Produced by the Catalyst AI discovery engine (genomics → ADC-feasibility → modality-matching → target-biology gate), drug-hunter-reviewed, BD-critiqued, and current-landscape verified (2025–26). Asset-review structure: decision first, traceable evidence, honest risk. Engine numbers trace to source (Ch. 10).
**DLL3 is the cleanest *clinically-validated delivery antigen* in SCLC / neuroendocrine
cancer — tumour-high, normal-near-zero. The target risk is retired (tarlatamab now carries
full FDA approval). The question is no longer "can DLL3 be hit?" but "which modality wins."
Our answer: lead with a DLL3 radioligand (the most differentiated, best-target-fit lane);
position a next-gen ADC as a Best-in-Class / dual-payload / ISAC / TCE-combination play —
NOT as First-in-Class, because that lane is now actively contested.**
| Rank | Modality | Narrative | Competition (2026) |
|---|---|---|---|
| A | DLL3 radioligand / RIT (Ac-225 / Lu-177) | DLL3's ~0 normal expression natively removes RLT's #1 risk (irradiating normal tissue); most differentiated lane | early clinical (Novartis ETN029 Ph1; ABD-147) |
| B | DLL3 next-gen ADC — Best-in-Class / combo | Rova-T exposed *old PBD* window; ZL-1310 re-validated → lane moving "revival → competition" | frontrunner: ZL-1310 in Ph3 |
| C | DLL3 TCE | validated + regulatory-confirmed, crowded | tarlatamab (full approval) + ≥4 followers |
| — | DLL3 CAR-T/NK | emerging; solid-tumour CAR risk | early |
Engine: ADC-feasibility GO, biology validated, selectivity clean (no
sensitive-organ landmine). Confidence: biology HIGH; modality differentiation
MEDIUM (RLT best-fit but not virgin; execution risk = isotope/payload/format, not target).
The pitch, in one line: *DLL3 isn't a "can we drug it" bet — it's a "which warhead
wins" bet on a regulator-validated, exquisitely selective antigen. Lead radioligand; make
the ADC a best-in-class/combination asset, not a me-too first-in-class claim.*
surface-trafficked** in ASCL1-driven SCLC / high-grade neuroendocrine tumours. Surface
presentation is itself a tumour-defining event.
DLL3×CD3 TCE — accelerated approval 2024-05, then **TRADITIONAL (full) FDA approval
2025-11-19 for platinum-progressed ES-SCLC, on the confirmatory Ph3 DeLLphi-304**
(mOS 13.6 vs 8.3 mo vs chemo; 40% reduction in death risk). A target whose entire
mechanism is DLL3 surface binding now carries *full* approval — the target is validated
by clinic and regulator.
Open Targets antibody tractability = surface-accessible.
(fraction_dependent ≈ 0.001) — the cancer doesn't *need* DLL3 to survive. That is the
*correct* profile for a delivery modality (RLT / ADC / TCE), which exploits a clean
surface address label, not pathway addiction. For EGFR/FGFR2/RET you ask "does
inhibiting the pathway kill the cell?"; for DLL3 you ask "can we deliver a warhead
precisely to the SCLC cell?" — and the answer is yes.
accumbens 11.4, hypothalamus 8.6, amygdala 8.3) — behind the BBB, which a full-size
ADC/radioconjugate doesn't meaningfully cross. Peripheral normal ≈ 0.
intracellular-normal / surface-tumour duality *is* the window.
FGFR2 → GI/skin + FGFR-pathway tox + must be the FGFR2b isoform; **DLL3 → tumour~100 /
normal~0–1.** For a cytotoxic-payload or radioligand modality, that is the dream profile.
trap, cardiotox, shed antigen, or haematopoietic expression — unlike KIT, Ch. 8).
