See the nav bar on the left to look through my work for How To Biomanufacture Almost Anything (Fall 2026)!
HTBAA Homework #1 | A discovery tool for cryptic, low-affinity HLA-E ligands
| Question | Proposal |
|---|---|
| What I will make | A 20-member research-use panel of soluble HLA-E complexes, each paired with a different β2-microglobulin (β2m) variant and supplied as a peptide-receptive capture reagent. |
| Living chassis | Suspension HEK293 cells that secrete correctly folded, glycosylated HLA-E–β2m complexes into defined medium. |
| Automation | AlphaFold/Rosetta triage, robotic liquid handling for expression and peptide-loading screens, and automated sample preparation and data analysis for LC–MS/MS. |
| Meaningful scale | 100 discovery kits: 20 variants × 0.10 mg per variant × 100 kits, plus 25% manufacturing overage = 250 mg finished protein. |
Central hypothesis. Changing the β2m partner can tune HLA-E complex persistence enough to preserve weakly bound peptides that are normally lost during handling. The engineered β2m is a discovery aid—not proof that a recovered peptide is physiologically presented. Every candidate peptide would therefore be re-tested with wild-type human β2m and functional NK-cell assays.
I propose to manufacture a research-use HLA-E peptide-capture panel containing 20 soluble HLA-E–β2m complexes. Each complex would use the same HLA-E heavy-chain ectodomain but a different β2m partner selected from natural orthologs and affinity-engineered mutants.
Each reagent would carry a purification handle and be formulated in a peptide-receptive state so that the panel can be challenged with defined peptide libraries or peptide pools isolated from cells. Bound peptides would then be recovered and identified by LC–MS/MS.
HLA-E is a non-classical MHC class I molecule with limited polymorphism. It presents peptides to CD94/NKG2 receptors on natural killer (NK) cells and can also be recognized by HLA-E-restricted T cells. Structural studies established a constrained canonical binding mode, while later studies demonstrated broader peptide tolerance and showed that peptide identity can change immune-receptor recognition [1–3].
Low-affinity complexes are intrinsically difficult to study. If they dissociate during immunoprecipitation, washing, concentration, or LC–MS sample preparation, the experiment preferentially reports the most stable ligands. My goal is to lower this stability barrier so that weak and normally undetectable HLA-E ligands can be captured.
The panel is intended for immunopeptidomics, NK-cell biology, antigen-discovery, and immunotherapy laboratories. One kit would let a laboratory compare twenty β2m contexts under matched conditions.
At 5 μg of protein per capture reaction, 0.10 mg of each variant supports approximately twenty assays:
A single highly stabilizing β2m could create a strong but biased peptide repertoire. A panel exposes the trade-off between recovery and distortion: variants with different stabilization strengths can be compared against wild-type β2m, negative controls, and known HLA-E peptides.
The deliverable is therefore both a reagent set and an experimentally calibrated map of which β2m contexts recover which peptide classes.
I would use suspension-adapted HEK293 cells for transient prototype production and later create stable producer pools for repeat manufacturing. HEK293 cells are widely used for rapid transient expression of recombinant proteins and can perform mammalian secretory-pathway folding and glycosylation [4].
This is useful for HLA-E because its heavy chain must fold with both β2m and peptide. A bacterial refolding process could eventually reduce cost, but mammalian secretion is the lower-risk starting point for producing a diverse variant panel.
Inputs
Biological process
The cells transcribe and translate the HLA-E and β2m constructs. Signal peptides route the chains into the endoplasmic reticulum, where they associate, fold, acquire mammalian post-translational modifications, and enter the secretory pathway.
Harvest and purification
The clarified culture medium is processed by affinity chromatography, buffer exchange, and size-exclusion or polishing chromatography. UV absorbance, SDS–PAGE or SEC, and a conformation-sensitive binding assay are used for release testing.
Outputs
The process produces twenty purified HLA-E–β2m variants with known concentration, purity, monomer content, and peptide-loading competence. When the reagents are used for peptide capture, the workflow also produces an LC–MS/MS dataset of recovered peptides.
The workflow has two different automation regimes: high-throughput automation for choosing and testing variants, and process automation for manufacturing the selected panel. Automated design–build–test–learn workflows can accelerate construct optimization, whereas manufacturing requires robust unit operations and process control [5,6].
