Section A — Comparing cell-free reaction compositions
1. Why do K-glutamate and Mg-glutamate concentrations differ between compositions, and which is more sensitive?
The reason they're tuned so differently comes down to what each ion is actually doing mechanistically, not just "ionic strength."
Potassium glutamate is there to mimic the bulk ionic environment of the E. coli cytoplasm — it's a high-concentration bystander ion, sitting around 130–200 mM across formulations with PANOx using 200 mM potassium glutamate and PANOx-SP settling closer to 130 mM potassium glutamate. Because glutamate (unlike chloride) is the native counter-ion E. coli actually uses to balance intracellular charge, swapping in glutamate salts instead of something like potassium chloride isn't cosmetic — it's recreating the real cytoplasmic osmotic and ionic backdrop the translation machinery evolved in. K+ mostly buffers general ribosome/enzyme activity across a fairly wide tolerance band.
Magnesium glutamate is the one doing the real precision work. Mg²⁺ isn't a bystander — it's structurally load-bearing: it stabilizes ribosome conformation, coordinates with the phosphate backbone of mRNA and tRNA, and is a required cofactor for essentially every polymerase and many of the enzymes driving the energy-regeneration system. That's why you see it dialed in so tightly and why it shifts formulation to formulation — 16 mM with magnesium acetate versus a different optimum with magnesium glutamate (23.3 mM optimal with glutamate vs. 19.3 mM with acetate) — the counter-ion itself changes the magnesium optimum. This is the classic "Goldilocks" cofactor: too little and ribosomes/polymerase can't function properly; too much and it starts non-specifically bridging phosphates and gumming up folding/translation, or causing precipitation. Olsen et al. make this explicit too — they found new optimum magnesium and phosphate concentrations had to be determined for each new lysate batch, and that protein yields were more sensitive to lysate volume in their optimized formulation than in PANOx-SP, reiterating a fine metabolic balance.
Key takeaway: magnesium glutamate is the more sensitive, protein/mechanism-specific dial. Potassium glutamate is the coarser, bulk-environment dial.
2. PEP+NTP vs. Ribose+NMP — what's actually different?
PEP+NTP (the PANOx-SP approach): You supply the fully phosphorylated building blocks directly — 0.86 mM each of GTP, UTP, and CTP plus 1.2 mM ATP — and you regenerate spent ATP using phosphoenolpyruvate (PEP) as a high-energy phosphate donor, 33 mM phosphoenol pyruvate being typical. PEP is a "substrate-level phosphorylation" energy source: a single enzymatic transfer regenerates ATP from ADP, fast and robust, which is exactly why PANOx-SP reliably produces the highest yields in head-to-head comparisons. But PEP is brutally expensive — Olsen's group flags phosphorylated energy substrates like PEP as costing $2,000–3,000 per liter of CFE reaction, contributing over 75% of all material costs associated with cell-free expression.
Ribose+NMP (the low-cost route): Instead of buying expensive pre-phosphorylated NTPs, you supply cheaper nucleotide monophosphates (NMPs) and let the lysate's own endogenous kinases do the phosphorylation work using glycolytic/metabolic flux (fed in part by substrates like ribose) rather than an exogenous phosphate donor. Olsen's group found that their active minimal reagent formulation was comprised only of potassium glutamate, nucleotide monophosphates, and amino acids — no PEP at all. Mechanistically this is a completely different ATP regeneration philosophy: instead of one dedicated phosphate-donor enzyme doing a direct transfer, you're leaning on the lysate's native central-carbon metabolism (something closer to how the actual cell regenerates ATP) to slowly rebuild the phosphate groups. That's reflected in the metabolite data too — RF_opt sustains ATP at roughly 250 μM for the whole reaction instead of the sharp PANOx-SP spike-then-crash, and keeps phosphate three-fold lower throughout because phosphate isn't being dumped in from PEP hydrolysis.
