🔬 Bispecific Antibody Format Engineering

Choosing a Bispecific Architecture — Fc-Bearing IgG-Like Formats versus Fc-Free Tandem scFv

Bispecific antibody engineering begins not with protein chemistry but with a format decision that constrains every downstream step. More than 100 distinct bispecific formats have been described in the literature, but they cluster into a few functional classes: IgG-like heterodimers that retain an Fc domain for long serum half-life and effector function, Fc-free tandem scFv (BiTE) constructs optimized for compact size and potent T-cell engagement, and Fab-arm exchange platforms (DuoBody) that assemble bispecifics post-translationally from two parental antibodies.

• 100+: Known bispecific formats (reviewed in Brinkmann & Kontermann 2017)
• ~150 kDa: IgG-like format size (full Fc, ~3 week serum half-life)
• ~55 kDa: BiTE format size (no Fc, ~2 hour serum half-life)
• >14: FDA-approved bispecifics (as of 2024, oncology + immunology)

Matching architecture to biological requirement

The choice of bispecific format is driven primarily by three factors: the desired pharmacokinetic profile, the mechanism of action, and manufacturability.

IgG-like heterodimeric formats (e.g., knob-into-hole, CrossMAb): • Retain a full or engineered Fc domain — FcRn recycling confers serum half-life of 1-3 weeks, comparable to a conventional monoclonal antibody • Enable Fc effector functions (ADCC, ADCP, CDC) if desired, or these can be silenced (e.g., LALA-PG mutations) if not • Bivalent-plus-bivalent or 2+1/1+1 valency configurations possible • Manufacturing complexity: requires solving both heavy-chain heterodimerization and light-chain mispairing (Stages 2-3) • Typical use: dual-target receptor blockade (e.g., emicizumab bridging Factor IXa/X), or T-cell engagers requiring extended dosing intervals (e.g., mosunetuzumab)

Tandem scFv / BiTE format (Blinatumomab prototype): • Two scFv units (VH-VL of each specificity) joined by a short Gly-Ser linker, no Fc, ~55kDa • Small size enables deep tissue/tumor penetration and rapid renal clearance — but this short half-life (~2h) necessitates continuous IV infusion in the clinic • No Fc effector function or FcRn recycling — potency comes entirely from forced proximity between a CD3+ T cell and a tumor antigen+ target cell • Manufacturing simplicity: single polypeptide chain, standard microbial or mammalian expression, no heterodimerization problem at all

Fab-arm exchange (DuoBody, Genmab platform): • Two parental IgG4-like antibodies, each carrying complementary CH3 mutations (F405L / K409R), separately expressed then mixed under mild reducing conditions in vitro • Half-molecule exchange reaction reconstitutes a bispecific IgG post-translationally — avoids co-expression mispairing entirely by expressing each arm separately at full purity first • Highly efficient (>95% exchange) and CMC-friendly since each parental antibody is a conventional, well-behaved biologic

For this case study we follow the IgG-like knob-into-hole route (most broadly deployed platform, e.g., in emicizumab-class molecules) through the remaining four stages, since it surfaces the full set of heterodimerization and light-chain pairing engineering problems.

Engineering the CH3-CH3 Interface — Steric Complementarity Drives Heavy-Chain Heterodimerization

When two different heavy chains are co-expressed in a single cell, random pairing at the CH3-CH3 dimerization interface produces a statistical mixture: roughly 50% of each possible homodimer and only 50% of the desired heterodimer (following simple binomial pairing, ~25% HH, ~25% LL and ~50% HL for two chains H and L expressed equally — the numbers here refer to two distinct heavy chains, not heavy/light). Knob-into-hole (KiH) technology, introduced by Genentech in 1996, re-engineers this interface with a steric "lock and key" that penalizes homodimer formation and favors the desired heterodimer.