Rovalpituzumab tesirine (Rova-T, AbbVie) — DLL3 ADC — discontinued 2019. A BD
reviewer will challenge "how do you *prove* it was just the payload?" — so we don't claim
that. Evidence:
photosensitivity, peripheral oedema. Ph3 MERU (maintenance): no survival benefit →
AbbVie terminated the program.
| Candidate cause | Weight |
|---|---|
| Target (DLL3) | NOT the cause — later vindicated by tarlatamab's full approval |
| PBD-dimer payload | PRIMARY — narrow therapeutic index, characteristic toxicities |
| Linker stability / DAR | contributory |
| Dose window | contributory |
| Patient selection (DLL3-high enrichment immature) | contributory |
| DLL3 heterogeneity / antigen escape | contributory |
| ADC PK exposure + fast SCLC progression | contributory |
Honest conclusion: Rova-T's failure does not negate DLL3, but it also can't be
*reduced* to "just payload." The defensible statement: *the target was vindicated; the
old-generation PBD ADC exposed a multi-factor therapeutic-window problem (payload primary).*
This is what makes the next-generation story credible — and testable.
| Modality | Fit to DLL3 | 2026 status | Call |
|---|---|---|---|
| Radioligand / RIT | Best. RLT's worst fear = radiation into normal tissue; DLL3's ~0 normal expression natively avoids it. Doesn't depend on internalisation efficiency; α-emitters (Ac-225) give potent short-range double-strand breaks ideal for a surface-restricted antigen. | early clinical | A — flagship |
| Next-gen ADC | Strong, but contested. Modern camptothecin/topo-1 payload + better linker fix the Rova-T problem. | ZL-1310 in Ph3 | B — best-in-class / combo, NOT FIC |
| TCE / BiTE | Validated, payload-free (CRS / T-cell-fitness risk instead). | full-approved + crowded | C — saturating |
| CAR-T / NK | Clean CAR antigen, but solid-tumour delivery/persistence unsolved. | early | exploratory |
Why RLT first: an ADC has three failure points (internalisation + payload + linker);
a radioligand needs binding + tumour-restricted expression — exactly DLL3's twin
strengths — and is far more forgiving of internalisation. For a target whose single
greatest asset is *selectivity*, RLT converts that selectivity most directly into
therapeutic index.
RLT vs ADC — head-to-head decision table (DLL3 context):
| Dimension | ADC | RLT / RIT |
|---|---|---|
| Internalisation dependence | High — payload must be internalised + processed | Not absolute — radiation acts at/near the surface |
| Tolerance of antigen heterogeneity | Medium — relies on bystander payload diffusion | Higher — α/β crossfire kills antigen-low neighbours (covers the DLL3-low ~15%) |
| Dominant toxicity | payload / linker (the Rova-T failure mode) | kidney / marrow / liver radiation dose |
| Patient selection | IHC-based | theranostic imaging ([111In]/89Zr) — *built in* |
| Differentiation space (2026) | pre-empted by ZL-1310 (Ph3) | still earlier — the open lane |
| Hardest engineering problem | payload / linker / DAR | isotope / chelator / format / dosimetry |
This table is the spine of the RLT-first call: on a target whose weaknesses are
heterogeneity and a crowded ADC lane, RLT's crossfire, built-in patient selection, and
earlier competitive stage are *exactly* the levers that matter.
followers: obrixtamig (BI 764532), peluntamig (PT217, DLL3×CD47), QLS31904,
multispecific HPN328.
(Zai Lab, TMALIN platform) — humanised anti-DLL3 + cleavable linker + camptothecin
(topo-1) payload (not PBD). Ph1 2L ORR 68% (1.6 mg, n=53); Global Ph3 ZL-1310-003**
initiated in 2L+ SCLC; 1L SCLC + NEC heading registrational in 2026; FDA Fast Track;
Ph1b combination with tarlatamab ± durvalumab (Zai/Amgen). Also SHR-4849, FZ-AD005.
→ a Gen-3 ADC must now win best-in-class (payload/linker/DAR/bystander) or
combination, not "first."
macrocyclic peptide with a DOTA chelator, a theranostic pair ([111In]In-ETN029
SPECT imaging + [225Ac]Ac-ETN029 therapy), Ph1 NCT07006727 in advanced DLL3+ solid
tumours (lung/pancreatic/prostate/NEC); ABD-147 (Abdera, Ac-225); 89Zr/177Lu
theranostic work in the literature. → early but NO LONGER EMPTY. The opportunity is
not First-in-Class by absence, but **differentiation by format, isotope, chelator,
dosimetry, and DLL3-high patient-selection strategy** — i.e. an early competitive zone
where the best-engineered radioligand wins, not an uncontested one.
Net: the ADC lane has moved from "revival" to "competition" (ZL-1310); the radioligand
lane is an early competitive zone with the best target-fit. The flagship is the
radioligand, won on engineering/selection — not on being first.
