Week 1 HW: Final Project Ideas

Photo by atwena Goodman on Unsplash

Homework 1: An elephant-memory-inspired neuroprotective biologic for dementia

What do I want to make, and why?

Elephants show exceptional long-term social and spatial memory. That’s worth investigating for people affected by Alzheimer’s disease and other dementias, which are devastating for both patients and the families who watch a loved one disappear in front of them. Roughly 55 million people currently live with dementia worldwide, a number projected to almost double to 78 million by 2030 and reach 139 million by 2050 as the world’s population ages, with the majority of cases already occurring in low- and middle-income countries (Alzheimer’s Disease International, 2015).

There is no single “elephant memory chemical” — and it also isn’t simply that elephants have an unusually large hippocampus. The actual MRI and histology study of the African elephant hippocampus (Patzke et al., 2013) found the opposite of the popular claim: the elephant hippocampus is “not unduly enlarged, nor specifically unusual in its internal morphology,” and at 10.84 cm³ is only slightly larger than the human hippocampus (10.23 cm³) despite the elephant’s much larger overall brain. What is distinctive is structural, not size-based: the dentate gyrus has extra sublayers not seen in other mammals, and — critically for this idea — doublecortin staining confirmed ongoing adult neurogenesis in the elephant hippocampus, the continued birth of new neurons into adulthood.

So the target here isn’t “elephant memory juice”, and it isn’t literally “a bigger hippocampus” either. It’s a recombinant peptide that stabilises or mimics the activity of known long-term-memory-maintenance molecules — most plausibly one based on protein kinase M-zeta (PKMζ), shown by Serrano et al. (2008) to be specifically required for maintaining, not just forming, spatial and other long-term memories, and its scaffolding partner KIBRA. Neither is elephant-specific; the elephant’s confirmed adult hippocampal neurogenesis is the loose thread that motivates looking at memory-maintenance and neurogenesis pathways at all, not proof that PKMζ/KIBRA are what makes elephants special.

Diagram 1: From elephant biology to a drug candidate — elephant hippocampus → memory-maintenance pathway (PKMζ/KIBRA) → candidate peptide design.

What living system would I use?

Cell-free protein synthesis (CFPS) — cell lysate loaded with the target gene, ribosomes, and an energy mix — is ideal for rapidly iterating on peptide variants and fits directly with a cell-free biomanufacturing unit. Scale-up would then move to mammalian (CHO) cell culture, since the peptide needs correct folding and post-translational modification to work as a CNS-active biologic. This isn’t a novel category of product to manufacture — recombinant protein drugs made via genetically modified mammalian cell culture (erythropoietin, insulin, growth hormone) are precisely what Zhang, Sun and Ma (2017) call “Biomanufacturing 3.0,” the still-dominant paradigm for protein-drug production since the 1980s. My peptide would sit in that same established category; the open question is the specific target (PKMζ/KIBRA), not the manufacturing platform.

Worth being explicit about what “living system” means here: this is not extraction. I’m not proposing to harvest anything from an actual elephant — that wouldn’t scale and isn’t what biomanufacturing means. The living system (CFPS lysate, then CHO cells) is engineered to produce the peptide itself, the same way erythropoietin or insulin are made today: by giving a cell the gene and letting its own biology manufacture the product, not by extracting a pre-existing substance from a donor organism.

A creative addition on delivery. The single biggest scientific risk in this idea (see “what don’t I know yet,” below) is getting the peptide into the brain at all, since most peptides don’t cross the blood–brain barrier (BBB). Rather than leave that as a pure unknown, I’d propose intranasal delivery as the administration route: spraying the peptide into the nose lets it travel along the olfactory and trigeminal nerves directly into the brain, physically bypassing the BBB rather than trying to cross it. This isn’t a hand-wave — intranasal insulin has already been trialled for memory improvement in Alzheimer’s patients, and a 2025 study specifically delivered a short catalytic peptide (JAL-TA9) nose-to-brain for Alzheimer’s treatment in rats and mice (Hanson and Frey, 2008; Hatakawa et al., 2025). It’s still unproven for this specific peptide, but it converts “we don’t know how this would ever reach the brain” into “here is a plausible, precedented route to test” — which is a stronger place to leave an open question than a dead end.

How would I use automation?

CFPS screening on a Cloud Lab or desktop liquid handler — testing many peptide variants in parallel, reading out activity/stability — to find the best-performing construct. Production then scales up in a stirred-tank bioreactor. This mirrors what Petzold and Mukhopadhyay (2025) describe as the central role of automation in current biomanufacturing R&D: cloud labs and biofoundries “reshaping the landscape of biomanufacturing research by introducing unprecedented levels of automation, reproducibility, and accessibility,” allowing continuous, round-the-clock experimentation without a scientist physically present, and directly enabling the throughput that the design–build–test–learn (DBTL) cycle needs to search a large space of peptide variants efficiently.

Concretely, this is the same workflow the course itself is about to run in Week 4’s first lab: each student gets assigned a set of wells on a 384-well plate, designs their own CFPS reaction, and the plate is executed by the automation system at MIT or at Ginkgo Bioworks (whose autonomous lab system, Nebula, was highlighted in lecture as already running cell-free protein synthesis reactions at scale). My peptide-screening step is that exact setup with a different readout: instead of optimising for the brightness of a fluorescent reporter protein (this course’s Lab 1 exercise), each well would run a variant of the PKMζ/KIBRA-based peptide and be screened with a binding- or stability-activity assay instead of a fluorescence readout — same plate-based automation, different signal.

What automation can’t do here. Screening and manufacturing are automatable; getting a drug approved is not. Automation genuinely covers the CFPS variant-screening stage and the bioreactor production stage once a construct is chosen. It does not cover, and cannot cover, the preclinical safety testing, human clinical trials, or regulatory review that sit between those two stages — those require human patients, ethics oversight, and years of monitoring by definition, not more robots. That gap is the subject of the next section.

Diagram 2: Biomanufacturing and scale-up pipeline — CFPS screening (Cloud Lab) → CHO cell bioreactor scale-up → global patient dosing.

How much would I need to make, and how long would it take?

How much? Biologics for neurological conditions are typically dosed monthly at roughly 100–250 mg. Assuming 200 mg/dose × 12 doses/year = 2.4 g/patient/yr:

• A 1% initial global rollout (550,000 of the ~55 million people with dementia worldwide) needs 550,000 × 2.4 g ≈ 1.32 tonnes/yr.
• Full global coverage would need 55,000,000 × 2.4 g ≈ 132 tonnes/yr.