• T366W: Knob mutation (CH3 domain, one heavy chain)
• T366S/L368A/Y407V: Hole mutations (CH3 domain, other heavy chain)
• ~57%: Heterodimer purity, KiH alone (without stabilizing disulfide)
• >90%: Heterodimer purity, +disulfide (S354C/Y349C interchain bond)

Structural basis of steric complementarity and its limits

CH3 domain background: the two CH3 domains of an IgG heavy-chain homodimer pack against each other through a largely hydrophobic interface. Threonine 366 sits near the center of this interface on both chains in the wild-type sequence.

Knob design: substituting Thr366 with the bulky tryptophan (T366W) on one heavy chain ("chain A") creates a protruding "knob" that sterically clashes with a second copy of itself — a Trp366/Trp366 homodimer is strongly disfavored due to steric overlap.

Hole design: on the partner heavy chain ("chain B"), three substitutions — T366S, L368A, Y407V — remove bulk from the corresponding region, creating a complementary cavity ("hole") sized to accommodate the knob. A Ser/Ala/Val hole-hole homodimer is also disfavored, though less strongly, because it leaves an underpacked, energetically unfavorable void at the interface.

Net result: knob-chain-A/knob-chain-A homodimer is sterically blocked; hole-chain-B/hole-chain-B homodimer is thermodynamically disfavored by poor packing; knob-A/hole-B heterodimer is the only combination achieving good shape complementarity, and is therefore strongly favored.

Quantitative outcome: KiH mutations alone typically shift heterodimer yield from the ~50% expected by chance to ~90-95% by peptide mapping / mass spec in early reports, though many groups report substantially lower purity (50-70%) without additional stabilization, especially at high expression titers where mass-action effects favor homodimer re-formation. In this case study, initial KiH-only heterodimer purity is 57%, prompting the addition of a stabilizing interchain disulfide.

Stabilizing disulfide: introducing a second, non-natural disulfide bond at the CH3-CH3 interface (S354C on the knob chain, Y349C on the hole chain) covalently locks the heterodimer once formed, kinetically trapping the desired species and pushing measured heterodimer purity above 90% even before chromatographic purification. This disulfide sits adjacent to, not in place of, the native inter-heavy-chain hinge disulfides.

Residual byproducts: even with KiH + stabilizing disulfide, small amounts of homodimer and half-molecule species persist due to imperfect expression stoichiometry between the two heavy-chain plasmids — this residual mispairing is resolved downstream by chromatography (Stage 4), not by protein engineering alone.

Knob-into-hole is a purely steric/thermodynamic solution to heavy-chain pairing — it does not address the separate and equally serious problem of light-chain mispairing, which arises whenever two different heavy chains are co-expressed alongside two different cognate light chains. That problem is solved independently in Stage 3.

Solving the Light-Chain Mispairing Problem — Common Light Chain and CrossMAb Strategies

Knob-into-hole forces the correct pairing of the two heavy chains, but a full bispecific IgG also requires two different light chains to pair correctly with their cognate heavy chains — and light chains show comparatively little heavy-chain selectivity when co-expressed, generating a scrambled mixture of correctly and incorrectly paired half-molecules. A single, engineered or naturally cross-reactive "common light chain," compatible with both heavy-chain arms, sidesteps this problem entirely by removing the light-chain choice altogether.

• 10: Possible mispaired species (2HC+2LC) (without any pairing control)
• 3: Possible species with common LC (heterodimer + 2 homodimers)
• phage display: Common LC discovery method (dual-target panning, e.g. CrossMAb)
• 89%: Purity after common LC + KiH (before chromatographic polishing)

Combinatorics of the mispairing problem and engineering solutions

The combinatorial mispairing problem: co-expressing heavy chain A (HA), heavy chain B (HB), light chain A (LA) and light chain B (LB) in one cell, with random heavy-heavy pairing (addressed by KiH) and independently random heavy-light pairing, yields up to 10 distinguishable half-molecule and full-molecule species by mass, most of which are non-functional or only monospecific. Even after KiH restricts heavy-chain pairing to HA-HB heterodimers, each heavy chain can still independently pick up either LA or LB, giving four possible HA-HB heterodimer light-chain combinations, only one of which (HA/LA + HB/LB) is the desired bispecific.