A pharma reviewer's first question: *why DLL3 over B7-H3, SEZ6, TROP2, CEACAM5?*
| Target | Normal-tissue expr | SCLC expr | Validation | ADC | RLT fit | Biggest risk |
|---|---|---|---|---|---|---|
| DLL3 | very low (CNS-only, BBB) | high, ASCL1 subtype | TCE full-approved | high (ZL-1310) | very high (clean window) | heterogeneity / antigen escape |
| B7-H3 (CD276) | broader / normal background | high + uniform across subtypes | many programs | high | medium | normal-tissue background → window |
| SEZ6 | neural | highest + most SCLC-specific (ASCL1/NeuroD1) | early (ABBV-706) | med-high | medium | CNS / neural expression |
| TROP2 (TACSTD2) | substantial normal | mid-high | ADC validated (Dato-DXd) | medium | low-med | tox / crowded |
| CEACAM5 | GI background | partial | validated (tusamitamab) | medium | medium | GI on-target tox |
Same-cohort hard data (George 2015 SCLC, n=81, cBioPortal RNA-seq — we ran it): %near-
negative (<1 RPKM) — B7-H3 0%, SEZ6 12%, DLL3 15%, TROP2 17%, CEACAM5 36%.
B7-H3 is the most *uniform* antigen; DLL3 carries a ~15% negative tail; CEACAM5 is the most
heterogeneous. Quantifies the trade-off below.
Read: DLL3 wins on therapeutic window (B7-H3 is more uniform but dirtier in normal;
SEZ6 is most specific but neural; TROP2/CEACAM5 carry normal-tissue tox). DLL3's *one*
relative weakness is heterogeneity (Ch. 7) — which is exactly why **bispecific delivery
(DLL3×B7-H3, DLL3×SEZ6) is an emerging hedge** and why RLT (with bystander crossfire) is
attractive. This chapter is what turns "DLL3 is good" into "DLL3 is the priority delivery
antigen, eyes open on heterogeneity."
The questions that decide real commercial value:
(U Cologne, *Nature*; n=81; cBioPortal RNA-seq RPKM): **DLL3 median 6.2, range 0–37.2;
~51% ≥ median, but ~20% DLL3-low (<25% of median) and ~15% near-negative (<1 RPKM).**
That ~15–20% is a real primary-resistant fraction → companion-diagnostic patient
selection is not optional. The RLT theranostic pair ([111In]/89Zr DLL3 PET-SPECT)
*natively provides* that selection — a structural advantage. DLL3 is predominantly the
ASCL1 transcriptional subtype.
CELLxGENE Census, SCLC, 246,800 single cells: **59% of the malignant *epithelial* cells
are DLL3⁺ (and ~28% DLL3-high) — i.e. the tumour-cell-positive fraction lands in the same
~50–60% band as the George bulk ~51% ≥median**, now corroborated across two independent
platforms (bulk RNA-seq *and* single-cell) instead of one n=81 cohort. **Crucially, DLL3 is
near-absent from the non-tumour compartment — ~1% of T cells, ~2% of CD8 memory T cells, ~0%
neutrophils — confirming DLL3 is tumour-epithelial-restricted within the TME**, not on
immune/stromal cells. This sharpens two things: (a) the ~40–60% treatable fraction is robust,
not a single-cohort artifact; (b) the selectivity is clean at single-cell resolution, which
the bulk GTEx normal-tissue atlas (CNS-only) could not see inside the tumour microenvironment.
81 patients* — B7-H3 (CD276) 0% negative / 2% low (uniformly high, dirtier window);
SEZ6 12% neg; TROP2 17% neg; CEACAM5 36% negative (most heterogeneous). This is the
quantitative basis for DLL3×B7-H3 bispecific delivery — cover the DLL3-low 15–20%
with the uniformly-expressed antigen. (Caveat: RNA proxy, not surface protein — confirm
by IHC/flow; single cohort.)
it). George 2015, Spearman: DLL3 ~ ASCL1 ρ = 0.71 (DLL3 median 10.2 in
ASCL1-high vs 2.5 in ASCL1-low — a 4× split); DLL3 ~ POU2F3 ρ = −0.37**, DLL3 ~ YAP1
ρ = −0.22 (DLL3 is LOW in the non-NE POU2F3 / YAP1 subtypes). **So the DLL3-negative ~15%
are predictably the non-ASCL1 subtypes** — patient selection is *biologically tractable*
(subtype-stratify, or image), not a coin toss.
