Protein Engineering: Foundations, Therapeutics, Generation, Characterization, and De Novo Design

Comprehensive practice flashcards covering protein structure and folding, monoclonal antibody therapeutics, display technologies, purification, characterization, sybodies, SARS-CoV-2 neutralizing antibodies, GPCR-targeting miniproteins, and de novo antibody design with RFdiffusion.

What are the five primary applied fields in which proteins are utilized beyond their physiological roles?

The five applied fields are biotherapeutics (antibodies, protein therapeutics), biomaterials (silk, collagen, engineered scaffolds), diagnostics (ELISA, lateral flow tests, biosensors), research tools (GFP, CRISPR-Cas9, affinity tags), and biocatalysis (industrial enzymes).

What is the assessment weighting scheme for the Protein Engineering course, and what special rule applies to the final exam grade?

The course assessment consists of: Final Exam ($45\%$), Continued assessment ($30\%$, which includes the Partial Exam and questions asked during presentations), Presentation ($20\%$), and Participation ($5\%$). If the final exam grade is $< 4.5$, the final course grade is strictly the grade of the final exam.

In a Ramachandran plot, what secondary structural conformations correspond to the major allowed regions shown in the upper-left and lower-left quadrants?

The upper-left quadrant contains the sterically allowed conformational regions for antiparallel $\beta$-sheets, parallel $\beta$-sheets, collagen triple helices, and right-twisted $\beta$-sheets. The lower-left quadrant corresponds to right-handed $\alpha$-helices. The isolated region in the upper-right quadrant corresponds to left-handed $\alpha$-helices (frequently populated by Glycine).

How does the local protein microenvironment shift the $pK_a$ of an amino acid side chain via charge-charge interactions, charge-dipole interactions, and the Born/desolvation effect?

Charge-charge interactions stabilize opposite charges (e.g., Asp in the catalytic triad shifts the His $pK_a$ from $\sim 7.5$ up to $10-12$). Charge-dipole interactions stabilize charges near helix termini (the positive N-terminus of an $\alpha$-helix stabilizes negative charges, lowering Asp/Glu $pK_a$). The Born/desolvation effect occurs when a group is buried in a hydrophobic core with a low dielectric constant ($\varepsilon \approx 2-4$ vs $\varepsilon \approx 80$ in water), destabilizing charged states and causing $pK_a$ to increase for acids and decrease for bases.

What chemical reaction produces a disulfide bond between two cysteine residues, and what are its thermodynamic consequences on protein folding?

Two cysteine thiol groups undergo oxidation: $2\,\text{R-SH} \rightarrow \text{R-S-S-R} + 2\text{H}^+ + 2\text{e}^-$. Disulfide bonds provide covalent cross-links between distant segments of a polypeptide chain, greatly reducing the conformational entropy of the unfolded state and locking the folded tertiary structure.

How do kosmotropic and chaotropic agents of the Hofmeister series differ in their effects on protein folding and solubility?

Kosmotropic ions (e.g., $\text{SO}_4^{2-}$, $\text{HPO}_4^{2-}$, $\text{NH}_4^+$) increase water surface tension, making cavity formation harder, decreasing hydrocarbon solubility, salting out proteins (promoting aggregation/folding), and increasing conformational stability. Chaotropic agents (e.g., urea, guanidinium, $\text{SCN}^-$, $\text{ClO}_4^-$, $\text{Ca}^{2+}$) decrease surface tension, increase hydrocarbon solubility, salt in proteins, and promote denaturation.

In Differential Scanning Calorimetry (DSC), what does the unfolding peak represent, and how does pH affect the melting temperature ($T_m$) of lysozyme?

Thermal unfolding of a protein domain is an endothermic transition characterized by a peak in specific heat capacity ($C_p$); the temperature at the peak maximum represents the melting temperature ($T_m$). For lysozyme, decreasing the pH destabilizes the protein, shifting $T_m$ downward from $\sim 79\,^{\circ}\text{C}$ at $\text{pH } 4$ to $\sim 66\,^{\circ}\text{C}$ at $\text{pH } 2.5$ and $\sim 55\,^{\circ}\text{C}$ at $\text{pH } 2.0$.

