Ubi Titer

Core finding

A non-blocking anti-CD2 antibody was engineered into a HER2×CD2 bispecific that, combined with an EpCAM×CD3 T cell engager at sub-toxic doses, produced complete tumor remission in 8 of 9 mice — compared to 1 of 9 with the TCE alone. Cytokine release was markedly lower than a HER2×CD28 bispecific comparator, while anti-tumor activity was comparable. The CD2 approach delivers costimulation to CD28-negative CD8 T cells, a significant population in elderly patients and solid tumors where CD28-targeted strategies fail.

What's novel

First demonstration of CD2-directed costimulatory bispecific in combination with CD3 TCE for solid tumors. The non-blocking CD2 antibody preserves TCR function while adding signal 2. CD28-independent costimulation of CD8 T cells has been underexplored as a rescue strategy for solid tumor TCE programs.

Limitations

Fully preclinical xenograft models; does not recapitulate tumor-immune microenvironment. HER2 and EpCAM are proof-of-concept targets, not a proposed clinical pairing. Manufacturing complexity of co-administering two bispecifics is not addressed.

In context

CD3 T cell engagers have largely failed in solid tumors due to dose-limiting cytokine toxicity and insufficient costimulation. Papers combining TCEs with costimulatory arms (CD28, 4-1BB, IL-2) are multiplying; CD2 is a less-traveled path that sidesteps CD28-related toxicity concerns. Tarlatamab is the only approved solid-tumor TCE to date.

Core finding

Sigvotatug vedotin (anti-integrin beta-6, MMAE payload) was evaluated across three dosing regimens in 117 advanced NSCLC patients (median 3 prior systemic therapies; 96% prior platinum, 91% prior PD-1/L1 inhibitor). Overall ORR 19%, 11.3-month median DOR. Taxane-naive nonsquamous subset: 29% ORR. IB6 expression detected in 92% of evaluable biopsies regardless of histology. Optimal regimen: 1.8 mg/kg adjusted ideal body weight Q2W.

What's novel

First peer-reviewed Phase I dose-expansion readout validating IB6 as an ADC target in NSCLC. The 11.3-month median DOR in this heavily pre-treated, post-immunotherapy population is a clinically meaningful signal. IB6 expression uniformity (92%) across histologies suggests a broad biomarker-selected population.

Limitations

Phase I, single-arm, no comparator. Heavily pre-treated, performance-status-filtered population reduces generalizability. MMAE payload; increasing clinical preference for TOP1i payloads not addressed.

In context

NSCLC ADC landscape is crowded: TROP2, HER2, MET, Nectin-4, B7-H4. IB6 offers distinct biology — expressed in tumor cells including squamous histology where other targets are sparse. JCO publication reflects confidence in the signal.

Core finding

Hybridoma and phage display screens identified anti-TFPI antibodies blocking the TFPI-FXa interaction via the TFPI K2 domain. High-affinity candidates showed rapid clearance due to target-mediated drug disposition. The optimized candidate, marstacimab at ~5.72 nM affinity, achieved durable plasma circulation while maintaining hemostatic efficacy. CDR-grafted humanization followed by affinity adjustment — not maximization — produced the clinical candidate now approved as Hympavzi.

What's novel

Retrospective discovery narrative for an approved drug providing a concrete case study in deliberate affinity de-optimization for PK benefit. The specific antibody engineering path — from screen to affinity tuning — is documented comprehensively for the first time.

Limitations

Retrospective publication on an already-approved drug; the novel contribution is the engineering narrative, not a new principle. Generalizability depends on TMDD kinetics for the target of interest.

In context

Hemophilia non-factor replacement therapies are an active class: emicizumab, fitusiran, marstacimab. The affinity-PK tradeoff lesson applies broadly to any mAb with significant target-mediated clearance.

Core finding

NMR spectroscopy combined with rheometry and DLS reveals how common excipients modulate mAb viscosity at clinically relevant concentrations. Arginine and lysine substantially reduce self-assembly and viscosity. Proline and glycine promote clustering. Sucrose binds antibody surface sites — confirmed by NMR — but does not reduce viscosity, showing that excipient-mAb binding is necessary but not sufficient.

What's novel

First systematic NMR-based pipeline for excipient-mAb interaction characterization at formulation-relevant concentrations. The sucrose finding — binding without viscosity reduction — resolves a long-standing formulation paradox. Provides mechanistic grounding for empirical excipient rankings previously established by rheometry alone.

Limitations

Single antibody characterized in depth; generalizability to other mAb charge profiles requires validation. NMR at high protein concentrations is technically demanding and not yet suitable for high-throughput screening.

In context

High-concentration formulation for subcutaneous delivery is a major bottleneck for biologics. Prior work ranked excipients empirically; this paper provides the first mechanistic picture using NMR. The arginine/lysine superiority over proline is now mechanistically grounded.