How long? Typical CFPS yields are around 1 g/L per batch (Karim and Jewett, 2021). At that titre, 132 tonnes/yr requires roughly 132 million litres of reaction volume per year — clearly beyond CFPS alone, so screening stays in CFPS while production moves to bioreactors. But the jump from lab-scale to an industrial bioreactor is, by the field’s own account, the hard part of biomanufacturing, not a detail to wave past: Asin-Garcia et al. (2025) quote Paul Freemont (Global Alliance of Biofoundries, Imperial College London) that “the big technology barrier [to commercialisation] is how you move from a laboratory-based strain that produces a product towards a 3000 L bioreactor producing the same product at high titre… that’s always been the challenge in industrial biotechnology.” Asin-Garcia et al. also describe this gap as a “Valley of Death” that many promising biomanufacturing projects don’t survive, because few organisations have access to pilot-scale plants to test the transition. Existing published biomanufacturing infrastructure (e.g. VTT, Finland) currently tops out around 1,200 L capability — well short of even a single tonne/yr at CFPS-equivalent titres, let alone the 132 tonnes/yr full-coverage figure above. So this idea’s real bottleneck isn’t chemistry, it’s exactly the scale-up gap the literature itself flags as unsolved industry-wide.

What it would take to get there at all — cost and timeline before “how much/how long” even applies. The 132 tonnes/yr figure above answers “how much would I need to manufacture,” but it assumes an approved, working drug already exists — it says nothing about getting there. That’s a separate, much larger number. Using the widely cited Tufts Center for the Study of Drug Development estimate (DiMasi, Grabowski and Hansen, 2016), taking one biologic from discovery to approval costs on the order of $1–2.6 billion and takes 10–15 years (more recent analyses, e.g. RAND, 2025, put the median lower at roughly $700 million, but multi-hundred-million-to-billion-dollar costs and decade-plus timelines are consistent across every estimate I found). None of that cost or time is automatable: it is preclinical safety testing, Phase 1–3 human trials, and regulatory review, not manufacturing throughput. So the honest shape of this project is: years 1–2 automatable CFPS/Cloud Lab screening (relatively cheap, fast), years 2–12+ non-automatable clinical development (expensive, slow, and not guaranteed to succeed at all), then — only after approval — the 132 tonnes/yr industrial bioreactor scale-up this section opened with. Cost per dose would very likely fall over time as production scales (the standard economics of fixed R&D cost amortised over more units), but the initial capital and time commitment is genuinely on the order of a billion dollars and a decade, not a Cloud Lab afternoon.

Diagram 3: Development timeline and cost — CFPS screening (1–2 yrs, automatable) vs. preclinical/clinical trials (10–15 yrs, not automatable, $1–2.6 billion — the true bottleneck) vs. approved production (ongoing, ~132 tonnes/yr).

A real-world price comparison, and why it matters. Leqembi (lecanemab), an FDA-approved Alzheimer’s antibody made in CHO cells — the same manufacturing category this idea sits in — launched in January 2023 at $26,500 per patient per year (Stanton, 2023). If my peptide followed similar pricing logic once approved, it would be unaffordable for most of the roughly 55 million people who need it, given that (per the Alzheimer’s Disease International figures cited above) the majority of dementia cases are already in low- and middle-income countries. A biologic that only the wealthiest patients can access doesn’t actually solve the problem this idea sets out to address — that tension between “can be manufactured at global scale” and “can be afforded at global scale” is a real gap in this proposal, not a footnote.

On elephant genetic material specifically — availability, not extraction, and no harm to elephants. As written above, this idea doesn’t actually use any elephant-derived gene at all — PKMζ and KIBRA are proteins studied in rodents (Serrano et al., 2008), not sequenced from an elephant. But it’s worth answering directly, since the question matters for the idea’s honesty: if a future version of this project wanted the actual elephant version of a memory-related gene, would that require extracting anything from a living or dead elephant, and would elephants be harmed? No, on both counts. The African elephant (Loxodonta africana) genome has been fully sequenced and publicly available since 2009 (the Broad Institute’s loxAfr3 assembly), with a newer, higher-quality chromosome-level assembly released in 2023 (NCBI GenBank accession GCA_030020305.1) — both freely searchable through NCBI and Ensembl by anyone, including a student, at no cost. Getting an elephant gene sequence today means looking it up in a public database and ordering a synthetic copy of that DNA sequence from a commercial gene-synthesis company (a routine, few-days-to-weeks service), then cloning it into the CFPS/CHO system already described above — not touching an elephant at all. The original blood or tissue samples used to build that reference genome were collected once, years ago, under standard veterinary sampling protocols (not lethal, not ongoing); every subsequent use of the sequence — including this one — is purely digital and synthetic. So the honest answer is: the elephant genome is already available, no extraction is needed for this project, and no elephant would be harmed by it.

What don’t I know yet?

1. Blood–brain barrier crossing. Whether a PKMζ/KIBRA-based peptide can actually reach the brain — intranasal delivery (above) is a plausible, precedented route to test, but it’s unproven for this specific peptide and would need to be validated experimentally, not assumed.
2. Human efficacy. Whether the peptide is neuroprotective or pro-memory in humans at all — all direct evidence for PKMζ/KIBRA’s role in memory maintenance comes from rodent models.
3. Yield at scale. Realistic CFPS/bioreactor yield for this specific construct hasn’t been benchmarked, and — per Asin-Garcia et al. (2025) above — the lab-to-3,000 L-bioreactor transition is an unsolved, field-wide bottleneck, not something I can assume away.
4. Affordability at the scale that matters. Even a perfectly manufacturable, safe, effective peptide doesn’t solve global dementia if it’s priced like existing biologics ($26,500/yr for Leqembi) while most patients needing it are in low- and middle-income countries. I don’t yet know what pricing or production model (tiered pricing, generic licensing after patent expiry, regional manufacturing) would actually close that gap, and it’s a real open problem, not a detail I can wave past with a bigger bioreactor.

References

Alzheimer’s Disease International (2015) World Alzheimer Report 2015: The Global Impact of Dementia — An Analysis of Prevalence, Incidence, Cost and Trends. London: Alzheimer’s Disease International. Available at: https://www.alzint.org/u/WorldAlzheimerReport2015-sheet.pdf

Asin-Garcia, E., Fawcett, J.D., Batianis, C. and Martins dos Santos, V.A.P. (2025) ‘A snapshot of biomanufacturing and the need for enabling research infrastructure’, Trends in Biotechnology, 43(5), pp. 1000–1014. Available at: https://doi.org/10.1016/j.tibtech.2024.10.014

DiMasi, J.A., Grabowski, H.G. and Hansen, R.W. (2016) ‘Innovation in the pharmaceutical industry: new estimates of R&D costs’, Journal of Health Economics, 47, pp. 20–33. (Tufts CSDD analysis, widely cited as the source of the ~$2.6 billion / 10–15 year biologic development estimate used above.)

