Week 1 HW: Final Project Ideas

1. What do you want to make? → An engineered trehalose-overproducing rhizobacterium, applied as a dry seed coating on rainfed wheat, that manufactures drought-priming signal continuously at the root instead of delivering it once as a chemical.

2. What living system could make it? → Azospirillum brasilense (or a Bacillus backup for shelf life) carrying a bifunctional yeast TPS–TPP fusion under an osmotically inducible promoter. In: seed, soil water, root exudate carbon. Out: a colonized, stress-primed root system.

3. How could automation help you make it? → Prototyping: an Opentrons OT-2 screening 24 promoters × 8 osmotic potentials × 3 replicates in one afternoon to find the right induction threshold. Manufacturing: automated fed-batch control, continuous centrifugation, drying, and drum seed coating with in-line viability QC.

4. What would it take to make it at a meaningful scale? For 3 Mha of Central Anatolia: 1.5 × 10¹⁹ CFU/year, which is 1,500 m³ of broth at 10¹³ CFU/L — 30 batches in a 50 m³ fermenter, about 45 days. Allow one log of shelf-life loss and it becomes 15,000 m³ and two lines running most of the year. Roughly $1–5/ha, so break-even needs under 1% yield response.

Premise:

Exogenously applied trehalose is a well-supported drought-priming agent for cereals, but it is also a commodity chemical produced industrially since 1995, and it is cleared from plant and soil within days by ubiquitous trehalase activity. Both facts point away from manufacturing the molecule itself. This proposal therefore relocates the biomanufacturing target: instead of producing trehalose and applying it once at imbibition, I propose to manufacture the organism that synthesises it continuously at the root interface — an engineered, trehalose-overproducing plant-growth-promoting rhizobacterium delivered as a dry seed coating on rainfed wheat.

The design rests on a specific mechanistic claim, supported in the evidence section below: the benefit of bacterial trehalose is not bulk osmoprotection of the plant but a signalling effect, in which small quantities reaching the root upregulate stress-response pathways. If that is correct, continuous low-level production at the rhizosphere is the appropriate delivery mode, and a single large dose is the wrong one.

(1) What you would like to biomanufacture and why?

The product. A seed-applied, engineered trehalose-overproducing rhizobacterial inoculant for rainfed bread wheat, manufactured as a spore-based dry seed coating. The manufactured object is living cells, not a chemical: roughly 10⁶ colony-forming units deposited on each seed, which then replicate in the rhizosphere and synthesize trehalose in situ under water deficit.

Who it is for. Rainfed wheat growers on the Central Anatolian plateau. Türkiye plants approximately 7.1 million hectares of wheat at an average yield near 2.61 t/ha, concentrated in Central Anatolia and Southeast Anatolia. These are largely unirrigated fields where yield is set by spring rainfall, and the damaging stress is terminal drought during grain filling. Turkish average wheat yields sit well below those of comparable temperate producers, and the gap is substantially a water gap rather than a genetics gap.

Why this, and why me. My motivation is utilitarian and regional. I am not proposing to eliminate drought loss — no seed treatment can — but to buy a partial physiological buffer at a cost per hectare low enough that a smallholder adopts it without a subsidy. The target is a single-digit dollar cost per hectare against roughly $650/ha of gross wheat revenue, which means even a modest yield response clears break-even by a wide margin. My own background is in microbial bioprocess development and fermentation scale-up, so the manufacturing half of this problem is one I can assess concretely rather than hand-wave.

Why it is interesting as biomanufacturing. Trehalose is already an industrial commodity, produced biologically since 1995 and available at roughly $3.50–6/kg. Manufacturing it myself would be a purchase order dressed up as a project. The interesting engineering problem is not the molecule but the delivery of continuous synthesis — which converts a chemical-supply question into a genuine chassis, construct, and formulation question, and which addresses the central weakness of exogenous trehalose application: persistence.

(2) What living system would you use?

Chassis. Two credible options, with an explicit trade-off I have not yet resolved experimentally.

Azospirillum (e.g. A. brasilense or A. argentinense) carries almost all of the published efficacy evidence for trehalose-mediated drought protection in cereals, and it is an established, commercially deployed cereal root colonizer. Bacillus sporulates, which is decisive for shelf life and dry seed coating — a non-sporulating inoculant loses roughly three log units of viability over six months of storage, which as shown in Section 4 is the single largest driver of required plant size.

I propose Azospirillum as the scientifically justified primary chassis, because efficacy evidence is harder to substitute than formulation engineering, and I flag the shelf-life penalty as a named assumption to be tested rather than assumed away. A Bacillus backup chassis would be carried in parallel.

Construct. A bifunctional yeast TPS–TPP fusion (trehalose-6-phosphate synthase fused to trehalose-6-phosphate phosphatase), rather than otsA alone. This choice follows directly from the maize data: the strain carrying the bifunctional fusion outperformed the strain carrying ReOtsA alone, giving biomass 1.7-fold greater than non-inoculated plants. Fusing the two activities also channels the T6P intermediate, limiting accumulation of a metabolite that is itself a potent regulator of central carbon metabolism.

Regulation — my one deliberate departure from the literature. The published constructs used a constitutive lac promoter. I propose instead a stress-inducible promoter — an osmotically regulated element such as proU, osmY, or the native otsBA promoter. The rationale is metabolic burden: constitutive overproduction taxes the cell throughout the non-stressed majority of the growing season, when trehalose confers no benefit, and burden is a well-documented cause of impaired rhizosphere colonization fitness. The desired transfer function is minimal expression at field capacity and sharp induction below a defined soil water potential. Tuning this is the core prototyping task in Section 3.

Process description.

• In: coated seed; soil water at imbibition; root exudate carbon (organic acids, sugars) as the ongoing feedstock once colonization is established.

• Process: seed imbibes → bacteria germinate from spores or resuscitate from dried cells → rhizosphere colonization over the first 1–3 weeks → under declining soil water potential the osmoresponsive promoter fires → TPS–TPP converts UDP-glucose and glucose-6-phosphate to trehalose-6-phosphate and then to trehalose → trehalose accumulates intracellularly, protecting the bacterium, and a small quantity reaches the root.

• Out: a colonized, stress-primed root system; upregulated plant stress-response, carbon and nitrogen metabolism gene expression; measurable outcomes are relative water content, biomass, and ultimately grain yield under water deficit.

Assumptions I am making, stated explicitly.

1. That trehalose or T6P crosses from bacterium to root in a physiologically meaningful quantity. This is the field’s working hypothesis, not a measured flux, and it is Unknown #1 in Section 5.
2. That results obtained in maize transfer to wheat, despite differences in root architecture and in the timing of the critical stress window.
3. That a promoter tuned against PEG or NaCl in liquid culture fires appropriately at real field matric potentials — laboratory osmotic stress and field drying are not the same stimulus.
4. That the engineered strain colonizes wheat roots at wild-type efficiency. This is supported by the observation that altering trehalose metabolism in Azospirillum did not impair root colonization under either normal or drought conditions.

(3) How would you use automation?

Prototyping: promoter tuning on a liquid-handling robot

The critical design parameter is the promoter’s transfer function, and finding it is a combinatorial search. I would use an Opentrons OT-2 to build and screen a library of candidate stress-responsive promoters driving a fluorescent reporter in the chosen chassis.

• 24 candidate promoters (native otsBA, proU, osmY, synthetic variants with mutated operator spacing)

• × 8 osmotic potentials (a PEG-6000 or NaCl ladder spanning field capacity to severe deficit)

• × 3 biological replicates

• = 576 wells, or two 384-well plates, read on a plate reader in a single afternoon

The selection criterion is not maximum expression but the right shape: low basal output at high water potential, sharp induction below a defined threshold, and no runaway expression at extreme stress. Screening this by hand is the difference between a week and a semester, and manual pipetting introduces exactly the well-to-well variance that would obscure a transfer-function shape.