The tradeoff is stark and is basically the whole plot of the Olsen paper: PANOx-SP produced the highest sfGFP yield of all eight formulations tested, but the Ribose/NMP-style non-phosphorylated systems cost 22–92% less per gram of protein, and after a full optimization campaign the NMP-based RF_opt actually beat PANOx-SP by 87% in yield while costing 99% less — $55/g_protein versus roughly orders of magnitude more for PANOx-SP.
Key takeaway: PEP+NTP is "pay for the answer directly." Ribose+NMP is "give the lysate the raw materials and let its own metabolic machinery do the expensive synthesis step for free." For a coopetition scored on cost-per-gram, that's basically the entire ballgame — expensive pre-made parts vs. cheap raw materials + the lysate's own enzymatic labor.
3. Comparing two literature formulations — PANOx-SP vs. Olsen's RF_opt
| Comparison | PANOx-SP (Jewett & Swartz 2004) | RF_opt (Olsen et al. 2026) |
|---|---|---|
| Energy source | PEP (33 mM), phosphorylated | NMPs + ribose, non-phosphorylated |
| NTPs/NMPs | ATP 1.2 mM, GTP/UTP/CTP 0.86 mM each | NMPs (capped ~5 mM to avoid precipitation) |
| K-glutamate | 130–170 mM | Present, re-optimized per lysate batch |
| Mg-glutamate | 16 mM (optimum ~23 mM w/ glutamate) | Re-tuned per lysate batch |
| Extra cofactors | NAD, CoA, folinic acid, spermidine, putrescine, oxalate | Stripped out — NAD/CoA/tRNA deemed unnecessary |
| Component count | ~15+ | As low as 6 in the minimal version, 12 in RF_opt |
| sfGFP yield | Highest of 8 tested formulations | 87% higher than PANOx-SP after optimization |
| Cost | $2,000–3,000+/L reagents, ~$1,000s/g protein | $133/L CFE, $55/g protein (99% cheaper) |
What jumps out comparing these two side by side is that they're not just "different recipes" — they're different design philosophies. PANOx-SP is a maximalist, kitchen-sink formulation: throw in every cofactor known to support transcription/translation (NAD, CoA, spermidine, putrescine, folinic acid, oxalate) and pay for a fast, reliable energy donor (PEP) to guarantee yield. It's the "if it might help, include it" school of formulation design, from an era where cost-per-gram wasn't the primary optimization target.
RF_opt is the opposite instinct: Olsen's team explicitly notes many of the more expensive reagents — tRNA, coenzyme A, NAD+ — have previously been shown to be unnecessary for cell-free gene expression when cell lysates are not dialyzed, so they didn't include them. Rather than assume more reagents = more insurance, they ran 1,231 different reaction combinations screening 58 components to find the minimum viable formulation, then optimized within that minimal set. This is a classic "subtract first, then tune" engineering approach versus "add everything, then see what you can afford to cut" — and for a coopetition scored explicitly on cost-per-gram, the subtract-first philosophy is clearly where the leverage is.
Key takeaway: The practical lesson for our own 80 compositions: don't start from PANOx-SP and shave dollars off. Start minimal (glutamate salts + NMPs + amino acids) and add back only what the data says earns its cost — which is exactly the design-of-experiments logic Olsen's team used, and the same logic Ginkgo's staged 384→96→2mL screen is built to support at scale.
Section B — Reaction compositions for sfGFP, PETase, and Reteplase
Design philosophy for all three: RF_opt (Olsen et al. 2026) as the cost backbone — glutamate salts + NMPs + amino acids, since that formulation beat PANOx-SP on yield and cost by 84–99%. On top of that minimal base, I'm layering back in a couple of PANOx-SP's classic cofactors (folinic acid + spermidine) as a yield safety margin, since we only get a few shots at the 384-well stage and a totally bare-bones formulation is riskier before we've seen our own plate data. All volumes are in multiples of 25 nL, variable component total = 11,000 nL, water tops up to 20,000 nL.
Fixed in all three (not shown per-reagent below): 5,000 nL lysate, 2,000 nL DNA template, 2,000 nL base buffer.
