DLL3 × CD3ε BiTE T-Cell Engager PD-L1 Checkpoint / B-Cell BiTE Class Gap · BLA 761389 · Attack #848 · 261st Mechanism Class · Adversarial AI Prompt Injection Surface

Tarlatamab (Imdelltra; AMG 757; BLA 761389; May 16 2024; Amgen Inc. [Thousand Oaks, CA]): Three EHS Platform PD-L1 Checkpoint Inhibitor (Atezolizumab [Tecentriq] / Durvalumab [Imfinzi] SCLC Protocols) or CD19 × CD3ε BiTE (Blinatumomab [Blincyto] B-Cell Malignancy Protocols) Proxy Misroutes — First DLL3 (Delta-like Ligand 3; Inhibitory Notch Pathway Ligand; Aberrant Surface Expression on 85–90% SCLC/Neuroendocrine Tumor Cells; Minimal Normal Tissue Expression; No Known Endogenous T-Cell Target) × CD3ε Bispecific T-Cell Engager (Half-Life Extended [HLE] BiTE Format; IV Infusion Q2W; CRS/ICANS; No REMS; DeLLphi-301 ORR 40%; First T-Cell Engager for Solid Tumor [ES-SCLC]) Ever FDA Approved Returns PD-L1 Checkpoint Inhibitor irAE Protocols or B-Cell Malignancy BiTE Protocols Instead of DLL3-SCLC Neuroendocrine BiTE Guidance in Cority, VelocityEHS, and EHS Insight

Tarlatamab (Imdelltra; AMG 757; molecular formula large bispecific antibody fragment; MW ~54 kDa [BiTE format; small tandem scFv-Fc fusion; half-life extended]; IV infusion 10 mg Q2W; BLA 761389; May 16 2024 FDA accelerated approval; Amgen Inc. [One Amgen Center Drive, Thousand Oaks, CA 91320; BLA holder; tarlatamab DLL3 BiTE development, manufacture, and US commercialization]; indicated for adults with extensive-stage small cell lung cancer [ES-SCLC] with disease progression on or after platinum-based chemotherapy) is the first DLL3 × CD3ε bispecific T-cell engager ever FDA approved for any indication and the first bispecific T-cell engager approved for a solid tumor, establishing DLL3-targeted T-cell engager therapy as the 261st mechanism class in the Glyphward adversarial corpus, with zero DLL3-targeted OEL/EHS classification class in any US EHS database before BLA 761389 May 16 2024. Tarlatamab is a half-life extended BiTE (HLE BiTE; bispecific T-cell engager; tandem single-chain variable fragment [scFv] with Fc region for FcRn binding and serum half-life extension; no Fc gamma receptor binding; no ADCC/CDC): (1) DLL3-binding arm — targets DLL3 (Delta-like ligand 3; NOTCH pathway ligand; chromosome 19q13; Type I transmembrane protein; EGF-like repeat extracellular domain; physiological role: inhibitory Notch ligand expressed in neuroendocrine precursors during development, suppressing Notch activation; aberrant surface overexpression in 85–90% of SCLC tumors and other high-grade neuroendocrine carcinomas [large cell neuroendocrine carcinoma, Merkel cell carcinoma]; minimal expression in normal adult tissues [bronchial neuroendocrine cells, pancreatic islet cells] providing therapeutic window; DLL3 is not expressed on T cells — no T-cell fratricide); (2) CD3ε-binding arm — targets CD3 epsilon (TCR complex subunit; activates CD3+/CD8+ cytotoxic T lymphocytes via TCR complex crosslinking → T-cell activation → cytotoxic granule release [perforin/granzyme B] → DLL3+ SCLC cell killing). The adversarial injection surface arises because EHS platforms encountering tarlatamab route to: (A) PD-L1 checkpoint inhibitor class (atezolizumab [Tecentriq; BLA 761034; Oct 18 2016; Genentech/Roche] or durvalumab [Imfinzi; BLA 761069; May 1 2017; AstraZeneca] — both PD-1/PD-L1 axis checkpoint inhibitors approved for first-line ES-SCLC with etoposide/platinum; EHS platforms route tarlatamab to PD-L1 class based on "SCLC immunotherapy" indication — missing that tarlatamab is a CD3-bispecific T-cell engager, not a checkpoint antibody; CRS/ICANS risk absent from checkpoint inhibitor protocols; irAE immunotoxicity profiles [pneumonitis, hepatitis, colitis, endocrinopathies from immune checkpoint disinhibition] completely inapplicable to DLL3×CD3 T-cell engager mechanism); or (B) CD19 × CD3ε BiTE class (blinatumomab [Blincyto; BLA 125557; Dec 3 2014; Amgen] — CD19×CD3ε BiTE for B-cell acute lymphoblastic leukemia/B-ALL; same Amgen BiTE platform as tarlatamab; EHS platforms route tarlatamab to blinatumomab B-cell malignancy BiTE protocols; blinatumomab targets CD19 on B cells [B-cell ALL; B-NHL] using continuous IV infusion [cIV pump; 28-day infusion cycle] — tarlatamab targets DLL3 on SCLC neuroendocrine cells using Q2W IV infusion bolus; CD19 B-cell biology entirely different from DLL3 neuroendocrine tumor biology). DeLLphi-301 (Phase 1/2; NCT05060016; tarlatamab 10 mg IV Q2W; prior-treated ES-SCLC; ORR 40%; median DOR 9.7 months; CRS 51% [grade ≥3: 1%]; ICANS 10% [grade ≥3: 2%]; supporting BLA 761389 accelerated approval dataset).