A second automation target is the production fermentation itself: a parallel micro-bioreactor array under automated design-of-experiments to optimize carbon source, dissolved oxygen, feed rate, and (for a Bacillus backup) the sporulation trigger. This is where fed-batch process development compresses from months of sequential runs to weeks of parallel ones.

Manufacturing: entirely different equipment

At production scale the automation is conventional bioprocess and seed-treatment machinery, not laboratory robotics: automated fed-batch feed control on a 50 m³ fermenter, continuous-flow centrifugation for harvest, spray or freeze drying with in-line moisture control, and automated seed coating. Commercial seed treatment uses fluidized-bed equipment, augers, roller mills, rotostatic seed treaters and drum coaters — and the automation that matters most is in-line viability QC, because the product’s specification is live cells per seed and that number decays continuously from the moment of harvest.

(4) How much would you need to make, and how long would it take?

(a) How much?

Scale definition: the rainfed wheat area of Central Anatolia, taken as 3 million hectares (a conservative subset of Türkiye’s ~7.1 Mha total wheat area).

(b) How long?

Assume high-cell-density fed-batch fermentation reaching 1 × 10¹⁰ CFU/mL = 1 × 10¹³ CFU/L. This is deliberately conservative; optimised low-cost media for bioinoculants have been reported substantially higher.

Base case (no viability loss):

• Broth required: 1.5 × 10¹⁹ CFU ÷ 10¹³ CFU/L = 1.5 × 10⁶ L = 1,500 m³/year

• In a 50 m³ fermenter: 1,500 ÷ 50 = 30 batches

• At 24 h fermentation + 12 h turnaround = 1.5 days/batch: 45 days of fermentation per year

So a single 50 m³ line, running roughly six weeks a year, supplies an area the size of Central Anatolia.

Realistic case (1-log loss through drying, formulation and storage):

Regulatory frameworks typically require 10⁷–10⁹ CFU per gram or millilitre at time of manufacture, and liquid inoculants have been documented losing on the order of three log units over six months of storage. Allowing a single log of overage:

• Broth required: 15,000 m³/year → 300 batches → ~450 days, i.e. two lines running nearly continuously.

This tenfold swing is the dominant result of the whole calculation, and it is not a titer problem. Plant size here is set by whether the organism survives the shelf, not by how well it grows in the fermenter. That is the engineering argument for carrying a sporulating Bacillus chassis in parallel: spores would collapse the overage factor and with it the capital requirement.

Cost sanity check

At 1,500 m³ for 3 Mha, broth consumption is 0.5 L per hectare. At a fully loaded fermentation cost of $2–10/L, the active ingredient costs $1–5/ha — essentially identical to the purchased-trehalose route (0.27 kg/ha × ~$4/kg ≈ $1.10/ha), but delivering continuous rather than single-dose synthesis. Against roughly $650/ha of gross wheat revenue at 2.6 t/ha, break-even requires a yield response below 1%. The economics are not the risk in this project. The biology is.

5. What don’t you know yet?

Unknown 1 — Does anything actually cross from bacterium to root?

The entire design rests on bacterial trehalose reaching the plant as a signal, and this flux has never been directly measured. The literature’s own statement of the mechanism is explicitly hypothetical: that very small amounts of trehalose are translocated to the roots and signal stress-tolerance pathways in the plant. It remains possible that the observed benefit is mediated entirely by bacterial survival — a trehalose-protected bacterium simply persists better under drought and continues delivering unrelated plant-growth-promoting functions such as auxin secretion.

What I would measure: ¹³C- or ¹⁴C-labelled trehalose feeding with root-tissue tracing; quantification of T6P and trehalose in root tissue by LC-MS/MS comparing engineered, wild-type and trehalase-expressing strains; and transcriptomic confirmation that plant stress-response genes are induced in a trehalose-dependent manner. If the flux is negligible, the whole delivery rationale collapses and the project should revert to direct trehalose application.

Unknown 2 — Does the maize result transfer to wheat under terminal drought?

Every supporting study is a pot experiment scoring seedling survival and biomass, not a multi-site field trial scoring grain yield. The proximate endpoints reported — survival percentage, shoot and root fresh weight, relative water content, leaf and root length — are not yield. Wheat and maize differ in root architecture, in C3 versus C4 physiology, and critically in the timing of the stress window that determines yield: wheat’s vulnerability is terminal drought during grain filling, which no seedling-stage assay captures.

What I would measure: multi-site, multi-year replicated field trials in Central Anatolia with grain yield as the primary endpoint and managed terminal-drought treatments, plus rhizosphere colonization persistence through to anthesis.

Unknown 3 — Is the regulatory path longer than the technical one?

This is a deliberate environmental release of a genetically modified organism across millions of hectares. Turkey’s biosafety framework is restrictive, and no engineered PGPR of this type is approved for release there. It is entirely possible that the science works and the product never reaches a field. The fallback, and it is a serious one: screen for naturally high-trehalose-accumulating wild isolates rather than engineering them. A. argentinense Az19 is itself a wild osmotolerant strain, and the loss-of-function work demonstrates that its native trehalose accumulation is already functionally significant under drought. A selection-based rather than engineering-based product would likely show a smaller effect size but faces a dramatically shorter route to the field. I would run both tracks in parallel and let the regulatory timeline, not the effect size, decide which ships.

Supporting evidence

The proposal has a direct published precedent, and the evidence chain is unusually strong because it includes gain-of-function, extension to a cereal, and loss-of-function control.

Gain of function — Suárez et al. 2008

Overexpression of trehalose-6-phosphate synthase in the symbiotic bacterium Rhizobium etli, inoculated onto common bean:

• Plants given the overexpressing strain had more nodules, higher nitrogenase activity and greater biomass than plants given wild-type R. etli

• Plants given a trehalose-6-phosphate synthase mutant strain had fewer nodules, less nitrogenase activity and less biomass — a dose-response in both directions

• Under drought, plants inoculated with the overexpressing strain fully recovered, while those inoculated with wild-type or mutant strains wilted and died

• Yield under constant irrigation increased by more than 50%

• Macroarray analysis of 7,200 ESTs from nodules revealed upregulation of genes in stress tolerance and in carbon and nitrogen metabolism, suggesting a signalling mechanism for trehalose

That final point is the mechanistic basis for preferring continuous in-situ synthesis over a single applied dose.

Extension to a cereal — Rodríguez-Salazar et al. 2009

Azospirillum brasilense transformed with a trehalose-biosynthesis gene fusion from S. cerevisiae, inoculated onto maize:

• Transformed bacteria grew at up to 0.5 M NaCl and accumulated trehalose; wild type tolerated neither osmotic stress nor accumulated significant disaccharide

• 85% of inoculated maize plants survived drought stress versus 55% with the wild-type strain

• 73% increase in biomass relative to wild-type inoculation, with significant increases in leaf and root length

• The bifunctional yeast fusion construct outperformed ReOtsA alone, giving biomass 1.7× that of non-inoculated plants — the basis for the construct choice in Section 2

Loss of function — Azospirillum argentinense Az19

The strongest single piece of evidence, because it isolates the variable. The trehalase gene treF was constitutively expressed in Az19 to abolish trehalose accumulation:

• The resulting strain Az19F did not accumulate trehalose and was impaired in coping with salt, osmotic and UV stress

• Root colonization was unaffected under both normal and drought conditions — removing the obvious confound

• Both strains promoted maize growth equally under normal irrigation

• Under drought, Az19F was defective: reduced capacity to improve plant height and fresh weight, and it failed to raise relative water content as Az19 did Colonization held constant, trehalose removed, benefit disappears only under drought. This rules out the alternative explanation that the bacteria help through unrelated plant-growth-promoting mechanisms.