1. sfGFP — reducing-condition baseline
| Reagent | Category | Volume (nL) | Rationale |
|---|---|---|---|
| Potassium glutamate (10× stock) | Salt | 2,000 | Bulk ionic/osmotic environment |
| Magnesium glutamate (10× stock) | Salt | 1,000 | Tightly tuned cofactor; mid-range starting point for titration |
| NMP mix (AMP/GMP/CMP/UMP) | Building block | 2,500 | RF_opt backbone — cheapest route to NTPs via endogenous kinases |
| Amino acid mix (20 aa) | Building block | 2,500 | Required monomer pool |
| HEPES-KOH, pH 7.2 | Buffer | 1,000 | pH stability; reducing-condition default (no shift needed — sfGFP has no disulfides) |
| Spermidine | Cofactor | 500 | Cheap, PANOx-SP-derived yield booster |
| Folinic acid | Cofactor | 500 | Cheap, supports transcription efficiency |
| DTT | Reducing agent | 500 | Keeps the 2 free cysteines reduced; standard reducing-condition default |
| Nuclease-free water | — | 500 | Rounds out to 11,000 nL variable |
No CTP/GTP/UTP, no NAD, no CoA — all flagged as unnecessary once lysate isn't dialyzed (Olsen et al.), and they're among the most expensive line items in the whole reagent menu.
2. PETase (TfCut2) — reducing, with extra folding support
The logic: still a reducing-condition protein (DTT-compatible), but TfCut2's one disulfide bond and serine-hydrolase catalytic triad mean it's more fold-sensitive than sfGFP. I'm not pushing it into oxidizing/DSB territory — that would actively hurt a protein that still wants a broadly reducing environment — but I am giving it a bit more chaperone/folding support and slightly richer building-block supply, since we're measuring enzymatic activity, not just the presence of protein.
| Reagent | Category | Volume (nL) | Rationale |
|---|---|---|---|
| Potassium glutamate (10x stock) | Salt | 2,000 | Same bulk ionic role as sfGFP |
| Magnesium glutamate (10x stock) | Salt | 1,250 | Slightly higher — more demanding fold, want polymerase/ribosome margin |
| NMP mix | Building block | 2,500 | RF_opt backbone |
| Amino acid mix (20 aa) | Building block | 2,750 | Marginally richer pool — larger, more complex fold than sfGFP |
| HEPES-KOH, pH 7.2 | Buffer | 1,000 | Standard reducing buffer |
| Spermidine | Cofactor | 500 | Yield booster |
| Folinic acid | Cofactor | 500 | Transcription support |
| DTT | Reducing agent | 250 | Lower than sfGFP's — want free thiols available but not so reducing it discourages the single native disulfide from eventually forming post-translationally/during workup |
| Brij-35 | Detergent | 250 | Mild solubility aid — one more fold-demanding protein benefits from reduced aggregation |
| Nuclease-free water | — | 0 | Already at 11,000 |
3. Reteplase — oxidizing/DSB formulation, with added folding support
The logic: nine disulfide bonds is a completely different regime — standard DTT kills this protein's folding. This needs the full three-part DSB modification (modified lysate treated with IAM, redox buffer of GSSG/GSH, DsbC isomerase), plus a higher DsbC load and the pH-adjusted HEPES that Olsen's group found necessary to recover yield once glutathione addition drops reaction pH below 7.0.
Lysate pre-treatment (not part of the 11,000 nL variable budget, done upstream): lysate from BL21 Star (DE3) Δgor strain, treated with IAM (worked out at the lower 25 µM "lo" dose Olsen's group showed recovers yield with 20-fold less IAM once combined with the Δgor background, rather than the harsher 500 µM dose needed without it), quenched with equimolar cysteine.
