Week 1 HW: Final Project Ideas

Title: Engineering a hyperstable, secretion-optimized FGF2 for neural-cell and organoid culture: AI-guided protein design and chromatography-light biomanufacturing in Komagataella phaffii.

Date: Tuesday September 22

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

Target: a hyperstable, secretion-optimized FGF2 variant (in the design lineage of FGF2-STAB/G3), produced by secretory fermentation in Komagataella phaffii, sold as an animal-component-free culture supplement for neural stem cells, iPSC-derived neurons, and brain organoids (not initially an injectable therapeutic).

Why I'm interested: growth factors commonly make up 60–99% of defined stem-cell media cost, and wild-type FGF2's ~9–10 hour half-life at 37°C forces near-daily media replenishment, so a longer-lived variant multiplies potency per gram and directly lowers the cost barrier to neuroscience and regenerative-medicine research. Personally, the project is interesting because it treats protein stability and manufacturing economics as one inverse-design problem ("what sequence/strain/process/DSP combination gives the cheapest, most robust, biologically equivalent product?" rather than just "what mutation is most stable?").

Customers: academic and biotech labs doing neural-organoid, iPSC-neuron, and neural-stem-cell culture.

(2) What living system would you use?

Chassis: Komagataella phaffii (Pichia pastoris), secretory expression. Chosen over E. coli because it secretes correctly folded, disulfide-bonded protein without inclusion bodies (no refolding step needed), and over CHO because FGF2 doesn’t require complex mammalian glycosylation; Pichia also has a lower endotoxin profile.

Inputs: an engineered FGF2 gene (hyperstable mutant) fused to a secretion leader (α-mating-factor prepro or an ML-selected signal peptide), under a methanol-inducible (AOX1) or constitutive (GAP) promoter, in a host with protease knockouts (PEP4, PRB1) and chaperone co-expression (PDI, HAC1, Kar2/BiP) to reduce degradation and secretion bottlenecks; plus glycerol/methanol carbon feed, oxygen, and controlled pH/temperature in a bioreactor.

Process: batch growth on glycerol, then induction, during which cells secrete correctly folded FGF2 into the medium.

Output: clarified broth containing secreted FGF2, concentrated and purified mainly by tangential-flow filtration (with an optional heparin-affinity polishing step if needed), formulated as a freeze-dried supplement.

Assumption flagged: the actual secretory titer and proteolytic loss for this specific variant are unknown and must be measured experimentally, not assumed.

(3) How would you use automation?

Automation: a biofoundry using acoustic liquid handling (e.g., Echo) for barcoded DNA assembly, colony handling, and strain normalization, combined with a cloud lab (e.g., Emerald Cloud Lab) for standardized DNA, plate, and biochemical workflows once protocols are mature.

What's automated and why: the Build and Test steps of the design–build–test–learn cycle — DNA assembly, robotic microplate/deep-well secretion screening, and automated readouts (differential scanning fluorimetry for thermal stability, aggregation, phospho-ERK and neurite-outgrowth bioassays, accelerated 37°C stability) — because each DBTL round tests 24–96 designed sequences and needs consistent, LIMS-linked measurements to train the AI surrogate model reliably; doing this by hand would not scale to the number of rounds needed.

Specialized neural-stem-cell/organoid potency assays stay human-supervised (a “hybrid SDL,” not full robotic autonomy).

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

(a) How much?

Scale: enough hyperstable FGF2 to supply the defined-media needs of about 1,000 academic and biotech labs doing neural-organoid or iPSC-neuron culture worldwide, for one year (an order-of-magnitude proxy for the current global user base of bFGF-supplemented neural culture protocols).

Calculation:

• Assume an Essential-8-type medium at ~100 ng/mL FGF2 (the proposal’s cited ~$373/L E8-type formulation).
• Assume a lab consumes ~20 L of media per year (~400 mL/week).
• FGF2 per lab per year = 100 ng/mL x 1,000 mL/L x 20 L = 2,000,000 ng = 2 mg/lab/year.
• Total for 1,000 labs = 1,000 x 2 mg = 2,000 mg (2 g) of active FGF2 per year.

At the reference price of $2,700/mg this is roughly a $5.4M/year market; at the $196/mg low-cost benchmark, roughly $392,000/year — consistent with growth factors being high-value, low-mass products.

(b) How long?

Using the proposal's own Phase 1–2 translational milestone (≥100 mg/L secreted active protein at 5 L) and ≥80% downstream recovery (TFF-primary, chromatography-light):

• Recovered product per 5 L batch = 100 mg/L x 5 L x 0.80 = 400 mg/batch.
• Batches needed for the 2,000 mg target = 2,000 / 400 = 5 batches.
  • Fed-batch cycle time for methanol-induced Pichia: ~3–5 days growth + ~2–3 days induction + ~1–2 days clarification/TFF/formulation + turnaround ≈ ~9 days/batch.
  • Run serially in one 5 L reactor: 5 batches x 9 days ≈ 45 days (about 6–7 weeks).
• Alternative: run batches in parallel across multiple 5 L reactors (biofoundry/cloud-lab supported), or move directly to a 50 L pilot reactor once pilot infrastructure is qualified — a single 50 L batch at the same titer and recovery yields ~4 g, exceeding the 2 g target in one ~9-day campaign.

Note: this uses the proposal’s early-milestone titer (100 mg/L) rather than its later gram-per-liter stretch goals, so it is a conservative (slower) estimate.

(5) What don’t you know yet?

- Whether the stability-enhancing mutations preserve full FGFR1 binding and neural bioactivity — the central stability–bioactivity trade-off risk; this needs bioassay gating (phospho-ERK, neurite outgrowth, organoid potency) on every candidate before it advances.

• The actual secretory titer and proteolytic loss of this specific hyperstable variant in K. phaffii — published Pichia-secreted FGF2 titers range from under 1 mg/L to higher for other variants, so it is unknown whether the ≥100 mg/L at 5 L milestone is reachable without the PEP4/PRB1 knockouts and chaperone co-expression actually working as intended.
  • Whether TFF-only downstream processing can consistently hit ≥95–98% purity and endotoxin specifications at pilot scale, or whether the low-recovery (~6%) heparin-affinity polishing step becomes necessary more often than the chromatography-light strategy assumes.

AI

Claude Code (Anthropic, Sonnet 5 model) was used to help organize and draft answers to this worksheet from the author’s own project proposal (“03 project_proposal.md”), with the prompt: “using the information detailed in the project proposal, complete the questions in this doc.” All content was reviewed and understood by the author prior to submission.

Final Project