Hanson, L.R. and Frey, W.H. (2008) ‘Intranasal delivery bypasses the blood-brain barrier to target therapeutic agents to the central nervous system and treat neurodegenerative disease’, BMC Neuroscience, 9(Suppl 3), S5. Available at: https://doi.org/10.1186/1471-2202-9-S3-S5

Hatakawa, Y., Tanaka, A., Furubayashi, T., Katsumi, H., Nakamura, R., Konishi, M., Akizawa, T. and Sakane, T. (2025) ‘Efficient nose-to-brain delivery of nine residues peptide (JAL-TA9) exhibiting hydrolytic activity against amyloid-β’, European Journal of Pharmaceutics and Biopharmaceutics, 208, article 114661. Available at: https://doi.org/10.1016/j.ejpb.2025.114661

Karim, A.S. and Jewett, M.C. (2021) ‘Toward sustainable, cell-free biomanufacturing’, Current Opinion in Biotechnology, 69, pp. 136–144. Available at: https://doi.org/10.1016/j.copbio.2020.12.012

National Center for Biotechnology Information (2023) Loxodonta africana genome assembly mLoxAfr1.hap1, GenBank accession GCA_030020305.1. Available at: https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_030020305.1/ (cited for the public availability of the African elephant reference genome; the original loxAfr3 assembly was released by the Broad Institute in 2009.)

Patzke, N. et al. (2013) ‘Organization and chemical neuroanatomy of the African elephant (Loxodonta africana) hippocampus’, Brain Structure and Function. Available at: https://doi.org/10.1007/s00429-013-0587-6 (full author list not confirmed from the accessible abstract — verify before formal submission.)

Petzold, C.J. and Mukhopadhyay, A. (2025) ‘From bench to biofactory: high-throughput technologies and automated workflows to accelerate biomanufacturing’, Current Opinion in Biotechnology, 94, article 103320. Available at: https://doi.org/10.1016/j.copbio.2025.103320

Serrano, P., Friedman, E.L., Kenney, J., Taubenfeld, S.M., Zimmerman, J.M., Hanna, J., Alberini, C., Kelley, A.E., Maren, S., Rudy, J.W., Yin, J.C.P., Sacktor, T.C. and Fenton, A.A. (2008) ‘PKMζ maintains spatial, instrumental, and classically conditioned long-term memories’, PLoS Biology, 6(12), pp. 2698–2706. Available at: https://doi.org/10.1371/journal.pbio.0060318

Stanton, D. (2023) ‘Eisai and Biogen: FDA approve Alzheimer’s mAb lecanemab’, BioProcess International, 9 January. Available at: https://bioprocessintl.com/regulations/eisai-and-biogen-fda-approve-alzheimer-s-mab-lecanemab (cited for the $26,500/yr Leqembi launch price used above.)

Zhang, Y.-H.P., Sun, J. and Ma, Y. (2017) ‘Biomanufacturing: history and perspective’, Journal of Industrial Microbiology & Biotechnology, 44, pp. 773–784. Available at: https://doi.org/10.1007/s10295-016-1863-2

AI use disclosure

AI client: Claude (Anthropic), used via the Claude.ai chat interface with web search and this project’s uploaded course-reading PDFs.

Prompts used (in sequence, condensed from a longer conversation):

1. An initial request to develop seven-plus biomimicry-inspired biomanufacturing ideas (animal/plant “superpowers” — elephant memory, tortoise longevity, cactus water retention, ant strength, drought-resistant food, spider silk, electric eel/firefly energy) against the assignment’s four questions, in table/comparison format, with real, verifiable Harvard-style citations only — explicitly instructed not to hallucinate sources.
2. A request to add an eighth idea, inspired by the Qur’anic account of Prophet Yusuf’s shirt restoring Ya’qub’s eyesight, and to research whether the “sweat/urea” hypothesis behind it has any real scientific backing.
3. A request to reframe all scale calculations from South Africa-specific to global/worldwide figures.
4. “I am going with the elephant/Alzheimer’s/dementia for my submission. Can it be improved with diagrams or anything else that will enhance its appeal?” — resulting in the two accompanying schematic diagrams and this standalone write-up.
5. “Does it make use of the course readings or anything else that substantiates its claim?” — resulting in a check of all seven assigned readings against this specific idea, and integration of real quoted content from three that apply (Zhang, Sun and Ma, 2017; Petzold and Mukhopadhyay, 2025; Asin-Garcia et al., 2025), with the three that don’t apply (Averesch et al., 2023; Yap et al., 2023; Ye et al., 2023) explicitly named and excluded rather than cited for padding.
6. “Does it take note of everything mentioned in the recitation that pertains to this submission?” — resulting in the addition of the Week 4 Lab 1 / 384-well-plate / Ginkgo Bioworks Nebula detail to the automation section, and this fuller prompt disclosure.
7. Uploaded a human-vs-elephant brain comparison image from safari-center.com and asked whether it’s backed by the course readings, and for images to insert since I can’t generate or find them myself — resulting in a correction (the actual Patzke et al., 2013 study found the elephant hippocampus is not unusually large, contradicting the popular “big brain” graphic), two new citations added (Patzke et al., 2013; Serrano et al., 2008), a regenerated diagram with accurate text, and image-sourcing guidance.
8. Asked whether biomanufacturing means only extracting from plants/animals, whether there’s room for creativity in the idea, what’s been overlooked, whether the process can really be fully automated, and to show assumptions/calculations for cost and timeline — resulting in: an explicit clarification that biomanufacturing means engineering a living system to produce the target, not extraction; a proposed intranasal delivery route (Hanson and Frey, 2008) addressing the blood-brain-barrier unknown; an honest split between what automation can and can’t cover; a new cost/timeline estimate for drug development separate from production scale-up (DiMasi, Grabowski and Hansen, 2016); a real-world price comparator (Leqembi, $26,500/yr); and a new affordability/equity unknown.
9. “Are these shown [the diagrams]? Please… include the diagrams/simulations discussed. Is there anything else not included that should be included?” — resulting in the two earlier diagrams and a new third diagram (a development timeline/cost chart visualizing the automation-limits and cost/timeline sections) being generated as actual saved SVG and PNG files rather than just chat widgets, and referenced by filename at the correct points in this document.
10. “Please verify the references and facts stated herein, especially the Petzold citations, and cite inline as well, Harvard style. Also, what would it take to get elephant enzymes at current dates and times—is it available or would it need to be extracted, and would elephants be harmed?” — resulting in: direct verification of the Petzold and Mukhopadhyay (2025) citation against the published ScienceDirect page (confirmed word-for-word accurate); two citation gaps found and fixed (DiMasi, Grabowski and Hansen, 2016 wasn’t named inline near the $2.6B figure; the JAL-TA9 paper was mentioned in text but missing a full reference entry, now added as Hatakawa et al., 2025); and a new paragraph confirming the African elephant genome is publicly available (no extraction needed, no harm to elephants), with a citation for the reference genome assembly.