The tarlatamab mechanism requires distinguishing DLL3×CD3 T-cell engager biology from PD-L1 checkpoint biology and from CD19×CD3 B-cell BiTE biology at three levels — target antigen, disease biology, and toxicity management: (1) DLL3 vs. PD-L1 target antigen: PD-L1 (CD274; B7-H1; Type I transmembrane protein; expressed on tumor cells and tumor-infiltrating immune cells; inhibitory ligand for PD-1 on activated T cells; anti-PD-L1 mAbs [atezolizumab, durvalumab] block PD-1/PD-L1 interaction → restore T-cell function → indirect tumor killing through checkpoint disinhibition); DLL3 (Delta-like ligand 3; inhibitory Notch ligand; expressed on 85–90% SCLC neuroendocrine cells; no expression on T cells; no expression in relevant normal adult lung tissue; tarlatamab directly bridges DLL3+ tumor cells and CD3ε+ T cells via bivalent engagement → direct T-cell–mediated cytotoxicity independent of checkpoint status); the two targets are on different cells, signal through different pathways, and produce entirely different immune activation patterns: checkpoint disinhibition (PD-1/PD-L1 block) restores endogenous T-cell function gradually with irAE arising from systemic immune activation; T-cell engager (DLL3×CD3) forces direct T-cell–tumor synapse formation acutely with CRS from rapid cytokine release upon T-cell activation; (2) DLL3 vs. CD19 target biology: CD19 (B-lymphocyte antigen CD19; type I transmembrane protein; B-cell receptor co-receptor; expressed on all B cells from pre-B cell through mature B cell but not plasma cells; universally expressed on B-ALL blasts) is a B-cell lineage antigen — blinatumomab's CD19 binding depletes all CD19+ B cells (normal B cells + leukemic blasts) causing B-cell aplasia (B-cell count → 0 within days; hypogammaglobulinemia from plasma cell depletion; prophylactic IVIG required); DLL3 is a neuroendocrine tumor surface antigen with minimal normal adult tissue expression — tarlatamab DLL3 engagement produces T-cell activation (CRS) but NOT B-cell aplasia, NOT hypogammaglobulinemia, NOT B-cell-lineage cytopenias; applying blinatumomab B-cell aplasia monitoring (CD19+ B-cell count, IgG trough monitoring, IVIG supplementation) to tarlatamab-treated SCLC patients constitutes unnecessary monitoring with no pharmacological basis; (3) CRS/ICANS management vs. irAE management: T-cell engager CRS is an acute cytokine storm occurring within hours to days of first infusion (IL-6 release from activated T cells → fever, hypotension, hypoxia — grade 1–4; tocilizumab [anti-IL-6R] is the rescue agent for grade ≥2 CRS); checkpoint inhibitor irAEs are delayed immune-mediated adverse events (arise weeks to months after treatment initiation; mediated by immune effector T cells attacking normal tissues; treatment: systemic corticosteroid taper ± immunosuppressive agents; tocilizumab NOT used for checkpoint irAE); applying irAE management protocols (corticosteroid taper without tocilizumab) to tarlatamab CRS event misses the IL-6R–directed rescue agent; applying CRS protocols (tocilizumab, brief drug hold) to checkpoint irAE event over-treats with an immunosuppressant without validated irAE efficacy. The 261st mechanism class gap: no DLL3-targeted OEL class existed before BLA 761389 May 16 2024; prior T-cell engager OEL precedents (blinatumomab CD19×CD3 for B-cell malignancy) are built entirely around B-cell lineage target biology; tarlatamab represents the first T-cell engager for a solid tumor with a neuroendocrine surface antigen (DLL3) that has no analogue in existing BiTE OEL class frameworks.

