Week 3 HW: Cell-Free Reaction Design

HOMEWORK #3

A. Compare cell-free reaction compositions

1. Why do potassium glutamate and magnesium glutamate concentrations differ between compositions, and which is more sensitive?

They differ because the two ions do different jobs, and only one of them is coupled to the energy system.

Potassium sets bulk ionic strength. K⁺ is the dominant monovalent cation of the E. coli cytoplasm, and glutamate is its dominant counter-ion; the pairing was chosen originally to mimic that environment rather than to hit a sharp catalytic optimum. Because its role is to establish an ionic background compatible with translation, the response surface around it is broad, and formulations can sit anywhere across a wide band without collapsing.

Magnesium is a structural and catalytic cofactor. Mg²⁺ holds the ribosomal subunits together, stabilises tRNA tertiary structure, and is required by the polymerases. It is also the ion the rest of the reaction competes for: NTPs, PEP and inorganic phosphate all chelate Mg²⁺, so the free Mg²⁺ concentration — the quantity that actually matters — is not the amount you pipetted. Too little and ribosomes dissociate; too much and translational fidelity drops and subunits over-associate.

Magnesium is far more sensitive, and the six reference plates for this coopetition show it directly:

Five of the six plates sit at exactly 8.22 mM magnesium regardless of protein or energy system, while potassium is tuned freely between 200 and 345 mM. In absolute terms the magnesium window is roughly seventy-seven times narrower. The same pattern appears in the published formulations: PANOx-SP uses 130 mM K(Glu) and the related PANOx formulation 200 mM — a 70 mM difference nobody treats as important — while magnesium moves only a few mM between them.

Olsen et al. (2026) make the mechanism explicit: in their low-cost system, new optimum magnesium and phosphate concentrations had to be determined for each lysate concentration and each new lysate batch [2]. Magnesium and phosphate re-optimise together precisely because phosphate sequesters Mg²⁺. Potassium is not described as needing that treatment.

Practical consequence for this assignment: magnesium is the parameter not to improvise on. I kept it at the consensus 8.22 mM in two of my three designs and moved it only where the reference data itself showed a protein-specific optimum.

2. How does a PEP+NTP formulation differ from a Ribose+NMP formulation?

They differ in where the phosphate bond energy comes from, and that one choice cascades through the whole recipe.

PEP + NTP is the direct route. You supply the nucleotides already triphosphorylated (ATP 1.25 mM, CTP/GTP/UTP 0.875 mM each in these plates) and supply phosphoenolpyruvate as a high-energy phosphate donor. Pyruvate kinase in the lysate transfers phosphate from PEP onto ADP to regenerate ATP. It is fast and it works, which is why PANOx-SP has been the field’s reference for twenty years.

Its problem is cost and chemistry. PEP is the single most expensive line item in a classical formulation — Olsen et al. report it accounted for roughly 75 % of all material costs [2]. And every phosphoryl transfer releases inorganic phosphate, which accumulates, chelates Mg²⁺ and eventually stalls the reaction, which is part of why PEP systems are short-lived.

Ribose + NMP pushes the work back onto the lysate’s own metabolism. You supply the cheap monophosphates (AMP, CMP, GMP, UMP) plus ribose as a carbon and energy substrate, and let endogenous central metabolism and the nucleotide salvage pathways phosphorylate them up to triphosphates and regenerate ATP. Inorganic phosphate is supplied deliberately (15 mM in the sfGFP and PETase ribose plates) rather than accumulating as waste.

The knock-on effects are visible across the whole column of the reference plates. The ribose formulations drop PEP, all four NTPs, folinic acid, CoA, NAD, putrescine and oxalic acid entirely. In place of NAD ($33/mL) they add nicotinamide at 4 mM — the cheap precursor, letting the lysate build the cofactor itself. They raise amino acids from 2 to 5 mM and HEPES from 57.5 to 75 mM, because a reaction that now runs its own metabolism needs more buffering and more substrate. Olsen et al. push the same logic further, showing GMP can be replaced by guanine plus ribose with no yield penalty, and UMP by its base plus ribose at reduced yield [2].

The trade-off is not uniform across proteins, and this is the most useful thing in the reference data:

For the two easy proteins the cheap system costs you 14–25 % of yield, which a large cost saving will easily outrun. For Reteplase it costs you more than 92 % of the yield. No plausible reagent saving recovers a 13-fold loss, so the energy-system choice is effectively made for you by the folding difficulty of the target.

3. Comparison of two published formulations

PANOx-SP (Jewett & Swartz 2004) [3] versus RF_opt (Olsen et al. 2026) [2].

