Cell-Free Protein Synthesis: Lysate, PURE and the Open Reaction

Strip away the membrane and translation becomes a reaction you can pipette into. Ribosomes, elongation factors, tRNAs and synthetases keep working in a tube, and every component becomes addressable: you can impose a redox buffer, add a chaperone, feed isotope-labelled amino acids, or supply an acylated tRNA carrying a residue no genetic code encodes. The same openness is the liability. Nothing is in the tube unless the extract brought it or you put it there, and nothing is absent unless you took it out.

Two starting materials: an extract you inherit, or a system you assemble

A crude extract carries the translation apparatus along with everything else soluble in the cell — nucleases, proteases, disulfide-reducing enzymes, central metabolism, and the cell's own mRNA. The classic extracts differ mainly in which of those dominates. An Escherichia coli S30 extract is the workhorse and supplies most of the numbers below; wheat embryo, rabbit reticulocyte and insect extracts supply eukaryotic translation machinery.

Reticulocyte lysate is saturated with globin mRNA, so an untreated lysate translates its own message rather than yours. Pelham and Jackson's fix is the classic one: preincubate with CaCl₂ and micrococcal nuclease, then add excess EGTA to chelate the Ca²⁺ and inactivate the nuclease. Lysates treated this way have negligible endogenous amino acid incorporation, and 75% of the original lysate's activity is recovered on adding globin mRNA, with no residual nuclease activity detectable and the tRNA functionally unimpaired ^[1]. They reported several mRNA species translated into full-sized products up to about 200,000 molecular weight with no detectable accumulation of incomplete protein products ^[1] — established for their preparation, not a guarantee that an arbitrary lysate returns full-length chains.

Wheat embryo extract had a different problem. Madin and colleagues traced the time-instability of existing plant systems to endogenous ribosome-inactivating proteins: inactivation by tritin begins during extract preparation and continues through the synthesis incubation. Extract from extensively washed embryos, free of the endosperm that is the source of tritin, gave continuous translation for 4 h in batch; in a dialysis bag with continuous substrate supply and byproduct removal, translation proceeded for more than 60 h and yielded 1–4 mg of enzymatically active proteins, and 0.6 mg of a 126-kDa tobacco mosaic virus protein, per millilitre of reaction ^[2]. The authors set this against what they describe as the common belief that cell-free systems are inherently fragile. On their evidence the instability was a removable contaminant of that preparation, not a property of the translational machinery, so do not carry "cell-free is unstable" forward as a fact about the technique.

Insect extract supplies the smallest figure quoted here: extract from cultured Spodoptera frugiperda 21 cells, driven by a template carrying the baculovirus polyhedrin 5′ untranslated region as a translational enhancer, gives roughly 50 µg/mL of protein ^[3].

What the reconstituted system buys, and what it costs

The PURE system replaces the extract with recombinant, individually purified, tagged translation factors. In its 2001 form it produced protein at about 160 µg/mL/h in batch with no supplementary apparatus, and the product could be purified within 1 h by affinity chromatography that removes the tagged factors rather than capturing the product ^[4]. Two consequences follow: you know the composition, because you built it. And you can omit a component: in the same report, leaving out a release factor allowed efficient incorporation of an unnatural amino acid using a suppressor tRNA ^[4]. In a lysate that subtraction requires engineering the source strain.

The price is yield and cost. An optimised PURE system, with component concentrations varied systematically, synthesised 0.38 mg/mL GFP in batch and 3.8 mg/mL in dialysis mode, reaching 4.3 and 4.4 mg/mL for dihydrofolate reductase and β-galactosidase respectively in dialysis, with synthesised protein at 30% (w/w) of total protein ^[5]. Those are well-behaved reporter proteins, not peptides. On cost, Hong and colleagues compared crude S30 lysate from a genomically recoded release-factor-1-deficient strain against a PURE translation system also lacking RF1, and found the extract-based approach more cost-effective and higher yielding — roughly 1000-fold on a milligram-protein-produced-per-dollar basis ^[6]. Keep that inside its bounds: two specific systems, one modified sfGFP product, PURE configured without RF1.

