence, as worked through in the gate-to-gate versus cradle-to-gate comparison.
• The functional unit. A kilogram of purified drug substance, a gram, a patient dose and a course of therapy are four different denominators, and only the first two are interchangeable by arithmetic.

Welling and Ryding, examining how environmental performance is distributed across published assessments, reached the same conclusion from the statistics. Comparisons between studies are only defensible once the underlying inventories sit on a common footing. The NREL harmonisation programme described by Heath and Mann was built for exactly this reason in the electricity sector, where restating published results onto shared assumptions collapsed apparent disagreements that had persisted for years.

Reference figures: six processes, one basis

The table below is the useful half of this page, and it is the benchmark set the rest of the article works with. Six biomanufacturing processes, all run through the same model, all at 400 g CO₂e/kWh and all cradle-to-gate. Because the basis is identical, these CO₂e benchmarks are comparable with each other in a way that figures gathered from six separate publications are not.

Three things are worth reading off it before anything else.

The 21-fold range is real, and it is mostly a denominator story. The E. coli batch is not a cleaner process in any engineering sense. It is a process that makes 30.36 kg where the CHO batch makes 3.024 kg, in a tenth of the calendar time, so the cleanroom slot it occupies is divided across ten times the product. Expression system choice shows up here as an occupancy effect far more than as a metabolic one.

Electricity is 71% to 97% everywhere. Nothing else comes close. Media is the largest non-electrical line only in the microbial cases, at 16.3% for E. coli and 7.0% for Pichia. In the CHO cases the entire non-electrical inventory of media, water, polymer resin and incineration comes to 4.4%. That concentration is what makes the grid factor so decisive, and it is why the choice of emission factor source matters more than the rest of the inventory put together.

Perfusion is 2.45 times lower than fed-batch here, and stays there. Across all 26 US subregions the ratio only moves between 2.37 and 2.47, because both processes have similar electricity shares. Ratios between processes are far more portable than the processes' own levels, which is a theme that returns below.

The spread is mostly reporting, not process

Now hold the process completely still. Same vessel, same organism, same media, same cleanroom, same polymer inventory, same downstream yield. Change only two things that a published study would ordinarily just state in its methods: which grid the plant sits on, and what titer the campaign achieved.

Across the 26 US eGRID subregions the single-use CHO case moves from 1,453 kg CO₂e per kg on upstate New York's 134.3 g/kWh to 7,421 on MRO East's 761.8. That is 5.11-fold, for a plant that behaves identically. Titer from 1 to 10 g/L moves it from 11,940 to 1,194, exactly ten-fold, because everything in the numerator is titer-independent. Combine the two and the same unchanged process reports anywhere from 436 to 22,263 kg CO₂e per kg.

Fifty-one-fold, from reporting choices alone. The six genuinely different processes in Table 1 span 21-fold. How the number is reported moves it roughly two and a half times further than the choice of organism, scale and process mode combined.

The sharpest way to state the consequence is to invert it. Ask what process would have to be true for a study to report exactly 4,000 kg CO₂e per kg, and the answer is a whole family of processes rather than one.

A five-fold difference in titer is the difference between a struggling legacy campaign and a competitive modern one. If that entire difference can hide inside a single quoted figure, then the figure is not measuring what the reader thinks it is measuring.

Restating any figure onto your own grid

The fix is arithmetic rather than judgement, and it is exact rather than approximate.

Every emission source in the inventory is either proportional to electricity consumption or independent of it. So carbon per kilogram is a straight line in grid intensity.

The coefficients also explain why some processes are more portable than others. E. coli at 71.2% grid-dependence carries nearly a third of its footprint in media and steam, which do not move when the plant relocates. The stainless CHO case at 97.0% is almost pure electricity, so its published figure is almost purely a statement about the local grid.

What survives normalisation

Two things come through the restatement intact, and they are the parts worth quoting.

Energy intensity is grid-invariant by construction. Move the single-use CHO plant from the cleanest US subregion to the dirtiest and its carbon figure moves 5.11-fold while its energy intensity stays at exactly 9,510 kWh per kg. Across the six processes, energy intensity spans 337 to 9,629 kWh per kg, so it keeps full discrimination between processes while discarding the largest source of spurious variation. If a study reports one number, kWh per kg is the more useful one, and it happens to be the number an engineer can act on.

Rank order is mostly robust. There are 15 pairwise comparisons between the six processes. Fourteen of them give the same answer at every grid intensity between 134.3 and 761.8 g/kWh. Exactly one reverses inside that window: single-use against stainless, crossing at 479.2 g CO₂e/kWh, which is the finding worked through in the single-use versus stainless comparison. The three other crossovers that exist at all, the microbial pilot against both CHO cases and E. coli against Pichia, occur at 38 to 48 g/kWh, cleaner than any US subregion and reachable only on a hydro or nuclear supply.

So the honest summary is that comparisons between processes travel well and levels do not. A study saying "our perfusion route is 2.4 times lower than our fed-batch route" is telling you something durable. A study saying "our route is 1,600 kg CO₂e per kg" is telling you mostly about its electricity contract.