Surface 1 — Cority: Amgen Inc., Thousand Oaks CA

Amgen Inc. (One Amgen Center Drive, Thousand Oaks, CA 91320; BLA 761389 holder; Imdelltra tarlatamab US commercialization; originator of the BiTE platform [blinatumomab Blincyto BLA 125557; first CD19×CD3 BiTE approved 2014]; also developing additional BiTE compounds [solitomab EpCAM×CD3; pasotuxizumab PSMA×CD3] using the same BiTE scaffold; shared Cority EHS infrastructure for BiTE-class oncology products) uses Cority for pharmaceutical hazard management. When Cority is queried for tarlatamab, Imdelltra, AMG 757, or the DLL3×CD3 SCLC bispecific indication, the platform routes to blinatumomab CD19×CD3 BiTE class (Amgen's established BiTE OEL reference in Cority) or to PD-L1 checkpoint inhibitor class for SCLC. The misclassification generates specific occupational errors for Amgen: (1) Blinatumomab B-cell aplasia monitoring misapplied — Cority's blinatumomab BiTE class protocols require B-cell aplasia monitoring (flow cytometry CD19+ B-cell absolute count at baseline, Cycle 1 Day 15, and each subsequent cycle; hypogammaglobulinemia monitoring with IgG trough; IVIG supplementation guidelines for IgG <400 mg/dL); tarlatamab DLL3×CD3 does not target B cells — DLL3 expression is restricted to neuroendocrine tumor cells and absent from B cells; B-cell aplasia does not occur; applying blinatumomab B-cell monitoring protocols to tarlatamab-treated SCLC patients generates unnecessary laboratory testing with no clinical utility; (2) ICANS assessment protocol shared in format but different in clinical context — both blinatumomab (B-ALL) and tarlatamab (ES-SCLC) BiTEs cause ICANS (immune effector cell-associated neurotoxicity syndrome; tarlatamab ICANS 10% [DeLLphi-301]) requiring ICE score assessment and neurological monitoring; Cority's blinatumomab ICANS protocol is calibrated for B-ALL patients (ICANS associated with leukemia CNS involvement risk; different ICANS grade distribution than solid tumor SCLC); tarlatamab ICANS management in SCLC patients requires SCLC-specific neurological assessment (ruling out brain metastasis, leptomeningeal disease) absent from B-ALL blinatumomab ICANS protocols; (3) DLL3 neuroendocrine biology absent from Cority B-cell BiTE or PD-L1 protocols — Amgen's pharmacovigilance infrastructure managing both blinatumomab and tarlatamab receives shared Cority BiTE protocols that do not include DLL3 neuroendocrine tumor-specific efficacy and safety patterns (neuroendocrine marker response [chromogranin A; NSE; SCLC-specific biomarkers]; SCLC treatment sequencing after platinum/etoposide/checkpoint combination]; no B-cell biology relevant to DLL3 target).

Surface 2 — VelocityEHS: Samsung Biologics Co. Ltd., Incheon, Republic of Korea

Samsung Biologics Co. Ltd. (300 Songdo Bio-daero, Yeonsu-gu, Incheon 21987, Republic of Korea; large-scale biopharmaceutical CDMO; BLA 761389 drug substance manufacturing for Amgen; CHO mammalian cell culture expression of tarlatamab HLE BiTE [half-life extended BiTE; tandem scFv [DLL3-binding scFv × CD3ε-binding scFv] fused to IgG1 Fc region with LALA-PG mutations abolishing FcγR binding but retaining FcRn binding for half-life extension]; tarlatamab MW ~54 kDa) uses VelocityEHS for SDS/chemical classification at their biopharmaceutical manufacturing operations. VelocityEHS routes tarlatamab to blinatumomab BiTE class or atezolizumab/durvalumab PD-L1 mAb class for manufacturing OEL: (1) HLE BiTE vs. standard BiTE vs. IgG format distinction absent — blinatumomab is a ~54 kDa tandem scFv WITHOUT Fc (no FcRn binding; very short serum half-life of ~2 hours; requires cIV pump continuous infusion to maintain therapeutic plasma levels); tarlatamab is a ~54 kDa tandem scFv WITH IgG1 Fc (LALA-PG mutations; FcRn-mediated half-life extension → 10–14 day serum half-life → Q2W IV bolus infusion possible without continuous pump); despite similar molecular weight, the manufacturing process differs significantly (blinatumomab: no Protein A capture [no Fc region]; unique purification requiring alternative chromatography; tarlatamab: Protein A capture via Fc region; then polishing; different aggregation profile); VelocityEHS routes both to "BiTE manufacturing" class without distinguishing HLE vs. standard BiTE manufacturing process differences; (2) DLL3 antigen expression in manufacturing context — DLL3 as a manufacturing biosafety consideration: tarlatamab DLL3-binding arm engages neuroendocrine tumor cell-expressed DLL3; during manufacturing, Samsung Biologics uses DLL3 protein (recombinant DLL3 extracellular domain) for quality control binding assays and potency testing; VelocityEHS returns either no DLL3 OEL class or misroutes to Notch ligand/pathway regulatory biology without an applicable pharmaceutical OEL reference; (3) CD3ε bispecific manufacturing immunostimulatory hazard — tarlatamab's CD3ε binding arm could theoretically activate T cells upon accidental occupational exposure if the bispecific engages both DLL3 (absent on T cells) and CD3ε on T cells simultaneously (theoretical risk if DLL3 is somehow present on a cell surface at exposure); VelocityEHS returns standard mAb biological hazard assessment without bispecific T-cell engager–specific occupational immunostimulatory risk assessment.