Computational triage
I would model 100 candidate HLA-E–β2m complexes with AlphaFold and apply predefined interface-confidence filters:
The top 20 structurally credible candidates would then be evaluated for relative interface ΔΔG using Rosetta. Computation would not replace experimental testing; it would reduce the number of constructs that must be built.
Robotic liquid handling
An Opentrons Flex or a cloud laboratory could automate:
The initial 100 candidates would fit into two 96-well plates with controls. The selected 20 candidates could be tested in triplicate on one 96-well plate.
Automated analytics
Plate-based ELISA or fluorescence and thermal-shift measurements would quantify secretion and stability. For immunopeptidomics, automation could handle immunocapture, washing, acid elution, desalting, and sample tracking. A reproducible LC–MS/MS search and false-discovery-rate pipeline would analyze the resulting spectra.
Closed-loop selection
Variants would be ranked using a multi-objective score combining:
No candidate would advance based on predicted ΔΔG alone.
At kit scale, a benchtop liquid handler would no longer be the main production machine. I would use twenty parallel, instrumented 2-L single-use stirred vessels or shake-flask equivalents, with one vessel assigned to each variant.
Automated controllers would record temperature, pH, dissolved oxygen, agitation, viable cell density, and harvest time. An ÄKTA-style chromatography system would execute standardized capture and polishing methods. A liquid handler would normalize final concentrations, perform plate-based QC, and dispense labeled aliquots.
The LC–MS instrument would remain a likely bottleneck. Discovery samples would therefore be queued by priority instead of assuming that every well could be analyzed immediately.
I define a meaningful first scale as enough material to supply 100 research groups with one complete discovery kit. Each kit contains 20 variants at 0.10 mg per variant. This amount is small compared with a therapeutic dose but large enough for repeated analytical experiments.
| Quantity | Calculation | Result |
|---|---|---|
| Protein per kit | 20 variants × 0.10 mg | 2.0 mg |
| Finished protein for 100 kits | 100 kits × 2.0 mg/kit | 200 mg |
| Manufacturing target | 200 mg × 1.25 for QC, filling, and stability overage | 250 mg |
| Per-variant target | 250 mg ÷ 20 variants | 12.5 mg/variant |
| Effective recovered titer | 20 mg/L harvest titer × 50% recovery | 10 mg/L |
| Total culture volume | 250 mg ÷ 10 mg/L | 25 L |
| Culture volume per variant | 12.5 mg ÷ 10 mg/L | 1.25 L/variant |
Therefore, the base-case manufacturing requirement is:
20 variants × 1.25 L per variant = 25 L of total culture
The 20 mg/L harvest titer and 50% downstream recovery are planning assumptions for this specific HLA-E panel, not measured values. HEK293 systems support rapid recombinant-protein production, but achievable titer varies widely with the construct and process [4]. A pilot campaign must therefore measure titer and recovery for every variant before capacity is fixed.
The required culture volume is:
Culture volume = required finished protein ÷ (harvest titer × recovery)
| Scenario | Harvest titer | Recovery | Effective titer | Culture needed for 250 mg |
|---|---|---|---|---|
| Conservative | 5 mg/L | 50% | 2.5 mg/L | 100 L |
| Base case | 20 mg/L | 50% | 10 mg/L | 25 L |
| Lower recovery | 20 mg/L | 25% | 5 mg/L | 50 L |
| Optimized | 50 mg/L | 60% | 30 mg/L | 8.3 L |
The base case could be run as one parallel campaign using twenty 2-L vessels at approximately 1.25–1.5 L working volume each. If only four vessels were available, five sequential production waves would be required, substantially increasing the elapsed time.
| Operation | Base-case duration | Parallelization or bottleneck |
|---|---|---|
| Seed expansion and transfection | 3–4 days | Twenty variants prepared in parallel |
| Expression and secretion | 5–7 days | Parallel culture; harvest determined by viability and titer |
| Harvest and purification | 2–3 days | Chromatography may require staggered batches |
| QC, concentration, and formulation | 1–2 days | SEC, purity, concentration, and loading competence |
| Aliquoting and kit assembly | 1 day | Automated normalization and dispensing |
The physical production campaign would therefore require approximately 12–17 calendar days.
This estimate assumes that sequence-verified plasmids or established producer pools already exist and that all twenty variants can be cultured in parallel. It excludes initial construct design, cloning, and producer qualification.