References

Suárez, R., Wong, A., Ramírez, M., Barraza, A., Orozco, M. del C., Cevallos, M. A., Lara, M., Hernández, G., & Iturriaga, G. (2008). Improvement of drought tolerance and grain yield in common bean by overexpressing trehalose-6-phosphate synthase in rhizobia. Molecular Plant-Microbe Interactions, 21(7), 958–966. https://doi.org/10.1094/MPMI-21-7-0958

Rodríguez-Salazar, J., Suárez, R., Caballero-Mellado, J., & Iturriaga, G. (2009). Trehalose accumulation in Azospirillum brasilense improves drought tolerance and biomass in maize plants. FEMS Microbiology Letters, 296(1), 52–59. https://doi.org/10.1111/j.1574-6968.2009.01614.x

Maize drought protection by Azospirillum argentinense Az19 requires bacterial trehalose accumulation. (2024). Applied and Environmental Microbiology. PMC11680609.

Al Hinai, M. S., Rehman, A., Siddique, K. H. M., & Farooq, M. (2025). The role of trehalose in improving drought tolerance in wheat. Journal of Agronomy and Crop Science, 211, e70053. https://doi.org/10.1111/jac.70053

USDA Foreign Agricultural Service. (2020). Turkey MY2020/2021 near-record wheat yield. International Production Assessment Division.

Republic of Türkiye Ministry of Agriculture and Forestry. Seed Production. (National wheat area and seed volume figures.)

Course readings drawn on: Petzold & Mukhopadhyay (2025) on high-throughput technologies and automated workflows in biomanufacturing; Asin-Garcia et al. (2025) on enabling research infrastructure.

Declaration of AI use

Client: Claude Opus 5 and Undermind.ai

Prompt: I want to find primary research on exogenous trehalose application in wheat, barley, and maize—as seed priming, seed soaking, seed coating, or foliar treatment—to improve drought tolerance under agronomic-scale water-deficit conditions beyond seedling or PEG-based screening, including field, semi-field, or other production-relevant studies that report quantitative grain yield and/or yield-component outcomes. I want studies that specify trehalose concentration/dose and application method, ideally include dose–response relationships, and provide evidence relevant to whether the observed effects are due to osmoprotection versus trehalose-6-phosphate–related signaling, including data on the persistence or turnover of exogenously applied trehalose in planta given trehalase activity. Reviews are useful for orientation, but the priority is primary research with quantitative agronomic outcomes, excluding studies focused solely on transgenic TPS/TPP manipulation unless they also include an exogenous trehalose application comparison.

Week 2 HW: DNA Design

Part 0: Setup in Benchling

• Repository Name: Project/MIT HTBAA 2026

• Project Name: Engineered trehalose-overproducing rhizobacterial seed inoculant for rainfed wheat

Part 1: Expression Target

• What molecule am I expressing?

  A single bifunctional trehalose-6-phosphate synthase / trehalose-6-phosphate phosphatase (TPS–TPP) fusion enzyme of 892 amino acids: one polypeptide carrying two sequential catalytic activities. TPS (EC 2.4.1.15) condenses UDP-glucose and glucose-6-phosphate to trehalose-6-phosphate (T6P); TPP (EC 3.1.3.12) then dephosphorylates T6P to trehalose. Expressing this as one fusion rather than two genes is a deliberate design choice with three justifications, two of them drawn directly from the source literature:

1. It avoids T6P accumulation. Rodríguez-Salazar et al. state that the bifunctional construct was selected because it has strong TPP activity, thereby avoiding accumulation of free trehalose-6-phosphate in the cytosol, which is known to provoke pleiotropic effects (De Virgilio et al. 1994; Schluepmann et al. 2004). T6P is a regulator of central carbon metabolism, not an inert intermediate.

2. It is empirically the better performer. In the same study the BIF strain outperformed the single-enzyme ReOtsA strain at 0.5 M NaCl, accumulated 2× basal trehalose under salt stress versus 1.5× for ReOtsA, and produced the only statistically significant gains in plant survival (85%) and biomass (1.7× uninoculated).

3. It simplifies the genetic design. One promoter, one ribosome binding site, one coding sequence, and no stoichiometric balancing between separately expressed enzymes.

• What is the sequence, and where does it come from?

  The construct reproduces the chimeric BIF gene of Miranda et al. (2007), as deployed in A. brasilense by Rodríguez-Salazar et al. (2009).

  Domain	Accession	Region used	Notes	Sequence
  TPS (synthase)	Q00764	TPS1_YEAST residues 1–493 of 495	Full-length CDS minus the C-terminal Lys-Asn and the stop codon, removed by the cloning primer	MTTDNAKAQL… MTTDNAKAQLTSSSGGNIIVVSNRLPVTITKNSSTGQYEYAMSSGGLVTALEGLKKTYTFKWFGWPGLEIPDDEKDQVRKDLLEKFNAVPIFLSDEIADLHYNGFSNSILWPLFHYHPGEINFDENAWLAYNEANQTFTNEIAKTMNHNDLIWVHDYHLMLVPEMLRVKIHEKQLQNVKVGWFLHTPFPSSEIYRILPVRQEILKGVLSCDLVGFHTYDYARHFLSSVQRVLNVNTLPNGVEYQGRFVNVGAFPIGIDVDKFTDGLKKESVQKRIQQLKETFKGCKIIVGVDRLDYIKGVPQKLHAMEVFLNEHPEWRGKVVLVQVAVPSRGDVEEYQYLRSVVNELVGRINGQFGTVEFVPIHFMHKSIPFEELISLYAVSDVCLVSSTRDGMNLVSYEYIACQEEKKGSLILSEFTGAAQSLNGAIIVNPWNTDDLSDAINEALTLPDVKKEVNWEKLYKYISKYTSAFWGENFVHELYSTSSSSTSSSAT
  Linker	—	BamHI site GGATCC	Encodes Gly-Ser; contributes 2 residues, counted once in the fusion	GS
  TPP (phosphatase)	P31688	TPS2_YEAST residues 500–896 of 896	C-terminal 397 aa. Removing the first 504 residues of TPS2 leaves a 392-aa C-terminus that is still fully active, since it encompasses the homology blocks present in all TPP enzymes (Vogel et al. 1998); this construct retains slightly more than that minimum	AKSINMALKL… AKSINMALKLDKEEKSNLESKLWKEVPTIQDWTNKFLSSLKEQASSNDDMERKMTPALNRPVLLENYKQAKRRLFLFDYDGTLTPIVKDPAAAIPSARLYTILQKLCADPHNQIWIISGRDQKFLNKWLGGKLPQLGLSAEHGCFMKDVSCQDWVNLTEKVDMSWQVRVNEVMEEFTTRTPGSFIERKKVALTWHYRRTVPELGEFHAKELKEKLLSFTDDFDLEVMDGKANIEVRPRFVNKGEIVKRLVWHQHGKPQDMLKGISEKLPKDEMPDFVLCLGDDFTDEDMFRQLNTIETCWKEKYPDQKNQWGNYGFYPVTVGSASKKTVAKAHLTDPQQVLETLGLLVGDVSLFQSAGTVDLDSRGHVKNSESSLKSKLASKAYVMKRSASYTGAKV