Core finding

In urothelial carcinoma tumors resistant to enfortumab vedotin (NECTIN4-ADC), a cell subpopulation maintains high NECTIN4 surface expression while showing defective ADC internalization. AKR1C1 physically binds NECTIN4 and disrupts AP2M1-dependent clathrin-mediated endocytosis; the AKR1C1-WWP2 axis simultaneously promotes extracellular vesicle-mediated export of ADC payload. Pharmacologic AKR1C1 inhibition restores internalization and drug sensitivity in patient-derived preclinical models.

What's novel

Mechanistically resolves the high-antigen/resistant phenotype paradox. AKR1C1 is a new resistance gene for the ADC field, not previously implicated in NECTIN4 trafficking. Paired spatial transcriptomics before and after treatment enables the subpopulation identification that bulk sequencing would miss.

Limitations

Urothelial carcinoma-specific study; AKR1C1 resistance mechanism may not generalize to other ADC targets or tumor types. AKR1C1 inhibitor combination strategy is preclinical only with no dose-optimization or toxicity data. Spatial transcriptomics is resource-intensive and not routine in clinical trial design.

In context

The PNAS CTC paper (June 19 issue) showed at the clinical level that receptor expression does not predict ADC response; this Cancer Cell paper now explains the mechanism — defective endocytic trafficking in a high-antigen subpopulation. Together they make the case that ADC biomarker strategy needs to incorporate endocytic competence, not just receptor IHC.

Core finding

Three L1CAM-targeting antibodies were developed against the Ig-like and FNIII domains of the L1CAM extracellular region, then conjugated to PNU-159682 (nemorubicin metabolite) via a ThioBridge disulfide-rebridging platform with non-cleavable tri-glycine linker at DAR=4. In lung metastasis models of TNBC and LUAD (including two PDX lines), four weekly doses of 1 mg/kg completely eradicated detectable disease, with animals surviving >9 months without relapse. L1CAM knockdown abolished efficacy, confirming on-target mechanism.

What's novel

L1CAM is the first ADC target specifically chosen to eliminate metastasis stem cells (MetSCs) rather than the bulk tumor. Prior L1CAM antibodies failed clinically due to lack of cytotoxic mechanism; the ADC format leverages L1CAM's high intrinsic internalization rate to deliver payload selectively to the MetSC compartment that survives standard therapy. The non-cleavable linker outperforming the cleavable format is a mechanistically interesting finding suggesting cathepsin activity limitations in the lung tumor microenvironment.

Limitations

Purely preclinical. L1CAM is expressed in peripheral neuronal tissue and kidney tubules, raising off-target safety concerns that will require careful NHP toxicology. Therapeutic window was assessed only in mice using a mouse-cross-reactive antibody; extrapolation to humans needs NHP confirmation. L1CAM expression varies widely even within histological subtypes.

In context

Metastasis stem cells have long been proposed as the key therapeutic target for preventing relapse, but the field has lacked a validated surface marker with sufficient tumor-to-normal selectivity. The Massagué group's prior work establishing L1CAM as a pan-cancer MetSC marker (Nature Cancer 2019) laid the groundwork for this ADC approach. This study provides the first proof-of-concept that pharmacological elimination of L1CAM+ cells translates to durable metastatic control — a conceptual advance as significant as the specific molecule.

Core finding

GenSci143 combines a bispecific antibody targeting B7-H3 and PSMA with a TOP1 inhibitor payload via a plasma-stable linker. B7-H3 and PSMA are highly co-expressed in prostate cancer (confirmed by gene profiling). In vitro potency was maintained even in cells expressing only one target, and bystander killing extended activity to antigen-negative bystander cells. In CDX/PDX models of mCRPC, GenSci143 induced superior tumor regression versus single-target B7-H3 and PSMA benchmark ADCs. NHP PK studies confirmed favorable plasma stability.

What's novel

Bispecific ADCs represent a new structural class — the antibody component is bispecific rather than monospecific — specifically designed to address tumor antigen heterogeneity, a persistent cause of ADC resistance. GenSci143 is one of the first BsADCs to enter IND-stage development, validating the concept as a clinically viable strategy rather than a theoretical exercise.

Limitations

Industry-sponsored study with potential for positive result bias. Preclinical only — no Phase I pharmacokinetics or tolerability data in humans yet. The PSMA ADC space is crowded with clinical-stage programs; differentiation of bispecific format vs. monospecific PSMA-ADC needs clinical validation. Prostate cancer antigen heterogeneity may be more complex than two-antigen targeting can address.

In context

Tumor antigen heterogeneity — where a subset of cancer cells downregulate the ADC's target antigen — is increasingly recognized as a primary mechanism of ADC resistance. The field is pursuing three strategies: bystander killing payloads (membrane-permeable, already in approved ADCs), bispecific antibodies (two tumor targets, one ADC), and dual-payload formats. GenSci143 exemplifies the bispecific approach, and the near-simultaneous emergence of DXC014 (a competing B7-H3×PSMA BsADC) strongly validates the target combination.