Reconstitution also removes the folding help. Niwa and colleagues used a reconstituted, chaperone-free translation system to test the major E. coli chaperones against roughly 800 aggregation-prone cytosolic E. coli proteins. The DnaK and GroEL systems drastically increased solubility for hundreds of proteins with weak biases; trigger factor had only a marginal effect on solubility; and combined addition rescued a subset that no single chaperone system rescued ^[7]. The readout is solubility of bacterial proteins, not folding fidelity and not peptides — but chaperone identity is plainly a variable you set, and combinations can act where singles do not.

Energy regeneration, and why a batch reaction stops

Translation consumes ATP and GTP faster than a tube can hold them, so the reaction runs on a secondary energy source that regenerates them. Conventional secondary sources — phosphoenolpyruvate, creatine phosphate — carry high-energy phosphate bonds, and every turnover liberates inorganic phosphate. Kim and Swartz showed that accumulating phosphate inhibits protein synthesis, most likely by reducing the concentration of free magnesium ion; their alternative generated the high-energy bonds inside the reaction instead, using Pediococcus pyruvate oxidase with thiamine pyrophosphate and FAD to make acetyl phosphate from pyruvate and inorganic phosphate, with acetate kinase (acetyl kinase in the source) already present with sufficient activity in the S30 extract regenerating ATP ^[8]. Oxygen is required and the H₂O₂ byproduct is sufficiently degraded by endogenous catalase. Synthesis was extended up to 2 h, and with repeated addition of pyruvate and amino acids it continued for 6 h to a final yield of 0.7 mg/mL ^[8]. That magnesium mechanism is offered as likely, not proven.

The Cytomim approach pushed the same logic further, replacing high-energy phosphate compounds with pyruvate and more closely replicating cytoplasmic conditions; chloramphenicol acetyltransferase accumulated for up to 6 h in batch to 700 µg/mL, without phosphate accumulation, pH change, exogenous enzyme addition, or expensive high-energy phosphate compounds ^[9]. Reagent cost, inhibitory byproducts such as phosphate, and pH change: those are the three problems.

The timescale matters. In one direct comparison, reaction rate with PEP slowed considerably within 30–45 min, whereas 3-phosphoglycerate sustained synthesis beyond 2 h and produced twice the protein under otherwise identical conditions ^[10]. If your reaction has stopped at 40 minutes, interrogate the energy system before the template.

The other half of the fix is physical: dialysis and continuous-exchange formats replenish substrates and drain small byproducts across a membrane. Kigawa and colleagues combined creatine phosphate and creatine kinase for regeneration, condensed the S30 extract by dialysis against polyethylene glycol to raise the synthesis rate, and dialysed the running reaction against a low-molecular-weight substrate solution, raising chloramphenicol acetyltransferase yield to 6 mg/mL of reaction mixture ^[11]. The wheat system's 60-hour run came from the same manoeuvre ^[2].

The open reaction is the point

Non-canonical residues, without a membrane in the way

Amber suppression in a cell requires the non-canonical amino acid to cross the membrane and survive metabolism; in a tube you weigh it out. Albayrak and Swartz co-produced the orthogonal tRNA and the modified protein in the same cell-free reaction, accumulating 0.9–1.7 mg/mL of modified soluble sfGFP containing either p-azido-L-phenylalanine or p-propargyloxy-L-phenylalanine, corresponding to 50–88% suppression efficiency, with the orthogonal tRNA transcribed either from a linearised plasmid or from a crude PCR product ^[12]. Their survey across 12 sites in sfGFP suggested that modified protein yields and suppression efficiencies did not correlate with the trends that had been reported for position effects ^[12] — a reason to screen sites empirically, not a licence to quote any particular positional rule.