What is the Levinthal paradox, and how does the funnelled energy landscape model resolve it?

The Levinthal paradox states that if a 100-residue peptide chain with 3 conformations per residue randomly searched all $3^{100}$ conformations at $10^{12}\,\text{s}^{-1}$, folding would take $5 \times 10^{35}\,\text{s}$ ($\sim 2 \times 10^{28}\,\text{years}$, far exceeding the age of the universe, $1.38 \times 10^{10}\,\text{years}$). The funnelled landscape model resolves this by showing that folding is not random; the polypeptide descends a free-energy funnel toward the native state ($E_{\text{nat}}$) through progressive conformational bias via a molten globule intermediate ($Q = 0.27$), a transition state ($Q = 0.6$), and discrete intermediates.

What three major enzyme and chaperone families facilitate protein folding in vivo, and what specific reactions do they catalyze?

1. Molecular chaperones (e.g., GroEL-GroES): an ATP-dependent chamber that encapsulates unfolded polypeptides to allow folding while preventing aggregation. 2. Prolyl isomerases: catalyze the cis-trans isomerization of X-Pro peptide bonds (trans is the major conformer at $94\%$; isomerization is often rate-limiting in vitro). 3. Protein Disulfide Isomerases (PDI): reshuffle non-native disulfide pairs into native pairs using active thiols with a perturbed $pK_a$ of $7.3$.

What was demonstrated by Christian Anfinsen's classic denaturation and refolding experiment on ribonuclease A (RNase A)?

Anfinsen demonstrated that the primary amino acid sequence alone encodes all the information required for a protein to fold into its native three-dimensional structure. Denaturing RNase A in $8\,\text{M}$ urea and $\beta$-mercaptoethanol followed by simultaneous dialytic removal restored native catalytic activity, whereas removing urea first led to inactive 'scrambled' proteins with non-native disulfide pairs that required catalytic reducing agent to reshuffle.

What are the structural domains and components of a full-length Immunoglobulin G (IgG) antibody?

An IgG ($\sim 150\,\text{kDa}$, $\sim 10\,\text{nm}$ hydrodynamic diameter) consists of two heavy chains (Hc) and two light chains (Lc) held together by interchain disulfide bonds. It is organized into two Fab fragments ($V_H-C_H1$ and $V_L-C_L$) mediating antigen recognition via hypervariable complementarity-determining regions (CDRs: H1, H2, H3, L1, L2, L3), a flexible hinge region, and one Fc fragment ($C_H2-C_H3$) bearing an N-linked glycan on $C_H2$ that mediates effector functions.

What physiological mechanisms allow full-length human IgG antibodies to circulate in blood for weeks to months?

1. Lack of renal clearance: molecular mass of $\sim 150\,\text{kDa}$ is well above the glomerular filtration cutoff ($\sim 50-60\,\text{kDa}$). 2. Resistance to proteolysis: flexible loops and hinges have evolved to resist common serum proteases. 3. High structural stability. 4. FcRn (neonatal Fc receptor) recycling: endocytosed IgG binds FcRn in acidified endosomes ($\text{pH } \approx 6.0$), rescuing it from lysosomal degradation and routing it back to the cell surface where it dissociates at physiological $\text{pH } 7.4$.

What are the key functional differences and serum half-lives among human IgG subclasses (IgG1, IgG2, IgG3, and IgG4)?

IgG1 has strong ADCC and CDC with a half-life of $\sim 21\,\text{days}$ (the default backbone for killing targets). IgG2 has weak effector functions and a $\sim 21\,\text{day}$ half-life (used for neutralization/blocking). IgG3 has the strongest ADCC and CDC and a very flexible hinge, but a short half-life of $\sim 7\,\text{days}$. IgG4 is functionally 'silent' with minimal C1q and Fc$\gamma$R engagement (used when target-cell killing must be avoided, as in checkpoint inhibitors; engineered with S228P to prevent Fab-arm exchange).

How does the anti-CD20 monoclonal antibody Rituximab eliminate target B cells?

