Crystal structures of a llama VHH antibody BCD090-M2 targeting human ErbB3 receptor

Introduction

Receptor tyrosine kinases ErbB1-4 (HER1-4) receive inputs from growth factors and transmit signals to the cell nucleus, thus regulating key cellular processes such as growth, differentiation, migration, and apoptosis¹. Aberrations of ErbB signaling, caused by mutations or receptor overexpression, are associated with the development of a wide variety of cancers. The essential role of ErbB receptors in tumor development makes them a unique target in cancer therapy by monoclonal antibodies². Therapeutic antibodies often act on the first stage of signal transduction by inhibiting ligand binding or receptor dimerization.

The first two members of the family, ErbB1 (EGFR, HER1) and ErbB2 (HER2/neu), were early recognized as promising drug targets because of their frequent overexpression in tumors. The examples of successful application of anti-ErbB antibodies in cancer treatment include cetuximab (anti-EGFR) in head and neck cancer therapy and trastuzumab (anti-HER2) in breast cancer treatment. The role of the third member, ErbB3, has long been underestimated because it lacks intrinsic tyrosine kinase activity. However, its ability to form functional dimers with ErbB1-2 and to confer resistance to their inhibitors makes ErbB3 an important drug target³. Particularly, it was shown that inhibition of ErbB2 with lapatinib caused transcriptional up-regulation of ErbB3, which was then phosphorylated by residual ErbB2 kinase activity thus limiting antitumor effect⁴. A comprehensive clinical study revealed that ErbB3 overexpression was a significant marker of reduced survival in patients with breast cancer⁵.

This new data stimulated the development of anti-ErbB3 antibodies, which are at various stages of clinical trials⁶. The rational antibody design requires knowledge of molecular mechanism of ErbB3 inhibition. Recently, several structures of ErbB3-antibody complexes were solved^7–9. Surprisingly, these structures showed that antibodies target entirely different epitopes on the receptor: extracellular domain I⁷, domains II and IV⁸, or domain III⁹.

In Russia, anti-ErbB3 antibodies are developed by BIOCAD biotechnology company. The phage display selection of antibodies from immunized llamas and subsequent sequencing allowed the identification of several anti-ErbB3 single-chain antibodies. As a part of our ongoing effort to elucidate the molecular mechanism of ErbB3 inhibition and ultimately open up a possibility of therapeutic application of these antibodies, we study their thermodynamic stability¹⁰, functional properties, and structure. In this work, we describe the expression, purification, crystallization and structural analysis of the variable fragment of an antibody BCD090-M2, which demonstrated an affinity to the extracellular domain of ErbB3 in preliminary experiments.

Methods

Results and discussion

The llama VHH antibody BCD090-M2 was expressed in soluble form in the cytoplasm of E. coli SHuffle T7 Express cells as a SUMO fusion. The SHuffle strain¹³ has deletions of the genes trxB and gor, and constitutively expresses a chromosomal copy of the disulfide bond isomerase DsbC to promote formation of correct disulfide bonds in recombinant proteins. The system proved very efficient for single-chain antibody expression, the typical protein yield in our experiments was 50 mg of a fusion protein per liter of a bacterial culture after the first IMAC step. For the IMAC we used new chromatography media cOmplete (Roche), which withstands high EDTA and DTT concentrations and demonstrates a different binding strength and specificity compared to traditional Ni-NTA resins. Particularly, our histidine-tagged fusion protein eluted at 30 mM Imidazole concentration in gradient elution experiments, and addition of only 5 mM Imidazole to cell extract and wash buffer efficiently suppressed non-specific binding, resulting in greater than 90% purity after the first chromatography step. The protein was then cleaved by TEV protease to obtain untagged antibody BCD090-M2 with nearly native N-terminus, differing from the original sequence by a single additional glycine residue left from the TEV recognition site. The extent of cleavage was monitored by SDS-PAGE and typically was more than 70% at 1:80 enzyme ratio and more than 85% at 1:40 ratio, as shown in Figure 1 (left panel).

Figure 1. Expression and purification of BCD090-M2 VHH antibody.

