Decoding topoisomerase-I payload cross-resistance and re-specifying antigen, warhead, linker, and DAR
Background. Datopotamab deruxtecan (Dato-DXd) pairs an anti-TROP2 antibody with DXd, a topoisomerase-I (TOP1) inhibitor. Patients with triple-negative breast cancer (TNBC) who progress on Dato-DXd have, by definition, failed the TROP2/TOP1-inhibitor axis and are at risk of payload-class cross-resistance to the other approved TOP1-inhibitor ADCs — trastuzumab deruxtecan and sacituzumab govitecan.
Approach. A five-module engine (A: resistance decode; B: target selection with a tumor-versus-normal safety filter; C: payload/linker/DAR; D: structural assembly; E: competitive/regulatory landscape) was run end-to-end on live public data to specify a mechanistically orthogonal ADC.
Result. The engine recommends an anti–LIV-1 (SLC39A6) humanized IgG1 carrying MMAF (a tubulin inhibitor, mechanistically independent of the SLFN11/DNA-damage-response and P-glycoprotein-efflux resistance that defeats TOP1-inhibitor payloads) via a non-cleavable maleimidocaproyl linker at a site-specific DAR of 4, with a duocarmycin/PBD backup for low-density antigens. The design differentiates from the stalled LIV-1 precursor ladiratuzumab vedotin on warhead, linker chemistry, and conjugation homogeneity, and occupies an explicit non-TOP1i white space in the post-Dato-DXd line. Boltz-2 structure prediction (NVIDIA BioNeMo) confirmed a well-folded anti-B7-H3 Fab–antigen reference complex (interface ipTM 0.55) and, folding all three candidate ectodomains, revealed a structural axis that inverts the expression ranking — B7-H3 folds confidently (mean pLDDT 91) while the LIV-1 ectodomain is largely disordered (33), flagging epitope quality as the key risk to retire for the expression-leading target.
Antibody–drug conjugates (ADCs) have reshaped the treatment of metastatic breast cancer, but their sequencing has created a structural problem: three of the four ADCs approved in the breast setting carry the same payload mechanism. Trastuzumab deruxtecan and datopotamab deruxtecan both deliver DXd, and sacituzumab govitecan delivers SN-38 — all three are topoisomerase-I (TOP1) inhibitors of the camptothecin/exatecan class (Table 1). A patient who progresses on datopotamab deruxtecan (Dato-DXd) has therefore failed not only a specific drug but an entire payload class, and the biology that drives that failure — silencing of the DNA-damage sensor SLFN11, acquired TOP1 mutations, and drug efflux — is shared across the class.
The central design question is not "which ADC is best for TNBC" but "which ADC is orthogonal to the resistance the patient has already selected for." This is a per-patient, per-mechanism decision, and it motivates an engine that starts from the failed drug and works backward to a design whose antigen, payload mechanism, and chemistry share as little biology as possible with the resistance in play. Here we run that engine on the concrete case of TNBC refractory to Dato-DXd.
Table 1. Approved ADCs in the breast setting. Three of four share a topoisomerase-I payload (highlighted); a Dato-DXd–refractory patient has failed this axis.
| ADC | Target | Payload | Payload class | FDA approval | Application |
|---|---|---|---|---|---|
| Trastuzumab emtansine (T-DM1, Kadcyla) | HER2 | DM1 (maytansinoid, anti-microtubule) | microtubule (maytansinoid) | 2013-02-22 | BLA125427 |
| Trastuzumab deruxtecan (T-DXd, Enhertu) | HER2 / HER2-low | DXd (deruxtecan, TOP1 inhibitor) | TOP1 inhibitor | 2019-12-20 | BLA761139 |
| Sacituzumab govitecan (SG, Trodelvy) | TROP2 | SN-38 (TOP1 inhibitor) | TOP1 inhibitor | 2020-04-22 | BLA761115 |
| Datopotamab deruxtecan (Dato-DXd, Datroway) | TROP2 | DXd (deruxtecan, TOP1 inhibitor) | TOP1 inhibitor | 2025-01-17 | BLA761394 |
The engine is organized as five modules executed in a dependency chain (Fig. 1). Each module queries live public databases rather than fixed values, so the same pipeline generalizes to other indication/failed-drug pairs.