Release factor 1 is the competing reaction. Cell-free synthesis from a genomically recoded RF1-deficient strain gave 190 ± 20 µg/mL of modified soluble sfGFP carrying a single p-propargyloxy- or p-acetyl-L-phenylalanine, an improvement of more than 250% over the parent strain retaining RF1 ^[6]. Extending that platform by functionally inactivating negative effectors in recoded cells, the most productive extracts reached 1780 ± 30 mg/L sfGFP, and supported site-specific introduction of 40 identical p-acetyl-L-phenylalanine residues into an elastin-like polypeptide at an incorporation accuracy of at least 98% and a yield of 96 ± 3 mg/L ^[13]. Multi-site incorporation had been limited by release factor competition; removing the competitor is what changed it. The selection logic behind orthogonal pairs, and why a gel band at the right mass is not evidence of incorporation, are treated in the companion guide on genetic code expansion.

Reprogramming the code for peptides specifically

Peptide work uses a sharper version of the same move. Flexizymes are flexible tRNA acylation ribozymes that enable preparation of a diverse array of non-proteinogenic acyl-tRNAs; integrated with a custom in vitro translation apparatus they form the flexible in vitro translation (FIT) system, in which the acyl-tRNAs read vacant codons and yield nonstandard peptides bearing N-methyl amino acids, D-amino acids and macrocyclic scaffolds ^[14]. Vacancy is the open-system trick: withhold a canonical amino acid from a defined reaction and its codon becomes free to reassign. Altering the peptide sequence then means changing only the mRNA template ^[14].

Kawakami and colleagues combined flexizyme acylation with the reconstituted PURE system to reassign arbitrarily chosen codons to N^α-methylated amino acids, established which N-methylated residues are accessible to the ribosome, and incorporated several into one nascent chain to give linear and cyclic N-methyl peptides in high purities ^[15]. These are peptide-class results, and they are the ones to reason from when the product is a peptide — the milligram-per-millilitre figures elsewhere here were measured on reporter proteins.

The same reaction is the front end of display selection. Roberts and Szostak generated covalent mRNA-peptide fusions by in vitro translation of synthetic mRNAs carrying puromycin at their 3′ end, then enriched fusions displaying a myc epitope peptide from a pool of random-sequence fusions by immunoprecipitation ^[16]. Both a reprogrammed code and a display selection are built on the same in vitro translation reaction, so this guide and the library-screening guide covering phage and mRNA display describe adjacent halves of one workflow.

Sequences a cell will not tolerate, and labels a cell would scramble

A product that kills the host imposes no penalty on a reaction with no host. Watanabe and colleagues make the case for structural work: proteins toxic to cells cannot be prepared in quantity in vivo, and cell-free synthesis bypasses the toxicity problem ^[17]. They list three advantages of the wheat germ system for this: the gene sits under a plant translational signal, reducing toxic expression in a bacterial host; the system shows little discrimination between methionine and selenomethionine, so selenomethionylated protein is straightforward to prepare for SAD and MAD phasing; and transcription is uncoupled from translation, so toxicity of the translation product towards DNA or its transcription, if any, can be bypassed. Their demonstration was a 4-base-pair-cutter restriction enzyme, expected to be toxic to any genome-bearing cell ^[17].

Isotope labelling is the other case where feeding the reaction beats feeding a culture. Kigawa and colleagues achieved stable-isotope labelling with ¹³C/¹⁵N-labelled amino acids for NMR in the same optimised system that reached 6 mg/mL ^[11]. Label goes into the reaction directly rather than through a growing culture's biosynthesis, and consumption is capped at the reaction volume.

Redox control and disulfide formation

An E. coli extract is reducing, and disulfide-bonded products fail in it by default. Kim and Swartz found that disulfide-reducing activities in the extract initially prevented disulfide bond formation, and that a simple pretreatment of the extract with iodoacetamide abolished the reducing activity while leaving the extract active for protein synthesis even under oxidising conditions. A glutathione redox buffer coupled with DsbC disulfide isomerase and pH optimisation then produced 40 µg/mL of active urokinase protease in a simple batch reaction ^[18].