The headline is a reported 95 % average cost reduction [2]. What is striking is where it comes from: not from a cleverer version of the same recipe but from deleting most of it. PANOx-SP supplies finished high-energy intermediates and purified cofactors; RF_opt supplies cheap precursors and lets the lysate’s surviving metabolism assemble them. The entire cofactor block — NAD, CoA, folinic acid, tRNA — disappears, and the single most expensive component, PEP, is removed outright.

The cost is reproducibility. RF_opt is reported to be more sensitive to how much lysate is added than PANOx-SP is, and magnesium and phosphate must be re-optimised for every new lysate [2]. That is the real trade: PANOx-SP buys robustness by overpaying for finished reagents.

Where the money actually is in this assignment

One observation worth stating, because it changes what is worth optimising here. The brief notes that lysate is typically the largest single expense in CFPS. In this coopetition it is fixed at 5,000 nL and cannot be modified, so it is a constant, not a variable:

The nucleotides and cofactors in a PEP-type plate cost about 1.6 × the lysate, in one twentieth of the volume. Since the lysate cannot be touched, the entire optimisation space is the 11,000 nL of variable reagents — and within it, a handful of nucleotides and cofactors dominate. That is what makes the NMP substitution the main lever rather than a refinement.

B. Reaction composition designs

Design strategy

Each design anchors on the better-performing reference plate for that protein and makes a small number of directed changes, rather than proposing a composition from scratch. Three reasons:

1. The reference plates were actually run, so they are known to be feasible and known to produce protein. A design that is a small perturbation of a working plate is far more likely to return interpretable data than one assembled from first principles.
2. With one shot per protein in this round, a design that changes many variables at once cannot be attributed if it wins or loses. Few, well-separated changes give a readable result.
3. Each change is checked against the plate’s spare volume (the water row), so feasibility is verifiable rather than assumed.

sfGFP — anchor: sfGFP Ribose (0.46 g/L), spare volume 170 nL

Why the ribose system at all. sfGFP loses only 25 % of yield on the cheap formulation while shedding PEP, all four NTPs, CoA, NAD, folinic acid, putrescine and oxalic acid. Cost per gram is cost per reaction divided by yield, so a reagent saving well above 25 % wins. sfGFP has no disulfides and folds robustly, so it is the target least likely to be hurt.

Why drop potassium to 273 mM. 345 mM is the highest potassium of any of the six plates; sfGFP’s own PEP plate runs at 280 mM and PETase Ribose at 300 mM. Since the potassium optimum is broad (A1), this tests whether the ribose plate was simply above its optimum, and it frees 1.6 µL of volume. 273 mM is not an arbitrary number — it is the value the Reteplase Ribose plate uses, so it is known to support a ribose-based reaction.

Why add spermidine. Smith et al. (2026) report that adding small amounts of spermidine reduced cost per gram of protein [1]. The Reteplase Ribose plate already runs spermidine at 1.25 mM in a ribose system, so compatibility is demonstrated. Polyamines stabilise ribosomes and tRNA, and at 100 nL from a 250 mM stock the volume cost is negligible.

Why not touch magnesium. Five of six plates converge on 8.22 mM. This is the parameter where a wrong guess costs the most and the reference data offers no evidence that sfGFP wants something different.

PETase — anchor: PETase Ribose (274.8 µM), spare volume 230 nL

Why the ribose system. PETase loses only 14 % of yield on the cheap formulation — the smallest penalty of the three proteins. Its single disulfide is already handled by DsbC at 5 µM, which both PETase plates include.

Why raise magnesium to 10.1 mM. This is the one place in the entire reference set where a protein-specific magnesium optimum shows up: PETase PEP is the only plate that deviates from 8.22 mM, running at 10.1 mM. Every other plate, across both energy systems and all three proteins, sits at 8.22. If that deviation is real rather than noise, it says PETase prefers slightly more free Mg²⁺ — and the PETase Ribose plate may be under-supplied at 8.22. Given that A1 established magnesium as the sensitive parameter, a 1.88 mM move is exactly the kind of single directed change worth spending the one shot on.

Why spermidine. Same rationale as sfGFP.

Reteplase — anchor: Reteplase PEP (10.9 µM), spare volume 630 nL

Why the expensive energy system. This is the one case where cost optimisation has to lose. Reteplase on the ribose formulation yields 0.81 µM against 10.9 µM on PEP — a 13.5-fold collapse. The cheap formulation cannot save 13-fold on reagents, so cost per gram is worse no matter what. Nine disulfide bonds that must find their correct pairings need the fast, high-flux energy supply; a reaction running on its own metabolism apparently cannot sustain folding at that difficulty.