Rituximab binds CD20 on normal and malignant B cells and clears them primarily through Complement-Dependent Cytotoxicity (CDC, requiring hexameric/clustered Fc recruitment of C1q to activate the membrane attack complex MAC) and Antibody-Dependent Cellular Cytotoxicity (ADCC, mediated by NK cell Fc$\gamma$RIII activation and perforin/granzyme secretion), as well as Antibody-Dependent Cellular Phagocytosis (ADCP) by macrophages.

What molecular mechanisms of action distinguish the anti-cancer antibodies Cetuximab, Trastuzumab, and Bevacizumab?

Cetuximab binds the extracellular domain of EGFR to compete with EGF binding and prevent receptor dimerization, blocking downstream MAPK and AKT signaling. Trastuzumab targets HER2, preventing dimerization and shedding, inducing receptor endocytosis and degradation, and recruiting ADCC. Bevacizumab acts via ligand blockade, directly binding and sequestering vascular endothelial growth factor A (VEGF-A) to prevent activation of VEGFR1/2/3 and halt angiogenesis.

How do immune checkpoint inhibitors targeting CTLA-4 and PD-1/PD-L1 restore T-cell anti-tumor activity?

Without immunotherapy, CTLA-4 binds CD80/86 on antigen-presenting cells to suppress T-cell activation, and tumor PD-L1 binds T-cell PD-1 to induce T-cell exhaustion. Anti-CTLA-4 antibodies (e.g., Ipilimumab) block CD80/86 engagement to allow co-stimulation during initial priming, while anti-PD-1 (e.g., Pembrolizumab) or anti-PD-L1 antibodies block the inhibitory PD-1/PD-L1 interaction in the tumor microenvironment, revitalizing exhausted cytotoxic T cells to eliminate tumor cells.

What is Blinatumomab, what does it target, and what is the therapeutic rationale for its short serum half-life?

Blinatumomab is a Bispecific T-cell Engager (BiTE) consisting of two tandem scFvs that simultaneously bind CD3 on cytotoxic T cells and CD19 on precursor B-cell acute lymphoblastic leukemia (ALL). Because it lacks an Fc domain, it has a short serum half-life ($\sim 1-2\,\text{days}$) requiring continuous intravenous infusion; this short half-life is advantageous because stopping the infusion rapidly clears the drug to mitigate Cytokine Release Syndrome (CRS) and life-threatening hyperinflammation.

What is the humanization ladder, and why is humanization essential for murine-derived monoclonal antibodies?

The humanization ladder progresses from mouse antibody ($100\%$ murine, highest immunogenicity) to chimeric antibody ($\sim 65\%$ human, murine variable domains fused to human constant domains), humanized antibody ($\sim 90-95\%$ human, murine CDRs grafted onto human variable framework scaffolds), to fully human antibody ($100\%$ human). Progressive humanization diminishes the human anti-mouse antibody (HAMA) response and anti-drug neutralising antibodies.

What are the infection and replication steps of the filamentous M13 bacteriophage cycle?

1. Pilus binding: the pIII tip protein binds the F pilus of an E. coli host. 2. Uptake of circular ssDNA (+ strand) into the cytoplasm. 3. Synthesis of the complementary strand to form the double-stranded replicative form (RF). 4. Expression of viral proteins, including pII and pX, which nick and replicate the RF to generate progeny circular ssDNA. 5. Progeny ssDNA is bound by pV dimers to form the pV-ssDNA complex. 6. Transport to the inner membrane where pV is stripped and replaced by coat proteins (pVIII, pIII, etc.) during assembly and non-lytic secretion through the periplasm into the extracellular environment.

How does a standard biopanning cycle operate in M13 phage display?

1. A phagemid library encoding antibody fragments fused to pIII is packaged using helper phage in E. coli. 2. Target binding: phages are incubated with immobilized target antigen in solution or on a surface. 3. Washing: non-binders and low-affinity phages are washed away under increasing stringency. 4. Elution: bound phages are released (typically via low pH or trypsin protease cleavage). 5. Amplification: eluted phages infect E. coli for amplification. 6. The cycle is repeated for $3-5$ rounds before single clones are analyzed by ELISA and Sanger sequencing.