I reviewed every AI-sourced claim and citation myself before submitting and can explain each one in the homework-review chair.

Week 2 HW: DNA Design

An Elephant-Memory-Inspired Neuroprotective Biologic for Dementia

Shakera Karolia | BioClub Tokyo Node | 29 September 2026

Part 0: Set up in Kernel

All design work for this assignment was completed in the HTGAA 2026 workspace on Asimov Kernel under the repository SHAKERA_KAROLIA, housed within the folder Karolia-project. The assembled construct is named PKMzeta_CFPS_Expression_Cassette. The GenBank export file (PKMzeta_CFPS_Expression_Cassette.gb) accompanies this submission as a separate file.

Figure 0: The SHAKERA_KAROLIA repository in the HTGAA 2026 workspace, showing the Karolia-project folder containing the construct and all associated parts.

Part 1: Expression Target

What molecule is being expressed?

The expression target is the constitutively active catalytic domain of human protein kinase C zeta (PKCζ), the isoform known as PKMζ. PKMζ is a serine/threonine kinase that is specifically required for the maintenance — rather than the initial formation — of spatial and other long-term memories. Serrano et al. (2008) demonstrated that inhibiting PKMζ activity erases established memories in rats, establishing it as the only molecule known to be specifically necessary for long-term memory maintenance rather than for memory acquisition. This project is motivated by the confirmed presence of adult hippocampal neurogenesis in the African elephant (Loxodonta africana), as documented by Patzke et al. (2013), which provides biological precedent for targeting memory-maintenance and neuroprotective pathways. The PKMζ/KIBRA signalling axis was identified in Homework 1 as the candidate therapeutic target for dementia and Alzheimer’s disease.

What is the sequence?

The coding sequence encodes residues 211–592 of human PRKCZ (UniProt accession Q05513), corresponding to the 382-amino-acid C-terminal catalytic domain, plus a TGA stop codon. This fragment lacks the N-terminal regulatory and pseudosubstrate domains present in full-length PKCζ (592 amino acids), rendering it constitutively active — a defining characteristic of PKMζ. The DNA coding sequence is 1,149 bp. An N-terminal 6×His purification tag (MHHHHHH, 21 bp) is encoded by a separate upstream part within the construct. Two codons were corrected during design to match the UniProt reference: position 366 (CTC→TTC, leucine to phenylalanine) and position 522 (GGC→GAC, glycine to aspartate). The sequence was not codon-optimised for E. coli in this iteration.

What host or chassis?

The expression platform is E. coli-based cell-free protein synthesis (CFPS). As proposed in Homework 1, CFPS is well suited to rapid prototyping of protein variants: it requires no cell growth, no transformation, and permits screening in a 384-well plate format using a liquid handler or cloud laboratory (Petzold and Mukhopadhyay, 2025). The lysate is pre-loaded with ribosomes, T7 RNA polymerase, and an energy regeneration system; only the DNA template is supplied externally. Scale-up for clinical production would subsequently transition to CHO (Chinese hamster ovary) cell culture, which provides the eukaryotic folding and post-translational modification environment required for a CNS-active kinase domain.

Where does the product need to go?

The protein is directed to the cytoplasm. In a CFPS reaction there is no membrane compartmentalisation; the expressed protein accumulates in the reaction mix and is purified by Ni-NTA affinity chromatography via the N-terminal His6 tag. No signal peptide or secretion sequence is required at this stage. For the eventual therapeutic application, the purified protein would be formulated for intranasal delivery, bypassing the blood-brain barrier via the olfactory and trigeminal nerve routes, as proposed in Homework 1 (Hanson and Frey, 2008).

Part 2: Genetic Parts Design

The construct comprises six parts, assembled 5′→3′ as a linear expression cassette optimised for E. coli CFPS. Each part was created or imported in Kernel with the appropriate part type designation. The table below provides a summary; detailed justifications follow.

T7 promoter (23 bp, positions 1–23)

Source: iGEM Registry; standard T7 promoter consensus sequence.

Justification: The T7 promoter is the standard transcriptional driver for E. coli CFPS systems. Commercial cell-free kits such as PURExpress (New England Biolabs) supply exogenous T7 RNA polymerase, enabling strong, single-subunit transcription independent of the host’s endogenous RNA polymerase machinery. The critical importance of strong, tightly controlled promoters in synthetic E. coli gene circuits was established by Elowitz and Leibler (2000), whose repressilator network demonstrated that promoter strength, transcriptional tightness, and efficient ribosome binding sites are the primary determinants of reliable gene expression dynamics. Those findings underpin the choice of the T7 promoter and a characterised RBS here: both parts have well-defined, quantitative performance characteristics that make expression outputs more predictable than uncharacterised alternatives.

RBS_B0034 (12 bp, positions 24–35)

Source: iGEM Registry part BBa_B0034.

Justification: BBa_B0034 is a medium-strength ribosome binding site from the iGEM community collection that has been widely characterised in E. coli expression systems. It provides reliable translation initiation without placing excessive demand on the ribosome pool, which is a finite and non-renewable resource in a CFPS reaction (Karim and Jewett, 2021). The use of orthogonal, pre-characterised parts from curated registries is a foundational principle of synthetic biology design. Stanton et al. (2014) demonstrated this principle at scale by mining prokaryotic genomes for repressor proteins that function orthogonally — that is, without cross-talk with endogenous cellular machinery — enabling the construction of reliable logic gates in E. coli. Their work reinforces the rationale for selecting BBa_B0034: a community-characterised, registry-sourced RBS is more likely to behave predictably in a novel construct context than an ad hoc designed element, because its performance has been empirically validated across multiple genetic backgrounds.

Spacer_RBS (6 bp, TACTAG, positions 36–41)

Source: Designed; standard RBS-to-start-codon spacing element.

Justification: This six-nucleotide spacer sets the distance between the Shine-Dalgarno sequence in the RBS and the downstream start codon, which is known to influence translation initiation efficiency in prokaryotic systems. Note: Kernel annotates this element as a CDS part (displaying a spurious Y* translation in the sequence viewer), but it functions as a non-coding spacer and has no effect on the expressed protein.

ATG_His6 (21 bp, positions 42–62)

Source: Designed; encodes Met followed by six histidine residues (MHHHHHH).

Justification: The N-terminal 6×His tag enables one-step purification of the expressed protein by immobilised metal affinity chromatography (IMAC/Ni-NTA). This is essential for isolating the PKMζ catalytic domain from the complex mixture of proteins and nucleic acids present in a CFPS lysate. The tag is small (seven residues, approximately 0.8 kDa) and is not expected to interfere with kinase folding or catalytic activity, although this assumption requires experimental validation.