A first prototype program would add approximately 2–4 weeks for DNA preparation, small-scale expression, and assay qualification. Creating and banking stable producer pools could take several months, but it would shorten and standardize later manufacturing campaigns.
A robotic peptide-loading and stability screen could prepare and measure the twenty-variant panel in approximately 2–3 days. Immunopeptidomic sample preparation could require another 2–3 days.
LC–MS/MS capacity would be limiting. If one method requires 60 minutes per sample plus overhead, 60 samples representing 20 variants in triplicate would occupy approximately 3–4 instrument days. Data analysis and manual review would add about one week.
Thus, a complete design-to-ranked-peptide cycle would more plausibly be measured in weeks than in the 12–17 days required only to manufacture the protein.
A β2m variant may rescue weak ligands, but it may also change peptide exchange, loading hierarchy, or the structure recognized by CD94/NKG2 receptors.
I would compare multiple stabilization strengths, spike in peptides with known affinity, quantify overlap with wild-type β2m, and reconstitute every discovery hit using wild-type HLA-E and human β2m before making a biological claim.
Empty or weakly loaded HLA-E may aggregate or be lost during purification—the same instability that the product is intended to solve.
I need to measure secretion titer, monomer fraction, thermal stability, peptide-exchange kinetics, and peptide-loading yield for each construct. Native mass spectrometry, SEC, differential scanning fluorimetry, and conformation-sensitive antibodies could distinguish correctly assembled complexes from soluble aggregates.
Recovering more peptide by LC–MS/MS does not automatically imply stronger or weaker NK-cell regulation.
Candidate complexes must be tested for binding to CD94/NKG2A and CD94/NKG2C, followed by NK-cell degranulation or cytotoxicity assays. The key success criterion is not simply “more peptides,” but the identification of peptides that are reproducibly presented with wild-type β2m and measurably alter immune-receptor signaling.
This project turns a computational β2m-variant hypothesis into a biomanufacturing problem with a concrete product, chassis, automated workflow, and capacity estimate.
The 20-member HLA-E capture panel is intentionally a discovery tool rather than a therapeutic. Its value would come from making unstable peptide–HLA-E states experimentally visible, while the wild-type reconstruction step protects against the central risk that engineering the complex also engineers the answer.
O’Callaghan, C. A., et al. (1998). Structural features impose tight peptide binding specificity in the nonclassical MHC molecule HLA-E. Molecular Cell, 1(4), 531–541. https://doi.org/10.1016/S1097-2765(00)80053-2
Walters, L. C., et al. (2018). Pathogen-derived HLA-E bound epitopes reveal broad primary anchor pocket tolerability and conformationally malleable peptide binding. Nature Communications, 9, 3137. https://doi.org/10.1038/s41467-018-05459-z
Huisman, B. D., et al. (2023). High-throughput characterization of HLA-E-presented CD94/NKG2x ligands reveals peptides which modulate NK cell activation. Nature Communications, 14, 4809. https://doi.org/10.1038/s41467-023-40220-1
Tan, E., et al. (2021). HEK293 cell line as a platform to produce recombinant proteins and viral vectors. Frontiers in Bioengineering and Biotechnology, 9, 796991. https://doi.org/10.3389/fbioe.2021.796991
Asin-Garcia, E., Fawcett, J. D., Batianis, C., & Martins dos Santos, V. A. P. (2025). A snapshot of biomanufacturing and the need for enabling research infrastructure. Trends in Biotechnology, 43(5), 1000–1014. https://doi.org/10.1016/j.tibtech.2024.10.014
Petzold, C. J., & Mukhopadhyay, A. (2025). From bench to biofactory: High-throughput technologies and automated workflows to accelerate biomanufacturing. Current Opinion in Biotechnology, 94, 103320. https://doi.org/10.1016/j.copbio.2025.103320
Weitzen, M., et al. (2024). Deciphering the HLA-E immunopeptidome with mass spectrometry. Frontiers in Immunology, 15, 1442783.
AI client: OpenAI ChatGPT Work (Codex), accessed September 17, 2026.
I supplied the HLA-E/β2m project abstract and used AI to help translate it into the assignment format, perform transparent back-of-the-envelope calculations, identify relevant literature, and edit the document. I reviewed and take responsibility for the final scientific content and calculations.
Prompts used:
_index.md webpage.”