Designed Ankyrin Repeat Proteins

DARPin N3C and N2C libraries (Binz et al., 2004 (link)) were used and ribosome display (RD) selections were performed as described (Dreier and Plückthun, 2011 (link)). Green Fluorescent Protein (GFP) or mCherry, containing a C-terminal His-tag and an N-terminal avi-tag for in-vivo biotinylation during expression, were used as targets. Four rounds of RD with increasing stringency were carried out in solution with pull-down of the ternary complexes using streptavidin-coated magnetic beads. In the fourth round an off-rate selection with 300-fold excess of unbiotinylated GFP or mCherry over biotinylated target was performed. After the fourth round the enriched DNA pools were subcloned into the expression vector pQIq (Simon et al., 2012 (link)). For each selection, 192 single clones of each pool were screened by crude extract ELISA as described previously (Binz et al., 2004 (link); Zahnd et al., 2006 (link)). All single clones were also screened for binding to superfolder GFP (sfGFP). The nomenclature of the binders is as follows: the first number indicates the N2C or N3C library, respectively, the letters G or m indicate a DARPin specific for GFP or mCherry, respectively, and the last two- to three-digit number is a continuous numbering of the 192 clones that were screened. 3G86.1 and 3G86.32 come from the same initial clone that turned out to be a double transformant; single transformants were obtained by plasmid extraction and retransformation.

Data were collected at the Swiss Light Source beamline PX (http://sls.web.psi.ch/view.php/about/index.html) and processed using the program XDS [24 ]. The crystal belonged to space group P2₁2₁2₁, with a Matthews coefficient V_M of 3.8 ų/Da, corresponding to an estimated water content of 67.7%.
The crystal structure was solved by molecular replacement using the program PHASER [25 (link),26 (link)], with the structure of the AcrB monomer (Protein Data Bank code 1IWG [6 (link)]) used as search model. A PHASER search with the DARPin E3_5 (Protein Data Bank code 1MJO [13 (link)]) as search model did not yield a meaningful solution. The information obtained from the conventional PHASER protocol for the three AcrB monomers was sufficient to model one DARPin molecule into the resulting electron density with the program O [27 (link)]. The second DARPin also was positioned in O. Refinement of the structure was carried out through multiple cycles of manual rebuilding using the program Coot [28 (link)] and refinement using CNS [29 (link)] resulting in a final model with an R factor of 22.9 and an R_free factor of 27.9. The refined structure of the AcrB–DARPin complex was validated by the program PROCHECK [30 ]. Three-dimensional structural figures were prepared by using PyMOL [31 ].

The starting models for simulations were built from the respective XFEL-based time-resolved structures. We removed the DARPin protein and reconstructed the βM-loop (res 272-288) using the one in chain D of PDB 4I4T as a template by employing the Prime^{65 (link)} module of the Schrödinger 2020-4 suite. Analogously, Mg²⁺ ion in chain B (β-tubulin) was added by superimposition with the one of chain D of 4I4T structure. Calcium ions were removed from the model.
The final model consisted of an αβ-tubulin heterodimer in complex with azo-CA4, GTP, and GDP molecules bound to the α- and β-tubulin subunits, respectively, as well as their associated Mg²⁺ ions and their coordinating water molecules. The resulting protein structure possessed 437 out of 451 residues of α-tubulin (UniProtKB ID P81947) and 431 out of 445 residues of β-tubulin (UniProtKB ID Q6B856). Missing residues belonging to the intrinsically disordered C-terminal tails of α- and β-tubulin were not modeled and C-termini were capped with N-methyl amide (NME) groups. Residue protonation states were evaluated at pH 7.0 using the Protein Preparation Wizard tool^{66 (link)} implemented in the Schrödinger 2020-4 suite⁶⁷ . The αβ-tubulin heterodimer structure was solvated with the TIP3P-model^{68 (link)} for water molecules in a truncated octahedron box using 12 Å as the minimum distance between the protein and the box edges. The system was neutralized by adding Na⁺ ions resulting in a total of about 122k atoms. The atomistic force field Amber-ff14SB^{69 (link)} was used for all simulations. Parameters for Mg²⁺ ions and the GTP and GDP molecules were developed by Allner et al.^{70 (link)} and Meagher et al.^{71 (link)}, respectively.
Concerning the ligand, partial atomic charges were obtained with the Restrained Electrostatic Potential (RESP) method on a DFT-optimized trans conformation at the B3LYP/6-31 g* level. The geometric optimization was carried out with Terachem v1.94. The General Amber Force Field (GAFF) was used, with custom parameters for the three main dihedral angles: 1-1a-1b-1’, 6-1-1a-1b, 1a-1b-1’-6’.
To obtain such optimized parameters, we fitted the GAFF energy curves associated with a rigid torsion of the dihedral angles to the corresponding DFT energy curves. For each dihedral, we proceeded as follows: we started out with the geometrically optimized trans structure. We then performed a clockwise rigid rotation with 15° steps followed by an analogous counter-clockwise rigid rotation. For each rotation step we performed a DFT geometric optimization restraining the rotated dihedral, and we evaluated the energy. We retained the minimum energy among the clockwise and anticlockwise rotation cycles to overcome hysteresis. The three resulting DFT energy profiles consisting of 24 evaluation points were used as a target for the fitting procedure. DFT calculations were performed with Terachem v1.94, at the B3LYP/6-31 g* level. The fitting procedure was carried out separately for the three energy profiles with mdgx, part of AmberTools21. The αβ-tubulin heterodimer system was assembled with the LEaP tool implemented in the AmberTools21 software package⁶⁷ .
The designed models for QM calculations consisted of protein residues α(99-101, 178-181), β(237-242, 248-259, 314-321, 349-354, 378-380), with the respective N- and C- termini capped with acetyl (ACE) and NME groups, the azo-CA4 molecule, and its two closest water molecules. They resulted in 743 atoms.

For crystallization, the tubulin-DARPin D1 (TD1) complex was formed by mixing the respective components in a 1:1.1 molar ratio. The TD1 complex was crystallized using EasyXtal 15-well plates by the hanging drop vapor diffusion method (drop size 2 μl, drop ratio 1:1, 8 drops per well) at a concentration of 9.8 mg ml⁻¹ at 20 °C with a precipitant solution containing 21% (w/v) polyethylene glycol (PEG) 3000, 0.2 M ammonium sulfate, and 0.1 M bis-tris methane, pH 5.5. All drops were subsequently hair-seeded with crystalline material obtained in previous crystallization trials to increase the homogeneity and density of crystals. After 48 h, crystals were washed off the plates with precipitant solution, collected in 0.6 ml tubes, and vortexed for a few seconds. This procedure induces batch crystallization within the tubes and a sedimented crystal pellet was formed after 24 h of incubation at 20 °C. Initially, crystals grew as long needles that were broken into smaller fragments of approximately 20 × 20 × 5 μm³ during the sample preparation described below.

Ec1–LoPE fusion protein and DARPin Ec1 were produced as described previously [47 (link),61 (link)]. The monoclonal antibody seribantumab (MM-121) was purchased from Merrimack Pharmaceuticals (Cambridge, MA, USA). The molecular weights of the proteins used in this study were: Ec1–LoPE 43061 Da [49 (link)], Ec1 18348 Da [61 (link)], and MM-121 143151 Da.