Module A used ChEMBL and PubChem for drug/payload composition and mechanism, and PubMed/bioRxiv for resistance literature. Module B combined Open Targets (target–disease association), cBioPortal (TNBC cohort genomics), GTEx (normal-tissue expression) and the Human Protein Atlas (normal and tumor tissue, subcellular localization), with UniProt confirming plasma-membrane topology. Module C used ChEMBL and PubChem for payload potency, physicochemistry, and efflux-substrate status. Module D retrieved structures from the PDB and AlphaFold, and a Boltz-2 NIM (NVIDIA BioNeMo) was registered and run for the folding steps; the inverse-design steps (ProteinMPNN, backbone generation) are specified but were not executed (see §6). Module E queried ClinicalTrials.gov and Drugs@FDA. Identifiers cited below are those returned by these queries.
Dato-DXd (ChEMBL CHEMBL4297939) is a humanized anti-TROP2 (TACSTD2) IgG1 conjugated through a cleavable GGFG tetrapeptide linker to DXd, an exatecan-derived TOP1 inhibitor, at a drug-to-antibody ratio of approximately 4. DXd traps the TOP1–DNA cleavage complex, producing replication-associated double-strand breaks; the released warhead is membrane-permeable and produces a bystander effect on neighboring TROP2-negative cells. This mechanism, and the resistance it selects for, defines the constraints on any successor design.
Module A partitioned documented and mechanistically-expected resistance into five buckets, each mapped to a design lever (Fig. 2). The dominant, class-defining driver is at the payload level: silencing of SLFN11 — the canonical resistance biomarker shared across TOP1, TOP2, platinum, and PARP inhibitors — together with acquired on-treatment TOP1 mutations (reported in roughly one in eight ADC-progressing metastatic breast cancers)8 and upregulation of the DNA-damage response. Because these mechanisms are engaged by replication-blocking DNA damage per se, they confer cross-resistance to any TOP1-inhibitor payload, not merely to DXd. A second bucket, drug efflux via ABCG2 (BCRP) and ABCB1 (P-glycoprotein), pumps out DXd and SN-38 and compounds the class effect. TROP2 antigen downregulation, impaired lysosomal trafficking, and apoptosis evasion complete the picture.
Payload mechanism to avoid: topoisomerase-I inhibitors (DXd, SN-38, exatecan/camptothecins).
Cross-resistant ADCs not to reuse: trastuzumab deruxtecan (identical DXd payload) and sacituzumab govitecan (SN-38 payload and the same TROP2 antigen — a double overlap).
Orthogonal payload classes: microtubule inhibitors (auristatins, maytansinoids) and DNA minor-groove alkylators/crosslinkers (duocarmycin, PBD, IGN), which are treated as SLFN11-independent; and, as a novelty option, the RNA-polymerase-II inhibitor α-amanitin. TROP2 need not be abandoned in principle, but is de-prioritized here given antigen-loss risk in the refractory setting.
An ADC target is not simply a gene that is highly expressed in the tumor; it is a gene that is high in the tumor and low in the normal tissues whose damage would be dose-limiting. Module B scored twelve membrane-confirmed TNBC surface antigens on both axes, cross-referencing Open Targets association and Human Protein Atlas tumor expression against GTEx normal-tissue expression across heart, lung, liver, kidney, gastrointestinal tract, brain, marrow, and skin (Fig. 3).
The ranking (Table 2) places LIV-1 (SLC39A6) first. LIV-1 is a zinc transporter enriched in TNBC, confirmed as a multi-pass plasma-membrane protein with an extracellular ectodomain, and — critically — already antigen-validated by a clinical ADC. Its normal-tissue expression is moderate (skin and cerebellum in the ~50–60 TPM range), giving an acceptable rather than pristine window. B7-H3 (CD276) ranks second: broadly expressed across TNBC tumor and stroma with a moderate window, but its leading ADC uses a TOP1-inhibitor payload (see §6). ROR1 ranks fourth with the widest intrinsic safety window (oncofetal, near-absent in adult normal tissue) but lower and more variable TNBC prevalence. TROP2 itself falls to rank 7 — extremely high association but a narrow window and the antigen the patient has already failed.
Table 2. Top-ranked alternative antigens (of twelve evaluated), scored on TNBC association/prevalence against normal-tissue expression and dose-limiting-toxicity (DLT) organ risk. LIV-1 highlighted.
