Week 3 Cell Free

HTBAA Week 3 — Designing a Cell-Free Reaction Composition

Section 3 (A–C) report — for submission on the homework website (version 2, 13 October 2026)

Targets: sfGFP · PETase (TfCut2) · Reteplase Objective: lowest specific cost — $ per gram (sfGFP) or $ per unit of activity (PETase, Reteplase)

Reading notes. All costs use the coopetition price list. Items marked [VERIFY] are not yet confirmed and are listed in Section 11. The three designs are hypotheses with stated tests, not predicted winners. Companion files: the workbook HTBAA_HW3_CFPS_Costing.xlsx (every figure below is a live formula there) and the CFPS Design Toolkit web page (Section 8), which recomputes these designs and checks them against this report.

0. Summary

Three findings drive every decision:

1. For sfGFP, titer — not reagent tweaks — sets the cost. Lysate, DNA and base buffer cost about $1,215 per litre before any reagent is added.
2. For the two enzymes, purified DsbC dominates. At 5 µM it costs $281,600 per litre, about 99% of the reference plates.
3. The best energy system depends on the protein and on the DsbC level. Ribose+NMP is far cheaper for sfGFP; PEP+NTP wins for Reteplase at every DsbC level; for PETase ribose wins below about 0.24 µM DsbC.

Figure 1. How a known recipe is chosen for each protein, and the checks every design must pass. SVG version

0.1 Requirements checklist

1. Background and key terms

1.1 How the reaction works

Cell-free protein synthesis (CFPS) makes protein in a tube from a broken-open cell extract, with no living cells. The E. coli lysate (the clarified liquid left after the cells are broken and spun) supplies ribosomes, tRNAs and metabolic enzymes. It also supplies T7 RNA polymerase (a fast single-protein viral enzyme that copies DNA to mRNA, but only from a T7 promoter), because the BL21 Star (DE3) strain carries its gene. The reaction mix must supply building blocks (amino acids, nucleotides), an energy system that keeps ATP topped up, salts that keep ribosomes and RNA stable, and, for disulfide-bonded proteins, a redox system.

Figure 2. How a cell-free reaction makes protein, and which reagent groups feed each step. SVG version

1.2 Rules of the exercise

Reading the designer’s numbers. Each concentration the designer shows is a total in the 20 µL reaction, including what the base buffer supplies. I tested this: with that reading, my volume arithmetic reproduces the designer’s water row for all 12 reference plates, within 0.005 µL (rounding).

1.3 Terms (what each is, what it does, why it matters here)

2. Cost model

Volume added (nL)   V = (C_total − C_base) / C_stock × 20,000          rounded to 25 nL
Reactions per litre N = 1,000,000,000 nL / 20,000 nL = 50,000
Cost per litre ($/L) = V(nL) / 1,000,000 × price($/mL) × N
Total                = reagents + lysate + DNA + base buffer
Specific cost        = total ÷ titer   (or ÷ activity)
Break-even multiple  = (B + D) / B     B = cost without an additive, D = its cost

Fixed cost per litre: sfGFP $1,215.30 · PETase $1,229.80 · Reteplase $1,250.40.

Validation. My model’s specific cost versus Smith et al.’s published values: #77 $414 vs $422; #26 $419 vs $425; #35 $426 vs $430; #74 $441 vs $446; 1793119_36 $461 vs $466. The model reads 1–2% low consistently, which fits small unitemised costs. Oxalic acid has no list price and is modelled at $0, so PEP-plate totals are slightly understated [VERIFY].

Reference plates in the designer, costed with this model:

Implication. Because the fixed cost is about $1,215–1,250/L, a 10% titer gain cuts $/g by about 9%, while removing $75/L of reagents saves under 6%. A reagent earns its place if its share of the total cost is smaller than the titer gain it produces.

3. Part A — Comparing cell-free reaction compositions

A1. Why do potassium glutamate and magnesium glutamate differ between compositions? Which is more sensitive?

Evidence. Jewett & Swartz (2004), Table II, PANOx-SP, fraction of maximum chloramphenicol acetyltransferase (CAT) yield, one solute varied at a time:

The Cytomim system shows the same pattern: Mg optimum 8 mM (0.62 at 4 mM, 0.41 at 16 mM), K optimum 130 mM (0.95 even at 200 mM).

Answer.