• TPS-TPP Fusion Protein

  MTTDNAKAQLTSSSGGNIIVVSNRLPVTITKNSSTGQYEYAMSSGGLVTALEGLKKTYTFKWFGWPGLEIPDDEKDQVRKDLLEKFNAVPIFLSDEIADLHYNGFSNSILWPLFHYHPGEINFDENAWLAYNEANQTFTNEIAKTMNHNDLIWVHDYHLMLVPEMLRVKIHEKQLQNVKVGWFLHTPFPSSEIYRILPVRQEILKGVLSCDLVGFHTYDYARHFLSSVQRVLNVNTLPNGVEYQGRFVNVGAFPIGIDVDKFTDGLKKESVQKRIQQLKETFKGCKIIVGVDRLDYIKGVPQKLHAMEVFLNEHPEWRGKVVLVQVAVPSRGDVEEYQYLRSVVNELVGRINGQFGTVEFVPIHFMHKSIPFEELISLYAVSDVCLVSSTRDGMNLVSYEYIACQEEKKGSLILSEFTGAAQSLNGAIIVNPWNTDDLSDAINEALTLPDVKKEVNWEKLYKYISKYTSAFWGENFVHELYSTSSSSTSSSATGSAKSINMALKLDKEEKSNLESKLWKEVPTIQDWTNKFLSSLKEQASSNDDMERKMTPALNRPVLLENYKQAKRRLFLFDYDGTLTPIVKDPAAAIPSARLYTILQKLCADPHNQIWIISGRDQKFLNKWLGGKLPQLGLSAEHGCFMKDVSCQDWVNLTEKVDMSWQVRVNEVMEEFTTRTPGSFIERKKVALTWHYRRTVPELGEFHAKELKEKLLSFTDDFDLEVMDGKANIEVRPRFVNKGEIVKRLVWHQHGKPQDMLKGISEKLPKDEMPDFVLCLGDDFTDEDMFRQLNTIETCWKEKYPDQKNQWGNYGFYPVTVGSASKKTVAKAHLTDPQQVLETLGLLVGDVSLFQSAGTVDLDSRGHVKNSESSLKSKLASKAYVMKRSASYTGAKV

  Protein: 493 (TPS1) + 2 (Gly-Ser linker) + 397 (TPS2 C-terminus) = 892 aa, matching the ORF reported after sequencing of the published fusion.

  DNA: the published chimeric fragment is 2,873 bp, comprising a 1,526-bp TPS1 fragment (including 41 bp of 5′ leader) and a 1,353-bp TPS2 fragment (including 153 bp of untranslated 3′ end), with the shared BamHI site counted once: 1,526 + 1,353 − 6 = 2,873 bp. Subtracting the untranslated flanks gives 2,873 − 41 − 153 = 2,679 bp of coding sequence = 892 codons + stop. The protein and DNA arithmetic agree independently.

• What host or chassis?

  Azospirillum brasilense strain Cd, the plant-growth-promoting rhizobacterium proposed in Homework #1. Two organisms appear in the workflow: the plasmid is constructed in E. coli DH5α, mobilized by conjugation using E. coli S17-1, and expressed in A. brasilense. Conjugation rather than electroporation is the published route.

  One fact about this host drives much of the design: A. brasilense has no native trehalose biosynthesis genes — Rodríguez-Salazar et al. note that no homologous genes are available in databases from A. brasilense or closely related species, and that wild-type cells neither tolerate osmotic stress nor accumulate significant trehalose. The trehalose pathway here is genuinely introduced, not amplified, and there is no native ots promoter in this organism to borrow.

• Where does the product need to go?

  Cytoplasm. No signal peptide, no targeting sequence. Both substrates are cytoplasmic metabolites and the enzyme has no reason to leave the cell. Whether the trehalose product reaches the wheat root is a transport question, not a targeting question — and as Part 6 sets out, the published attempt to detect it in plant tissue failed. Adding a secretion signal to an 892-aa cytoplasmic enzyme would be a design error.

Part 2: Genetic Parts

The expression context is taken from Rodríguez-Salazar et al., who used the broad-host-range vector pBBR1MCS-5 (Kovach et al. 1995) with expression under the lac promoter, gentamicin selection at 30 µg/mL, and conjugal transfer from E. coli S17-1. Their construct is designated pBBR1M::BIF, where :: is the standard genetic notation for “carrying” — i.e. pBBR1MCS-5 carrying the BIF gene. The constructs described below are distinct from theirs and are named accordingly.

What is novel here, and what is deliberately reused The backbone is reused, and intentionally so. An origin of replication, a mobilization region and a resistance cassette are solved engineering problems; re-deriving a broad-host-range replicon that functions in an alphaproteobacterium would be a multi-year project rather than a design exercise. pBBR1MCS-5 is treated here as a chassis for the cassette, and what it provides is stated explicitly in the table above.

The novelty is in the cassette, and it is threefold:

(i) the coding sequence is codon-optimized for ~68% GC and is therefore a different molecule from the published BIF gene — roughly 75–80% nucleotide identity while encoding an identical protein;

(ii) the regulatory architecture replaces Plac with an RpoE2-dependent osmo-inducible promoter, which is the design contribution carried forward from Homework #1; and

(iii) an explicit terminator is added, where the published construct relied on vector context. The deployment construct is different again — a markerless mini-Tn7 chromosomal integrant rather than a plasmid at all.

Three plasmids are specified rather than one, so that each experimental arm changes exactly one variable. Without pAzT-1 and pAzT-2 it would be impossible to attribute any change in activity to codon optimization as opposed to the promoter swap.

Containment — named, and deliberately not built into this plasmid Homework #1 identified regulatory approval, not biology, as the likely binding constraint on this product. Plasmid-borne gentamicin resistance distributed across three million hectares is not a defensible release. The deployment strain should therefore be a markerless chromosomal integrant delivered by mini-Tn7, which also removes the plasmid-retention risk discussed in Part 6. The plasmid described here is a research tool for the promoter screen and the activity assays, not the product.

Part 3: Construct Assembly

Order of parts in pAzT-3, 5′→3′, in the pBBR1MCS-5 multiple cloning site. pAzT-1 and pAzT-2 differ only in the promoter and the codon usage of the CDS:

RNA polymerase copies DNA into mRNA. It starts at the promoter and runs until the terminator. Then the ribosome reads that mRNA into protein — but it does not start at the beginning of the mRNA. It docks at the RBS, slides to the ATG, and stops at TAA, well before the mRNA ends.

Native TPS1 ends with …AAG AAC TAA — Lys, Asn, then stop. Leave that stop in place and the ribosome falls off the mRNA before it ever reaches TPS2.

What the finished protein actually does T6P is not an inert intermediate — it is a potent regulator of central carbon metabolism, so letting it accumulate causes trouble of its own. Tethering the phosphatase to the synthase means the intermediate is handed straight from one active site to the next.

Part 4: Optimization and quality control

(a) Codon optimization

What I ran. The 892-residue fusion was back-translated against the Azospirillum brasilense codon usage table from the Kazusa database (species 192; 233 CDSs, 86,126 codons). The table’s published GC figures were recomputed independently and matched to within 0.02%, confirming the table was read correctly.

Constraints applied: exclude XhoI, KpnI, EcoRI, XbaI, SpeI and PstI entirely; permit exactly one BamHI site, at the fusion junction; cap GC at 75% in any 60 bp window; cap homopolymer runs at 6; cap repeated k-mers at 12 bp; and forbid internal Shine–Dalgarno-like motifs, which cause ribosomal pausing.

What I concluded. The most important result is a limit rather than a gain: this protein cannot reach the 68.2% GC of the host genome. Its ceiling is 63.5%, and the constrained build reaches 63.0% of that. The reason is the second codon position, which is fixed by the amino acid for 825 of the 892 codons — serine is the only residue in the whole fusion offering any choice there. 582 of the residues force an A or T at that position, lysine and asparagine worst of all with AA and no GC at all in the first two bases. Optimization can therefore only act on the third position, which it drives to 98.8%. Quoting a 68% target would misstate what is achievable for this sequence.

(b) Sequence search

Blastp result for TPS

Blastp result for TPP

(c) Biosecurity screen

Benchling has no native biosecurity screen. Rather than skip the step, run the Common Mechanism, the free open-source screening tool published by IBBIS (the International Biosecurity and Biosafety Initiative for Science) and available on GitHub.

The coding sequence itself should screen clean. Trehalose biosynthesis enzymes are metabolically innocuous, the donor is a food-grade yeast, and the product is a common disaccharide sold as a food ingredient. Two flags are plausible on the full plasmid rather than the insert: the aacC1 gentamicin resistance marker, because antimicrobial resistance determinants are routinely flagged, and the mob/oriT mobilization functions, because conjugative transfer is a horizontal-gene-transfer consideration.