| Rank | Gene | Protein | TNBC association / prevalence | Max normal TPM / DLT organ | Window | Existing ADC |
|---|---|---|---|---|---|---|
| 1 | SLC39A6 | LIV-1 / ZIP6 (Zinc transporter) | 0.07 / ~65-90% | 59.8 TPM · Skin & CNS (moderate ZIP6 in skin ~60 TPM, cerebellum/cord ~37-49 TPM) | Moderate | Ladiratuzumab vedotin (Ph1/2, TNBC) |
| 2 | CD276 | B7-H3 (CD276) | 0.04 / ~60-80% | 32.0 TPM · Broad low-level stromal/vascular + adrenal (~32 TPM), skin, nerve — diffuse expression widens off-tumor sink but no lethal single organ; endothelial expression a theoretical vascular risk | Moderate | Ifinatamab deruxtecan (I-DXd, Ph); vobramitamab |
| 3 | EGFR | Epidermal growth factor receptor | 0.32 / ~50% | 78.3 TPM · Skin (~78 TPM) → rash/cutaneous tox is the classic on-target DLT; nerve/esophagus moderate | Moderate | Multiple clinical (MRG003, losatuxizumab) |
| 4 | ROR1 | Tyrosine-protein kinase transmembrane receptor ROR1 | 0.08 / Subset | 6.8 TPM · Oncofetal — very low adult normal tissue (max ~7 TPM in colon/artery) | Wide | Zilovertamab vedotin (clinical) |
| 5 | NECTIN4 | Nectin-4 (PVRL4) | 0.02 / ~50-60% | 205.2 TPM · Skin (~205 TPM) & esophageal mucosa → skin toxicity is the enfortumab vedotin dose-limiting effect (cutaneous, incl | Narrow | Enfortumab vedotin (Padcev, approved urothelial) |
| 6 | MSLN | Mesothelin | 0.03 / ~30-35% | 78.3 TPM · Lung (~78 TPM) + serosal mesothelium (pleura/peritoneum/pericardium, under-sampled in GTEx) → serositis/pleuritis risk; otherwise restricted | Moderate | Anetumab ravtansine; others (clinical) |
| 7 | TACSTD2 | TROP-2 (Trophoblast cell-surface antigen 2) | 0.55 / ~80-90% | 1419.1 TPM · Esophagus (1419 TPM!), skin (~700), kidney → stomatitis/skin tox (seen with SG/Dato-DXd) | Narrow | Sacituzumab govitecan (Trodelvy, approved TNBC); Dato-DXd (INCUMBENT, progressed) |
The payload must share no biology with the resistance buckets in §3. The engine's key insight is that SLFN11/DDR resistance is engaged specifically by replication-blocking DNA damage; payloads that cause no DNA damage are therefore indifferent to SLFN11 status. Two classes have zero overlap with the DNA-damage axis — microtubule inhibitors (auristatins, maytansinoids) and the RNA-polymerase-II inhibitor α-amanitin — while DNA minor-groove alkylators (PBD, duocarmycin, IGN) are TOP1-orthogonal and clinically SLFN11-independent but formally share the replication-stress axis, making them a notch less orthogonal (Fig. 4).
Within the auristatins, the choice between MMAE and MMAF is decisive for the efflux bucket. MMAE (XLogP ≈ 4.1) is membrane-permeable and a P-glycoprotein substrate — vulnerable if ABCB1 drives resistance — whereas MMAF carries a charged C-terminal phenylalanine (XLogP ≈ 2.1), is cell-impermeable, and is a poor P-gp substrate. MMAF therefore resolves all four resistance buckets at once: its tubulin mechanism is orthogonal to buckets 1–3 (no DNA damage), and its physicochemistry is orthogonal to bucket 4 (efflux).
PRIMARY MMAF (PubChem CID 10395173) on a non-cleavable maleimidocaproyl (MC) thioether linker at site-specific DAR 4 (engineered-cysteine/THIOMAB or enzymatic/glycan conjugation). A non-cleavable linker releases only the charged Lys-MC-MMAF catabolite, which is not membrane-permeable — eliminating re-effluxable free drug and reducing off-target toxicity.
BACKUP Duocarmycin (seco-DUBA) or a PBD/IGN dimer on a cleavable Val-Cit (or GGFG) linker at low DAR (2–2.7), for heterogeneous or low-density antigens (B7-H3, ROR1) where a membrane-permeable payload is needed for bystander killing. Therapeutic index, not potency, is the constraint here.
WILDCARD α-amanitin (transcription arrest; XLogP ≈ −4.4, non-effluxed) — the highest-novelty option, with zero cross-resistance to any marketed ADC, limited by hepatotoxicity and early clinical maturity.
This configuration also differentiates cleanly from the one LIV-1 ADC that reached the clinic. Ladiratuzumab vedotin paired LIV-1 with MMAE on a cleavable vc linker and stalled on an auristatin-class tolerability ceiling — not on target failure. The present design changes the warhead (F vs E), the linker chemistry (non-cleavable vs cleavable), and the conjugation (homogeneous site-specific DAR4 vs heterogeneous), each of which independently addresses the plausible root cause of that tolerability ceiling: premature systemic payload release and re-effluxable free drug.
