Cell-Free Production of Difficult Targets for Structural Biology Sample Prep

Every downstream method in structural biology, X-ray crystallography, NMR, cryo-EM, depends on one thing upstream: a protein sample pure enough and plentiful enough to feed into the instrument. Cell-free protein synthesis (CFPS) matters to this field because it removes the specific failure points, toxicity, insolubility, isotope labeling constraints, and thin yields from hard targets, that make sample prep the place where structural biology projects quietly die.

Cryo-EM needs at least 50 micrograms of highly purified protein to get started. X-ray crystallography wants a milligram or more before a crystal tray is even worth setting up. These are not high bars set by fussy structural biologists; they're the floor. And a meaningful share of the targets structural biology actually cares about, membrane proteins, toxins, receptors, rare human proteins, can't clear that floor through ordinary cell-based expression no matter how much troubleshooting time gets thrown at it.

What cell-free protein synthesis is and how it removes the cell from the equation

CFPS runs protein synthesis outside a living cell. Take a cell extract (the cytosolic and organelle contents, with membranes stripped out), add a DNA template, and supply everything the ribosome needs on top: transcription and translation machinery, nucleotides, amino acids, an energy regeneration system, cofactors, and buffer. The reaction runs in a tube. There's no cell to keep alive, which turns out to be the whole point.

The DNA template can be a plasmid or a linear PCR product straight off the bench. No cloning into a vector, no transformation, no colony picking and screening. That alone cuts days out of testing a new construct.

Two broad platform designs dominate. Crude lysate systems, E. coli S30 extract being the standard example, cost a small fraction of a cent per microliter and scale from microliter reactions up to industrial volumes, though the extract composition is undefined: whatever was in the cell when it was lysed is in the reaction. Reconstituted systems, known generally as PURE systems, are built from purified, individually defined components. They're free of nucleases and proteases that would otherwise chew up template or product, and they give exceptional reproducibility from batch to batch.

Extract source is its own decision. E. coli, wheat germ, rabbit reticulocyte, insect cell (Sf21), CHO, and human cell extracts are all in use. Plant-based systems built on tobacco BY-2 lysate are commercially available. CHO and yeast lysates have been demonstrated in the literature too, though commercial access to them is thinner.

Toxic proteins: how CFPS changes the arithmetic of cell-based expression failure

Some proteins kill the cells asked to make them. A protein that jams a biosynthetic pathway or blocks cell division doesn't just express poorly, it gets selected against. The colonies that grow are the ones that silenced or mutated the construct, which means the plate is enriching for exactly the clones a structural biologist doesn't want.

This pattern appears constantly in the targets structural biology most wants solved: membrane proteins overexpressed in a foreign host, bacteriocins, restriction endonucleases, the toxin domains of toxin-antitoxin systems. GPCRs sit in this category too, alongside the tetracycline pump TetA and cytolethal distending toxin, both of which have been expressed successfully in CFPS specifically because there was no host cell around to object.

CFPS doesn't solve toxicity by engineering around it, the way a structural biologist might swap in a tighter promoter or a specialized strain. It just removes the thing toxicity acts on. There's no viable cell in the tube, so there's nothing for the protein to kill and nothing to select against. No strain engineering, no iterative rounds of induction timing and temperature tweaks that eat up weeks in a cell-based workflow. The reaction either makes protein or it doesn't, and that answer comes back in hours, not passages of subculturing.

Membrane proteins and insoluble targets: matching the chemical environment to the protein's needs

Membrane proteins are a separate headache from toxicity, though the two often overlap. Overexpression tends to be poorly tolerated, yields run low even when the cells survive, and in E. coli the protein frequently ends up misfolded into inclusion bodies, dense clumps of aggregated polypeptide that are functionally useless for structural work. Pulling a folded membrane protein back out of a cellular extract full of native lipids, other membrane proteins, and chaperones adds another layer of difficulty on top.