(independent)** in the same patients. B7-H3 is subtype-agnostic (uniform); DLL3 is
ASCL1-restricted. Their independence means B7-H3 covers exactly the non-ASCL1 tumours
DLL3 misses — complementary, not redundant. That is the mechanistic case for a
DLL3×B7-H3 bispecific, from real co-expression data. **Positioning (keep the story clean):
DLL3 RLT is the LEAD asset; the DLL3×B7-H3 bispecific is a FOLLOW-ON / heterogeneity-hedge
/ platform-expansion asset — not the current flagship.** Lead = DLL3 RLT; follow-on =
DLL3×B7-H3 bispecific (RLT or ADC).
fraction; quantify surface density + %positive cells by flow/IHC, not bulk RNA.
plasticity / DLL3-low subclone outgrowth is the central escape risk — and the key
question for sequencing a DLL3 ADC/RLT after a DLL3 TCE (does prior tarlatamab select
DLL3-low disease?). This is an open, decisive question for positioning.
RLT advantage), bispecific delivery to cover heterogeneity, and α-emitter **bystander/
crossfire** killing of antigen-low neighbours (an RLT strength an ADC lacks).
Treatable-population framework (the real market-size question). The commercial question
is not "does DLL3 express?" but "how many patients express it *high enough* to treat?" —
which is a *threshold* decision. A DLL3-high definition must be set as one of: **IHC
H-score ≥ X, or ≥ 50% of tumour cells membrane-DLL3⁺, or a PET/SPECT SUV ≥ X**
(the theranostic route). The choice swings the market several-fold. Illustrative
sensitivity from the George cohort (RNA-seq RPKM proxy — IHC/PET will shift absolutes, the
*shape* holds):
| DLL3-high threshold | % SCLC addressable |
|---|---|
| any positive (>1 RPKM) | 85% |
| ≥ 25% of cohort median | 80% |
| ≥ median | 51% |
| high-expresser (≥ 2× median) | 23% |
| *single-cell:* tumour (epithelial) cells DLL3⁺ | 59% (CELLxGENE, 247k cells) |
| *single-cell:* DLL3-high overall | 28% |
→ The program's single most value-determining decision is **where the DLL3-high bar is
set** (a permissive ≥-positive label = ~85% of SCLC but risks treating low-antigen
non-responders; a stringent high-expresser label = ~23% but enriches response). Defining
this cut-off — ideally via the theranostic SUV (treat what you can image) — *is* the
market-sizing and the registrational-enrichment strategy. (Caveat: RNA proxy; convert to
IHC/PET before quoting absolutes.)
The same engine that surfaced DLL3 ranked KIT #1 on a naïve read (surface, internalising,
low competition, high score) — but **missed that KIT sits on haematopoietic stem cells →
myelosuppression**: the AI-scoring trap where "where is it expressed in *normal*?" is the
question a score doesn't ask. We patched KIT/FLT3/PDGFRB into the toxicity-landmine set, and
we surface this openly: **the engine narrows and reasons; curated pharmacology + human
review still catch residual landmines.** That honesty is the product, not a flaw.
Known engine limits on *this* pick: tumour-overexpression is a bulk-RNA/IHC proxy (not yet
single-cell/protein); "open lane" is a lower bound; heterogeneity (Ch. 7) is under-quantified
until the Ch. 9 data pull.
(CELLxGENE / Tabula Sapiens / GEO) → per-cell %DLL3⁺ tumour vs normal, ASCL1-subtype
stratification, surface-vs-intracellular split, and post-treatment DLL3 dynamics from
published series. Quantifies the window *and* the heterogeneity risk (Ch. 7).
(the Rova-T patient-selection lesson).
supply-constrained) vs Lu-177 (β, safer, supply-mature); format = peptide / minibody
/ Fab vs full IgG (Novartis chose a macrocyclic peptide** for ETN029 — fast tumour
uptake + renal clearance, less marrow dwell; a deliberate, instructive choice).
not de-novo binder (validated anti-DLL3 binders exist).
Kill on cheap public data before any radiochemistry capex.