• Magnesium is far more sensitive. Its curve is narrow and falls steeply on both sides (0.39 at 28 mM), whereas potassium sits on a broad plateau (0.92 at 225 mM).
• Why values differ between compositions (hypotheses consistent with the data, not measured here):
  1. Mg²⁺ is bound by NTPs, phosphate and RNA, so the free-Mg optimum moves with the nucleotide and phosphate load. NTP-based PANOx-SP needs 20 mM; NMP-based systems run near 8 mM.
  2. The extract itself contributes K⁺, Mg²⁺ and acetate, so “added” amounts are not totals.
  3. Each lysate batch differs; Olsen re-optimised Mg and phosphate for every new lysate.
• Consequence for my design. The base buffer already supplies 200 mM K-glutamate, so potassium cannot go below 200 mM, which is already above Jewett’s PANOx-SP and Cytomim optima. The only available move is upward, where the curve is flat to falling (0.94 → 0.92), so I leave it. I also leave Mg unchanged and treat a ±0.5–1 mM fine scan as a second-round experiment, because changing the narrowest-optimum variable without a scan is a gamble.
• Open discrepancy. Smith’s Table 2 lists 273–313 mM K-glutamate for compositions that the designer shows at 200 mM. I follow the designer and ask the course team why [VERIFY].

A2. How is a PEP+NTP formulation different from a Ribose+NMP formulation?

Olsen and Smith both conclude that NTPs are not cost-effective for sfGFP; the Reteplase result is the exception because activity, not reagent price, is the limiting factor.

A3. Two literature compositions: reagents and concentrations compared

Contrast. PANOx-SP is a PEP-driven system with a heavy salt and nucleotide load and 20 mM Mg. Smith #35 swaps PEP and NTPs for sugars and NMPs, drops Mg to 8.8 mM, and adds a pH buffer and nicotinamide. Both keep a polyamine and glutamate salts. Yields are not directly comparable (different protein, scale and time); the comparison is of formulation logic and cost structure.

4. Part B — Design one reaction composition for each protein

The brief asks for novel designs; a first pass starts from a known formulation and changes a small number of variables, each with a stated mechanism and a test that could prove it wrong. All designs were checked for non-negative water and 25 nL steps. Enter values in the designer’s units (total final concentration).

B1a. sfGFP — start from Smith #35, change two things

Why #35 and not the cheapest, #77. From Smith Table 1 (tube cost; titers in g/L by format):

#77 is cheapest in tubes but weakest in 384-well plates, which is the stage that decides what advances. Choosing it would risk early elimination for an $8/g advantage. (384-well error bars are large, e.g. 0.72 ± 0.38, so this is a risk judgement, not a certainty.)

Changes

Composition (water 3,875 nL; reagents 7,125 nL)

Figure 3. Where each litre of the sfGFP design (D1) costs money: the fixed block is 94%. SVG version

At #35’s tube titer (3.04 g/L) this would be about $425/g; that is the baseline’s titer, not a prediction.

Test that could prove it wrong. D1 fails if its titer is below #35’s on the same plate. Because the design is cost-neutral, any real gain must come from titer.

Why not the alternatives

B1b. PETase — start from the PETase Ribose reference, change one thing

Change. DsbC 5 → 0.125 µM (25 nL). Everything else is identical to the reference.

Why. The brief states TfCut2 has one disulfide bond and that the standard reducing formulation remains compatible with it. At the reference 5 µM, DsbC is 99% of the cost. I confirmed from the supplied sequence that TfCut2 has two cysteines (one bond).

Composition (water 3,450 nL; reagents 7,550 nL)

Test that could prove it wrong, in cost terms

• To beat the reference’s specific cost, D2 needs only 3.0% of its activity (about 8.1 µM† of the reference’s 274.8).
• To justify any DsbC versus none (the same recipe without DsbC costs $1,334.34/L), D2 must be 6.3× as active as that no-DsbC version.

Why one variable only. A single change keeps any difference attributable to DsbC; bundling other changes would confound it. A group member is testing the PEP PETase reference at 3.75 µM, so my design does not duplicate that.

Why not the alternatives

B1c. Reteplase — start from the Reteplase PEP reference, change one thing

Change. DsbC 5 → 1 µM (200 nL). Everything else is identical to the reference.