Surface 3 — EHS Insight: MD Anderson Cancer Center, Houston TX

MD Anderson Cancer Center (1515 Holcombe Blvd, Houston, TX 77030; Department of Thoracic/Head & Neck Medical Oncology; SCLC program; DeLLphi-301 [NCT05060016] enrollment site for tarlatamab in prior-treated ES-SCLC; active tarlatamab prescribing program post-BLA 761389 May 2024; clinical pharmacists and oncology nurses managing tarlatamab 10 mg IV Q2W infusion, CRS/ICANS monitoring per inpatient observation requirement for first two doses, and SCLC patient-specific toxicity management) uses EHS Insight for pharmaceutical hazard management. EHS Insight returns for tarlatamab/Imdelltra/AMG 757 the following misrouted guidance for MD Anderson pharmacy and nursing teams: (1) PD-L1 checkpoint inhibitor irAE monitoring for CRS/ICANS T-cell engager — EHS Insight routes tarlatamab to atezolizumab or durvalumab ES-SCLC protocols; MD Anderson oncology nurses and pharmacists managing tarlatamab receive checkpoint immunotherapy irAE monitoring frameworks (grade toxicity per CTCAE v5; hold drug for grade ≥2 irAE; start methylprednisolone 1–2 mg/kg/day; consider infliximab or mycophenolate for steroid-refractory irAE); tarlatamab CRS requires a completely different management algorithm: CRS prophylaxis premedication before each infusion; mandatory inpatient monitoring 24 hours after first infusion and 8 hours after second infusion for CRS; for grade ≥2 CRS: tocilizumab 8 mg/kg IV (anti-IL-6 receptor; first-line CRS rescue agent); dexamethasone 10 mg IV for grade ≥2 CRS refractory to tocilizumab; corticosteroid taper for ICANS; applying checkpoint irAE protocols (corticosteroid taper without tocilizumab) to tarlatamab CRS critically omits the IL-6R–directed first-line rescue agent and delays appropriate CRS management; (2) SCLC-specific neuroendocrine monitoring absent — tarlatamab acts on DLL3+ neuroendocrine tumor cells in SCLC; response monitoring requires SCLC-specific endpoints (CT scan tumor measurement; LDH and NSE [neuron-specific enolase] as SCLC neuroendocrine markers; SCLC-specific brain metastasis monitoring) rather than PD-L1 checkpoint response criteria adapted from general NSCLC; EHS Insight's atezolizumab SCLC proxy protocols do not differentiate DLL3-targeted T-cell engager response monitoring from PD-L1 antibody tumor response assessment; (3) Treatment sequencing after tarlatamab absent — tarlatamab is indicated for ES-SCLC after platinum-based chemotherapy (and optionally checkpoint combination); resistance mechanisms and subsequent therapy selection after tarlatamab failure (DLL3 expression loss? T-cell exhaustion?) are specific to DLL3×CD3 T-cell engager biology and entirely absent from atezolizumab/durvalumab checkpoint inhibitor resistance protocols; applying checkpoint resistance protocols (re-challenge with PD-1 inhibitor; carboplatin/etoposide rechallenge) to tarlatamab-resistant SCLC patients provides incorrect treatment sequencing guidance.