A radioligand is not "an antibody with Lu-177/Ac-225 bolted on." The real questions:
| Question | Why it's a killer |
|---|---|
| Format: full IgG vs minibody / Fab / peptide? | Long-circulating IgG → high marrow/normal dose + slow; SCLC progresses fast. Novartis's peptide (ETN029) choice signals the field is going small/fast for renal clearance. |
| Dose-limiting organ — kidney / marrow / liver? | Peptides → renal dose-limiting (mitigate w/ amino-acid co-infusion); IgG → marrow. Defines the deliverable dose, hence efficacy. |
| Ac-225 (α) vs Lu-177 (β)? | Ac-225: potent short-range crossfire (ideal for heterogeneity) but supply-chain-constrained + daughter-recoil. Lu-177: safer, mature supply, but less potent and longer range. |
| Half-life vs disease speed | Isotope physical half-life must match tumour residence + SCLC kinetics; mismatch = dose to normal tissue or under-kill. |
| Theranostic pairing | An imaging companion ([111In]/89Zr) for patient selection is a structural advantage DLL3-RLT has *built in* — use it. |
Design reference — read Novartis ETN029 closely. ETN029 is *not* an antibody — it is a
macrocyclic peptide + DOTA chelator. That format choice is a deliberate, instructive
signal of where DLL3-RLT design is going, and the reasoning is the heart of the program:
tumour delivery. SCLC progresses fast — a slow, long-dwelling radioconjugate fights
the disease clock and irradiates the marrow while it waits.
less marrow dwell, a cleaner dose to tumour, and a kinetic match for Ac-225 (a potent
α-emitter you want delivered and cleared quickly, not parked in circulation).
(mitigated by amino-acid co-infusion / linker design), whereas IgG shifts the limit to
marrow. So the format decision *is* the dose-limiting-organ decision.
The design question for a differentiated DLL3-RLT is therefore not "antibody + isotope" but
"which fast-clearing format + which isotope + which dose-protection strategy" — the same
logic that made [177Lu]/[225Ac]-PSMA small-molecule ligands win in prostate. Lead with a
small, fast format; do not default to a full IgG.
Answering these is the flagship program's first 6 months.
Make the flagship concrete — three buildable formats, screened head-to-head before any
isotope commitment:
| Design | Format | Isotope | Strength | Biggest risk |
|---|---|---|---|---|
| A | macrocyclic peptide (ETN029-like) | Ac-225 (α) | fast in/out, renal clearance, potent crossfire, differentiated | renal dose / affinity |
| B | minibody / Fab | Lu-177 (β) | moderate PK, more forgiving dosimetry, mature supply | potency may be insufficient |
| C | biparatopic small biologic | Ac-225 (α) | higher affinity/avidity + selectivity | engineering complexity |
| D | DARPin / small protein / peptide | Pb-212 (α, ²¹²Pb→²¹²Bi) | intermediate t½ between Ac-225 and Lu-177; a live competitive route, not theoretical | chelator/in-vivo daughter retention; supply |
Pb-212 is now a formal option, not an implicit one. Molecular Partners / Orano's [212Pb]Pb-MP0712 (a DLL3-targeted DARPin) is in Phase 1 (NCT07278479) with disclosed imaging showing tumour uptake + limited normal-tissue uptake — i.e. real DLL3 × Pb-212 competitive data already exists. Any serious DLL3-RLT comparison must put Pb-212 alongside Ac-225 and Lu-177, not treat it as an afterthought.
Recommended first experiment: *screen peptide (A) vs Fab/minibody (B) biodistribution +
tumour retention BEFORE committing to an isotope.* Format dictates the dose-limiting organ
(peptide→kidney, IgG/large→marrow) and the isotope match — so decide format on biodistribution
data, then run the isotope comparison across all three relevant nuclides — Ac-225 (fast,
potent α crossfire), Lu-177 (safer, supply-mature β), and Pb-212 (intermediate α, MP0712-
validated). The choice is not "which isotope is strongest" but which **format-PK × half-life ×
kidney/marrow dose** combination delivers a tumour-lethal dose within DLT-organ limits. Do
not default to a full IgG.
The next step is not a longer narrative — it is a data package that can kill or prove
the program in 6 months.
GO if:
pre-set thresholds. *(Thresholds to be finalised by the dosimetry consultant before the
animal study; the **primary gating metric is the deliverable tumour absorbed dose without
exceeding kidney/marrow organ-at-risk limits** — not a single ratio in isolation.)*
blood clearance**.
overrun).
NO-GO / KILL if:
positioning.