| Reagent | Category | Volume (nL) | Rationale |
|---|---|---|---|
| Potassium glutamate (10x stock) | Salt | 1,750 | Bulk ionic role, slightly reduced to make room for redox components |
| Magnesium glutamate (10x stock) | Salt | 1,000 | Standard cofactor role |
| NMP mix | Building block | 2,250 | RF_opt backbone, still the cost-efficient route |
| Amino acid mix (20 aa) | Building block | 2,500 | Required monomer pool |
| HEPES-KOH, pH 7.5 (elevated) | Buffer | 1,250 | Olsen's key fix — GSSG/GSH addition drops pH below 7.0 in RF_opt-style formulations; pre-raising buffer pH to 7.5 avoided the pH crash and recovered yield |
| GSSG (oxidized glutathione) | Redox/disulfide | 500 | Establishes oxidizing environment — ~4 mM final, per Olsen's protocol |
| GSH (reduced glutathione) | Redox/disulfide | 250 | Controlled ratio with GSSG — ~1 mM final, tunes redox potential rather than going fully oxidizing (over-oxidizing increases misfolded/scrambled disulfides) |
| DsbC (purified) | Folding helper | 750 | Elevated vs. Olsen's baseline 5 µM — nine disulfide bonds is a much harder isomerization problem than the Fc/antibody constructs DsbC was validated on, so more isomerase capacity to correct mispaired bonds seems justified |
| Spermidine | Cofactor | 250 | Cheap yield support, retained from reducing-condition logic |
| Folinic acid | Cofactor | 250 | Transcription support |
| Nuclease-free water | — | 250 | Rounds to 11,000 |
Explicitly absent: DTT (would reduce the 9 target disulfides as fast as they form).
One flag for recitation: Olsen et al. found that adding GSSG/GSH to their RF_opt formulation reduced sfGFP production by 75–97% before the Δgor fix and pH correction — so the oxidizing modifications aren't "free" even when done right, they're actively hostile to a reducing-condition protein. That's exactly why I kept the sfGFP and PETase compositions untouched by any DSB components rather than trying to build one "universal" formulation across all three proteins — the paper's own data argues against that.
Community reagent contribution
Proposed reagent: Ribose (as a standalone, separately-addable component)
Rationale: Olsen et al. (2026) found that GMP — one of the more expensive nucleotide monophosphates — could be fully replaced by a combination of guanine + ribose without sacrificing protein yield, since the lysate's own salvage pathway enzymes can reconstruct the nucleotide from cheaper raw precursors. If ribose isn't already a standalone selectable reagent (separate from being bundled into a fixed NMP mix), making it independently titratable would let teams explore nucleotide-sparing substitutions as a cost lever across all three target proteins, not just sfGFP. This is exactly the kind of "subtract the expensive ingredient, let lysate metabolism do the work" lever the coopetition is scored on.
Source: Olsen et al., "Design-driven optimization of low-cost reagent formulations for reproducible and high-yielding cell-free gene expression" (2026), bioRxiv: biorxiv.org/content/10.1101/2025.08.01.668204 — see Discussion section, guanine+ribose GMP-replacement finding.
Forum post
Been working through Olsen et al. 2026's RF_opt data for the HTGAA submissions — one thing that stood out: they found NMP-based (vs PEP+NTP) formulations sustain ATP at a steady ~250 µM for the whole 20h reaction instead of PANOx-SP's early spike-and-crash, and keep phosphate 3x lower throughout. Makes me wonder if we should be timing our plate-reader measurements differently for NMP-backbone compositions vs PEP-backbone ones — a spike-crash kinetic profile vs. a sustained-low one might call for different "optimal read window" assumptions when we're comparing cost-per-gram across compositions that use fundamentally different energy systems. Curious if anyone's seen timing-sensitivity data on this.
Section C — T7-compatible plasmid design
Every CFPS plasmid in this system needs the same functional backbone regardless of payload, because the lysate's transcription/translation machinery is fixed (T7 RNAP supplied by the BL21 Star (DE3) source strain) — only the insert and its immediate regulatory elements change per protein. So I'm using one standardized expression cassette and swapping only the coding sequence and a couple of protein-specific tweaks:
T7 promoter → 5′ UTR/RBS → start codon → ORF → stop codon → T7 terminator, on a standard high-copy backbone (e.g., pET-style, ColE1 origin, antibiotic resistance marker for plasmid prep — not relevant inside the CFPS tube itself, but needed to produce DNA template at scale beforehand).