Solution 1 — Common light chain: If a single light chain (LC) can be identified or engineered that pairs correctly and retains full antigen-binding affinity for both the HA and HB heavy-chain variable domains, the light-chain mispairing problem vanishes: only one light chain species exists, so there is nothing to mispair. Common light chains are typically discovered by: • Screening a single fixed light chain (often a germline Vκ1-39 or Vκ3-20 framework, favored for developability) against phage or yeast display libraries of heavy-chain CDRs for each of the two target antigens independently, so both resulting heavy chains are pre-validated to work with the shared light chain • Alternatively, starting from two independently discovered antibodies and engineering one light chain to functionally replace the other via CDR grafting and affinity-restoring mutagenesis This strategy, used in molecules like emicizumab-class bispecifics, reduces the possible mispaired species after KiH from 4 down to essentially 1, since only HA/LC and HB/LC pairings exist.

Solution 2 — CrossMAb domain exchange: When a true common light chain cannot be found without sacrificing affinity, the CrossMAb strategy swaps the CH1/CL domain order on one arm only (e.g., replacing CH1-CL with CL-CH1, or exchanging the entire Fab domain architecture on the B arm), so that LA can only pair correctly with the structurally distinct HA, and LB can only pair correctly with the structurally distinct HB — restoring pairing fidelity through domain-architecture orthogonality rather than sequence commonality.

Measured outcome in this case study: switching from independent light chains (no control) to a validated common light chain, layered on top of the Stage 2 knob-into-hole heavy-chain engineering, raised overall correctly-assembled bispecific purity from 57% (heavy-chain pairing only) to 89% (heavy- and light-chain pairing both addressed), measured by native mass spectrometry of the crude expression supernatant before any chromatographic purification.

From Crude Co-Expression Supernatant to a Single, Homogeneous Bispecific Species

Even a well-engineered knob-into-hole, common-light-chain construct does not emerge from cell culture at 100% purity — residual homodimer, half-molecule, and aggregate species must be removed chromatographically before a bispecific antibody can be considered a defined chemical entity suitable for biophysical characterization or dosing. This stage combines optimized transfection stoichiometry with a purification train exploiting the small physicochemical differences the knob-into-hole mutations themselves introduce.

• 1:1:2: Plasmid ratio (HA:HB:LC) (optimized by DoE screen)
• 97%: Final SEC monomer purity (analytical SEC-HPLC)
• <2%: Residual homodimer (by native MS after polishing)
• 69°C: Measured Tm (DSC) (first unfolding transition, CH2)

Transient co-expression, chromatographic polishing, and analytical release testing

Transfection optimization: transient co-transfection in Expi293 or CHO-based systems requires balancing three plasmids (heavy chain A, heavy chain B, common light chain) simultaneously. A design-of-experiments (DoE) screen across plasmid mass ratios identified 1:1:2 (HA:HB:LC, favoring light-chain excess to drive complete heavy-chain occupancy and minimize unpaired heavy-chain aggregation) as optimal for maximizing heterodimer titer and minimizing aggregate byproduct.

Purification train: 1. Protein A affinity capture: binds all Fc-containing species (heterodimer, both homodimers) non-selectively via the CH2-CH3 Fc region; removes non-Fc host cell protein and light-chain-only byproduct. Elution at low pH (pH 3.5, citrate buffer). 2. Cation-exchange chromatography (CEX) or hydrophobic-interaction chromatography (HIC) polishing: the knob (Trp) and hole (Ser/Ala/Val) mutations subtly shift the local surface hydrophobicity and pI of each homodimer relative to the heterodimer, enabling a shallow linear salt or isopropanol gradient to resolve heterodimer from residual homodimer peaks — typically a partial but exploitable separation (resolution factor Rs ~0.8-1.2), often requiring a second orthogonal polishing step. 3. Size-exclusion chromatography (SEC): final polishing step removes high-molecular-weight aggregate and any residual half-molecule, and serves as the analytical purity release assay.