This is a commercial-positioning question, not only a biology one, and deserves its own
line because tarlatamab is becoming standard 2L SCLC. A DLL3-RLT must have an explicit answer
to *where it sits relative to a DLL3 TCE*:
on its home turf; needs a differentiation/biomarker rationale.
analyses show acquired tarlatamab resistance is frequently driven by DLL3 *downregulation*
via lineage plasticity (loss of ASCL1, emergence of YAP1; low DLL3 by multiplex IHC at
progression — medRxiv 2026.01; *J Thorac Oncol* 2026, "Low DLL3 expression… drive resistance
to tarlatamab"). So a DLL3-RLT after tarlatamab inherits the same antigen-loss problem.
Viable only in the subset that progresses with DLL3 retained — which must be re-biopsy /
PET-confirmed per patient, never assumed.
does not depend on T-cell fitness, so prior IO exhaustion, low T-cell infiltrate, or
TCE-contraindicated patients are a natural, non-competing niche.
Revised positioning (data-driven): the post-tarlatamab DLL3-loss data **moves the centre of
gravity to lanes 1 and 3 — DLL3-high, T-cell-independent settings — and makes PET/IHC
patient selection non-optional** (you cannot assume DLL3 is present post-TCE). The package's
post-tarlatamab DLL3-retention readout sizes the residual lane-2 subset; it does not open the lane.
Methods footnote (the cohort analysis — "we ran it", reproducibly):
`
Dataset: George et al. 2015 SCLC, U Cologne (cBioPortal study sclc_ucologne_2015)
Expression: RNA-seq, profile sclc_ucologne_2015_rna_seq_mrna (RPKM); not log-transformed
N: 81 tumours
Near-negative: < 1 RPKM ; Low: < 25% of cohort median ; High: >= median (or >= 2x median)
Correlation: Spearman rho, pairwise-complete samples (n>=10)
Caveat: RNA proxy — NOT membrane protein; confirm by IHC / flow / DLL3-PET before
quoting absolutes. Single cohort (clcgp/jhu are mutation-only, no RNA).
`
| Claim | Source |
|---|---|
| Surface antigen / intracellular-normal | Open Targets tractability + HPA subcellular |
| Normal expr CNS-only, BBB-protected | GTEx v8 median |
| Not a dependency (antigen) | DepMap CRISPR via Open Targets |
| Tarlatamab full approval 2025-11, DeLLphi-304 (13.6 vs 8.3 mo) | FDA / MSK / OncLive / ASCO Post 2025 |
| ZL-1310/zoci: TMALIN, camptothecin payload, Ph3, ORR 68%, Amgen combo | Zai Lab IR / AACR / CancerNetwork 2025 |
| ETN029: Novartis macrocyclic-peptide theranostic, Ph1 NCT07006727 | Novartis / ClinicalTrials.gov |
| SCLC target comparison + DLL3=ASCL1 heterogeneity | JTO/ScienceDirect 2026; npj Precision Onc; AACR |
| DLL3/B7-H3/SEZ6/TROP2/CEACAM5 SCLC expression %neg (hard data) | cBioPortal — George 2015 SCLC (U Cologne), RNA-seq RPKM, n=81 (we ran it) |
| DLL3~ASCL1 ρ=0.71, ~POU2F3 −0.37, ~B7-H3 −0.01 (subtype mechanism) | cBioPortal George 2015, Spearman (we ran it). NB: clcgp/jhu cohorts are mutation-only, no RNA |
| ADC GO / clean landmines / KIT contrast | Catalyst ADC-feasibility + target-biology gate |
Re-scored (founder calibration): biology 9/10 · therapeutic window 8.5 · TCE validation
9 (full approval) · ADC opportunity 7 (contested, best-in-class/combo) · **RLT opportunity
8.5 (flagship)** · competition risk 6.5 (ZL-1310 + RLT entrants + TCE crowding) · memo
completeness 8.5.
Final pitch: *DLL3 is the cleanest, regulator-validated delivery antigen in SCLC / neuroendocrine cancer. The TCE lane is validated and crowded; the ADC lane was re-opened by ZL-1310 but is now competitive; the most differentiated direction, and the one that best fits DLL3's biology, is a DLL3 radioligand / RIT. The program is won or lost not on the target but on format, isotope, chelator, dosimetry, and DLL3-high patient selection.* Next step is not more story — it is the 6-month data package (Ch. 10c) that can kill or prove the program: DLL3-PET dosimetry (tumour-to-kidney/marrow), the ≥50%-DLL3-high treatable floor, and peptide-vs-Fab biodistribution before isotope commitment. Lead asset = DLL3 RLT; follow-on = DLL3×B7-H3 bispecific.