The same variables carry to scale. An open cell-free synthesis platform was used to optimise extract activation, gene sequence and redox folding conditions for disulfide bond formation at microlitre scale, then run as batch reactions in standard bioreactors, producing a bioactive human cytokine at 700 mg/L in 10 h ^[19]. Separate the two: linear scalability is reported for the reaction format, while the 700 mg/L figure belongs to that protein under those redox conditions.

Linear templates and the exonucleases waiting for them

A PCR product going straight into the reaction skips cloning, which is much of the appeal for screening. The obstacle is that E. coli extracts degrade linear DNA. Michel-Reydellet and colleagues, describing linear DNA instability as what appears to be a major limitation of the system, modified the genome of strain A19 by removing endA (endonuclease I) and replacing the recCBD operon — in which recD encodes exonuclease V — with the lambda phage recombination system. Extract from that strain increased the stability of PCR products and gave chloramphenicol acetyltransferase yields from PCR templates comparable to plasmid, at 500–600 µg/mL in batch; PCR products amplified from genomic DNA and extended with a T7 promoter and terminator gave 550 µg/mL in 96-well plates ^[20].

If you cannot change the strain, protect the DNA in the tube. The Gam protein of phage lambda, an inhibitor of RecBCD (exonuclease V), protects linear PCR templates in vitro, and adding purified Gam to BL21 extract improved synthesis from PCR templates to a level comparable to plasmid template ^[10]. A cheaper variant uses the enzyme's own recognition sequence as a decoy: double-stranded DNA encoding χ sites — eight-base-pair sequences preferentially bound by the RecBCD machinery — stabilised linear DNA and enhanced expression in E. coli transcription-translation systems, with χ-site DNA and Gam yielding similar enhancements and as few as four χ sites sufficient for robust expression ^[21]. That was measured in TXTL preparations from E. coli; it says nothing about eukaryotic extracts, which carry their own nucleases.

A reconstituted system sidesteps the problem: assembled from purified components ^[4], it holds no host exonuclease unless you add one.

Scale, and what a yield number is telling you

Reaction format moves the number as much as the choice of system does: reconstituted PURE gave 0.38 mg/mL GFP in batch against 3.8 mg/mL for the same system in dialysis ^[5]; a dialysed, PEG-condensed S30 reaction reached 6 mg/mL of chloramphenicol acetyltransferase ^[11]; wheat embryo in a dialysis bag gave 1–4 mg/mL of active proteins ^[2]; insect extract gave about 50 µg/mL ^[3]. Reaction volumes span microlitres to standard bioreactors ^[19].

Every one of those was measured on a soluble reporter protein or a well-behaved enzyme. Superfolder GFP in particular is chosen because it folds and reports; it is a poor proxy for an aggregation-prone sequence and a worse one for a short peptide. Where modification load rises the cost is visible: the same work reports 1780 ± 30 mg/L for sfGFP and 96 ± 3 mg/L for the 40-site modified elastin-like polypeptide ^[13]. Those are different products, not one product at two modification loads.

Verify the product rather than assuming expression

Cell-free reactions are quantified by proxies, and each answers a narrower question than "yield" suggests. Fluorescence reports folded chromophore. Total amino acid incorporation reports label in acid-precipitable material. Both are separate measurements from full-length product, which is why Pelham and Jackson reported incorporation activity and the absence of detectable incomplete products as two distinct findings about their system ^[1].

The literature that gets this right reports two axes. Albayrak and Swartz give modified protein concentration and suppression efficiency separately, because a reaction can accumulate protein while failing to suppress ^[12]. Martin and colleagues report yield and incorporation accuracy separately, giving at least 98% accuracy of incorporation alongside 96 ± 3 mg/L for the 40-site product ^[13]. For a peptide the mass measurement is the primary evidence in any case, since a short chain will not resolve usefully on a gel.

Three checks before you believe a cell-free peptide preparation. Confirm intact mass, since a truncation or a canonical residue at a reassigned codon shifts it. Confirm the site, by fragment-ion coverage spanning any position you reassigned, because intact mass alone cannot localise a substitution. And run the no-template and no-added-residue controls, because a lysate already contains protein and a reconstituted system contains every factor you purified — neither background is zero.