Module D confirmed LIV-1/ZIP6 (UniProt Q13433) as a multi-pass membrane protein and delineated its ADC-accessible N-terminal ectodomain (residues 29–325, after the signal peptide), retrieving the AlphaFold model AF-Q13433-F1. No experimental anti–LIV-1 antibody structure exists in the PDB, so the LIV-1 binder must be designed de novo on a humanized IgG1/κ Fab framework. To ground the structural workflow, the module retrieved validated antibody–antigen complexes for the backup antigens — including an anti-B7-H3 Fab bound to the CD276 ectodomain (PDB 9LY5; Fig. 5) and an anti-ROR1 Fv–Kringle-domain complex (PDB 6BA5) — which serve both as backup-target scaffolds and as templates that anchor the co-folding protocol.
.cif files for this, the LIV-1 AlphaFold model, and the Boltz-2 predictions (Fig. 7) are provided as artifacts and open in the interactive 3D viewer.The specified protocol is: (1) fold the LIV-1 ectodomain; (2) co-fold the humanized Fab VH+VL with the ordered ectodomain epitope using Boltz-2 or Chai-1, ranking poses by ipTM and interface pLDDT; (3) redesign CDR/paratope residues with ProteinMPNN while holding the epitope fixed; (4) validate each design by re-folding and accepting only those that improve interface metrics; (5) graft accepted VH/VL into full IgG1 for conjugation.
Executed structural runs. A Boltz-2 service (NVIDIA BioNeMo NIM) was registered and run. Co-folding the anti-B7-H3 Fab (20G5) with the CD276 ectodomain reproduced a well-formed complex — interface ipTM 0.55 (above the 0.5 pass line), pTM 0.63, complex pLDDT 0.88 (Fig. 7) — confirming the pipeline end-to-end and providing a validated interface scaffold. Each of the three candidate-antigen ectodomains was then folded single-sequence to score the structural axis that the expression/safety ranking (§4) does not capture (Fig. 8, Table 4). The result is a genuine trade-off rather than a single winner: B7-H3 folds as ordered immunoglobulin domains (mean pLDDT 91, pTM 0.55), ROR1 is intermediate (44, 0.32), and the LIV-1 ectodomain is largely disordered (33, 0.20) — consistent with its low-complexity, His-rich sequence. Critically, this inverts the expression/safety ranking, where LIV-1 leads and B7-H3 is second.
The implication is that the engine's top target on expression and safety carries the weakest structural evidence: the LIV-1 ectodomain offers no confidently-folded region to anchor an epitope, flagging epitope quality as a real development risk. The design response is to define the LIV-1 binding epitope on its most ordered subsegments and to re-fold with a multiple-sequence alignment before committing to a paratope; where structural tractability is weighted heavily, B7-H3 becomes the stronger structural lead (with the countervailing caveat that its leading clinical ADC uses a cross-resistant TOP1i payload — see §7). Steps 3–5 of the protocol — de novo LIV-1 binder generation, ProteinMPNN interface redesign, and IgG1 grafting — require a backbone-design step (RFdiffusion-class) not available as a registered service in this run and are the immediate next experiment.
.cif structures are provided as artifacts and open in the interactive 3D viewer.
Table 4. Structural axis (Boltz-2, NVIDIA BioNeMo). Ectodomain ranges from UniProt topology; folds run single-sequence under identical parameters. Expression/safety rank is carried from Table 2.
| Target | Ectodomain (UniProt) | pTM | Mean pLDDT | Confidence | Expression rank | Structural read |
|---|---|---|---|---|---|---|
| B7-H3 (CD276) | 29–466 | 0.55 | 91 | 0.84 | 2 | Ordered Ig domains — strongest structural lead |
| ROR1 | 30–406 | 0.32 | 44 | 0.42 | 4 | Intermediate; mixed order |
| LIV-1 (SLC39A6) | 29–325 | 0.20 | 33 | 0.31 | 1 | Largely disordered — epitope-quality risk |
Reference complex: anti-B7-H3 Fab + CD276 co-fold — interface ipTM 0.55, pTM 0.63, complex pLDDT 0.88.
