Every week, we examine new research in antibody engineering, VHH design, ADC development, protein modeling, and biologics AI. Each analysis separates the core finding, what is novel, the limitations, and the broader field context.
Latest issueIssue #15
Counter-Selection: Off-Target Objectives, Dual-Antigen Engagers, and Sequence Variant Limits
Five papers on designing binders against off-targets, tri-specific T-cell engagers, AAV immunogenicity modelling, and sequence variant benchmarks.
This week's thesis: Five papers, and the common thread is counter-selection: each one is about specifying what a molecule or a process must avoid, not only what it must hit, and each shows that naming the off-target changes the outcome.
Optimising one binder sequence against on-targets and off-targets simultaneously raised peptide-MHC discrimination between complexes differing by a single residue from 6.0% to 14.2% of designs, at a cost in overall yield.
A tri-specific engager against CEACAM5, CDH17 and CD3 reached 94.5% tumour growth inhibition where a clinical-stage CEACAM5 bispecific analogue reached 29.0%, with moderate monovalent affinities and low cytokine release.
A PBPK model spanning five orders of magnitude of AAV dose shows anti-drug antibody clearance of the expressed product is dose-dependent, negligible at low dose and reaching maximum rate near day 20 at high dose.
Peptide mapping across 51 marketed antibodies found 3,006 sequence variant occurrences, 92.5% at or below 0.1% per site, supporting 0.2% as a pragmatic risk-alert benchmark for commercial material.
Empirical force field scanning identified one peptide position where hydrophobic substitution gains potency at C5aR1 without a matching gain at C3aR, and the prediction transferred across two scaffolds.
Training on affinity differences between antibody pairs rather than absolute values turns ~500 labelled points into ~125,000 training pairs, and produced an anti-EGFR variant at 66 pM from roughly 100 starting measurements.
A manually curated set of 690 non-redundant nanobody-antigen interfaces that corrects biological assembly errors, retains glycans and cofactors, and collapses 72 anti-SARS-CoV-2 RBD complexes into 8 genuinely distinct epitopes.
Docking-angle analysis predicted which TCR-mimetic clones would bind the HLA framework instead of the presented peptide, before any wet-lab testing, and the predicted failure mode appeared exactly as forecast after humanisation.
Across 81 bispecific VHH pairings, epitope choice governed IL-21 receptor agonism more than affinity, and framework mutations forcing VHH:VHH dimerisation lifted an almost inactive construct from 15% to 82% STAT3 activation.
Twenty-two TCR bispecifics across seven scaffolds and three T-cell recruiters: the most compact format paired with the weakest-binding recruiter gave the widest therapeutic window, with 5 of 6 mice in remission versus 0 of 6.
Replacing two hydrophobic CDR-H3 residues extended in vivo half-life from 27.7 to 71.9 hours and tripled the duration of pain relief, while binding affinity and inhibitory potency were essentially preserved.
Modelling bispecific engager binding in the actual synapse volume rather than in bulk solution shifts effective antigen concentration by six orders of magnitude, which is the difference between antigen density predicting nothing and predicting a 2.5-fold dose difference.
A three-compartment receptor occupancy model predicts tumour-tissue target engagement that the ongoing trial cannot measure, after calibration against mouse data moved the in vitro dissociation constant thirtyfold.
Models trained only on 559 IgGs predict VHH-Fc surface properties zero-shot (heparin binding rho 0.82), but thermostability does not transfer because VHH-Fc lacks the CH1-CL interface that drives IgG unfolding.
Direct preference optimization aligns a structure-conditioned protein language model to experimental stability data, then stabilizes H5N1 hemagglutinin by up to 17 C in one round while preserving broadly neutralizing antibody binding.
A picomolar CD276 minibinder made a worse CAR than weaker binders. Isoelectric point, not affinity, governed surface trafficking, with a working window of pI 6.5 to 8.5.
A linker cleaved by a nucleophile already sitting in the target binding pocket releases payload on binding rather than internalization, giving 5.9-fold higher tumour exposure and a 7.5-fold higher tolerated dose.
A pooled human-cell screen selects peptide-guided degraders by degradation activity rather than binding, and works on EWS::FLI1, a disordered fusion oncoprotein with no stable pocket.
511 AI-designed antibodies from 29 organizations tested blind: affinity maturation reached 95 pM, but AI ranking within sequence clusters beat the control less often than random clone picking.
An AI agent ran 24 to 48 hour binder design campaigns against 16 targets with no human input into any design decision; 354 of 1,320 designs bound, a 27 percent hit rate.
AlphaFold 3 modeled TCR mimic antibody-pMHC complexes accurately and passed mutagenesis validation, but all three AF3-designed affinity mutants failed experimentally.
Ranking a 20,000-sequence repertoire by how closely each antibody reproduces a cognate ligand's contact residues gave 11 binders from 31 picks, eight of them sub-nanomolar.
A designed 67-residue protein scaffold carrying a BCL-xL binder and an E3-recruiting motif degrades BCL-xL in cells and triggers apoptosis, approaching a clinical PROTAC.