CFPS reactions are open systems, and that's the lever here. Detergents, lipids, or nanodiscs can be added directly to the reaction as the protein is being made, giving it a folding environment on the spot rather than asking it to fold first and get rescued later. A paper in Nature Communications examining artificial membrane environments for cell-free synthesis stated that general rules for what makes membrane protein synthesis succeed are elusive and sometimes contradictory. That's not a knock against the method, it's a description of the problem. Every membrane target seems to want its own recipe, and an open reaction is what makes testing recipes fast.

The yield numbers back this up. Eukaryotic CFPS in a Sf21-based continuous exchange format has produced up to 285 micrograms per milliliter of epidermal growth factor receptor, and up to 700 micrograms per milliliter of the viral envelope protein gp67. For one receptor target, a Sf21-based system produced correctly glycosylated protein but at lower yield, while a mammalian-based system gave a higher yield with glycosylation that wasn't correct. Extract choice is a real trade-off a structural biologist has to weigh against what the downstream method can tolerate. It's a real trade-off a structural biologist has to weigh against what the downstream method can tolerate.

Isotope labeling and selenomethionine incorporation: where CFPS turns a painful workflow into a direct one

NMR structure determination runs on isotope-labeled protein, typically stable isotopes. Getting a cell to make that protein means shifting it from rich media to minimal media loaded with labeled precursors, and cells being cells, they metabolize those precursors into other compounds along the way. Labeling gets scrambled, incorporation is inefficient, and a chunk of the isotope budget, which isn't cheap, goes to waste. Selenomethionine substitution for crystallographic phasing has its own version of this problem: selenomethionine incorporation in cell-based systems adds another layer of complexity on top of an already fussy expression system.

CFPS sidesteps the scrambling because the reaction has no competing metabolic pathways to scramble through. Labeled amino acids or selenomethionine go into the tube directly, without a cell's competing metabolic pathways intercepting them first.

The Center for Eukaryotic Structural Genomics built a wheat germ cell-free platform for exactly this purpose: producing stable isotope-labeled eukaryotic proteins for NMR. The pipeline included small-scale screening upfront, a step meant to predict, before committing real material and instrument time, which targets would actually yield protein of the quantity and quality NMR structure work needs.

A related trick, non-canonical amino acid incorporation, shows how far the open-system idea stretches. Using a genomically recoded E. coli extract that lacks release factor 1, researchers site-specifically incorporated 40 identical copies of p-acetyl-l-phenylalanine into an elastin-like polypeptide, with incorporation accuracy at 98% or better and yield of 96 ± 3 milligrams per liter. The same platform separately produced 1,780 ± 30 milligrams per liter of superfolder GFP, a useful reminder that this kind of genetic code reprogramming doesn't come at the cost of gutting general output.

From cell-free reaction to structural data: what the sample prep-to-structure timeline looks like

Two approaches show what this looks like end to end. One uses a wheat germ cell-free platform with expression and affinity purification to move a construct through to high-resolution cryo-EM structure determination. That kind of compressed start-to-structure timeline is one that a cell-based pipeline dealing with a difficult target would struggle to approach.

The other approach skips purification almost entirely. Structures have been solved directly from cell-free expression mixtures using just 25 microliters of reaction volume, with minimal or no cleanup, in under 24 hours, and at resolutions better than 3 angstroms. This is the Evans et al. workflow, built on PURE E. coli-based transcription-translation, where the reaction mixture goes straight onto cryo-EM grids. The choice of PURE over crude lysate here wasn't arbitrary: proteomic analysis of crude lysate found roughly 3,000 unique proteins present before the target was even expressed, and the target was not detectable among the dominant protein components by abundance. Trying to pick a single particle out of that background on a cryo-EM grid would have been close to hopeless.

That's a real trade-off, not a quirk of one paper. PURE systems cost more per reaction and produce less total protein, but the payoff is a defined, low-background environment where the target isn't competing with thousands of other proteins for grid space. Crude lysate is cheaper and often higher-yielding, but that background complexity has to be reckoned with somewhere downstream. Which system to reach for depends on whether purification comes before or after the structural method gets applied.

