引擎输出 · CRO RFP

CRO Technical Package — Clean-IP anti-DLL3 VHH for Radioligand Therapy

Ready-to-send RFP for a CRO display campaign to produce a clean-IP, conjugation-ready anti-DLL3 single-domain antibody (VHH) as the targeting vector for a DLL3 radioligand therapy (RLT) in SCLC. Synthesised from a 4-expert panel (computational design / RLT-radiochem / antibody-discovery + FTO / pharma R&D). The AI engine supplies the strategy + epitope + FTO + in-silico maturation; the CRO supplies the wet-validated molecule. Together = a validated lead (the credible asset).

1. Objective

A monomeric, thermostable, sub-nanomolar VHH binding native cell-surface human DLL3 at a

non-DSL EGF-domain epitope, with a defined C-terminal conjugation handle, full IP assignment

to us, and an FTO-supporting epitope-bin map vs the known competitors.

2. Why VHH + this epitope (the design constraints — non-negotiable)

• Format = VHH/sdAb (~13–15 kDa). Same fast-clearance RLT physics as the peptide (Novartis

ETN029) and DARPin (Molecular Partners MP0712) competitors, but a **distinct composition-of-matter

class** outside their platform IP; clean C-terminal site-specific chelator conjugation; mature

CMC/humanisation.

• **Epitope = membrane-proximal EGF domains (target EGF4, res ~337–374; acceptable EGF5/EGF6),

NOT the DSL domain. Rationale: (a) the DSL domain is the most patent-dense** region —

US 9,334,318 claims an epitope at DSL G203/R205/P206; (b) DSL is the conserved Notch-binding

face; (c) membrane-proximal EGF epitopes give better RLT crossfire geometry and are

comparatively under-claimed white space. (Our AlphaFold analysis: EGF4 is the highest- and

most-uniformly-confident non-DSL EGF domain — mean pLDDT 80, min 66.)

3. Antigen / reagent package (CRO to produce or we co-supply)

• Human DLL3 ECD (full, positive control) and EGF4–6 sub-domain fragment(s) for

epitope-steered panning + DSL/MNNL counter-selection.

• Cynomolgus DLL3 ECD for tox-species cross-reactivity.
• Cell lines: DLL3-high SCLC (e.g. NCI-H82 / NCI-H69) and a DLL3-null line for specificity.

4. Campaign

Epitope-steered synthetic / naïve VHH phage or yeast display (lead), with **immune (llama/alpaca)

VHH as a parallel biological backstop. Pan on EGF4–6 fragment; counter-select against DSL/MNNL**

to force the binders off the patent-dense epitope.

5. Deliverables (write into the SOW)

• ≥ 5–10 sequence-unique VHH leads, each epitope-binned (SPR/BLI cross-block) against

reference binders SC16/Rova-T, tarlatamab DLL3 arm, ETN029-class, MP0712-class — a written

FTO-relevant bin map; keep only non-overlapping binders.

• Affinity (KD by SPR/BLI), monomeric, target single-digit → sub-nM.
• Binding to native cell-surface DLL3 (flow on DLL3⁺ SCLC) and no binding to DLL3-null.
• Cyno DLL3 cross-reactivity.
• Developability mini-panel: expression titre, thermostability (Tm), SEC % monomer,

polyreactivity (PSR), no surface-Cys conflict with the conjugation site.

• Engineered conjugation handle (C-terminal Cys / sortase LPETG / unnatural aa) on the top

2–3 leads, expressed + purified, conjugation-ready for the radiochemistry team.

• Full sequences + work-for-hire IP assignment (no platform reach-through royalty).

Optional add-ons: affinity maturation on the top lead; albumin-binder fusion for half-life

tuning (to match Ac-225/Lu-177); internalisation/residualisation assay (matters for the ADC follow-on

and for residualising-radiometal dose amplification).

6. Rough cost / timeline

~$80–180k, 3–5 months to conjugation-ready leads (first binders ~6–8 wk; the

epitope-binning + developability + tagging tail fills out to 3–5 mo). Affinity maturation +~$30–60k /

2–3 mo if needed.

7. What the AI engine does alongside the CRO (our differentiated contribution)

• Target + epitope selection (done): EGF4 membrane-proximal, confidence-gated, IP-white-space.
• In-silico FTO pre-screen of CRO hit sequences vs published claim CDRs (EPO OPS / USPTO ODP /

Lens) — flag anything near a Markush range before counsel review.

• Developability + humanisation optimisation of the leads.
• Affinity-maturation library design (in-silico-guided).
• Cross-model structural validation of each lead:DLL3 complex (AF3 + Boltz-2 consensus) — a far

stronger triage than single-model i_pTM.

• Conjugation-site selection to preserve affinity.

8. Next make-or-break step after a lead exists (cheapest de-risk)

One surrogate-isotope imaging biodistribution study ([¹¹¹In]/[⁶⁸Ga]/[²⁰³Pb]-labelled lead) in

DLL3-high and DLL3-low models — reads tumour uptake, kidney, marrow, and blood clearance

in a single study (10–50× cheaper than a therapy-efficacy build). This gates format + isotope before

any therapeutic build.

*Scope note: FTO conclusions require a biologics patent attorney; dosimetry thresholds require a

nuclear-medicine dosimetrist; isotope supply (Ac-225 / Pb-212) is the actual scarce resource and

needs a radiochem/isotope partner. The engine de-risks and narrows; it does not replace those.*