Deliberately weakening the CD3 arm gave a greater than 50-fold bias toward BCMA and near-identical potency against cells differing 23-fold in target density.
The first degrader built against the CLIP1-LTK fusion clears the whole protein at picomolar potency and keeps working against resistance mutations that raise kinase inhibitor potency by more than a thousandfold.
A two-parameter rule combining solvent exposure and local electrostatic potential predicts antibody CDR tryptophan oxidation risk with 79% accuracy, and distal charge-reversing mutations cut oxidation roughly in half without touching the liability residue.
ApexFold predicts how a peptide's secondary-structure composition redistributes across water, co-solvent, and membrane-mimetic environments, outperforming static single-conformation structure predictors on that task.
ARID-sf combines classical force-field physics with protein language model embeddings to rerank antibody-antigen docking poses, nearly doubling success rates over raw physics-based scoring while running far faster than existing antibody-specific alternatives.
Deliberately weak IL-2Rβ/γ agonist nanobodies fused to an anti-PD-1 IgG produce cytokine signaling that switches on only where PD-1 is engaged, avoiding the toxicity that has limited IL-2 therapy.
An EGFR×HER2 bispecific ADC outperformed DS-8201 across 22 patient-derived xenografts, with the advantage traced to receptor trafficking rather than broader target coverage.
A single T-cell engager recognizes a chemical hapten rather than a tumor antigen, so swapping the attached small-molecule ligand redirects it to a new target without re-engineering the antibody.
A screen across seven E3 ligases found a compound that only becomes a molecular glue after cells chemically modify it, revealing a new activation mechanism and a general handle on an underused ligase.
Antibody clearance in myeloma falls by half during treatment because tumor-secreted protein competes for the receptor that recycles antibodies, meaning exposure follows response rather than driving it.
Pairing a generative model with reward-guided search at inference time produced experimentally validated binders across 127 targets, including picomolar PDGFR binders and the first de novo proteins that bind a free carbohydrate.
A TROP2 antibody delivering a BRD4 degrader through an azoreductase-cleavable linker gave 62.2% tumor growth inhibition where the free PROTAC gave 7.7%, while sparing BRD4 in liver and kidney.
A single-cell QSP model across 14 tumor lines places half-maximal T-cell activation at 2.12 to 4.6 trimers per T cell, and shows target density alone does not predict killing.
A linker screening pipeline that eliminates PROTAC linkers geometrically incapable of productive ubiquitination, with 93% recall across 112 compounds and 8 E3-target systems.
Sequentially dosed antibody-TCO and ADC-tetrazine conjugates ligate inside the tumor via click chemistry, lifting trastuzumab uptake in HER2-ultralow tumors 3.2-fold and driving complete remissions in models that resist conventional ADCs.
A dual-specificity switch arm keeps IL-12 masked in circulation and unveils it only where the tumor matrix antigen FN-EDB competes it off, a trans-activation logic gate tuned with a QSP model and validated in cell assays.
Across 19,265 antibody structures, variable-domain N-glycosylation shows up at only 7.82%, about half the rate from prior enriched datasets; glutamine at the sequon X-position is not a suppressor after all, and N-X-T sequons are 3.84-fold more likely to be glycosylated than N-X-S.
One antibody against the GI-cancer antigen CDH17 is deployed two ways, as an MMAE/MMAF ADC and as a bispecific engaging a rare pan-Vδ1/Vδ2 gamma-delta T-cell agonist nanobody, exploiting the abundance of gamma-delta T cells in gut epithelium.
A monovalent PD-1/CTLA-4 bispecific engineered to steer CTLA-4 blockade onto PD-1-positive T cells reported first-in-human data in 86 patients: 19.8% objective response, 17.5-month median duration, and T-cell activation beyond approved combination regimens.
A mechanistic model jointly tracks AAV vector, expressed antibody, target antigen, and antibody-target complex across brain subregions, predicting that intra-cisterna magna or intrastriatal delivery beats high-dose IV for CNS target engagement.
One anti-FAP nanobody engineered into three formats maps how valency and molecular weight trade off in PET imaging: the monomer clears fast, the low-MW bivalent dimer maximizes tumor-to-blood ratio, and the Fc-fusion gives the highest, most durable uptake.
A phase 1 trial of DYP688, an ADC that delivers a Gq/11 signaling inhibitor otherwise too toxic to dose systemically, showed 19.7% objective response and 82% disease control in GNAQ/GNA11-mutant melanomas.
Researchers built a PROTAC that recruits a degradation enzyme to a disordered region of estrogen receptor alpha by routing through its structured partner protein 14-3-3, revealing an entirely new protein interface by cryo-EM.
A phase 1b trial tested an anti-PD-L1 antibody deliberately engineered for a short, roughly four-day half-life, dosed intermittently in early Alzheimer's patients, and found it safe with early signs of a neuroprotective biomarker effect.
A spatial computational model of liver cancer, calibrated directly against real tumor imaging data, shows that fibroblasts physically wall off tumors from immune cells and predicts which patients will respond to combination immunotherapy.
A synthetic antibody library engineered specifically for machine learning readout recovered antibodies against targets that a standard screen missed entirely.