Core finding

ANDD integrates 15 public sources into a single quality-controlled dataset of 48,683 antibody and nanobody sequences with structural data for 24,941 entries, antigen sequences for 12,575 entries, and 9,557 affinity values — the largest integrated collection to date. The VHH/nanobody slice alone covers 30,119 entries. A standardized format with explicit provenance tags resolves the format inconsistency that has fragmented antibody ML datasets across OAS, PDB, SAbDab, and smaller databases.

What's novel

First resource to systematically combine sequence, structure, antigen context, and experimentally-grounded binding affinity for both conventional antibodies and VHH nanobodies in a unified, quality-controlled dataset. The affinity coverage — augmented with ANTIPASTI predictions where experimental data was absent, with clear flags — is the largest single Ab/Nb affinity collection assembled.

Limitations

A substantial fraction of affinity values are ANTIPASTI-predicted rather than experimental, which is useful for training but not a substitute for measured Kd/IC50 in validation contexts. Dataset covers static sequence and structure; conformational ensemble data is absent.

In context

Benchmark fragmentation has been a persistent complaint in the computational antibody design community — cross-model comparison is confounded when each method uses different training/test splits from different databases. ANDD is the first resource credible enough to serve as a community benchmark. The Zenodo release (records/16894086) makes direct download accessible.

Core finding

NbBayesLM wraps ESM-2 protein language model embeddings in a Bayesian neural network where physicochemical amino acid properties serve as prior distributions over model weights rather than fixed input features. On nanobody melting temperature prediction, this outperforms point-estimate PLMs and classical feature-only models while producing calibrated uncertainty estimates that tell engineers when a prediction should not be trusted.

What's novel

First Bayesian neural network specifically for nanobody thermostability prediction that produces calibrated confidence intervals rather than point estimates. Using physicochemical properties as Bayesian priors — rather than concatenated features — is a principled integration of domain knowledge with learned representations that improves both accuracy and calibration simultaneously.

Limitations

Performance on nanobodies outside the camelid VHH scaffold is not evaluated. Training set size in nanobody Tm data is limited, which may inflate cross-validation performance. CDR attribution is sequence-based; structural context is not integrated.

In context

Computational nanobody stability prediction has been approached by fine-tuned PLMs, biophysical feature models, and graph neural networks. NbBayesLM's addition of calibrated uncertainty addresses the decision-making gap: the confidence interval tells engineers when to trust a prediction, not just what it says.

Core finding

Molecular dynamics simulations across a diverse nanobody panel show CDR3 loops are intrinsically disordered in solution — sampling multiple conformations before committing to a bound state upon epitope contact. CDR1 and CDR2 act as relatively rigid structural anchors while CDR3 mediates binding through adaptive disorder. The longer CDR3 loops characteristic of VHH are selected because disorder enables access to concave epitopes that rigid loops cannot reach.

What's novel

First systematic characterization of CDR3 intrinsic disorder across a diverse nanobody panel using MD simulations. Reframes CDR3 conformational freedom as an adaptive feature rather than a structural liability, with direct implications for computational nanobody design tools that model CDRs as single energy-minimized conformations.

Limitations

MD simulation timescales may not fully sample the slow conformational dynamics of longer CDR3 loops. Panel diversity may not capture the full range of therapeutic VHH CDR3 lengths. Experimental validation of disorder-to-order transitions in binding kinetics is not provided.

In context

The field has known nanobody CDR3 loops are longer and more variable than conventional VH CDR3, but the functional interpretation has been debated. This paper establishes adaptive disorder as the structural logic behind unique epitope access, positioning nanobody design tools to incorporate ensemble-based CDR3 scoring.

Core finding

High-affinity 5T4-specific VHHs were linked to a Vδ2-TCR-specific VHH to generate a bispecific T cell engager. In 2D and 3D patient-derived tumor models across multiple solid malignancy types, the bsTCE triggered strong Vγ9Vδ2-T cell cytokine production and tumor lysis. When tested against healthy tissues expressing 5T4 at physiological (low) levels, cytotoxicity was not observed — demonstrating that the oncofetal upregulation of 5T4 provides a functional tumor-selectivity window.

What's novel

Combining an oncofetal antigen (5T4) with gamma-delta T cell recruitment is a strategically elegant approach to the toxicity problem that has stalled solid tumor bispecific T cell engagers. Previous Lava Therapeutics work used this gamma-delta platform for hematological malignancies; this extends it to solid tumors via a target antigen expressed preferentially on malignant cells.

Limitations

Preclinical study only. Vγ9Vδ2-T cells constitute roughly 1-5% of peripheral blood T cells and their frequency and function vary significantly across patients. No clinical data available. The 5T4 expression landscape across solid malignancy subtypes requires full characterization before patient selection criteria can be established.

In context

Solid tumor bispecific T cell engagers have faced persistent challenges from on-target off-tumor toxicity and poor tumor penetration. The field is responding with two strategies: tumor-preferential targets (5T4, here) and alternative T cell subsets (gamma-delta T cells, innate-like with lower activation thresholds). This paper is one of several recent studies exploring whether gamma-delta T cells can overcome the limitations of conventional CD3-redirecting bispecifics in solid tumors.