Both flags are expected, and the response to both is the same: each is legitimate in a laboratory research plasmid and not defensible in an environmental release across three million hectares. That is precisely why Part 2 specifies the deployment strain as a markerless mini-Tn7 chromosomal integrant rather than this plasmid. The screen does not change the design; it confirms the reasoning already built into it.

Part 5: From design to build

How would I build it?

The CDS is 2,679 bp; the cassette with promoter, RBS and terminator is roughly 3 kb; the finished plasmid is approximately 7.5–8 kb. That size favours fragment-based assembly: order the codon-optimized CDS as two or three synthetic fragments of about 1 kb and join them into PCR-linearized pBBR1MCS-5 in one Gibson assembly reaction. Gibson et al. (2009) demonstrated one-pot isothermal assembly of overlapping fragments scaling to hundreds of kilobases, so a three- or four-piece join is well within the method, and it removes any dependence on unique restriction sites in a sequence that has just been rewritten.

This is a deliberate departure from the published route, which used restriction cloning via XhoI and KpnI into pBBR1MCS-5 with an internal BamHI junction. Restriction cloning was appropriate for the unoptimized yeast sequence; after codon optimization at ~68% GC the risk of introducing new internal sites makes a sequence-independent method the safer choice. A single 2.7 kb synthetic block was considered and rejected: a GC-rich block of that length is exactly what commercial synthesis struggles with, and splitting it means a failed synthesis costs one fragment rather than the whole gene.

What does synthesis constrain?

There is a genuine tension in this design: optimizing toward 68% GC pushes the sequence into hard-to-synthesize territory. GC-rich windows reduce oligonucleotide coupling efficiency and promote hairpins that stall assembly, so the same change intended to improve expression degrades manufacturability. The 75% windowed-GC cap in Part 4 is the compromise between those two pressures.

Chip-synthesized oligonucleotides carry error rates high enough that across a 2.7 kb assembly most clones will contain at least one error. Carr et al. (2004) demonstrated protein-mediated error correction using MutS to remove heteroduplex-containing molecules from synthesis pools; Kosuri et al. (2010) showed scalable gene synthesis by selectively amplifying subpools from high-fidelity microchips, building on the accurate multiplex synthesis from programmable microchips of Tian et al. (2004). For a single construct the practical choice is to buy error-corrected fragments or to sequence-verify several clones; at library scale, error correction stops being optional.

Prototype versus scale

For this product, DNA is not the thing that scales. The construct is built and sequence-verified once; from then on the process manufactures cells. The Homework #1 production figures — 1,500 m³ of fermentation broth per year in the base case, 15,000 m³ allowing one log of shelf-life loss — involve no further DNA synthesis at all.

Chip-based pooled synthesis earns its place at the prototyping stage instead. The promoter screen in Appendix A — 24 promoters × 8 osmotic potentials × 3 replicates — is exactly the case for ordering one multiplexed oligo pool rather than 24 separately synthesized constructs, which is the economic argument Kosuri et al. and Tian et al. make. The difference between prototype and scale here is not DNA volume but kind: pooled combinatorial synthesis for the design search, then a single sequence-verified chromosomal integrant for manufacturing.

Part 6: What don’t I know yet

1. Does trehalose actually reach the plant? The published attempt to detect it failed.

Rodríguez-Salazar et al. measured trehalose in maize roots and leaves of inoculated plants and reported that no significant levels of the disaccharide could be detected in any analysed tissue. In the discussion they state plainly that they could not detect the disaccharide in plant tissues and hypothesise that very small amounts translocate and signal stress-tolerance pathways, adding that there is no clear evidence as to how bacterial trehalose could regulate plant genes.

Three explanations remain open, and they are not equivalent for the product:

(i) trehalose does transfer, below the HPLC detection limit, and acts as a signal;

(ii) trehalose never leaves the bacterium, and the benefit comes from a better-surviving, better-colonizing bacterium delivering unrelated plant-growth promotion such as auxin;

(iii) T6P or another metabolite, not trehalose, is the transferred species. Under (ii), optimizing trehalose output is optimizing the wrong variable.

How I would find out: isotope tracing rather than bulk quantification — feed ¹³C-labelled glucose to the bacteria and look for labelled trehalose or T6P in root tissue by LC-MS/MS, which is orders of magnitude more sensitive than the HPLC method used. In parallel, discriminate hypothesis (ii) directly by inoculating with a trehalose-null strain that is otherwise identical, the design used in the A. argentinense Az19F work, where colonization was unaffected but the drought benefit disappeared.

2. Will a codon-optimized 892-aa yeast fusion still fold and stay active?

The published fusion works in A. brasilense — but with native yeast codon usage, not a GC-optimized version. Codon optimization alters co-translational folding kinetics even when the protein sequence is unchanged, so optimizing for expression level can reduce active yield. This construct therefore departs from the one datum that proves the design functions.

How I would find out: assay TPS and TPP activities separately in lysates (coupled-enzyme assay for TPS, phosphate release for TPP), quantify intracellular trehalose by HPLC against a purified standard, and inspect the insoluble fraction for inclusion bodies — with the unoptimized published sequence built in parallel as the reference arm, which is the only way to attribute any loss to optimization. Design variants if it fails: revert to native codons; co-express GroESL; weaken the RBS to slow translation; or substitute the bacterial otsA–otsB fusion of Seo et al. (2000), which preserves bifunctionality with a prokaryotic protein.

3. Will the inducible promoter behave, in a host with no trehalose regulon?

Because A. brasilense has no native trehalose genes, the osmo-inducible promoter must be heterologous — most plausibly an RpoE2/EcfG-dependent promoter from S. meliloti or R. etli. Whether the corresponding sigma factor and activation cascade are present and similarly tuned in Azospirillum is untested. There is also a subtler problem: the authors suspect their lac promoter may itself be stress-induced in these hosts, which would mean the published “constitutive” baseline was already partly inducible — and that my control arm is not the control I think it is.

How I would find out: the reporter screen in Appendices A and B — promoter–GFP or luxAB fusions across an eight-point osmotic ladder, selecting on the shape of induction rather than its maximum. Include Plac in that screen explicitly to test whether it is stress-responsive in this host.

Appendix A — Promoter panel for the induction screen

Appendix B — Osmotic ladder for the induction screen

References

Miranda, J. A., Avonce, N., Suárez, R., Thevelein, J. M., Van Dijck, P., & Iturriaga, G. (2007). A bifunctional TPS–TPP enzyme from yeast confers tolerance to multiple and extreme abiotic-stress conditions in transgenic Arabidopsis. Planta, 226, 1411–1421.

Rodríguez-Salazar, J., Suárez, R., Caballero-Mellado, J., & Iturriaga, G. (2009). Trehalose accumulation in Azospirillum brasilense improves drought tolerance and biomass in maize plants. FEMS Microbiology Letters, 296(1), 52–59.

Suárez, R., Wong, A., Ramírez, M., Barraza, A., Orozco, M. del C., Cevallos, M. A., Lara, M., Hernández, G., & Iturriaga, G. (2008). Improvement of drought tolerance and grain yield in common bean by overexpressing trehalose-6-phosphate synthase in rhizobia. Molecular Plant-Microbe Interactions, 21(7), 958–966.

Kovach, M. E., et al. (1995). Four new derivatives of the broad-host-range cloning vector pBBR1MCS, carrying different antibiotic-resistance cassettes. Gene, 166, 175–176.

Vogel, G., Aeschbacher, R. A., Müller, J., Boller, T., & Wiemken, A. (1998). Trehalose-6-phosphate phosphatases from Arabidopsis thaliana: identification by functional complementation in the yeast tps2 mutant. Plant Journal, 13, 673–683.