The metastatic TNBC ADC market is saturated with TOP1-inhibitor payloads (Table 1). A patient refractory to Dato-DXd has failed the TROP2/TOP1i axis and is at risk of cross-resistance to the remaining TOP1i ADCs — including the leading B7-H3 agent, ifinatamab deruxtecan (I-DXd), which is also DXd-based. The white space is therefore a mechanistically orthogonal ADC: a non-TOP1i payload against a non-TROP2/HER2 antigen (Table 3).
| Target | Key agents | Representative trials (NCT) | Competitive read |
|---|---|---|---|
| LIV-1 (SLC39A6) | ladiratuzumab vedotin (SGN-LIV1A, MMAE) - Seagen/Pfizer BRY812 (LIV-1 ADC, early) | NCT01969643, NCT03310957, NCT04032704, NCT01042379 | LOW-MODERATE (validated antigen, incumbent stalled on tolerability, few active competitors) |
| B7-H3 (CD276) | ifinatamab deruxtecan (I-DXd, DS-7300, TOP1i/DXd) - Daiichi/Merck MGC018/vobramitamab duocarmazine (duocarmycin) - MacroGenics (TERMINATED) IDE034 | NCT06330064, NCT06780085, NCT07630974, NCT03729596 | HIGH (crowded, but dominated by TOP1i payload - cross-resistance caveat) |
| ROR1 | zilovertamab vedotin (MK-2140, MMAE) - Merck/VelosBio NBE-002 (anthracycline ADC) - NBE-Therapeutics (TERMINATED) TQB2101 | NCT03833180, NCT04504916, NCT06717347, NCT05139017 | MODERATE in heme (Phase 3), LOW/unproven in TNBC solid-tumor setting |
| TROP2 (index / cross-resistance reference) | datopotamab deruxtecan (Dato-DXd, DXd) sacituzumab govitecan (SN-38) | NCT05104866, NCT05374512, NCT05629585, NCT03401385 | SATURATED (two approved TROP2/TOP1i ADCs - the space we are exiting) |
Table 3. Competitive landscape by target. Trials shown are representative NCT identifiers returned by ClinicalTrials.gov. LIV-1 is validated but relatively open (its lead agent stalled on tolerability, not target failure); B7-H3 is crowded but dominated by TOP1i chemistry; ROR1 is validated in heme malignancies but largely unproven in solid tumors.
LIV-1 fits the white space precisely: antigen-validated in TNBC, its only advanced incumbent (ladiratuzumab vedotin) stalled on tolerability rather than target failure, and no site-specific-DAR MMAF-payload LIV-1 ADC has been registered. The differentiation thesis operates on three axes at once — a distinct antigen (LIV-1, avoiding the saturated TROP2/HER2 space), a distinct payload mechanism (MMAF, orthogonal to the failed TOP1i class), and engineering that targets the incumbent's therapeutic-index problem (homogeneous site-specific DAR4 plus a non-cleavable MC linker).
The design is a mechanistically-reasoned hypothesis generated from live public data, not an experimentally validated asset. Several risks are explicit and must be retired prospectively:
The most important caveat is that the benefit of a non-TOP1i payload in TOP1i-refractory TNBC is mechanistically rational but not yet clinically proven; it requires a biomarker-selected, prospectively enrolled post-Dato-DXd trial with an operationalized refractory population. The LIV-1 target-validation gap (no LIV-1 ADC has shown registration-grade efficacy) is likewise unretired. Finally, the structural work was partially executed: Boltz-2 co-folding was run (validating the B7-H3 interface scaffold and revealing a low-confidence, likely-disordered LIV-1 ectodomain — an epitope-quality flag), but the de novo LIV-1 binder generation, ProteinMPNN interface redesign, and IgG1 grafting were not, because a backbone-design service was not available in this run. Those steps define the immediate next experiment.
Running the engine end-to-end on TNBC refractory to Dato-DXd converts a vague clinical problem — "the ADC stopped working" — into a specific, defensible design: an anti–LIV-1 humanized IgG1 carrying MMAF via a non-cleavable MC linker at site-specific DAR4, orthogonal to the topoisomerase-I resistance in play on antigen, payload mechanism, and chemistry, with a duocarmycin/PBD backup for low-density antigens and B7-H3/ROR1 as backup targets. The executed Boltz-2 folds add a decision the expression ranking alone could not surface: the three candidate antigens separate on a structural axis that inverts their expression ranking, so the recommendation is best read as a two-dimensional map — LIV-1 as the expression/safety lead with an epitope-quality flag to retire, B7-H3 as the structural lead conditioned on keeping a non-TOP1i payload. The same pipeline, driven by live data, generalizes to other indication/failed-drug pairs.
Primary resistance-biology citations retrieved by Module A (PubMed/DOI). Drug, target, structural, trial, and regulatory identifiers are cited inline in the text and tables (ChEMBL, PubChem, UniProt, PDB, NCT, and FDA application numbers).