How does vector design dictate polyvalent versus monovalent display in M13 phage libraries?

A phage vector containing an intact viral genome with a fusion gene III displays the fusion protein on all $\sim 5$ copies of pIII, yielding polyvalent display (high avidity, ideal for capturing rare or weak binders). In contrast, a phagemid vector encoding the fusion gene III combined with helper phage infection (which provides wild-type pIII) results in mostly wild-type pIII and an average of $\le 1$ copy of fusion protein per virion, yielding monovalent display (selection driven strictly by monomeric binding affinity, $K_D$).

What is the mechanism of ribosome display, and what enables its exceptionally high library diversity?

Ribosome display is a cell-free in vitro selection technology. The mRNA library lacks a stop codon and contains a 3' spacer/stem-loop, causing the ribosome to stall at the end of translation without releasing the nascent protein or mRNA, forming a stable ternary mRNA-ribosome-protein complex. Because it bypasses the physical bottleneck of bacterial transformation (which limits phage/yeast libraries to $10^9 - 10^{10}$), ribosome display can achieve diversities of $10^{12} - 10^{15}$ unique members.

How does Fluorescence-Assisted Cell Sorting (FACS) operate, and what do forward scatter (FSC) and side scatter (SSC) measure?

Cells in a fluid stream pass single-file through a focused laser beam, where forward scatter (FSC) measures cell size and side scatter (SSC) measures internal granularity or complexity, while fluorescence detectors measure fluorophore signals (e.g., antigen binding and expression tags). Droplets containing single cells are given an electrostatic charge by an electrode based on predefined gates and deflected into collection tubes by positive and negative high-voltage deflection plates.

What is the fundamental genetic difference between mammalian cell display and yeast surface display?

Yeast surface display relies on autonomous, stable episomal plasmid replication (typically using an Aga1p-Aga2p cell-wall tether). In contrast, mammalian cells do not stably maintain episomal plasmids without selection over time, requiring stable genomic integration of the transgene (using retroviral/lentiviral systems, transposases, or targeted nuclease/recombinase integration).

What are the characteristic far-UV Circular Dichroism (CD) spectra of $\alpha$-helices, $\beta$-sheets, and random coils?

$\alpha$-helices exhibit two distinct negative minima at $208\,\text{nm}$ and $222\,\text{nm}$ and a strong positive peak at $\sim 193\,\text{nm}$. $\beta$-sheets exhibit a single negative minimum at $\sim 217-218\,\text{nm}$ and a positive band near $195-198\,\text{nm}$. Random coil (unfolded) conformations exhibit a strong negative minimum below $200\,\text{nm}$ (around $195\,\text{nm}$) and minimal ellipticity above $210\,\text{nm}$.

How does a thermal shift assay (differential scanning fluorimetry) monitor protein thermal denaturation?

A fluorescent dye (such as SYPRO Orange) that is quenched in aqueous solution is mixed with the protein. As the temperature rises, the protein unfolds, exposing buried hydrophobic patches. The dye binds to these hydrophobic regions, causing a sharp increase in fluorescence. As the temperature continues to rise past unfolding, aggregation causes dye displacement and signal decay. The inflection point of the initial fluorescence transition marks the melting temperature ($T_m$).

How does Surface Plasmon Resonance (SPR) measure binding kinetics and affinity ($K_D$)?

A ligand is immobilized on a sensor chip (typically gold-coated). When analyte in buffer flows over the surface, binding increases the mass and refractive index near the sensor surface, altering the resonance angle of reflected light and generating a response curve in Response Units (RU). The association phase yields the association rate constant ($k_{\text{on}}$), the steady-state plateau reflects binding quantity and affinity, and the dissociation phase during buffer washing yields the dissociation rate constant ($k_{\text{off}}$). The equilibrium dissociation constant is calculated as $K_D = \frac{k_{\text{off}}}{k_{\text{on}}}$.

What is the standard industrial purification chromatography workflow for therapeutic full-length monoclonal antibodies?