Seo, H.-S., et al. (2000). Characterization of a bifunctional enzyme fusion of trehalose-6-phosphate synthetase and trehalose-6-phosphate phosphatase of Escherichia coli. Applied and Environmental Microbiology, 66(6), 2484–2490.

Sauviac, L., Philippe, H., Phok, K., & Bruand, C. (2007). An extracytoplasmic function sigma factor acts as a general stress response regulator in Sinorhizobium meliloti. Journal of Bacteriology, 189(11), 4204–4216.

Gibson, D. G., Young, L., Chuang, R.-Y., Venter, J. C., Hutchison, C. A., & Smith, H. O. (2009). Enzymatic assembly of DNA molecules up to several hundred kilobases. Nature Methods, 6, 343–345. [assigned reading]

Kosuri, S., Eroshenko, N., LeProust, E. M., Super, M., Way, J., Li, J. B., & Church, G. M. (2010). Scalable gene synthesis by selective amplification of DNA pools from high-fidelity microchips. Nature Biotechnology, 28, 1295–1299. [assigned reading]

Carr, P. A., Park, J. S., Lee, Y.-J., Yu, T., Zhang, S., & Jacobson, J. M. (2004). Protein-mediated error correction for de novo DNA synthesis. Nucleic Acids Research, 32, e162. [assigned reading]

Tian, J., Gong, H., Sheng, N., Zhou, X., Gulari, E., Gao, X., & Church, G. (2004). Accurate multiplex gene synthesis from programmable DNA microchips. Nature, 432, 1050–1054. [assigned reading]

iGEM Registry of Standard Biological Parts. BBa_B0034 (RBS), BBa_B0015 / BBa_B0010 / BBa_B0012 (terminators), BBa_R0010 (lac promoter), BBa_J23100–J23119 (Anderson constitutive series), BBa_E0040 (GFPmut3b). https://parts.igem.org

UniProt Consortium. TPS1_YEAST (Q00764), Saccharomyces cerevisiae alpha,alpha-trehalose-phosphate synthase, 495 aa, EC 2.4.1.15; TPS2_YEAST (P31688), trehalose-phosphatase, 896 aa, EC 3.1.3.12.

Declaration of AI use

Client: Claude, Opus 5.5

Week 3 HW: Cell-Free Protocols

Section A

A1. Why K-glutamate and Mg-glutamate concentrations differ — and which is more sensitive

• Potassium Glutamate (K-Glu, ~100–200 mM) — Bulk Counterion: Serves as the primary salt to set the ionic strength and osmolarity to mimic the cellular environment (E. coli cytoplasm) and stabilize ribosome/tRNA structures. It has low sensitivity because its optimal range is broad and forgiving (typically ±30–50 mM).

• Magnesium Glutamate (Mg-Glu, ~4–20 mM) — Catalytic Cofactor: Acts as a vital, highly active catalytic cofactor needed to coordinate NTPs, bridge the ribosomal subunits, and drive transcription and translation. It has extremely high sensitivity with a very narrow and sharp optimal range (typically ±2 mM).

Why Magnesium is More Sensitive

1. The Free vs. Bound Balance: Because almost everything in the reaction (NTPs, energy sources like PEP, nucleic acids) tightly binds (chelates) Mg²⁺, the actual available (“free”) magnesium is highly dynamic. Changing these inputs shifts the free Mg²⁺ without touching the magnesium level itself.

2. The Molecular Compromise: Coupled reactions must balance transcription (which thrives on higher Mg²⁺) and translation (which requires lower Mg²⁺ to avoid misfolding and miscoding).

3. Dynamic Depletion: As energy is consumed, accumulating inorganic phosphate binds and precipitates Mg²⁺, causing the active pool to drop over the course of the reaction.

The Role of Glutamate Used as the natural dominant cytoplasmic anion (Jewett & Swartz, 2004), glutamate doesn’t just provide osmotic support—it actively fuels the lysate’s metabolism (specifically the TCA cycle) to extend reaction lifetime and boost yields.

A2. PEP + NTP vs. Ribose (or maltodextrin) + NMP

PEP + NTP (Phosphoenolpyruvate + Nucleoside Triphosphates)

This formulation acts as a stoichiometric, high-energy system that supplies nucleotides pre-charged and uses PEP as a direct phosphate donor.

• Mechanism: Relies on direct substrate-level phosphorylation (e.g., via endogenous pyruvate kinase transferring phosphate from PEP to ADP).

• Kinetics: Initiates rapidly with zero lag time, yielding a very high initial rate of protein synthesis.

• Longevity: Reactions are short-lived (typically 1–2 hours). Each PEP molecule donates one phosphate and is consumed, leading to a rapid accumulation of inorganic phosphate. This phosphate tightly chelates free Mg²⁺, inhibiting the ribosome and halting translation.

• pH Stability: Prone to rapid acidic pH drift as PEP is converted into pyruvate and subsequently into lactate or acetate.

• Lysate Dependency: Highly robust and protein-agnostic. It functions even in highly clarified or structurally depleted lysates because it does not rely on complex membrane-bound metabolic pathways.

• Cost: Prohibitively expensive at scale due to the high cost of both the PEP energy substrate and the purified triphosphates.

Ribose (or Maltodextrin) + NMP (Nucleoside Monophosphates)

This formulation acts as a catalytic system, feeding a cheap carbon source into the lysate’s surviving metabolic network to build and charge nucleotides dynamically.

• Mechanism: Feeds a carbon substrate into endogenous glycolysis, the TCA cycle, and oxidative phosphorylation. Endogenous kinases phosphorylate the cheap NMPs into functional NTPs in situ.

• Kinetics: Exhibits a distinct initiation lag phase while the enzymatic pathways spin up and the nucleotide pool charges.

• Longevity: Reactions can run for many hours. Because the system recycles phosphate rather than dumping it, free Mg²⁺ levels remain stable and active for much longer.

• pH Stability: Maintains pH much better over time, especially if a complex polymer like maltodextrin is used to throttle the glycolytic flux and prevent the rapid accumulation of organic acids.

• Lysate Dependency: Highly sensitive to the lysate preparation method. Because it relies heavily on oxidative phosphorylation, the lysate must retain intact, inverted inner membrane vesicles (IMVs). Aggressive centrifugation (e.g., S120 extracts) will strip these out and break the system.

• Cost: Drastically reduces cost per gram of protein. NMPs cost an order of magnitude less than NTPs, and sugars cost pennies, making this the standard approach for volumetric scale-up.

A3. Comparing Two Cell-Free Reaction Compositions

Compositions selected: PANOx-SP (Jewett & Swartz, 2004) and RF_opt (Olsen et al., 2026).

These two represent the endpoints of a 22-year arc in CFPS formulation design: a high-performance, chemically-supplied system versus a cost-optimized system that offloads work onto the lysate’s endogenous metabolism. Critically, Olsen et al. re-measured PANOx-SP in their own laboratory (Table 1), so the comparison below is head-to-head under a single lysate batch and a single assay, rather than a cross-paper comparison of reported yields.

Side-by-side composition

All values in mM unless noted; folinic acid and tRNA in mg/mL. Source: Olsen et al. (2026), Table 1.

PANOx-SP contains more than 20 components. RF_opt contains exactly 12.

Contrast 1: Energy system – stoichiometric vs. catalytic

PANOx-SP uses 30 mM PEP, a high-energy phosphate donor consumed stoichiometrically: each molecule donates one phosphate to ADP via endogenous pyruvate kinase and is spent. RF_opt replaces it with 50 mM ribose + 10 mM glucose + 15 mM inorganic phosphate, feeding central metabolism so that ATP is regenerated catalytically with phosphate recycled rather than accumulated.