Proteins that come out of a cell-free reaction go into the same affinity tag capture, size exclusion chromatography, and other standard purification steps as anything made in a cell. CFPS removes the upstream expression failures that used to keep protein from ever reaching a purification column in usable form. It's removing the upstream expression failures that used to keep protein from ever reaching a purification column in usable form.

High-throughput construct screening before committing to milligram-scale prep

Getting a construct right, the domain boundaries, the fusion tag, the truncations, the surface mutations meant to help crystal packing, is a huge share of the work in structural biology, and doing that testing in cells is slow. Each variant means another cloning step, another transformation, another round of colony screening before there's even protein to look at.

CFPS collapses that cycle. A construct variant made straight from a PCR product can go into a reaction the same day, with no cloning step standing between the DNA and a folded protein to test. Running ten or twenty variants in parallel becomes a normal afternoon's work rather than a week of bench time spread across incubators.

CFPS has been used to rapidly screen protease variants generated by machine learning models. An ML model is only as useful as the training data behind it, and that training data usually has to come from real proteins that were actually expressed and tested. Purification is the bottleneck standing between a model's output and usable functional data, and CFPS is one of the more direct ways to relieve it. The same speed applies to more mundane optimization work: figuring out which expression host, which chaperone set, what temperature, and what codon optimization gets the best result for a given construct, all questions structural biology has to answer anyway before committing to a milligram-scale prep.

Choosing the right CFPS platform for a structural biology target

The choice of platform comes down to a handful of real decisions. Bacterial proteins generally do fine in E. coli-derived CFPS. Complex eukaryotic targets that need disulfide bonds or glycosylation need a eukaryotic extract instead, whether that's yeast, insect, or another mammalian-derived source.

Defined versus crude is its own axis. PURE, reconstituted systems earn their cost when reproducibility matters, when a low cryo-EM background is the point, or when the genetic code itself is being reprogrammed for non-canonical amino acids. Crude lysates earn their keep on cost-sensitive, milligram-scale runs where background complexity isn't going to interfere with the downstream method.

Membrane protein work still comes down to per-target trial and error on the lipid, detergent, or nanodisc environment, and glycosylation, if the target needs it, only comes correctly from specific extracts, not from every eukaryotic system interchangeably. Format matters too: batch reactions are simple and fine for screening, while continuous exchange cell-free (CECF) formats push yield higher, which is where those 285 and 700 microgram-per-milliliter membrane protein numbers came from.

When a target is proving difficult, knowing exactly what's in the reaction, which components, at what concentration, from which lot, is not a nice-to-have. It's the difference between troubleshooting systematically and guessing at a black box. Sepia Biosciences' OpenCFPS platform is built around that idea directly, offering documented formulations and documented formulations across multiple modes, with pricing meant to keep high-throughput screening from being gated by reagent cost.

What CFPS cannot yet do

CFPS is not a universal replacement for cell-based expression, and treating it as one would overstate what the current evidence shows. Total protein yield per dollar still tends to favor established cell-based fermentation for targets that express well in the first place, the ordinary soluble proteins that were never the problem to begin with. Scale-up beyond milligram quantities into gram-scale production, the kind biomanufacturing needs for therapeutic proteins, remains more mature and more thoroughly worked out in cell-based systems.

Certain post-translational modifications are also still extract-dependent in ways that aren't fully solved. Correct glycosylation, as the receptor example showed, only comes from specific eukaryotic extracts, and matching a target's exact modification requirements to the right extract is still a matter of testing, not a settled formula. For targets that express cleanly in a standard host, at the yield structural biology needs, without toxicity or aggregation getting in the way, cell-based expression remains the more practical route.

CFPS earns its place specifically at the point where cell-based expression already failed, or was expected to. That's a narrower claim than saying it replaces cells outright, but it's the claim the evidence actually supports, and for the targets structural biology cares most about, the toxic ones, the membrane-bound ones, the isotope-labeled ones, it's the claim that matters.