Reading the published mAb band

The benchmark most often quoted for monoclonal antibodies is a band of roughly 4,000 to 20,000 kg CO₂e per kg. The modelled single-use CHO case at 3,980 sits just at its lower edge, which is deliberate. The model's cleanroom and buffer inventories were calibrated so that a standard case lands there rather than an order of magnitude below, which is where a bioreactor-only boundary puts you.

Given everything above, the band should be read as a range of reporting positions rather than a range of process performance. Two observations make that concrete.

First, moving the one unchanged process across the 26 US subregions puts it inside the 4,000 to 20,000 band in 17 of 26 cases and below it in the nine cleanest. Whether a plant is "in the published range" is therefore substantially a statement about where it is plugged in.

Second, reaching the top of the band at a representative grid needs a titer that would be unusual today. At 400 g/kWh, 20,000 kg CO₂e per kg corresponds to 0.60 g/L, and 12,000 corresponds to 0.99 g/L. Those are plausible figures for the older studies in the literature and implausible for a modern fed-batch platform. Much of the width at the top of the band is the history of titer improvement, not a difference between contemporary facilities.

Set against sector-level data this is consistent. Belkhir and Elmeligi found the pharmaceutical industry's emission intensity per unit of revenue to be higher than the automotive sector's, and also found very wide variation between individual companies reporting on comparable terms. Wide dispersion is the normal condition of this data, and the fix is normalisation rather than a tighter average.

For the primary process-level literature, Pietrzykowski and colleagues' comparison of single-use and conventional monoclonal antibody manufacture remains the most-cited process study, Budzinski and colleagues established the mass intensity metric for biologics and later published a streamlined assessment of single-use technologies, and Amasawa and colleagues placed environmental impact and operating cost on the same axes for cultivation scenarios. Read alongside each other, their boundaries differ enough that their headline figures should not be subtracted from one another.

Per gram and per dose

Kilograms of drug substance is the right unit for a plant and the wrong one for almost every conversation outside it. The conversion is trivial and worth having to hand.

Two warnings attach to the right-hand column. It is drug substance only, so fill-finish, cold chain, packaging, distribution and administration are all outside it, and for a biologic the cold chain is not negligible. And a quantity of drug substance is not a dose. Dosing is weight-based for many molecules, and a course of therapy may be many administrations. Quote the per-gram figure and let the clinical side supply the multiplier.

It is also worth keeping the scale separate from commodity fermentation. Bulk fermentation products sit between roughly 1 and 18 kg CO₂e per kg, as tabulated in the fermentation carbon footprint guide. Biologics are two to three orders of magnitude above that, for the straightforward reason that the denominator is grams rather than tonnes. Comparing the two directly is a category error, and a surprisingly common one in sustainability decks.

Publishing a figure someone else can use

If you are on the reporting side, the following is enough to make a figure reusable, and it is a short list.

• Carbon intensity with its grid factor, side by side. Never the carbon figure alone. "3,980 kg CO₂e/kg at 400 g CO₂e/kWh" is a usable sentence. The first half on its own is not.
• Energy intensity in kWh per kg. This is the restatement coefficient a, and it is what lets a reader move your figure onto their grid exactly.
• The non-electrical residue. Coefficient b, in kg CO₂e per kg. Together with the line above, your figure becomes portable.
• Titer, downstream yield and batch success rate. The three inputs that set the denominator.
• The boundary, stated as a list of what is in. Cleanroom HVAC in or out is worth an order of magnitude on its own, and purified water generation is routinely forgotten, as the water footprint analysis sets out.
• Whether the factors are average or marginal. The attributional and consequential framings answer different questions and are not interchangeable, and the choice can reverse the value of a siting decision entirely.

Reporting a figure this way costs six lines of a methods section and converts a number that can only be admired into one that can be used. If the figure is going into a corporate disclosure rather than a paper, the same six lines are what make it auditable, and the scope 1, 2 and 3 allocation sits on top of them rather than replacing them.

What these numbers cannot settle

Four limits belong with any use of the figures on this page.

• They are screening estimates from one model. Every figure comes from a single calculator with an open, published factor set. That makes them internally consistent and reproducible, which is the point, but it does not make them a verified assessment of any real facility, and none of them should reach an environmental product declaration or a regulatory submission.
• One indicator is not a verdict. The same matched pair of processes gives different answers depending on whether you look at climate, water or mass, which is why ISO 14044 prohibits publicly weighting indicators into a single score. The impact assessment stage is where that divergence is made explicit.
• Single-use polymer data is the weakest input. Resin production and incineration together carry 3.7% of the single-use CHO footprint and rest on the lowest-confidence factors in the set. They matter more at pilot scale, where they reach 14.9%.
• A benchmark is not a target. Matching a published figure tells you nothing about whether the achievable reduction was taken, and the achievable reductions are mostly in titer, yield and batch occupancy rather than in any of the inputs people instinctively reach for.