Olsen et al. measured the consequence directly (Fig. 2a,b):

• PANOx-SP produces an ATP surge within 30 min, followed by a collapse to near zero by the end of the reaction, with phosphate rising steeply as PEP is consumed.
• RF_opt reaches comparable ATP within 30 min despite containing no exogenous ATP at all, then sustains ~250 uM for the full 20 h – well above the ATP affinity threshold for cell-free gene expression.
• RF_opt phosphate holds at 10–15 mM, three times lower than PANOx-SP, explicitly “because free phosphate is not liberated from PEP.”

This is the mechanistic basis for the longer productive lifetime: phosphate accumulation chelates free Mg²⁺ and eventually precipitates as magnesium-phosphate, which is a principal reason PEP-based batch reactions stall.

Contrast 2: Nucleotides – triphosphate vs. monophosphate

PANOx-SP supplies all four nucleotides pre-charged as triphosphates (3.75 mM total). RF_opt supplies only monophosphates (9.45 mM total), relying on endogenous NMP and NDP kinases to phosphorylate them in situ.

Note the direction of travel: total nucleotide concentration increases 2.5x while cost falls. Olsen et al. cap NMPs at 5 mM each to prevent formation of insoluble precipitates, and still identify NMPs as the single most expensive reagent group remaining in the minimal system ($62/g protein).

Contrast 3: Cofactors – supplying the product vs. supplying the precursor

PANOx-SP supplies NAD (0.4 mM) and CoA (0.27 mM) directly; both are among the most expensive reagents by volume. RF_opt eliminates both and instead supplies 4 mM nicotinamide, the cheap biosynthetic precursor, allowing the lysate’s own salvage pathway to generate NAD. This is the most economically elegant substitution in the formulation: pay for the precursor, let the extract do the chemistry.

Contrast 4: Deletion of legacy additives

RF_opt removes tRNA, folinic acid, spermidine, putrescine, oxalic acid, and ammonium glutamate with no yield penalty. Two of these deletions are mechanistically forced rather than empirical:

• Oxalate exists solely to inhibit PEP synthetase and prevent a futile ATP-consuming cycle. With no PEP in the formulation, it has no function.
• The polyamines (spermidine, putrescine) were originally introduced in PANOx-SP to replace PEG-8000 as a macromolecular crowding/stabilizing agent.

The remainder were identified empirically across a campaign of 1,231 formulations tested in triplicate.

The central finding: cheaper is not the same as less

RF_opt increases four concentrations and still costs 68x less per gram. Amino acids 2.5x up, potassium 2.8x up, total nucleotide 2.5x up, buffer 1.3x up. The savings derive entirely from substituting inexpensive molecules for expensive ones, not from reducing concentrations.

The 68x improvement also decomposes cleanly:

68x  =  32x (cheaper reagents per litre)  x  2.2x (higher protein yield)

Because cost per gram is a ratio, both the numerator and the denominator are legitimate optimization targets – and Olsen et al. attacked both. Any formulation design that reduces reagent cost without modelling the yield consequence is optimizing only half of the objective function.

Limitations reported by the authors

RF_opt is not uniformly superior, and the paper is explicit about where it is not:

1. Magnesium and phosphate had to be re-optimized for every new lysate batch (Supplementary Fig. 9) – consistent with free Mg2+, not total Mg2+, being the controlled variable.
2. RF_opt is more sensitive to lysate loading than PANOx-SP (Fig. 3b), which the authors attribute to the fine metabolic balance of a system that depends on endogenous metabolism rather than supplied high-energy phosphate.
3. RF_opt is more sensitive to surface-area-to-volume ratio, because oxidative phosphorylation requires oxygen. Supplying a 100% O2 feed in a 4-mL membrane bioreactor increased yield by 45% to 3.7 +/- 0.2 g/L ($39/g protein).
4. RF_opt underperformed PANOx-SP on complex therapeutic proteins – protein D, CRM197, and trastuzumab – and neither formulation produced the trastuzumab Fc dimer, because disulfide bond formation requires an oxidizing system (iodoacetamide-pretreated lysate plus a glutathione redox buffer).

Limitation 4 is directly relevant to formulation design for Reteplase: a composition optimized against sfGFP does not transfer to a multi-disulfide therapeutic protein, and the authors demonstrate this on a protein class far closer to Reteplase than to sfGFP.

A note on formulation drift

The composition circulating today as “PANOx-SP” is not the one published in 2004. Jewett & Swartz (2004), Table I, specifies PANOx-SP as 175 mM K-glutamate, 20 mM Mg-glutamate (plus 3.3 mM Mg-acetate carried over from the extract), 10 mM NH4-glutamate, 33 mM PEP, 1.5 mM spermidine, 1 mM putrescine, 2.7 mM sodium oxalate – with no HEPES and no PEG, both of which were present in PANOx and deliberately removed.

The modern “PANOx-SP” tabulated by Olsen et al. uses 130 mM K-glutamate, 8 mM Mg-glutamate, 57 mM HEPES, 30 mM PEP, and 4 mM oxalate. The 130/8 ionic condition is in fact inherited from the 2004 Cytomim column, not from the 2004 PANOx-SP column.

This drift is itself informative: the two ions most frequently re-tuned between publications are exactly potassium and magnesium, which is what one expects if free Mg2+ depends on the rest of the formulation and must be re-titrated whenever that formulation changes. When citing PANOx-SP, state which version is being used.

References

• 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. DOI: 10.1002/bit.20026. [Table I, p. 20]
• 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, 17:3478. DOI: 10.1038/s41467-026-69605-8. [Table 1, p. 2; Figs. 2–4]
• 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. DOI: 10.1038/s41576-019-0186-3. [Context for the energy-system landscape; contains no composition table.]

Section B

Reaction Designs & Community Reagent

Design philosophy

The objective is $ per gram of functional protein. That is a ratio, so it has two levers: reagent cost (numerator) and functional yield (denominator). Olsen et al. (2026) improved it 68-fold over PANOx-SP, and that decomposes as 32x cheaper reagents x 2.2x higher yield — both halves were attacked.

Critically, the correct balance between those levers is different for each of the three proteins, because a formulation that yields zero active protein has infinite cost per gram regardless of reagent price. My three designs therefore differ substantially from one another rather than being one mixture with minor variations:

1. sfGFP — Cost-floor design

https://coopetition.htgaa.org/cfps?design=50cv839371xn9p5

Backbone: RF_opt (Olsen et al. 2026), measured at 2.39 +/- 0.28 g/L, $60/g protein in BL21 Star (DE3) at 30 degC — the same strain and temperature used in this project.

Adjustments from the PANOx-SP baseline, and why:

• Energy: PEP removed, ribose + glucose + phosphate added. PEP is a stoichiometric donor; each molecule gives one phosphate and is spent, while liberated P_i accumulates, chelates free Mg2+ and eventually precipitates. Olsen measured the difference: PANOx-SP ATP surges then collapses to near zero, whereas RF_opt holds ~250 uM ATP for 20 h with phosphate steady at 10-15 mM — three-fold lower — “because free phosphate is not liberated from PEP.”
• Nucleotides: all four NTPs replaced by NMPs. Total nucleotide rises from 3.75 to 9.45 mM while cost falls, because monophosphates are far cheaper and the lysate’s NMP/NDP kinases charge them in situ. Ribose is the mechanistic partner here: it feeds the pentose phosphate pathway, which supplies the ribose moiety for nucleotide salvage.
• Cofactors: NAD and CoA both removed; 4 mM nicotinamide added. Buying the biosynthetic precursor rather than the product, and letting the lysate’s salvage pathway do the chemistry.
• K-glutamate raised 130 -> 362 mM. Potassium glutamate is among the cheapest reagents available and this was simply free yield left on the table in classical formulations.
• Amino acids raised 2 -> 5 mM. Deliberately against the instinct to cut. Amino acids set the yield ceiling; starving them lowers the denominator of $/g and therefore raises cost per gram.
• Deleted with no yield penalty (per Olsen’s 1,231-formulation campaign): tRNA, folinic acid, spermidine, putrescine, NH4-glutamate, cAMP. Oxalate’s deletion is mechanistically forced — it exists only to inhibit PEP synthetase, and there is no PEP.