A systematic study of 160 bispecific antibodies shows that some developability properties inherit predictably from the parent antibodies, while others only emerge once the two arms are combined.
A newly characterized molecular glue degrades a cancer-linked protein by having cereblon grip a surface tyrosine rather than the degron motif every prior cereblon-based glue has relied on.
A low-cost, high-throughput plate assay for antibody self-association turns out to predict in vivo clearance about as well as the field's leading specialized assays, and can now be predicted directly from sequence.
A proteome-wide screen of 174 approved and clinical-stage antibodies found that more than a quarter bind at least one unintended human protein, and shows these liabilities can sometimes be engineered away.
AbbVie shows that single-atom changes to a molecular glue degrader can reprogram which cellular disposal machinery it recruits — turning ligase choice into a tunable design parameter.
A new antibody blocks a cancer-linked receptor by gripping two separate loops on its surface at once, offering an early proof-of-concept for a target that's historically been hard to drug with antibodies.
A population PK analysis across 2,744 patients shows that a widely-used eye biologic's extended dosing interval comes from a genuine formulation effect on ocular clearance, not just from giving a bigger dose.
Germinal co-optimizes CDR sequence (IgLM) and structure (AlphaFold-Multimer) to design epitope-targeted nanobodies de novo, achieving 4-22% binding success with 43-101 designs tested per target — no starting binder required.
A VHH-VHH bispecific (TCRm Bi-NbTE) simultaneously engages CD3ε and pMHC-I tumor complexes, redirecting T cells to kill tumor cells via intracellular antigens — with in vivo efficacy in CDX and PDX xenografts.
Local fine-tuning of a PLM on 240k yeast-display CDR-H3 variants raises polyreactivity prediction success from 0% (base model) to 66.6% — practical demonstration that foundation models need domain adaptation for narrow drug-optimization tasks.
pH-dependent VHH engineering — 227-fold difference in dissociation rate between neutral and acidic pH — extends GH mimetic activity from 6 days to >15 days in rats, achieving ultra-long-acting PK via FcRn recycling logic in a VHH format.
T-Bren (BL-M07D1, DAR-8, cathepsin B-cleavable, topo I payload) achieves 81.5% ORR in HER2-positive and 69.5% in HER2-low breast cancer across 253 patients; RP2D 4.4 mg/kg Q3W.
ASO cargo inverts the TfR1 affinity-transcytosis relationship in Brainshuttles; a bispecific TfR1×CD98hc format rescues BBB delivery by engaging two distinct trafficking pathways simultaneously.
Ustekinumab + guselkumab variable regions combined into biparatopic 1+1 IgG and 2+2 Fab-IgG formats. Picomolar SPR affinities with avidity enhancement; dual-epitope engagement confirmed by bridging SPR and mass photometry. Functional neutralization equivalent to equimolar antibody mixture — a clean engineering framework without synergistic potency gain.
RNA Pol II miRNA targeting FUT8 in CHO: >90% afucosylation (HILIC-HPLC), no fucosylated product by intact MS, no titer loss. Beats shRNA (U6 promoter) which causes Dicer saturation → residual fucosylation + reduced titers. Clean solution to the afucosylation-vs-yield tradeoff in antibody manufacturing.
117 heavily pre-treated NSCLC patients, 19% ORR (29% taxane-naive nonsquamous), 11.3-month median DOR. IB6 expressed in 92% of biopsies. First JCO readout for an IB6-directed ADC.
Phage + hybridoma screen → TFPI-blocking mAb optimized to ~5.72 nM affinity for PK (not the tightest binder). Now FDA-approved as Hympavzi. Case study in deliberate affinity de-optimization as a PK lever.
ADCs targeting L1CAM — a marker of metastasis-initiating stem cells — achieve complete tumor regression and 100% survival in PDX lung metastasis models at 1 mg/kg with a non-cleavable linker to PNU-159682.
Bispecific ADC simultaneously targeting B7-H3 and PSMA outperforms single-target ADCs in prostate cancer CDX/PDX models by exploiting bystander killing and dual-antigen redundancy against heterogeneous tumors. FDA IND cleared.
MD simulations show nanobody CDR3 loops are intrinsically disordered in solution — not just flexible. Adaptive disorder enables binding to concave epitopes. CDR1/2 are rigid anchors; CDR3 is a conformationally malleable binding arm.
VHH-based bispecific T cell engager co-engaging oncofetal antigen 5T4 and Vδ2-TCR kills solid tumor cells in patient-derived 3D models without cytotoxicity against healthy 5T4-expressing tissue.
An AI-designed PROTAC beats kinase inhibitors on resistance mutants, electrostatics predicts antibody oxidation risk, and a model tracks peptide structure.
Three papers on a hypoxia-gated antibody-PROTAC conjugate for TNBC, a QSP model quantifying CD3-bispecific trimer thresholds, and PROTAC linker screening.
Four papers on targeted ADC delivery, disordered-protein degradation, short-lived anti-PD-L1 for Alzheimer's, and spatial QSP for liver-cancer immunotherapy.