Used with those limits in view, the table at the top of this page does one job well. It gives a set of internally comparable CO₂e benchmarks for biologics, generated on a stated basis, that you can restate onto your own conditions in a single line of arithmetic. That is a more useful thing to own than a remembered number with no stamp on it.

Frequently asked questions

What is a typical CO₂e benchmark for producing a monoclonal antibody?

On a modelled 2,000 L single-use CHO fed-batch at 3 g/L titer, 70% downstream yield and a 400 g CO₂e/kWh grid, the cradle-to-gate figure is 3,980 kg CO₂e per kg of drug substance, or 3.98 kg CO₂e per gram. Published monoclonal antibody studies most often fall in a 4,000 to 20,000 kg CO₂e per kg band. That band is wide because it mixes different grids, titers and system boundaries rather than because the processes differ that much.

Why do published carbon footprints for the same product differ so much?

Because three reporting choices move the number more than the process does. Holding one modelled CHO process completely unchanged and varying only the grid region across the 26 US eGRID subregions and the titer between 1 and 10 g/L produces figures from 436 to 22,263 kg CO₂e per kg, a 51-fold spread. The six genuinely different processes modelled here span only 21-fold at a common grid. Reporting choices therefore account for more variation than the choice of organism, scale and process mode combined.

What information has to accompany a CO₂e figure for it to be usable?

Four items. The grid emission factor in g CO₂e/kWh, the titer and downstream yield behind the product mass, the system boundary, and the functional unit. Without the grid factor a reader cannot restate the number onto their own electricity supply, and electricity is 71% to 97% of the total in every process modelled here. Without titer they cannot separate process efficiency from denominator inflation.

Can I compare my facility's number to a published benchmark?

Only after restating both onto a common grid. A carbon figure is a straight line in grid intensity, so it can be restated exactly with two coefficients: the energy intensity in kWh per kg and the non-electrical residue in kg CO₂e per kg. For the modelled single-use CHO batch those are 9.5104 and 176.0, giving 3,980 kg CO₂e per kg at 400 g CO₂e/kWh and 2,462 at California's 240.4. Comparing unrestated figures mostly compares electricity supplies.

Is energy intensity a better benchmark than CO₂e per kg?

For comparing processes, yes. Across all 26 US eGRID subregions the carbon figure for one unchanged process moves 5.11-fold while its energy intensity stays at exactly 9,510 kWh per kg. Energy intensity removes the single largest source of spread while keeping full sensitivity to titer, yield and batch occupancy. Report kWh per kg alongside the carbon number and the carbon number becomes reproducible.

Does a low carbon figure mean an efficient process?

Not on its own. A headline of 4,000 kg CO₂e per kg is equally consistent with a 1.09 g/L titer on the cleanest US grid and a 5.57 g/L titer on the dirtiest, a 5.1-fold difference in real productivity behind one identical number. Rankings survive much better than levels: of the 15 pairwise comparisons between the six processes modelled here, 14 hold at every US grid intensity and only the single-use against stainless pair reverses, at 479.2 g CO₂e/kWh.

References

1. Welling S., Ryding S.-O. (2021). Distribution of environmental performance in life cycle assessments: implications for environmental benchmarking. The International Journal of Life Cycle Assessment 26:275–289. doi:10.1007/s11367-020-01852-3
2. Heath G.A., Mann M.K. (2012). Background and Reflections on the Life Cycle Assessment Harmonization Project. Journal of Industrial Ecology 16(S1). doi:10.1111/j.1530-9290.2012.00478.x
3. Belkhir L., Elmeligi A. (2019). Carbon footprint of the global pharmaceutical industry and relative impact of its major players. Journal of Cleaner Production 214:185–194. doi:10.1016/j.jclepro.2018.11.204
4. Pietrzykowski M., Flanagan W., Pizzi V., Brown A., Sinclair A., Monge M. (2013). An environmental life cycle assessment comparison of single-use and conventional process technology for the production of monoclonal antibodies. Journal of Cleaner Production 41:150–162. doi:10.1016/j.jclepro.2012.09.048
5. Budzinski K., Blewis M., Dahlin P., D'Aquila D., Esparza J., Gavin J., et al. (2019). Introduction of a process mass intensity metric for biologics. New Biotechnology 49:37–42. doi:10.1016/j.nbt.2018.07.005
6. Budzinski K., Constable D., D'Aquila D., Smith P., Madabhushi S.R., Whiting A., Costelloe T., Collins M. (2022). Streamlined life cycle assessment of single use technologies in biopharmaceutical manufacture. New Biotechnology 68:28–36. doi:10.1016/j.nbt.2022.01.002
7. Amasawa E., Kuroda H., Okamura K., Badr S., Sugiyama H. (2021). Cost–Benefit Analysis of Monoclonal Antibody Cultivation Scenarios in Terms of Life Cycle Environmental Impact and Operating Cost. ACS Sustainable Chemistry & Engineering 9:14012–14021. doi:10.1021/acssuschemeng.1c01435