Screening priorities for round 1: (i) Mg x K grid, not a 1-D Mg scan — the two ions compete for the same condensed-counterion sites on RNA, so the optimum runs on a diagonal and a 1-D scan finds the best point on its own line rather than the peak; (ii) lysate fraction, the largest single cost line; (iii) total NMP, which Olsen identifies as the most expensive group remaining at $62/g.

2. PETase (TfCut2) — Cost-led, folding-aware

https://coopetition.htgaa.org/cfps?design=91m7bo8mt3h3247

Key decision: standard lysate, no redox buffer, no DsbC.

This is a deliberate commitment rather than a hedge. The assignment brief states that the reducing environment “is appropriate for sfGFP and remains compatible with TfCut2,” and TfCut2’s single disulfide is structurally stabilizing rather than catalytically essential (the catalytic triad is Ser170-Asp216-His248). Adding a glutathione redox buffer to an untreated lysate is worse than useless: the endogenous thioredoxin and glutaredoxin systems, plus glutathione reductase, collapse the GSSG pool within minutes, so the buffer is paid for and does nothing. Either commit to the IAM-treated, delta-gor DSB lysate or commit to reducing conditions. For a one-disulfide enzyme, reducing is the right and cheaper bet.

Adjustments relative to the sfGFP design:

• K-glutamate lowered 362 -> 300 mM. 362 mM was optimized against sfGFP; a hydrolase with a folding requirement is given slightly more conservative ionic strength pending its own titration.
• Amino acids held at 5 mM, not reduced. PETase is more sensitive to amino acid starvation than sfGFP.
• Betaine 100 mM added as a cheap chemical chaperone / compatible solute, suppressing aggregation without the cost of DsbC (~$5,000/mL). Flagged as a hypothesis: betaine and the already-high K-glutamate may be redundant rather than additive, which the screen will reveal.

3. Reteplase — Yield-first design

https://coopetition.htgaa.org/cfps?design=v0u9hlu89fn21b8

This design deliberately does not minimize reagent cost, for two reasons.

(a) The denominator dominates. Reteplase requires nine disulfide bonds in correct pairings. If the formulation yields no active protein, cost per gram is infinite and no reagent saving can recover it. Olsen et al. report that RF_opt underperformed PANOx-SP on complex therapeutic proteins (protein D, CRM197, trastuzumab), and that neither produced the trastuzumab Fc dimer because oxidizing conditions are required. A formulation optimized against sfGFP does not transfer to a multi-disulfide therapeutic — demonstrated on a protein class far closer to Reteplase than to sfGFP. Cost reduction is deferred to round 2, after an activity signal is secured.

(b) Mechanistic conflict between IAM treatment and glycolytic energy. This is the central design argument. The DSB lysate is pretreated with iodoacetamide to irreversibly cap free thiols and deactivate reductases. But IAM is not selective for reductases — it alkylates accessible cysteines generally, and GAPDH is a catalytic-cysteine enzyme at the centre of glycolysis. A low-cost formulation such as RF_opt depends entirely on endogenous glycolytic and pentose-phosphate flux to regenerate ATP from ribose and glucose. Pairing an IAM-treated lysate with a metabolism-dependent energy system therefore risks disabling the exact pathway the energy system relies on. PEP bypasses this: ATP regeneration proceeds by direct phosphate transfer via pyruvate kinase, independent of the glycolytic steps IAM may have compromised.

I have not found this interaction addressed in the formulations surveyed, and I would propose it as a specific, testable hypothesis for the screen: run the DSB lysate against PEP, against ribose/glucose, and against a 50:50 blend. If the glycolytic arm underperforms specifically in DSB lysate and not in standard lysate, the interaction is real — and it constrains how cheap any oxidizing CFPS formulation can become.

Other adjustments:

• GSSG:GSH = 4:1 oxidized-to-reduced, matching the Zawada et al. formulation as tabulated in Olsen et al. (2026), Table 1. A 4:1 ratio in the reduced direction would poise the reaction reducing and prevent the disulfide formation the design exists to achieve.
• DsbC retained at 2 uM despite cost. With nine bonds, the limiting problem is mis-pairing, not failure to oxidize — an isomerase, not just an oxidant, is required. 2 uM in 20 uL = 40 pmol = 400 nL of a 100 uM stock, which is 16 Echo droplets at 25 nL resolution.
• L-arginine 100 mM as an aggregation suppressant. Reteplase is produced industrially as inclusion bodies and refolded in a separate chemical step; arginine is the standard additive in that refolding chemistry and is inexpensive. Applying refolding-buffer chemistry inside the synthesis reaction is, as far as I can find, not standard practice in CFPS.
• K-glutamate at 180 mM, not 362. The high-potassium optimum was found against sfGFP in a PEP-free system; with PEP restored, the chelation balance and therefore the free Mg2+ / K+ optimum are different and must be re-found.

Section 3B(ii) — Community Reagent contribution

Primary proposal: D-pantothenate (vitamin B5). Secondary proposal: riboflavin (vitamin B2).

How I arrived at these

The single most economically elegant move in Olsen et al. (2026) is the replacement of NAD (~$33/mL) with 4 mM nicotinamide, its cheap biosynthetic precursor, with no yield penalty — the lysate’s own salvage enzymes perform the conversion. That logic was applied exactly once and then abandoned.

Pantothenate extends it to the most expensive small molecule in the inventory. Coenzyme A costs ~$104/mL — roughly three times NAD. E. coli retains the complete five-step CoA biosynthesis pathway (CoaA, CoaBC, CoaD, CoaE) from pantothenate, consuming ATP, CTP and cysteine, all of which are present in the reaction. The pathway is slower than simply adding CoA, but cell-free reactions run 20 h. I searched for prior application of this substitution in CFPS and found none.

Two distinct claims, which should be tested separately:

1. Versus a CoA-containing formulation (e.g. PANOx-SP, 0.26 mM CoA), pantothenate is a direct ~$104/mL saving.
2. Versus RF_opt, which contains no CoA at all, the question is different: can cheap restoration of CoA capacity push yield above 2.39 g/L? Also worth knowing, but it is an addition, not a substitution.

Riboflavin addresses a limitation the authors identified but did not act on. Olsen et al. show that RF_opt’s yield is oxygen-limited: a 100% O2 feed in a 4-mL bioreactor raised sfGFP by 45%, to 3.7 +/- 0.2 g/L ($39/g), and they attribute this to oxidative phosphorylation being active and required in crude lysates. Oxidative phosphorylation runs on the respiratory chain retained in inverted membrane vesicles, and that chain is built from flavins (FMN/FAD), quinones and heme. Reviewing the eleven formulations in Olsen’s Table 1, none supplies any respiratory-chain cofactor. Every published formulation optimizes the translation reaction and treats the lysate’s energy metabolism as a black box, while the authors’ own oxygen data shows that black box is rate-limiting.

Riboflavin is converted to FMN and then FAD by RibF, which is retained in lysate, and costs essentially nothing. Thiamine (-> TPP, required by transketolase at the centre of the pentose phosphate pathway that RF_opt’s 50 mM ribose enters, and by pyruvate dehydrogenase) and 5-aminolevulinic acid (-> heme, for the cytochrome bo3 / bd terminal oxidases) follow the same logic and would be my next two proposals if slots allow.

Supporting research

• 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 17:3478. https://doi.org/10.1038/s41467-026-69605-8 — Table 1 (eleven formulations, costs, yields); Fig. 2d,e (oxygen dependence); the nicotinamide-for-NAD substitution that motivates the precursor logic.
• Jewett, M. C. & Swartz, J. R. (2004). Mimicking the E. 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

Final Project