Why PEP+NTP. The two Reteplase references differ 13-fold in activity (10.9 vs 0.81 µM†) while their costs without DsbC differ only 2.6-fold ($3,557 vs $1,349 per litre). Even with DsbC at zero, PEP is about 5× cheaper per unit activity. Caveats: single plates, and the µM† unit is unconfirmed.

Why 1 µM DsbC. DsbC is 94% of this design’s cost. At 1 µM it costs one-fifth as much as at 5 µM. The literature level quoted on the course forum from Yin & Swartz (2004) is 75 µg/mL, which at DsbC’s 25.5 kDa is about 2.9 µM. So lowering from 5 µM has support, but 1 µM is below that level. A peer states DsbC saturates near 1 µM, but the passage quoted does not say so, so I treat it as a hypothesis. D3 is a deliberate cost test; 2.5 µM is the safer alternative but needs about 51% of the reference activity to break even versus 21% at 1 µM.

Composition (water 1,400 nL; reagents 9,600 nL)

Figure 4. Where each litre of the Reteplase design (D3) costs money: DsbC is 94%. SVG version

Test that could prove it wrong. D3 needs about 21% of the reference’s activity (2.3 µM† of 10.9) to match the reference’s specific cost. Below that, 1 µM does not pay back.

Why keep everything else (CoA, NAD, NTPs, oxalate, polyamines). It is the known formulation. CoA + NAD + NTPs total about $1,600/L, under 3% of D3, so removing them cannot decide the outcome now; they are second-round tests.

Why not the alternatives

B1d. DsbC break-even (from the workbook)

Figure 5. Total cost per litre against DsbC level for both enzyme designs (log scale); D2 and D3 are marked. SVG version

Backbone crossover (assuming activity does not change with DsbC level, a simplification): setting (C_ribose + d)/T_ribose = (C_PEP + d)/T_PEP gives 0.24 µM DsbC for PETase (ribose cheaper below it, PEP above it) and a negative value for Reteplase, so PEP wins at every DsbC level.

B2. Community reagent: Bis-Tris

Decision. I upvote the existing Bis-Tris proposal [confirm it is on the Community Reagents list and add your vote]. If it is not listed, I propose it with the rationale below.

What it is. Bis-Tris is a buffer with a pKa of about 6.8 (near-neutral pH), which resists pH change in the 6–7.2 range. It is a stable small molecule with no activity to lose, no cold chain, and no protein-sourcing delay.

How I chose it. I screened candidates against three criteria: cost-effective, stable, and useful for more than one protein.

Evidence and links

• Jewett & Swartz (2004): PEP-based reactions fall below pH 6.5 by the end of the reaction, and they suggest a buffer with a pKa near 6.8, such as Bis-Tris, as an approach. https://doi.org/10.1002/bit.20026
• Olsen et al. (2026): glutathione acidified the ribose formulation below pH 7.0, and correcting the pH recovered yield. https://doi.org/10.1038/s41467-026-69605-8
• Smith et al. (2026): buffering or pH control can materially improve specific cost. https://www.biorxiv.org/content/10.64898/2026.02.05.703998v1

Caveat. Olsen found that adding a buffer on its own to the minimal formulation did not improve yield, so Bis-Tris is a plausible low-risk test, not a proven gain.

One reagent or several? I kept it to one. The brief asks for at least one with a rationale, I must be able to defend every choice at recitation, and reagents come from a limited shared pool where duplicating an existing proposal wastes a slot.

B3. Forum

• I confirm access to the cfps-coopetition channel [tick when done].
• Idea to post: run a DsbC dose-response (0, 0.125, 0.25, 0.5, 1, 2.5 µM) across plates for Reteplase, because DsbC is about 94% of the cost of my Reteplase design and a TA reported ~95% in theirs.

B4. Submission notes (from the forum)

5. Part C (optional) — Asimov Kernel plasmids and helper plasmid

C1. T7-compatible plasmids

Check of the supplied sequences. I counted residues and cysteines:

The lengths match the overview document (240, 309 and 399). Tags in the sequences: a Strep II tag (WSHPQFEK) at the C-terminus of sfGFP; an N-terminal Strep II tag + linker + TEV site on PETase and Reteplase; and a C-terminal HiBiT tag (VSGWRLFKKIS) on both.

Layout to reproduce (from Ginkgo’s own plasmids, 2,511 / 2,696 / 2,966 bp): backbone (origin and selection marker, for propagation only) → T7 promoter (TAATACGACTCACTATAGGG) → ribosome binding site and spacer → ATG, coding sequence, stop codon → T7-type terminator → backbone.