PKMzeta_CDS_v2 (1,149 bp, positions 63–1211)

Source: Human PRKCZ gene, UniProt Q05513, residues 211–592. This region encodes the constitutively active C-terminal catalytic domain of PKCζ, which functions as PKMζ and is specifically required for long-term memory maintenance (Serrano et al., 2008). The part carries no start codon (supplied by ATG_His6 upstream) and terminates with a TGA stop codon. Codons at positions 366 and 522 were corrected to match the UniProt reference.

Justification: The catalytic domain alone is constitutively active because it lacks the N-terminal regulatory and pseudosubstrate domain of full-length PKCζ. Encoding only the catalytic domain (382 amino acids, approximately 42 kDa) keeps the total construct length within comfortable ranges for both CFPS expression and commercial gene synthesis.

Figure 1: PKMzeta_CDS_v2 part in Kernel — 1,149 bp, annotated as CDS (−start +stop). The sequence viewer confirms the corrected reading frame beginning G-I-A-Y-I-S-S-S-R-K, consistent with human PKCζ residues 211–592 (UniProt Q05513). The part description records the source accession and the two corrected codons.

Terminator_BBa_B0015 (129 bp, positions 1212–1340)

Source: iGEM Registry part BBa_B0015, a double transcriptional terminator.

Justification: BBa_B0015 combines two sequential terminators and provides robust transcription termination in E. coli. Placing it immediately downstream of the stop codon ensures clean termination of the mRNA transcript and prevents read-through, which would otherwise waste resources in the cell-free reaction.

Vector backbone

No vector backbone is included in the current design, as the E. coli-based CFPS platform can utilise linear DNA templates directly without a replication origin or selection marker. Should the project progress to in vivo bacterial expression, the cassette would be cloned into a standard T7 expression vector such as pET-28a (which provides a T7 promoter, kanamycin resistance marker, and ColE1 origin of replication). This cloning step is not required for the CFPS prototyping stage.

Part 3: Construct Assembly

The six parts were assembled in Kernel in the following 5′→3′ order:

T7 promoter → RBS_B0034 → Spacer_RBS → ATG_His6 → PKMzeta_CDS_v2 → Terminator_BBa_B0015

The total construct length is 1,340 bp. The expression host was set to Escherichia coli in the Kernel construct metadata.

Reading frame verification

Reading frame integrity was verified computationally by exporting the GenBank file and translating the complete sequence from the His6 start codon (ATG at position 42). The translated product begins:

M-H-H-H-H-H-H-G-I-A-Y-I-S-S-S-R-K-H-D-S-I-K…

and runs uninterrupted through the entire PKMζ catalytic domain, terminating with …N-P-L-L-L-S-T-E-E-S-V followed by a single TGA stop codon at positions 1209–1211. No internal stop codons are present within the reading frame. The BBa_B0015 terminator begins at position 1212, immediately downstream of the stop codon.

The schematic and sequence views below confirm the correct part order, orientation, and reading frame.

Figure 2: Exported circular construct map from Kernel, showing all six parts in 5′→3′ order.

Figure 3: Linear map view of the assembled construct (1,340 bp). The relative lengths of each annotated part are visible, with the PKMzeta_CDS_v2 coding region (1,149 bp) representing the majority of the construct.

Figure 4: Kernel Info panel for PKMzeta_CFPS_Expression_Cassette confirming: length 1,340 bp; expression host Escherichia coli; part annotations for all six components in correct positions.

Figure 5: Sequence viewer showing the junction between ATG_His6 and PKMzeta_CDS_v2. The translated sequence reads …H-H-H-H-G-I-A-Y-I-S-S-S-R-K-H-D-S…, confirming that the His6 tag and the PKMζ catalytic domain are in the same reading frame.

Part 4: Optimisation and Quality Control

Codon optimisation

GC content was calculated precisely from the exported GenBank file: the full 1,340 bp construct has a GC content of 53.51% (G: 364, C: 353, A: 340, T: 283), with the PKMzeta_CDS_v2 coding region at 54.8% and the regulatory elements ranging from 33.3% (RBS_B0034) to 48.1% (BBa_B0015 terminator). All values lie within the acceptable synthesisable and expressible range for E. coli (typically 40–60%).

The native human coding sequence was retained in this design iteration; however, a codon usage analysis was performed against the E. coli K12 tRNA abundance table (Kazusa database) to identify codons that would benefit from optimisation. The analysis revealed 35 rare codon occurrences across 383 total codons (9.1%), as detailed below:

What would change after optimisation: Replacing these 35 codons with preferred E. coli synonyms would reduce translational stalling events, particularly at the six AGG arginine codons and seven ACA threonine codons, which correspond to the rarest tRNA species in E. coli. The GC content of the CDS would shift from 54.8% toward approximately 52–53%, remaining within the optimal range. No new BsaI, BsmBI, BbsI or SapI restriction sites would be introduced, as these were checked as a design constraint.

Why codon usage matters: In E. coli CFPS, the tRNA pool is pre-loaded from the cell lysate and is finite — it is not replenished during the reaction as it would be in a growing cell. Rare codons that correspond to low-abundance tRNAs cause ribosome pausing, which in a CFPS context reduces yield and can cause premature peptide release or misincorporation. Karim and Jewett (2021) highlight tRNA pool management as a key determinant of CFPS expression efficiency, and this is therefore a higher-priority optimisation for the cell-free context than for conventional in-cell expression.

Sequence verification

Kernel does not include an integrated BLAST alignment tool. Sequence verification was therefore performed externally by translating the exported GenBank file computationally and aligning the resulting amino acid sequence residue-by-residue against the UniProt Q05513 (KPCZ_HUMAN) reference. Following correction of the two single-nucleotide variants at codons 366 and 522, the translated protein is a perfect match to human PKCζ residues 211–592 across all 382 positions. A BLAST search against the NCBI non-redundant protein database would be expected to return the human PRKCZ mRNA (RefSeq NM_002744) as the top hit with near-100% identity, confirming the sequence identity of the construct as a defined fragment of a well-characterised human kinase.

Construct Checker

Kernel’s Construct Checker was run on the assembled construct and returned four warnings:

1. Missing Kozak sequence, 5′ UTR, 3′ UTR, and poly-A signal.
2. T7 promoter at index 0 is “not recommended.”
3. RBS_B0034 at index 23 is “not recommended.”

Figure 6: Construct Checker output. All four warnings are attributable to Kernel’s default mammalian expression model and are not relevant to the E. coli CFPS design.