1. Protein A affinity capture: selectively binds the IgG Fc domain at neutral pH without requiring epitope tags. 2. Low-pH elution: elutes IgG from Protein A while simultaneously inducing viral inactivation. 3. Ion exchange chromatography: removes charged impurities, DNA, host cell proteins, and trace leachates. 4. Size Exclusion Chromatography (SEC): polishes the sample by removing high-molecular-weight (HMW) aggregates and low-molecular-weight (LMW) fragments.

Why is deglycosylation necessary prior to intact mass spectrometry analysis of a monoclonal antibody?

Therapeutic antibodies carry complex, heterogeneous N-linked oligosaccharide structures on the $C_H2$ domains of their Fc regions. This glycan heterogeneity splits the mass spectrometric signal into multiple overlapping glycoform peaks; enzymatic deglycosylation (e.g., using PNGase F) removes this heterogeneity to generate a single, sharp, unambiguous mass peak for verifying the antibody's primary mass.

In an Antibody-Drug Conjugate (ADC), what is the Drug-to-Antibody Ratio (DAR), and what is the bystander effect?

The DAR is the average number of cytotoxic drug payload molecules chemically conjugated to each antibody molecule (typically $\sim 4$; higher DAR increases potency but causes hydrophobicity-driven aggregation and rapid clearance). The bystander effect occurs when the released cytotoxic payload is membrane-permeable, allowing it to diffuse out of the targeted antigen-positive cell and kill adjacent antigen-negative tumor cells within the tumor mass.

What paratope shapes and structural designs define the three synthetic single-domain antibody (sybody) libraries developed by Zimmermann et al.?

1. Concave library: short CDR3, forming a concave cleft ideal for inserting protruding antigen surfaces ($342\,\text{bp}$). 2. Loop library: medium-length CDR3 forming a protruding, flexible loop ($360\,\text{bp}$). 3. Convex library: long 16-amino acid CDR3 supported by an extended hydrophobic core to limit loop flexibility and form a compact convex paratope ($372\,\text{bp}$).

Why was Fragment Exchange (FX) cloning utilized in the sybody generation protocol, and what restriction enzyme does it employ?

FX cloning uses the Type IIS restriction enzyme BspQI (or its isoschizomer SapI), which cleaves outside its non-palindromic recognition sequence. This allows efficient, seamless subcloning of enriched sybody pools between phagemid (pDX_init) and expression vectors (pSb_init, pBXNPH3/M3) without PCR amplification, preventing PCR-induced CDR shuffling across highly conserved framework regions.

How was the bacterial ABC exporter TM287/288 trapped in an outward-facing conformation using sybodies by Hutter et al.?

A mutation was introduced into the consensus Walker B motif (E517A) of TM287/288 to allow ATP binding while completely blocking ATP hydrolysis. Sybody selection was then conducted in vitro in the continuous presence of saturating ATP and $\text{Mg}^{2+}$, resulting in the selection of the convex sybody Sb_TM#35, which binds to an extracellular wing, clamps open the outer gate, and enabled solving the outward-facing crystal structure at $3.2\,\text{Å}$.

What dominant heavy-chain V gene was identified in the patient-derived SARS-CoV-2 neutralizing antibody STE90-C11 (COR-101), and what was its neutralization potency?

STE90-C11 uses the IGHV3-66 heavy chain germline gene (combined with IGKV1-9), exhibiting a very high germinality index ($98.6\%$ VH, $100\%$ VL). It demonstrated an $EC_{50}$ of $0.2-0.3\,\text{nM}$ by ELISA on RBD, S1, and S1-S2, and neutralized authentic, live SARS-CoV-2 with a subnanomolar $IC_{50}$ of $0.56\,\text{nM}$ in a plaque reduction neutralization assay.

What structural features and mutation tolerances were revealed by the $2.0\,\text{Å}$ crystal structure of STE90-C11 Fab in complex with the SARS-CoV-2 RBD?

