Week 3 HW: CFPS

A. Comparing compositions

1. Potassium vs magnesium glutamate

Magnesium glutamate is more sensitive.

Potassium glutamate just sets the ionic background, approximating the cytoplasm the ribosome evolved in. The optimum is broad, so published values scatter from 130 mM in PANOx-SP to 362 mM in Olsen’s RFopt. Magnesium is a catalytic cofactor with a narrow window: free Mg2+ drives 30S/50S association, holds tRNA tertiary structure together, and is the catalytic metal for T7 polymerase and the synthetases. Too little and ribosomes don’t assemble, too much and elongation slows and fidelity drops. Across those same formulations it barely moves, 7 to 10 mM everywhere.

The giveaway is how Olsen treats them: magnesium and phosphate re-optimized for every lysate batch, every lysate loading and every source strain, while potassium sat at 362 mM throughout. Where the magnesium number does drift it’s chelation, not a different requirement. Phosphorylated species bind Mg roughly stoichiometrically, so the optimum tracks the chelating load, and in a PEP system the free phosphate climbs as PEP is consumed and precipitates magnesium phosphate. Change the energy system and the magnesium optimum moves with it.

2. PEP + NTP vs Ribose + NMP

Both formulations have to end up with the same thing, a supply of ATP, GTP, CTP and UTP. They differ in whether those arrive ready-made or get built in the reaction from cheaper precursors.

PEP + NTP is the PANOx-SP lineage. Pyruvate kinase hands PEP’s phosphate straight to ADP, so ATP appears within half an hour and the nucleotides arrive ready to use, along with the classical supplements: folinic acid, tRNA, NAD, CoA, putrescine, oxalate. It’s fast and it burns out. Every PEP consumed releases a phosphate, phosphatases waste more of it, and the accumulated phosphate drags magnesium out of solution. It’s also the expensive route, with CTP, GTP and UTP at $28 to $39/mL.

Ribose + NMP has no NTPs and no phosphorylated energy substrate. Ribose feeds the lysate’s own pentose phosphate and glycolytic enzymes and endogenous kinases phosphorylate the NMPs up to triphosphates, so phosphate has to be added deliberately at 5 to 15 mM. ATP comes up slower but holds near 250 µM for the full 20 hours instead of spiking and crashing, and the expensive supplements can just be deleted because undialyzed crude lysate already has them. Nicotinamide replaces NAD for pennies. RFopt gets reagent cost to $133/L.

3. PANOx-SP vs RFopt

With the best GPT-5 composition as a third column, since all three are the same lineage twenty-two years apart.

Deletion did most of the work. RFopt drops tRNA, folinic acid, NAD, CoA, putrescine, oxalate, DTT and ammonium glutamate outright, because undialyzed crude lysate already supplies them. That’s three of the four priciest items in the classical mix, and it cost nothing: RFopt still made 87% more sfGFP. Magnesium meanwhile didn’t move at all while potassium nearly tripled, which is question 1 again. Buffering turned out to be a yield reagent rather than background, and GPT-5 put a number on it: median 0.057 g/L with no added HEPES against 0.177 g/L with it.

GPT-5’s gain came from different levers than Olsen’s. Once reagent cost was down, lysate and DNA were over 90% of the total, so it went after the fixed side: lysate 30% to 25%, DNA 8.7 to 5 nM, NMPs cut two to fourfold. Then it bought titer back with spermidine, median 0.253 against 0.131 g/L. The lesson for part B is that lysate and DNA are fixed by the rules, so specific cost is driven almost entirely by titer.

B. The three designs

The 5,000 nL of lysate costs $0.0200 a reaction and the template is fixed, so neither is mine to optimize. In the 11,000 nL I do control, only six things carry real cost: DsbC at $5,632/mL, CoA at $104/mL, CTP, GTP and UTP at $28 to $39/mL, NAD at $33/mL and folinic acid at $9/mL. Everything else is fractions of a cent. So each design is a decision about which of those six I can do without. Final concentrations below include the base buffer background, which I back-calculated from the designer’s example JSON: 200 mM K glutamate, 2.6 mM Mg glutamate, 30 mM HEPES, and 1 mM each of the amino acid mix, tyrosine and cysteine.

sfGFP. Ribose and NMP backbone with none of the expensive six, so the cost is basically all lysate and titer is the whole game. HEPES at 75 mM because it’s the best-evidenced titer lever on the menu and costs nothing; phosphate at 15 mM because the salvage kinases need it and the window is narrow; NMPs cut with adenosine and guanine filling in, following both papers, though CMP and UMP can’t be swapped for their nucleosides; spermidine because it roughly doubled median titer in Smith’s data. Catalase because sfGFP is the one target where oxygen is mechanistically required, since chromophore maturation ends in an O2-dependent step. That last one is speculative and deserves a minus-catalase arm.

PETase. Same backbone, cysteine up to 2 mM, no DsbC. Leaving DsbC out is the whole design. At $5,632/mL the default 5 µM is $5.63 a reaction against $0.02 of lysate, so a reference PETase reaction spends 280 times more on DsbC than on cell extract and over 99% of its reagent budget on one protein. And an isomerase doesn’t have much to do here: DsbC reshuffles wrong pairings, but TfCut2 has one disulfide, so there’s only one possible pairing and nothing to reshuffle. The caveat is that DsbC also has holdase activity, so it isn’t guaranteed inert. A 0 / 0.5 / 1.25 / 2.5 µM ladder would settle it and is the highest-value block of wells on a PETase plate. Everything else I held identical to sfGFP on purpose, so any difference is down to the protein rather than a dozen simultaneous changes.

Reteplase. Keep PEP and the NTPs, pay for them by deleting the cofactor package, since reteplase doesn’t survive the cheap backbone at any price. DsbC genuinely is needed, but the default overshoots the literature: Yin and Swartz, where this construct comes from, got 200 µg/mL total vtPA and almost no activity in a standard system. Glutathione plus iodoacetamide raised soluble protein to about 50 µg/mL and raised active protein not at all. Only DsbC produced activity, about 26 µg/mL, and bovine PDI did nothing, so it’s DsbC specifically. Their working concentration was 75 µg/mL, about 2.9 µM, while the platform default of 5 µM is roughly 128 µg/mL. I set 2.5 µM, inside the range they used, which halves the dominant cost. HEPES goes to 85 mM because 5 mM of glutathione is what drops reaction pH and buffer is the cheapest fix. No added phosphate, since in a PEP system phosphate is a product and above 25 mM gives essentially zero titer, which is also why PEP is at 25 rather than 30 mM. GSSG to GSH at 4:1 is Yin and Swartz’s measured optimum across five ratios. Since DsbC is about 99% of the cost, the rest of a reteplase plate should be a DsbC ladder: at 5 µM the reference works out near $0.52 per µM AMC, so anything holding above a third of reference activity at 1.25 µM wins.