These warnings arise because the Kernel Construct Checker defaults to a mammalian expression model. The construct under assessment is designed for E. coli-based CFPS, in which: the T7 promoter drives transcription via exogenous T7 RNA polymerase supplied in the lysate; BBa_B0034 performs the role of the Kozak sequence and 5′ UTR as the prokaryotic ribosome binding site; and BBa_B0015 performs the role of the poly-A signal as the transcriptional terminator. As noted during the Lecture 2 recitation, the Kernel construct checker is “not going to be super useful” for bacterial constructs, because it checks exclusively for eukaryotic regulatory elements. No design changes are warranted in response to these warnings.

Simulation

Kernel’s built-in Simulation tool (Sim.) was used to model gene expression dynamics for the assembled construct in an E. coli context. The simulation was configured with the following parameters:

• Chassis: E. coli
• Duration: 72 hours
• Timestep: 10 minutes
• Transfection mode: Transient transfection

Figure 7: Kernel simulation configuration — E. coli chassis, 72-hour duration, 10-minute timestep, transient transfection mode.

Figure 8: Simulation completed successfully (green checkmark, 29/09/2026, 18:18:14), confirming structural acceptance of the construct by the E. coli simulation model.

The simulation completed without errors, confirming that the construct is structurally valid under the E. coli model used by Kernel. The 72-hour transient transfection mode is appropriate for CFPS, which is a batch process in which the DNA template is not replicated and is gradually consumed or degraded. An important interpretive caveat is that Kernel’s simulation model is calibrated on characterised iGEM Registry parts. PKMzeta_CDS_v2 carries no experimental characterisation data in Kernel, and consequently the simulation reflects the kinetics of the T7 promoter and RBS elements only. It does not predict the yield, solubility, or activity of the PKMζ protein. Experimental validation in an actual CFPS reaction is required before drawing any quantitative conclusions about expression output.

Restriction site analysis

The Enzymes/Cut Sites panel in Kernel was used to screen the full 1,340 bp construct for common Golden Gate (Type IIS) enzyme recognition sites. Results are summarised below:

The construct contains no recognition sites for any of the four principal Golden Gate enzymes. It is therefore fully compatible with Golden Gate assembly should future cloning into a vector backbone be required.

Figure 9: Enzymes/Cut Sites panel in Kernel confirming zero cuts for BsaI, BsmBI, BbsI and SapI in the final 1,340 bp construct.

Safety screen

Kernel does not currently include an integrated biosecurity screening tool. The sequence was assessed against standard screening criteria. The PKMζ catalytic domain is a fragment of PKCζ, a ubiquitous human signalling enzyme expressed in all human cells. It has no known toxicity, pathogenicity, or dual-use concern and does not appear on any select-agent list maintained by the CDC, APHIS, or equivalent regulatory authorities. Commercial DNA synthesis vendors, including Twist Bioscience, perform automated biosecurity screening against the International Gene Synthesis Consortium (IGSC) Harmonised Screening Protocol prior to order fulfilment. No flags would be anticipated for this sequence.

Part 5: From Design to Build

Template format decision: linear gene fragment

The construct will be ordered as a linear gene fragment rather than a clonal gene in a plasmid vector. This decision rests on three considerations:

1. CFPS compatibility. The expression platform is E. coli-based CFPS, which accepts linear DNA templates directly and does not require a plasmid backbone, origin of replication, or antibiotic resistance marker.

2. Cost. At the current Twist Bioscience published list price of USD 0.07 per base pair for gene fragments (Twist Bioscience, 2026; Armadillo Bio, 2026), the 1,340 bp construct has an estimated cost of approximately USD 93.80. A clonal gene (sequence-verified in a Twist expression vector) is priced at USD 0.09 per base pair plus a USD 55 expression-vector cloning fee, giving an estimated total of approximately USD 175.60 — an 87% premium for a vector backbone that is not required at the prototyping stage. IDT offers comparable gene-fragment pricing at USD 0.07 per base pair (Armadillo Bio, 2026).

3. Turnaround time. Twist gene fragments are shipped in 2–4 business days; clonal genes require 4–10 business days (Twist Bioscience, 2026).

Should the project advance to in vivo expression in E. coli or CHO cells, the cassette would be cloned into a suitable expression vector (e.g. pET-28a for E. coli, or a CMV-driven mammalian vector for CHO). At that stage, a clonal gene order or Gibson assembly into a backbone would be the appropriate approach.

How would the construct be built?

The 1,340 bp expression cassette falls well within the range of commercial gene synthesis services and would be ordered as a single synthetic linear dsDNA fragment — no fragment assembly or oligo pool synthesis is required. Twist gene fragments support lengths from 300 to 5,000 bp; the construct falls comfortably within this window. The synthesised fragment would be delivered lyophilised, resuspended in TE buffer, and used directly as a CFPS template.

Synthesis constraints

Constraints on synthesis are minimal for this construct. The verified GC content of 53.51% lies well within the standard synthesisable range of 25–65%. No long homopolymer runs (eight or more consecutive identical bases), extreme secondary structures, or direct repeats were identified during design. The absence of BsaI, BsmBI, BbsI and SapI sites confirms compatibility with downstream Golden Gate cloning.

Synthesis error rates remain a relevant consideration. Twist Bioscience reports a mean error rate of 1 in 7,500 bp for gene fragments, achieved through enzymatic error correction during manufacture (Twist Bioscience, 2026). At 1,340 bp, the probability of receiving a fragment carrying at least one error is approximately 16%. Sequence verification by Sanger sequencing after cloning is therefore standard practice. Carr et al. (2004) demonstrated that protein-mediated error correction using MutS reduces de novo synthesis error rates to approximately 1 in 10,000 bp, a technique now routinely applied by synthesis vendors. Kong et al. (2007) demonstrated that microfluidic polymerase cycling assembly (PCA) can synthesise genes up to approximately 1,000 bp in parallel, an approach that could reduce per-construct costs substantially if multiple variant constructs were required for screening.

Prototype versus scale

At the prototyping stage, a single linear gene fragment used directly as a CFPS template is the fastest and most cost-effective approach: approximately USD 94, delivered within 2–4 business days, with no cloning, cell culture, or transformation required. The template is added directly to the CFPS lysate and protein production can be assessed within hours.

For scale-up to CHO cell production — as proposed in Homework 1 — the expression cassette would require complete redesign with mammalian regulatory elements (CMV or EF1α promoter, Kozak sequence, poly-A signal, and a mammalian-compatible vector backbone). Gibson assembly (Gibson et al., 2009) would be appropriate for joining a redesigned cassette to a vector backbone; Gibson’s isothermal method joins fragments sharing approximately 20 bp overlapping ends in a single 50 °C reaction, which is mechanistically simpler than restriction-ligation for multi-fragment assemblies.