STE90-C11 binds directly at the ACE2-RBD interface with a buried surface area of $1113\,\text{Å}^2$ ($60\%$ contributed by VH, $40\%$ by VL), engaging all six CDR loops. Its extensive light-chain contact network enables it to tolerate major emerging RBD mutations including E484K, N439K, and L452R (and B.1.617 L452R+E484Q). However, it loses binding to K417N/K417T due to the loss of a key salt bridge with heavy-chain residue D101, and exhibits reduced binding to N501Y.

Why was the Fc region of STE90-C11 (COR-101) engineered with silencing mutations for clinical development?

COR-101 was engineered with a 'silenced Fc' containing mutations (E233P, L234V, L235A, $\Delta\text{G236}$, D265G, A327Q, and A330S) to abolish binding to Fc$\gamma$ receptors and C1q. This eliminates Fc-mediated effector functions (ADCC, CDC) to prevent antibody-dependent enhancement (ADE) and avoid exacerbating acute lung inflammation in hospitalized COVID-19 patients.

What results were observed when STE90-C11 was tested in Syrian hamster and K18-hACE2 transgenic mice challenge models of SARS-CoV-2?

In Syrian hamsters infected with $1 \times 10^5\,\text{pfu}$ authentic virus, treatment with $37\,\text{mg/kg}$ STE90-C11 reduced lung viral load from $1.5 \times 10^6\,\text{pfu}$ (PBS control) down to $8.3 \times 10^3\,\text{pfu}$ at day 3, significantly attenuating body weight loss and speeding recovery. In K18-hACE2 mice infected with 1,000 or 2,000 pfu, doses between $6\,\text{mg/kg}$ and $120\,\text{mg/kg}$ completely cleared detectable virus from the lungs in nearly all animals by day 5 and prevented infection-induced weight loss.

How does the Optical Pooled Screening - Receptor Diversion (OPS-RD) assay operate to identify GPCR-targeting miniproteins in human cells?

Human cells co-express a GFP-tagged target GPCR and a library of barcoded, ER-retained (KDEL-tagged) miniprotein designs fused to mCherry. In the absence of binding, the GPCR traffics to the plasma membrane while the binder remains in the ER. Upon high-affinity binding, the miniprotein traps the GPCR within the secretory pathway, creating a subcellular GFP-mCherry colocalization phenotype. Positive hits are genotyped at the single-cell level via in situ sequencing of the non-coding DNA barcode.

What two computational strategies did Muratspahić et al. use to design de novo miniprotein binders targeting GPCRs?

1. Motif-guided RFdiffusion: docks a 5-residue de novo peptide motif deep into the recessed orthosteric pocket to engage key hotspot residues, keeps the motif fixed to generate a full miniprotein backbone, and refines designs via iterative partial diffusion. 2. MetaGen: docks 8,920 structurally diverse, monomeric scaffolds extracted from the AlphaFold structural metaproteome into the receptor binding site using Rosetta RifDock, followed by ProteinMPNN sequence design and AlphaFold2 filtering.

What were the pharmacological and structural properties of the de novo designed MRGPRX1 agonists mM1_068 and mM1_034_F12W_Y27F?

Agonist mM1_068 is a proline-kinked three-helix bundle that activates MRGPRX1 in calcium mobilization assays; its cryo-EM structure with MRGPRX1 and mini-Gq was solved at $3.29\,\text{Å}$ resolution, aligning with the computational design model with a $0.7\,\text{Å}$ C$\alpha$ RMSD. Site-saturation mutagenesis of hit mM1_034 yielded mM1_034_F12W_Y27F, which demonstrated a 12-fold affinity improvement to an $EC_{50}$ of $42\,\text{nM}$, matching the potency of the endogenous peptide agonist BAM(8-22).

How did the designed CXCR4 antagonist dCX1_001 perform in vivo compared to the clinical drug AMD3100 (plerixafor)?

dCX1_001 ($IC_{50} = 24\,\text{nM}$, $pA_2 = 7.6$) mobilized haematopoietic stem and progenitor cells (Lin-SCA1+KIT+ / LSK cells) and colony-forming units (CFUs) in CD46-transgenic mice at levels comparable to AMD3100 ($5\,\text{mg/kg}$ subcutaneous injection). Notably, dCX1_001 caused significantly less unwanted leukocytosis and did not induce systemic inflammatory cytokine release, showing equal mobilization efficacy with reduced toxic side effects.