Analytical characterization: • Analytical SEC-HPLC: monomer purity 97%, <1% aggregate, <2% low-MW species • Native mass spectrometry: confirms the intact heterodimer mass to within 5 Da of theoretical, and quantifies residual homodimer contamination directly by peak area (<2% each species) • Differential scanning calorimetry (DSC): first unfolding transition (typically the CH2 domain) at Tm=69°C, consistent with the parental IgG1 Fc and confirming the KiH/disulfide mutations did not destabilize the Fc scaffold • Peptide mapping (LC-MS/MS after tryptic digest): confirms 100% sequence identity to the designed construct, including correct disulfide connectivity at the engineered S354C-Y349C bond

Proving Both Arms Work Simultaneously — Dual-Target Bridging and Synapse Formation Assays

Biochemical purity and correct assembly do not guarantee function: a bispecific antibody could in principle be perfectly assembled yet have one arm sterically or allosterically impaired by proximity to the other. The final validation stage directly tests the molecule's defining property — the ability to engage two different antigens simultaneously, in trans, bridging two different cell surfaces or crosslinking two different receptor targets on the same cell.

• 0.31 nM: EC50, arm 1 (target antigen) (cell-surface ELISA, target+ cells)
• 0.24 nM: EC50, arm 2 (effector antigen) (cell-surface ELISA, CD3+ T cells)
• 92%: Synapse formation rate (live-cell confocal imaging, 2h co-culture)
• 8 pM: Redirected cytotoxicity EC50 (51Cr / LDH release assay, 18h)

Confirming trans-bridging function beyond simple monovalent binding

Monovalent arm validation (necessary but not sufficient): each binding arm is first confirmed independently by cell-surface ELISA or flow cytometry against cells expressing only the corresponding single antigen, yielding independent EC50 values for arm 1 (target antigen, e.g., a tumor surface marker) and arm 2 (effector antigen, e.g., CD3 on T cells) — 0.31nM and 0.24nM respectively in this case study, each within 2-fold of the parental monospecific antibody's affinity, confirming neither arm was materially disrupted by the bispecific engineering.

The critical trans-bridging test: monovalent EC50 values alone cannot confirm that both arms function simultaneously on the same molecule; a molecule could retain both binding sites but be conformationally restricted such that engaging one antigen sterically blocks engagement of the other. Two orthogonal assays address this directly:

1. Immunological synapse imaging: target-antigen-positive tumor cells (fluorescently labeled, e.g., CellTrace Violet) are co-cultured with CD3-positive T cells (separately labeled) in the presence of the bispecific molecule. Live-cell confocal microscopy over a 2-hour window scores the fraction of tumor-T-cell conjugate pairs showing polarized accumulation of the bispecific at the cell-cell interface (the synapse) — 92% of conjugate pairs showed clear synapse formation in this case study, versus <5% background conjugation without added bispecific.

2. Redirected cytotoxicity assay: the ultimate functional readout for a T-cell engaging bispecific — target cells pre-loaded with a cytotoxicity reporter (51Cr release or a luciferase/LDH-based alternative) are co-cultured with purified human T cells at a defined effector:target ratio (typically 10:1) across a dose range of bispecific antibody. EC50 for redirected killing was 8pM in this case study, and maximal specific lysis reached 78% at 18 hours — confirming the bispecific drives potent, antigen-dependent, T-cell-mediated killing that requires both arms (a monospecific anti-CD3 control at matched concentration produced <5% killing, since CD3 engagement alone without target-cell crosslinking does not efficiently activate cytotoxicity).

The redirected cytotoxicity EC50 (8 pM) is roughly 30-fold more potent than either arm's standalone monovalent binding EC50 (0.24-0.31 nM) — a hallmark signature of productive bispecific bridging, since forced physical proximity between effector and target cells dramatically lowers the antigen density and antibody concentration required for functional engagement compared to either arm acting alone. A bispecific that fails to show this potency gain despite good individual-arm affinities is a red flag for steric interference between the two binding domains and typically sends the molecule back to Stage 1 for format reselection (e.g., increasing linker length or repositioning the antigen-binding domains).