The antibody was expressed in the cytoplasm of E. coli SHuffle cells as His₆-SUMO fusion, purified by IMAC and cleaved with TEV protease at 1:80 or 1:40 enzyme:substrate ratio. The bottom band marked with a red arrow corresponds to BCD090-M2, the middle band marked with a blue arrow is a histidine-tagged SUMO, and a top band is an uncleaved protein. After the cleavage, BCD090-M2 was further purified by negative IMAC and cation-exchange chromatography on MonoS, the SDS-PAGE analysis of the purified antibody is presented in the right panel.

The bottom band with apparent molecular mass 14 kDa corresponds to processed BCD090-M2, the middle band corresponds to His₆-SUMO (13 kDa) and migrates anomalously slow probably due to positively charged histidine tag, and the top band is the uncleaved protein. The processed BCD090-M2 was separated from His₆-SUMO and intact fusion protein by negative IMAC chromatography in batch regime, and then polished by an additional step of cation exchange chromatography on a MonoS column. After elution from MonoS, the VHH antibody was almost pure as judged by SDS-PAGE shown in Figure 1 (right panel).

The surface plasmon resonance experiment confirmed that the recombinant VHH antibody was functional and efficiently bound to the immobilized receptor. The representative experimental binding sensograms are shown in Figure 2A, and the processed data fitted with an equilibrium binding model is shown in Figure 2B. The experiments showed little variation and yielded mean equilibrium dissociation constant for monovalent binding K_D=15±1 nM. Although the monovalent affinity is a fundamental characteristic of antibody-antigen interaction, the avidity of a full-length bivalent antibody can be much higher. Typically, the enhancement of avidity due to a bivalent interaction is 3–4 orders of magnitude and depends strongly on the surface concentration of antigen¹⁴.

Figure 2. Binding of BCD090-M2 VHH antibody to the extracellular domain of human ErbB3 receptor analyzed by surface plasmon resonance.

The purified extracellular domain (residues 21-643) was immobilized on CM5 chip surface. A) binding sensograms measured for different concentrations of BCD090-M2; B) fitting of the measured response as a function of concentration of free BCD090-M2 in solution with a simple bimolecular equilibrium binding model and averaging over three experiments gives K_D=15±1 nM.

The protein was successfully crystallized in two different forms: in space group C2 with a single copy of BCD090-M2 in the asymmetric unit and in P1 with two molecules and two cadmium ions in the unit cell. After data collection and processing, all further data analysis procedures, including phasing, model building, and refinement, were conducted in Phenix software suite v. 1.11.1_2575¹⁵. To solve the structures by molecular replacement, we selected a set of single-domain antibodies from PDB with the highest homology with BCD090-M2. We then processed the search models with Sculptor¹⁶ to delete residues that were not aligned with the target and to prune sidechains by the Schwarzenbacher algorithm¹⁷, and performed molecular replacement using Phaser v. 2.7.16¹⁸. The best molecular replacement solutions for both datasets were obtained with a nanobody targeting complement receptor Vsig4, PDB:5IMK¹⁹. Then we used phenix.autobuild²⁰ to automatically build the framework regions of the BCD090-M2 and Coot v. 0.8.6.1²¹ to manually fit the missing CDRs into the experimental electron density. The structures were refined using phenix.refine²² (see Table 1 for refinement statistics) and deposited to the Protein Data Bank under entry codes 6EZW and 6F0D. All figures were generated using PyMOL.

The overall structure of BCD090-M2 crystallized in space group C2 is shown in Figure 3. The framework regions are drawn as grey ribbons, and the CDRs are colored in orange (H1), blue (H2), and red (H3). The CDRs were defined according to Kabat²³ as shown in Figure 4.

Figure 3. Ribbon diagram of the crystal structure of BCD090-M2 in space group C2 at 1.6 Å resolution (PDB:6EZW).

Framework regions are grey and CDRs are colored.

Figure 4. Amino acid sequence of BCD090-M2.

The numbering is the same as in the PDB files 6EZW and 6F0D. The CDR regions according to Kabat definition are indicated with color and frames. In the case of CDR H1, the AbM definition (dashed frame, residues 27-36) was used.