What structural and pharmacological characteristics define the de novo designed CGRPR antagonists dC2_049 and mC2_022?

dC2_049 (designed by RFdiffusion and partial diffusion) is a competitive antagonist of CGRPR with an $IC_{50}$ of $4.5\,\text{nM}$ and a $pA_2$ of $8.4$ ($3.9\,\text{nM}$), whose cryo-EM structure at $3.18\,\text{Å}$ shows it sterically blocks the C-terminus of $\alpha\text{CGRP}$. mC2_022 is a disulfide-stapled MetaGen antagonist ($pA_2 = 7.9$). When mC2_022 was fused to an Fc domain, its in vivo clearance in mice dropped $>200\text{-fold}$ and its elimination half-life extended from $0.4\,\text{hours}$ to $25\,\text{hours}$.

How was RFdiffusion modified and fine-tuned by Bennett et al. to achieve de novo, epitope-specific antibody design?

RFdiffusion was fine-tuned on antibody-antigen complexes by conditioning the network on a user-specified therapeutic antibody framework (provided invariant to the global coordinate frame as a 2D pairwise distance/dihedral template track) and one-hot encoded epitope hotspot residues. This enabled the network to de-noise Gaussian coordinate noise to simultaneously sample novel CDR loop backbones and explore diverse rigid-body docking orientations against the target epitope.

What were the structural resolution and atomic accuracy of the de novo designed VHH_flu_01 in complex with influenza haemagglutinin?

Cryo-EM resolved VHH_flu_01 bound to the stem domain of natively glycosylated trimeric influenza H1 haemagglutinin (strain A/USA:Iowa/1943 H1N1) at $3.0\,\text{Å}$ resolution. The structure revealed an overall VHH backbone RMSD of $1.45\,\text{Å}$ and a CDR3 backbone RMSD of $0.8\,\text{Å}$ relative to the computational design model, with binding requiring local reorientation of the glycan at residue N296.

How did Bennett et al. overcome single-CDR design failure when designing two-chain scFvs containing six de novo CDR loops?

They used a structure-aware combinatorial assembly strategy: designs targeting the same epitope with nearly identical binding orientations (high pairwise TM scores) were grouped into clusters ($\ge 100$ designs). PCR subpools were used to combinatorially mix heavy and light chains across the cluster, generating high-diversity libraries ($\sim 10^7$ members matching yeast transformation capacity). This allowed functional combinations to complement a suboptimal CDR on one chain with an optimal CDR on the partner chain.

What were the structural features and CDR backbone accuracies of the de novo designed scFv6 bound to Clostridium difficile toxin B (TcdB)?

Cryo-EM resolved scFv6 bound to the Frizzled-binding epitope of the TcdB delivery and receptor-binding domain (DRBD) at $3.6\,\text{Å}$ resolution ($K_D = 72\,\text{nM}$ as scFv, $68\,\text{nM}$ as full IgG1). The structure matched the composite computational design with an overall RMSD of $0.9\,\text{Å}$ and near-atomic precision across all six CDR loops: CDRH1 ($0.4\,\text{Å}$), CDRH2 ($0.3\,\text{Å}$), CDRH3 ($0.7\,\text{Å}$), CDRL1 ($0.2\,\text{Å}$), CDRL2 ($1.1\,\text{Å}$), and CDRL3 ($0.2\,\text{Å}$).

How did the de novo designed scFv B1.2.1 target the neuroblastoma-associated PHOX2B peptide-MHC complex?

scFv B1.2.1 was designed with RFdiffusion to target the QYNPIRTTF peptide presented on the common HLA-C*07:02 allotype by focusing on the solvent-exposed R6 residue. It bound the disulfide-stabilized complex with an affinity of $400\,\text{nM}$ (SPR) and $1\,\mu\text{M}$ (ITC). Binding was highly peptide-specific, showing completely undetectable binding to the mutant complex containing a PHOX2B(R6A) peptide substitution.