How to Design a DLL3 × CD3 Functional Screening Workflow

A DLL3×CD3 T-cell engager (TCE) must activate T cells, kill DLL3+ tumor cells, and do so without an unacceptable cytokine profile. This guide makes each decision in that pipeline explicit. Clinical anchor: in DeLLphi-301, tarlatamab delivered a 40% confirmed ORR at 10 mg Q2W in relapsed SCLC, with CRS the most common adverse event — functional screening is where therapeutic index starts to be defined.

Decision 1 — Cell system: which DLL3+ lines and controls

DLL3 is surface-exposed on SCLC and neuroendocrine tumors but largely intracellular in normal adult tissue — the basis of the on-target/on-tumor window. Typical DLL3-high SCLC lines include SHP-77, NCI-H82, NCI-H524 and NCI-H2171; a DLL3-low line (NCI-H69-class) gives an intermediate density tier. Quantify surface density by flow cytometry with a reference anti-DLL3 antibody and report receptors/cell, because potency tracks density.

Choose X when… choose a DLL3-high line (typ. ~5,000–50,000 receptors/cell; titrate per system) as the primary efficacy matrix; add a DLL3-low line when ranking molecules by density sensitivity; always include a target-negative control — a non-neuroendocrine line or, better, an isogenic DLL3-knockout — to define the specificity floor. Heterogeneity caveats: surface DLL3 varies across and within lines and drifts with passage and media; lock passage ranges, bank a validated reference lot, re-measure density at assay start.

Decision 2 — T-cell activation readout choice

CD69 and CD25 are the standard activation markers — CD69 peaks at 4–24 h, CD25 at 24–48 h after engagement. Primary-cell flow readouts give translationally relevant potency; NFAT-luciferase Jurkat reporters give throughput. CD107a adds a degranulation proxy linking activation to killing in a 4–6 h window.

Choose X when… choose the NFAT-Jurkat reporter for high-throughput rank-ordering of dozens of leads; choose CD25/CD69 on primary cells (pan-T or CD8-enriched) when the readout must predict killing potency — CD69 at 4–24 h for early kinetics, CD25 at 24–48 h for the plateau; add CD107a for a killing-linked signal in the same window. Gate live lymphocytes and subtract effector-only and antibody-only background.

Decision 3 — Redirected killing assay design (E:T, time, detection)

Endpoint LDH release is the workhorse: label-free, plate-reader-compatible, simple. Calcein-AM is more sensitive but needs labeling and shows higher background. Real-time platforms (impedance such as xCELLigence, or Incucyte imaging) resolve kinetics and separate fast from slow killers. Typical: E:T 5:1–10:1 (range 1:1–20:1); 16–24 h for LDH endpoints; 24–72 h for real-time runs; titrate per system.

Choose X when… choose LDH at 16–24 h, E:T 5:1–10:1, for a fast first screen; calcein-AM for shorter, more sensitive windows; real-time killing when kinetic resolution or sub-pM potency discrimination matters, or when spontaneous target death biases endpoint reads. Mandatory controls: target-only (spontaneous), effector-only, Triton X-100 maximum release, and a CD3-only or untargeted TCE arm to prove lysis is DLL3-dependent. Report corrected specific lysis = (test − spontaneous)/(max − spontaneous).

Decision 4 — Cytokine release and CRS-relevant readouts

IFNγ, TNFα and IL-2 are the core T-cell outputs; add IL-6, IL-10 (and IL-8) when building a CRS-relevant profile. Clinically, CRS was the most common adverse event in DeLLphi-301 (~51% of patients at 10 mg, predominantly grade 1–2) — screening cytokine spikes early, not just killing, flags therapeutic-index risk. Detect with single-plex ELISA for cheap throughput, or MSD/Luminex multiplex for the full panel from one well.

Choose X when… choose IFNγ single-plex as the default screening readout; move to a multiplex panel once leads pass the potency benchmark, to expose high TNFα/IL-6 profiles that predict CRS risk. Compare candidates at matched DLL3 density and E:T, since cytokine output scales with both.

Decision 5 — Potency benchmark and go/no-go thresholds

Killing and activation EC50s for this TCE class typically land in the pM range (single-digit to tens of pM), with killing usually more sensitive than cytokine release. Treat these as typical ranges — titrate per system — and anchor every run to a reference bispecific.

Choose X when… set a go threshold of, e.g., killing EC50 within ~10-fold of your reference (or an absolute pM target you define); require specificity — activity on the target-negative line below ~10–20% of maximum or no shift vs effector-only; advance only molecules meeting both, then confirm the top 2–3 in primary T-cell activation and the multiplex cytokine panel. Reject molecules whose lysis EC50 is good but whose cytokine/lysis ratio implies an unfavorable profile.

Tool set this implies

• Reagents: recombinant DLL3 ECD and CD3ε (or anti-CD3 such as OKT3) for binding/blocking controls and reporter engagement.
• Effector cells: PBMCs (pan-T or CD8-enriched) and/or an NFAT-Jurkat reporter line.
• Target lines: DLL3-high, DLL3-low, and DLL3-negative (wild-type non-neuroendocrine or DLL3-KO).
• Detection: LDH or calcein-AM kits; MSD/Luminex panels; flow antibodies for CD25, CD69, CD107a and anti-DLL3 for density; real-time imager or impedance system if kinetics matter.

Escalate from binding to functional when… affinity or on-cell binding IC50 no longer discriminates lead pairs; on-cell binding looks strong but relative potency is unpredictable; or you need a potency rank order for the in vivo pick. Binding answers "does it engage"; functional answers "does it kill, and at what cost".

References

• Ahn MJ, et al. Tarlatamab, a First-in-Class DLL3-Targeted Bispecific T-Cell Engager, in Recurrent Small-Cell Lung Cancer (DeLLphi-301). *N Engl J Med* 2023;389:1693–1704. https://www.nejm.org/doi/full/10.1056/NEJMoa2307980 — clinical context: 40% ORR; CRS most common AE.
• Vendor technical notes: LDH (Promega CytoTox-96), calcein-AM, real-time killing (Agilent xCELLigence; Sartorius Incucyte), flow activation panels (BD, BioLegend), multiplex cytokine (MSD).