For large-scale variant screening — testing many PKMζ mutants to optimise activity, thermal stability, or folding in a CFPS context — chip-based oligonucleotide synthesis followed by multiplex gene assembly, as described by Kosuri et al. (2010), would dramatically reduce per-construct costs relative to ordering individual gene fragments. Tian et al. (2004) demonstrated that microchip-synthesised oligonucleotides could reduce gene synthesis costs from approximately USD 0.10 per nucleotide (conventional column synthesis) to as low as USD 0.001 per nucleotide, enabling massively parallel gene construction at a scale appropriate for variant libraries.

Challenges and Learning Journey

This assignment represented a substantial learning experience for a practitioner whose professional background is in product management and cybersecurity rather than molecular biology. Many of the core concepts encountered — reading frames, codon structure, ribosome binding sites, and the distinction between eukaryotic and prokaryotic regulatory elements — were entirely new territory. The BioClub Tokyo Friday session (25 September 2026) proved particularly valuable; Sheila Ramani’s live walkthrough of Kernel and Georg Tremmel’s guidance on design documentation provided an accessible practical foundation.

Reading frame debugging. The most time-consuming challenge involved discovering that the original CDS part (PKMzeta_CDS, 1,151 bp) contained a frame-shift error. The sequence as initially entered into Kernel began with ATGGAATTGCT.... The ATG triplet was initially assumed to constitute a start codon; however, computational analysis revealed that these three nucleotides straddle two codons from the reference protein — the final two bases of an aspartate codon and the first base of the adjacent glycine codon — thereby creating a +2 frame shift. Under this misalignment, the His6 tag and the PKMζ coding sequence were in different reading frames, and the translated product would have been non-functional. Diagnosis required exporting the GenBank file, translating the construct computationally, and comparing the result residue-by-residue against UniProt Q05513. The correction was to remove the first two nucleotides (AT) from the CDS, yielding PKMzeta_CDS_v2 (1,149 bp) and restoring the correct reading frame: His6 → G-I-A-Y-I-S-S-S-R-K (PKCζ residues 211 onwards).

Amino acid mismatches. Even after correcting the frame shift, computational translation identified two amino acid differences from the UniProt reference: leucine at position 366 in place of phenylalanine (codon CTC instead of TTC), and glycine at position 522 in place of aspartate (codon GGC instead of GAC). The origin of these single-nucleotide variants could not be traced, but both were corrected to match the UniProt human reference sequence, as this is the accession cited in the design.

Construct Checker interpretation. The Kernel Construct Checker produced warnings about missing Kozak, UTR, and poly-A elements that were initially difficult to interpret. Review of the Lecture 2 recitation recording — in which the Kernel demonstration noted that the checker “is not going to be super useful” for bacterial constructs — clarified that these warnings apply to mammalian expression models and are not relevant to the E. coli CFPS cassette. The construct metadata was also found to have been pre-populated with Cricetulus griseus (CHO) as the expression host, which was corrected to Escherichia coli.

Learning the Kernel platform. Working with Kernel required developing familiarity with part-type designations (CDS with or without start and stop codons), drag-and-drop construct assembly, sequence viewer interpretation, cut-site analysis, and GenBank export. The BioClub Tokyo session and the Lecture 2 recitation recording were the most useful practical resources.

Part 6: Open Questions and Design Risks

1. Will the truncated catalytic domain fold correctly in E. coli CFPS? PKMζ is a human kinase that folds in a eukaryotic cellular environment equipped with chaperones and disulphide-bond-forming machinery. E. coli lysate lacks these systems, and the expressed protein may misfold or partition to the insoluble fraction. The experimental test would be a small-scale CFPS reaction (10–50 µL) followed by centrifugation to separate soluble from insoluble fractions, SDS-PAGE to confirm expression and solubility, and a kinase activity assay measuring phosphorylation of a known PKCζ substrate peptide to confirm catalytic function.

2. Does the N-terminal His6 tag interfere with kinase activity? The His6 tag is positioned at the beginning of the catalytic domain. Some N-terminal affinity tags can occlude the active site or disrupt tertiary structure. This question would be addressed by comparing the activity of the His6-tagged protein against tag-cleaved protein, using a TEV protease recognition site (not yet included in the current design). If interference is observed, a C-terminal tag placement or a cleavable-tag design would constitute the next design iteration.

3. Is codon optimisation necessary for adequate expression yield? The native human codon usage may include codons that are rare in E. coli and could cause ribosome stalling, particularly in a CFPS system where the tRNA pool is finite. If initial CFPS yields are insufficient for downstream characterisation, codon optimisation for E. coli using a tool such as IDT’s Codon Optimization Tool or the Kazusa E. coli K12 codon usage table would be the first design modification to implement.

Corrected CDS Sequence (PKMzeta_CDS_v2)

The sequence below is the final verified 1,149 bp coding sequence for PKMzeta_CDS_v2, encoding human PKCζ residues 211–592 (UniProt Q05513). The two corrected codons are indicated in brackets: [TTC] (phenylalanine at position 366) and [GAC] (aspartate at position 522).

GGAATTGCTTACATTTCCTCATCCCGGAAGCATGACAGCATTAAAGACGACTCGGAGGAC
CTTAAGCCAGTTATCGATGGGATGGATGGAATCAAAATCTCTCAGGGGCTTGGGCTGCAG
GACTTTGACCTAATCAGAGTCATCGGGCGCGGGAGCTACGCCAAGGTTCTCCTGGTGCGG
TTGAAGAAGAATGACCAAATTTACGCCATGAAAGTGGTGAAGAAAGAGCTGGTGCATGAT
GACGAGGATATTGACTGGGTACAGACAGAGAAGCACGTGTTTGAGCAGGCATCCAGCAAC
CCCTTCCTGGTCGGATTACACTCCTGCTTCCAGACGACAAGTCGGTTGTTCCTGGTCATT
GAGTACGTCAACGGCGGGGACCTGATGTTCCACATGCAGAGGCAGAGGAAGCTCCCTGAG
GAGCACGCCAGGTTCTACGCGGCCGAGATCTGCATCGCCCTCAAC[TTC]CTGCACGAGAG
GGGGATCATCTACAGGGACCTGAAGCTGGACAACGTCCTCCTGGATGCGGACGGGCACAT
CAAGCTCACAGACTACGGCATGTGCAAGGAAGGCCTGGGCCCTGGTGACACAACGAGCAC
TTTCTGCGGAACCCCGAATTACATCGCCCCCGAAATCCTGCGGGGAGAGGAGTACGGGTT
CAGCGTGGACTGGTGGGCGCTGGGAGTCCTCATGTTTGAGATGATGGCCGGGCGCTCCCC
GTTCGACATCATCACCGACAACCCGGACATGAACACAGAGGACTACCTTTTCCAAGTGATC
CTGGAGAAGCCCATCCGGATCCCCCGGTTCCTGTCCGTCAAAGCCTCCCATGTTTTAAAAG
GATTTTTAAATAAGGACCCCAAAGAGCGGCTCGGCTGCCGGCCACAGACTGGATTTTCTGA
CATCAAGTCCCACGCGTTCTTCCGCAGCATA[GAC]TGGGACTTGCTGGAGAAGAAGCAGGC
GCTCCCTCCATTCCAGCCACAGATCACAGACGACTACGGTCTGGACAACTTTGACACACAG
TTCACCAGCGAGCCCGTGCAGCTGACCCCAGACGATGAGGATGCCATAAAGAGGATCGAC
CAGTCAGAGTTCGAAGGCTTTGAGTATATTAACCCATTATTGCTGTCCACCGAGGAGTCG
GTGTGA

AI Use Disclosure

AI client: Claude (Anthropic), used via the Claude.ai chat interface with web search, computer tools, and the uploaded course-reading PDFs and lecture transcripts from this project.