Kernel steps: (1) codon-optimise each protein for E. coli; (2) remove stray restriction sites for the chosen cloning method; (3) avoid internal terminator-like hairpins and strong internal RBS motifs; (4) confirm the reading frame across the tags and a single stop codon; (5) check GC content and repeats for synthesis; (6) translate the final sequence back and match it to the brief. I have not run the Kernel; this step must be done in the tool.

C2. Bonus — T7 helper plasmid for functional Reteplase (speculative)

Design. A second T7 plasmid expressing the mature E. coli DsbC (without its periplasmic signal peptide, so it is made in the cytoplasm-like reaction), with the same RBS and terminator as the target plasmid, and a tag (for example HiBiT or Strep II) to follow it.

Why DsbC and not the alternatives. DsbC is the enzyme that fixes wrong pairings, which is Reteplase’s bottleneck. DsbA only forms disulfides, and GSSG already does that; Yin & Swartz (as quoted on the forum) found PDI, GroEL/GroES, DnaK/DnaJ/GrpE and DsbA gave no significant benefit.

Cost logic (a calculation, not a result).

• Exogenous 1 µM DsbC costs $56,320/L.
• 1 µM × 25.5 kDa = 25.5 µg/mL = 0.0255 g/L, about 0.8% of the 3.04 g/L sfGFP yield, so it is plausible in theory for the reaction to make that much.
• A helper plasmid at 1 nM out of the 5 nM total DNA would cost about 1/5 of the DNA line, roughly $46/L (DNA at $228/L for 5 nM).

Why it might not work.

• The model allows exactly one DNA template (2,000 nL, 5 nM), so a helper would have to be co-formulated into that stock; this reduces the target gene dose and may not be permitted [VERIFY with the course team].
• It competes with Reteplase for ribosomes and ATP and adds a lag, so DsbC may arrive after misfolding has begun. A peer argues that overexpressing DsbC in the production strain before lysis is the better route (out of scope here).
• DsbC must be in its reduced, active form to act as an isomerase; whether it stays active in a GSSG-rich mix is untested.

6. Cybersecurity, legislation and quality

Manufacturing and quality principles applied: start from a known baseline; change one or two variables; write the hypothesis and the failure criterion in advance; replicates (n = 4) and controls on every plate; compare only within the same plate format; keep traceable records (design ID, file hash, date); cost every option before choosing it.

7. Constraints, risks and boundary-pushing

This section lists what could limit or break the designs, and where they push beyond what the sources show. The cost model cannot see any of it.

7.1 Timing and scoring unknowns

7.2 Wet-lab risks the cost model cannot see

7.3 Where the designs push beyond the evidence, and where they stay conservative

7.4 Ideas considered but not evaluated

• Redox tuning instead of purified DsbC (adjusting GSH and GSSG). D3 already contains both, and I have not costed or sourced a version that replaces DsbC, so I make no claim about it.
• Helper plasmid that makes DsbC in the reaction (Section 5): possible in theory, but only one DNA template is allowed and permission is unclear.
• Overexpressing DsbC in the lysate strain before lysis (a peer’s idea): outside this exercise.

7.5 Integrity and ethics

• AI assistance is disclosed (Section 10). Numerical claims were checked against the course PDFs, the draft coopetition overview and the designer CSV.
• Peer and forum statements are quoted as peer statements; Smith et al. (2026) is a preprint.
• Public posts contain design values only, with no credentials, emails or file paths.

8. Supporting tools

Figures. Figures 1–5 are supplied as PNG and SVG files in the images/ folder beside this report; keep the folder structure when uploading to the website.

How the toolkit was checked. When it loads it recomputes D1–D3 and compares them with this report: $1,290.93, $8,374.33 and $59,877.49 per litre, and water rows of 3.875, 3.45 and 1.40 µL. It also tests its own SHA-256 routine against a known value. Its DsbC ladder reproduces the break-evens in B1d.

What it is not. It is a planning model of the designer, not the designer. Its optimiser trades cost against departure from a known recipe and cannot predict yield. Its dispensing figures use placeholder error values. The final designs must be entered in the live designer, where the water row should match.