For the CDR H1, we also used AbM definition, a combination of Kabat and Chotia²⁴ definitions, which better matches the loop in protein structure. The high resolution of the dataset and good quality of the electron density maps allowed us to build a precise atomic model of the antibody. We then used a PyIgClassify²⁵ to analyze the structures of BCD090-M2 CDRs using a comprehensive database of antibody CDR loop conformations. The CDR H1 and H2 belonged to two large common clusters, H1-13-1 and H2-10-2 respectively. The CDR H3 was not assigned to any known structure cluster, probably due to size (18 residues) and complex structure of the loop, which has a cis-proline and four aromatic residues. It is consistent with an observation that CDR H3 loops are very diverse in structure and a few clusters of significant size are present in the database.

The structure of the BCD090-M2 dimer crystallized in space group P1 with cadmium ions is shown in Figure 5. Two cadmium ions in the unit cell are pictured as green spheres, and two symmetry-related ions from the neighbor unit cell are shown in dots. Each cadmium ion is bound by residues Asp100, Glu114, and Asp116 belonging to CDR H3, and by N-terminal glycine residue of an antibody molecule from neighbor unit cell. Thus, intermolecular interaction through cadmium ions effectively define the lattice of a crystal. This finding is consistent with a previous general observation that cadmium can induce the formation of protein crystals or improve their quality²⁶.

Figure 5. Ribbon diagram of the crystal structure of BCD090-M2 in space group P1 at 1.9 Å resolution (PDB:6F0D).

Asymmetric unit contains two copies of the VHH antibody and two cadmium ions which are pictured as green spheres, two symmetry-related ions from the neighbor unit cell are shown in dots. Framework regions are grey and CDRs are colored.

Finally, we analyzed the structural variability of the overall antibody fold and the putative paratope by comparing the three solved structures 6EZW, 6F0D (A), and 6F0D (B). We first performed the structural alignment based on C^α atoms of the framework regions, which is shown in Figure 6A.

Figure 6. Structural comparison of BCD090-M2 crystallized in different space groups.

A) the three structures, 6EZW, 6F0D chain A, and 6F0D chain B, were superimposed, the structural alignment was based on C^α atoms of framework regions. CDRs H1 and H3 showed little structural variation, while CDR H2 adopted different conformations in 6EZW and 6F0D, which are shown in blue and green color. B) Electron density map (F-obs, Phi-model, 1.0 σ) for CDR H2 (residues 51-67) of 6EZW. The quality of electron density maps allowed us to unambiguously trace the loop and place amino-acid side chains.

The overall fold of the antibody had a very low structural variability, the RMSD between framework C^α atoms in 6EZW and 6F0D was 0.19 Å. The conformational mobility of the short CDR H1 was also low, with C^α RMSD between 6EZW and 6F0D 0.28 and 0.31 Å for chains A and B, respectively. Surprisingly, the most complex CDR H3 had a rigid conformation, which was not significantly altered by binding of a cadmium ion and intermolecular interaction in the asymmetric unit of 6F0D. The C^α RMSD for CDR H3 between two structures was 0.61 (chain A) and 0.57 Å (chain B). The largest structural variation was observed in the loop H2, as seen in Figure 6A. The C^α RMSD was 1.81 and 1.77 Å for chains A and B, respectively, and the largest difference was observed in the region Arg53-Leu57. Despite the probable flexibility of this loop, in both crystals it was well-resolved in the electron density maps, as shown in Figure 6B, allowing us to unambiguously place all residues. As a result of the observed structural change, the CDRs H2 in 6F0D were attributed to a different from 6EZW distant structural clusters H2-10-6 (chain A) or H2-10-7 (chain B) by PyIgClassify²⁵.

Conclusions

In conclusion, here were present expression and purification, functional and structural analysis of a new single-domain llama antibody against human ErbB3. We crystallized the antibody in two different forms and solved high-resolution structures, giving an insight to the organization of the putative ErbB3 binding paratope. We believe this data may facilitate further studies of mechanisms of ErbB3 inhibition by single-chain antibodies. Besides, the solved structures will contribute to datasets that are required to develop new robust computational methods for antibody modeling and design.

Data availability

The atomic coordinates and structure factors can be accessed under PDB codes 6EZW and 6F0D.

Dataset 1: Uncropped gel from Figure 1 and raw output data from Biacore software. DOI, 10.5256/f1000research.13612.d209631²⁷

The plasmids and recombinant proteins used in this study are available from Igor Eliseev (corresponding author) upon request.