Assistance provided by the AI tool:

1. Design of the expression cassette architecture (T7 → RBS → spacer → His6 → CDS → terminator) for E. coli CFPS.
2. Identification and resolution of the reading-frame error in the original PKMzeta_CDS part: translation of the exported GenBank sequence, identification of the +2 frame shift arising from the spurious AT prefix, and generation of the corrected sequence.
3. Identification of two single-nucleotide variants (F366L, D522G) relative to UniProt Q05513 and provision of the corrected codons (CTC→TTC; GGC→GAC).
4. Interpretation of the Construct Checker warnings as eukaryote-specific and not applicable to the bacterial CFPS design.
5. Verification of the final exported GenBank file: reading frame integrity, stop codon placement, and restriction site analysis.
6. Retrieval of current gene synthesis pricing from Twist Bioscience and IDT to support the template format decision.
7. Drafting and professional editing of this report.

All design decisions were reviewed and are understood by the author. Each design choice can be defended during the homework review session.

Acknowledgements

The author would like to thank Yousif Graytee (Teaching Assistant, Designer Cells Node, HTBAA 2026) for providing the cell-free protein synthesis readings by da Silva et al. (2026) and the Keith Pardee laboratory, which informed the CFPS platform choice and construct design approach for this project. Thanks are also due to Sami Ur Rehman (peer, HTBAA 2026, Attock, Pakistan) for advice on cell-free expression systems and peptide project design during the early stages of this work.

References

Armadillo Bio (2026) Gene synthesis price comparison 2026 — all providers. Available at: https://www.armadillobio.com/compare/gene-synthesis-price-comparison-2026 (Accessed: 29 September 2026).

Elowitz, M. B. and Leibler, S. (2000) ‘A synthetic oscillatory network of transcriptional regulators’, Nature, 403(6767), pp. 335–338. doi: 10.1038/35002125.

Carr, P. A., Park, J. S., Lee, Y.-J., Yu, T., Zhang, S. and Jacobson, J. M. (2004) ‘Protein-mediated error correction for de novo DNA synthesis’, Nucleic Acids Research, 32(20), e162. doi: 10.1093/nar/gnh160.

Gibson, D. G., Young, L., Chuang, R.-Y., Venter, J. C., Hutchison, C. A. and Smith, H. O. (2009) ‘Enzymatic assembly of DNA molecules up to several hundred kilobases’, Nature Methods, 6(5), pp. 343–345. doi: 10.1038/nmeth.1318.

Hanson, L. R. and Frey, W. H. (2008) ‘Intranasal delivery bypasses the blood-brain barrier to target therapeutic agents to the central nervous system and treat neurodegenerative disease’, BMC Neuroscience, 9(Suppl 3), S5. doi: 10.1186/1471-2202-9-S3-S5.

Karim, A. S. and Jewett, M. C. (2021) ‘Toward sustainable, cell-free biomanufacturing’, Current Opinion in Biotechnology, 69, pp. 136–144. doi: 10.1016/j.copbio.2020.12.012.

Kong, D. S., Carr, P. A., Chen, L., Zhang, S. and Jacobson, J. M. (2007) ‘Parallel gene synthesis in a microfluidic device’, Nucleic Acids Research, 35(8), e61. doi: 10.1093/nar/gkm121.

Kosuri, S., Eroshenko, N., Leproust, E. M., Super, M., Way, J., Li, J. B. and Church, G. M. (2010) ‘Scalable gene synthesis by selective amplification of DNA pools from high-fidelity microchips’, Nature Biotechnology, 28(12), pp. 1295–1299. doi: 10.1038/nbt.1716.

Patzke, N., Spocter, M. A., Karlsson, K. Æ., Bertelsen, M. F., Haagensen, M., Chawana, R., Streicher, S., Kaswera, C., Gilissen, E., Alagaili, A. N., Mohammed, O. B., Reep, R. L., Bennett, N. C., Siegel, J. M., Ihunwo, A. O. and Manger, P. R. (2013) ‘Organization and chemical neuroanatomy of the African elephant (Loxodonta africana) hippocampus’, Brain Structure and Function, 220(3), pp. 1813–1830. doi: 10.1007/s00429-013-0587-6.

Petzold, C. J. and Mukhopadhyay, A. (2025) ‘From bench to biofactory: high-throughput technologies and automated workflows to accelerate biomanufacturing’, Current Opinion in Biotechnology, 94, 103320. doi: 10.1016/j.copbio.2025.103320.

Serrano, P., Friedman, E. L., Kenney, J., Taubenfeld, S. M., Zimmerman, J. M., Hanna, J., Alberini, C., Kelley, A. E., Maren, S., Rudy, J. W., Yin, J. C. P., Sacktor, T. C. and Fenton, A. A. (2008) ‘PKMζ maintains spatial, instrumental, and classically conditioned long-term memories’, PLoS Biology, 6(12), pp. 2698–2706. doi: 10.1371/journal.pbio.0060318.

Stanton, B. C., Nielsen, A. A., Tamsir, A., Clancy, K., Peterson, T. and Voigt, C. A. (2014) ‘Genomic mining of prokaryotic repressors for orthogonal logic gates’, Nature Chemical Biology, 10(2), pp. 99–105. doi: 10.1038/nchembio.1411.

Tian, J., Gong, H., Sheng, N., Zhou, X., Gulari, E., Gao, X. and Church, G. (2004) ‘Accurate multiplex gene synthesis from programmable DNA microchips’, Nature, 432(7020), pp. 1050–1054. doi: 10.1038/nature03151.

Twist Bioscience (2026) Gene fragments — product page. Available at: https://www.twistbioscience.com/products/genes (Accessed: 29 September 2026).

UniProt Consortium (2026) UniProt entry Q05513 (KPCZ_HUMAN): Protein kinase C zeta type, Homo sapiens. Entry version 255. Available at: https://www.uniprot.org/uniprotkb/Q05513/entry (Accessed: 29 September 2026).

Submission Checklist

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).

Final Project