Design fingerprints (SHA-256 of the toolkit’s canonical JSON for each design, with label “D1 sfGFP”, “D2 PETase” and “D3 Reteplase” and date 2026-10-13; recomputed independently in Python and matching):

A hash shows that the content later differs; it does not show who changed it. The toolkit link is private until its owner shares it, so share it before putting a link on a public page.

9. Defence pack

9.1 What I did, why, and how (short version)

What. I designed one 20 µL cell-free reaction for each target by taking a known recipe and changing as little as possible: sfGFP (Smith #35 with guanine replacing GMP and spermidine at 1.5 mM), PETase (the ribose reference with DsbC cut from 5 to 0.125 µM), and Reteplase (the PEP reference with DsbC cut to 1 µM). For the community reagent I chose Bis-Tris.

Why. The score is cost per gram or per unit of activity. Lysate and DNA fix about $1,215–1,250/L, so for sfGFP the lever is titer, and for the two enzymes the lever is purified DsbC, which at 5 µM is 99% of the reference cost. Each change targets either titer or the DsbC line, and each comes with a written hypothesis and a result that would prove it wrong.

How. I built a cost model from the price list and the reaction rules, checked it against all 12 designer reference plates (water row within 0.005 µL) and against Smith’s published $/g (within 2%), then used it to compare options, compute break-even activity for DsbC and choose baselines. Every rejected option has a numeric or evidence-based reason.

9.2 Likely questions

10. Use of AI

AI client: Claude (Anthropic), Claude Sonnet 5.5, used through a claude.ai Project, to read the brief, papers and lecture material, build the cost model and workbook, and draft this report. I verified the numerical claims against the course PDFs, the draft coopetition overview and the designer’s CSV export. The toolkit’s code was also written with Claude and tested in a headless browser against the report’s totals; I did not independently audit the code.

Prompts used (replace with your exact wording before submitting):

1. [Prompt: how to approach the assignment in an original, analytical way, with reasoning for choices and rejected alternatives, and costing]
2. [Prompt: the costing workbook request]
3. [Prompt: draft proposed solution in markdown with diagrams, formulations and calculations]
4. [Prompt: explain terms, alternatives not chosen, and a short defence description]
5. [Prompt: apply relevant group-chat items]
6. [Prompt: explain what the assignment is]
7. [Prompt: a cost-effective, stable, multi-use community reagent; one or several?]
8. [Prompt: write the report for the webpage]
9. [Prompt: other constraints, limitations or boundary-pushing aspects I am not considering]
10. [Prompt: can I build a tool useful for this assignment, and a wider set of tools]
11. [Prompt: write the updated markdown report with all requirements]

11. Limits and open items

• Oxalic acid has no list price; modelled at $0.
• µM† is unconfirmed, so PETase and Reteplase specific costs are relative rankings only.
• Smith’s Table 2 lists K-glutamate at 273–313 mM where the designer shows 200 mM.
• Smith et al. 2026 is a preprint (not peer reviewed). Peer statements from the forum are quoted as peer statements and were not independently verified.
• Titers from different plate formats are not comparable.
• Plate, labour and assay costs are excluded.
• Part C was not built in the Kernel.
• The toolkit’s dispensing-error parameters (3% CV, 15 nL) are placeholders, and its optimiser has no yield model.
• The toolkit is a model of the designer, not the designer; its link is private until shared.
• I could not confirm the scoring metric, the exact meaning of µM†, or the course rule on late work (Section 7.1).
• Confirm: the community reagent list and your vote; forum access; the live designer’s water row for D1–D3; the PEP export after submitting; and whether late submission is accepted.

12. References

• Smith, A. A., et al. (2026). Using a GPT-5-driven autonomous lab to optimize the cost and titer of cell-free protein synthesis. bioRxiv. https://www.biorxiv.org/content/10.64898/2026.02.05.703998v1 (preprint)
• Olsen, M. L., et al. (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
• 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
• 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
• Rasor, B. J., et al. (2021). Toward sustainable, cell-free biomanufacturing. Current Opinion in Biotechnology, 69, 136–144. https://doi.org/10.1016/j.copbio.2020.12.012
• Yin, G., & Swartz, J. R. (2004). Enhancing multiple disulfide bonded protein folding in a cell-free system. Biotechnology and Bioengineering. https://doi.org/10.1002/bit.10827 (cited as quoted on the course forum; not independently checked)
• Ginkgo Bioworks. CFPS coopetition overview and guidelines (draft), HTBAA copy; designer composition table (CSV export).