Why more DsbC. Reteplase’s failure mode is specifically mispairing, not absence of oxidation — GSSG supplies the oxidising potential, but with nine bonds the number of wrong pairings vastly exceeds the one correct arrangement. DsbC is the isomerase whose job is to reshuffle incorrect pairs, so it is the component most directly matched to the bottleneck. 5 µM is the same concentration used for PETase, which has one disulfide; it is not obvious that the protein with nine should get the same amount. A 50 % increase is the largest move the spare volume allows.

Why add catalase. The Reteplase Ribose plate includes catalase at 188 U/mL; the PEP plate does not. This is a cross-over between sibling plates, so it is known to be compatible with Reteplase. Catalase removes hydrogen peroxide, and in an oxidising formulation peroxide accumulation can oxidise residues off-target and compete with the intended GSSG-driven chemistry. At 75 nL it is nearly free in volume terms.

Why not touch the redox buffer. The 4.04 : 1 GSSG : GSH ratio is doing the central work and is the thing most likely to break if moved blind. Changing DsbC and GSSG : GSH in the same shot would also make the result unattributable.

Community reagent proposal: DsbA

The gap. The reagent inventory includes DsbC, the disulfide isomerase that reshuffles incorrect pairings, but no DsbA, the disulfide oxidase that introduces bonds in the first place. In the E. coli periplasm the two work as a pair: DsbA forms disulfides rapidly but with low fidelity, and DsbC proofreads them. The current oxidising formulation has the proofreader without the writer, leaving bond formation entirely to chemical oxidation by GSSG.

Why it matters for this coopetition. Reteplase is the protein that fails worst and the one the whole cost-per-gram metric is most sensitive to — at 10.9 µM it is roughly 30-fold below PETase in molar yield. Its bottleneck is explicitly disulfide formation and pairing. Adding a dedicated oxidase alongside the existing isomerase is the most direct intervention available on that bottleneck, and it is cheap to test because DsbC is already in the inventory as a precedent for supplying a purified Dsb protein.

Alternative if DsbA is declined: the GroEL/GroES chaperonin system, which addresses general misfolding rather than disulfide chemistry specifically and would apply to PETase and Reteplase both.

Notes and caveats

On volume arithmetic. Working from the stock concentrations in the designer, my computed totals for the reference plates came to roughly 26,000 nL rather than 20,000, so some part of the volume accounting is not visible in the interface. I therefore verified every proposed change incrementally against the spare volume shown in the water row of each reference plate, which is the quantity that actually constrains an edit. All three designs fit with margin. Each should still be checked in the designer before submission — if the water row goes negative, the mix is over budget.

On what would make these designs better. With more than one shot per protein, the obvious next move is a magnesium series around 8.22 mM for each protein, since A1 identifies it as the sensitive parameter and the reference set contains exactly one piece of evidence that it is protein-specific.

References

[1] Smith, A. A., et al. (2026). Using a GPT-5-driven autonomous lab to optimize the cost and titer of cell-free protein synthesis (Ginkgo Bioworks × OpenAI). bioRxiv. https://doi.org/10.64898/2026.02.05.703998

[2] Olsen, M. L., Copeland, C. E., Sundberg, C. A., Aw, R., Shaver, Z. M., Rao, G., Swartz, J. R., Karim, A. S., & Jewett, M. C. (2026). Design-driven optimization of low-cost reagent formulations for reproducible and high-yielding cell-free gene expression. Nature Communications. https://doi.org/10.1038/s41467-026-69605-8

[3] Jewett, M. C., & Swartz, J. R. (2004). Mimicking the Escherichia coli cytoplasmic environment activates long-lived and efficient cell-free protein synthesis. Biotechnology and Bioengineering, 86(1), 19–26. https://doi.org/10.1002/bit.20026

[4] Silverman, A. D., Karim, A. S., & Jewett, M. C. (2020). Cell-free gene expression: an expanded repertoire of applications. Nature Reviews Genetics, 21, 151–170. https://doi.org/10.1038/s41576-019-0186-3

[5] Rasor, B. J., Vögeli, B., Landwehr, G. M., Bogart, J. W., Karim, A. S., & Jewett, M. C. (2021). Toward sustainable, cell-free biomanufacturing. Current Opinion in Biotechnology, 69, 136–144. https://doi.org/10.1016/j.copbio.2020.12.012