- T7 promoter: the canonical strong consensus sequence (TAATACGACTCACTATAG), since the lysate's T7 RNAP is a single robust polymerase with no need for promoter-strength tuning — consistent, high transcription is the goal for all three targets.
- 5′ UTR / RBS: a strong, well-characterized ribosome binding site (T7 gene 10 leader or a consensus Shine-Dalgarno, e.g., AAGGAGG spaced ~7-9 nt upstream of the start codon) — kept identical across all three to isolate protein-folding differences as the actual experimental variable, rather than confounding results with translation-initiation differences.
- T7 terminator: standard T7 terminator downstream of the stop codon, to prevent read-through transcription from destabilizing the mRNA pool or wasting NTPs/NMPs on non-coding RNA — directly relevant to our cost-per-gram objective, since every ribonucleotide spent on junk transcript is ribonucleotide not spent on the protein we're scoring.
Per-protein design decisions
1. sfGFP — the sequence is clean: no N-terminal tag beyond the coding sequence itself, with a C-terminal Strep-tag II (WSHPQFEK) already built in for affinity purification/detection if needed downstream. No cleavage site is included, because sfGFP's function (fluorescence) doesn't depend on tag removal — the tag sits far from the chromophore-forming Thr65-Tyr66-Gly67 motif and won't interfere with autocatalytic chromophore maturation or β-barrel folding. Plasmid is simplest of the three: promoter → RBS → ORF (with built-in C-term Strep-tag) → stop → terminator.
2. PETase (TfCut2) — the sequence shows a more deliberate architecture: an N-terminal Strep-tag II (MEKK-WSHPQFEK) followed by a GGSG linker, then a TEV protease cleavage site (ENLYFQ↓S), then the mature TfCut2 catalytic domain, ending in a second, C-terminal Strep-tag II. This dual-tag, TEV-cleavable design makes sense functionally: the N-terminal tag + TEV site allows affinity capture and then tag removal if a native-like N-terminus is needed for activity assays (the catalytic triad Ser170-Asp216-His248 is sensitive to steric interference near the active site), while the C-terminal tag stays permanently fused for a second purification/detection handle that doesn't need removing since it's distal to the catalytic cleft. Plasmid: promoter → RBS → ORF exactly as given (N-tag/TEV/catalytic domain/C-tag) → stop → terminator — no redesign needed, the construct is already built for correct folding and functional assay compatibility.
3. Reteplase — same dual-tag, TEV-cleavable N-terminal architecture as PETase (Strep-tag → linker → TEV site), which matters even more here: Reteplase's kringle-2 + serine protease domains and nine disulfide bonds mean the N-terminus is close to functionally important structural elements, so having a clean TEV-cleavable removal option preserves the option to assay activity without a bulky N-terminal fusion interfering with correct disulfide-pairing topology or substrate access to the catalytic cleft. The C-terminal Strep-tag again serves as the permanent purification/detection handle. Because this construct is destined for the oxidizing/DSB reaction condition (not the standard reducing CFPS mix), there's a design note worth flagging in the writeup even though it doesn't change the DNA sequence: unlike sfGFP and PETase, this plasmid's transcript is translated in a lysate that has been IAM-treated and Δgor-background, with GSSG/GSH and DsbC added — all downstream-of-DNA conditions, but worth stating explicitly in the writeup since it's the whole reason this construct needs different reaction-level handling even though the plasmid architecture itself (promoter/RBS/ORF/terminator) is structurally identical to the other two.
One general design note across all three: none of these constructs need an internal T7 RNAP gene — that's supplied constitutively by the lysate's BL21 Star (DE3) genomic background, which is precisely why the same minimal plasmid architecture (just promoter + ORF + terminator) works for all three proteins despite their very different folding demands. The complexity in this system lives in the reaction composition (Section B), not the plasmid design — the DNA template is deliberately kept simple and interchangeable, which is the whole point of CFPS's claimed programmability.