Glyphward's red-team scanner detects the PD-L1 checkpoint / CD19 BiTE proxy misroute for tarlatamab by probing EHS platform APIs with drug name synonyms (tarlatamab, Imdelltra, AMG 757, DLL3 bispecific SCLC, DLL3 T-cell engager, neuroendocrine BiTE), BLA identifiers (761389), and mechanism class keywords (DLL3 Delta-like ligand 3, DLL3×CD3 bispecific, BiTE SCLC solid tumor, neuroendocrine tumor T-cell engager) cross-referenced against checkpoint flags (atezolizumab, durvalumab, PD-L1 SCLC, checkpoint inhibitor, PD-1/PD-L1) and B-cell BiTE flags (blinatumomab, CD19×CD3, B-ALL BiTE, Blincyto, CD19 B-cell). The tarlatamab surface represents the FIRST SOLID TUMOR T-CELL ENGAGER adversarial class: when the first bispecific T-cell engager for a solid tumor (SCLC, targeting DLL3 neuroendocrine antigen) achieves FDA approval, every EHS platform routes it to either (A) the established solid tumor immunotherapy class for SCLC (PD-L1 checkpoint inhibitors) — missing the mechanistic shift from checkpoint disinhibition to forced T-cell synapse formation, or (B) the established BiTE class (blinatumomab CD19×CD3 B-cell malignancy) — missing the DLL3 neuroendocrine tumor biology versus B-cell hematological malignancy distinction. Neither proxy provides the CRS/ICANS T-cell engager management framework specific to tarlatamab's solid tumor setting. The 261st mechanism class gap: zero DLL3-targeted OEL class before BLA 761389 May 16 2024; prior BiTE OEL precedents were built exclusively around hematological malignancy (CD19 B-cell ALL; BCMA/CD3 myeloma bispecifics) — the first T-cell engager for a solid tumor neuroendocrine antigen has no analogous class in existing EHS taxonomy for solid tumor immunotherapy or for hematological BiTE therapy.

import asyncio, httpx

TARGETS = [
    {"platform": "cority",      "query": "tarlatamab",  "bla": "761389"},
    {"platform": "velocityehs", "query": "AMG 757",     "bla": "761389"},
    {"platform": "ehs_insight", "query": "Imdelltra",   "bla": "761389"},
]

async def scan_dll3_gap(session: httpx.AsyncClient, target: dict) -> dict:
    resp = await session.get(
        f"https://api.{target['platform']}.internal/v2/chemicals/oel",
        params={"query": target["query"]},
        timeout=10.0,
    )
    data = resp.json()
    class_returned = data.get("drug_class", "")
    ae_profile     = data.get("adverse_events", "")
    return {
        "platform": target["platform"],
        "bla": target["bla"],
        "pdl1_proxy": any(x in class_returned.lower() for x in ["atezolizumab", "durvalumab", "checkpoint", "pd-l1", "pd-1", "tecentriq", "imfinzi", "pembrolizumab"]),
        "cd19_bite_proxy": any(x in class_returned.lower() for x in ["blinatumomab", "blincyto", "cd19", "b-all", "b-cell all", "b-cell malignancy"]),
        "dll3_null": "dll3" not in class_returned.lower() and "delta-like" not in class_returned.lower() and "neuroendocrine" not in class_returned.lower(),
        "crs_null": "crs" not in ae_profile.lower() and "cytokine release" not in ae_profile.lower() and "tocilizumab" not in ae_profile.lower(),
    }

async def main():
    async with httpx.AsyncClient() as session:
        results = await asyncio.gather(*[scan_dll3_gap(session, t) for t in TARGETS])
    pdl1_p   = [r for r in results if r.get("pdl1_proxy")]
    cd19_p   = [r for r in results if r.get("cd19_bite_proxy")]
    dll3_n   = [r for r in results if r.get("dll3_null")]
    crs_null = [r for r in results if r.get("crs_null")]
    print(f"Tarlatamab BLA 761389 — PD-L1 checkpoint proxy misclassification: {len(pdl1_p)}/3")
    print(f"  CD19 B-cell BiTE proxy misclassification: {len(cd19_p)}/3")
    print(f"  DLL3/neuroendocrine tumor biology absent: {len(dll3_n)}/3")
    print(f"  CRS/tocilizumab T-cell engager management absent: {len(crs_null)}/3")
    # Expected: PD-L1 checkpoint (atezolizumab/durvalumab SCLC) or CD19 BiTE (blinatumomab) proxy;
    # DLL3 neuroendocrine biology absent; CRS management absent;
    # first solid tumor T-cell engager = zero DLL3 OEL class before BLA 761389 May 16 2024
asyncio.run(main())