Shared ligand-blocking mechanism but distinct conformational modulation by α5-targeting antibodies BIIG2 and MINT1526A

Mouse models:

Harlan Sprague Dawley female athymic nude mice that are recessive for the nu mutation in the Foxn1 gene (Envigo) was used. This mouse strain exhibits a non-functional thymus and are unable to generate cytotoxic effector cells and do not elicit a graft-versus-host response. Mice selected for experimentation were 6 to 7 weeks old, weighing approximately 20 to 30 grams. Their housing conditions involved a 12-hour light/dark cycle with ad libitum access to food and water.

Cell lines:

Human tissue samples:

A 23 core normal human formalin-fixed paraffin-embedded (FFPE) tissue microarray was purchased from Bio SB (BSB 0298). FFPE tissue of patient-derived nevi and melanoma were obtained through an institutional tissue registry. The Mayo Clinic Institutional Review Board approved the human investigations performed in this study following the Department of Health and Human Services requirements, where appropriate.

2D-tube angiogenesis formation assay:

Endothelial tube formation assays experiments were pursued through Creative Bioarray project CBADY0508. HUVECs (human umbilical vein endothelial cells) were seeded at 5×10⁵ cells/well in 2 mL of complete culture medium in 6-well plates and cultured in a 37°C and 5% CO₂ incubator. Prior to experimentation, the culture medium was removed. Cells were subsequently treated with 2 mL of culture medium (negative control), 15 ng/mL VEGF (positive control), and BIIG2 or MINT1526A (1, 10, and 100 nM) supplemented with VEGF (15 ng/mL), then incubated at 37°C for 24 hours. Each condition was performed in triplicate. HUVECs were harvested and suspended in serum-free medium and then seeded in Matrigel-coated wells at a density of 1.5×10⁵/1 mL/well. The cells were incubated at 37°C for 6 hours to allow tube formation.

Spheroid sprouting assay:

Experiments were pursued through Reaction Biology under project PQ20385, with modifications of a previously published protocol ⁸⁶. HUVECs were seeded at 1 x 10⁴ cells/cm². HUVEC’s were subcultured every 3-5 days by splitting 1 volume of cells to 3 volumes of ECGM/ECBM media (endothelial cell growth and basal medium, PromoCell) at 37°C with 5% CO₂. Experiments and spheroid preparation are adapted from previous descriptions ^86,87. 400 HUVEC were pipetted in a hanging drop to allow overnight spheroid aggregation. 50 HUVEC spheroids were then seeded in 0.9 mL collagen gel and pipetted into a 24 well plate to allow polymerization. The inhibitors combined with 25 ng/mL of VEGF-A were added after 30 minutes by pipetting 100 μL of a 10-fold concentrated working solution on top of the polymerized gel. Plates were incubated at 37°C for 24 hours before being fixed with 4% paraformaldehyde PFA. Sprouting intensity of HUVEC spheroids were captured using an inverted microscope and the digital imaging software NIS-Elements BR 3.0 (Nikon) by determining the cumulative sprout length per spheroid (CSL). 10 randomly selected spheroids were measured and treated as an individual data point.

Mouse tumor experiment:

The mice were randomized into treatment groups and monitored daily. Tumor cells (M12 or M15) embedded in Matrigel were injected subcutaneously into the flank of the athymic mice. In efficacy studies, groups of mice (10 per group) with flank tumors were administered treatment once the tumor volumes became palpable (approximately 50 to 100 mm³). The flank tumors were measured three times a week. Treatment was administered twice weekly at a dose of 10 mg/kg. The experiments were concluded when the average tumor volume in the control group exceeded 1,000 mm³. All experimental procedures adhered to the Principles of Laboratory Animal Care (NIH) and were approved by the Institutional Animal Care and Use Committee (Mayo Clinic).

Immunohistochemistry:

Proteins of interest were detected using the Leica Biosystems BOND-MAX autostainer platform in combination with primary antibody (complete list in Key Reagents Table), post-primary antibody, and a BOND Polymer Refine Red Detection system (DS9390; Leica Biosystems). Heat-induced antigen retrieval was by BOND EDTA-based pH 9.0 epitope retrieval solution 2 (AR9640; Leica Biosystems) applied for 20 minutes. All stains were digitized with a Leica Aperio ScanScope system.

Flow cytometry:

Normal human melanocytes, M12 and M15 melanoma cells were stained using primary antibodies (complete list in Key Reagents Table) against cell surface proteins as shown and PE-labeled secondary antibodies. The BD Bioscience FACScaliber flow cytometer was utilized for data acquisition, and the acquired data was analyzed using FlowJo software.

Quantitative Microfluidic PCR:

Our approach to PCR was as previously described ⁸⁸. RNA from cells was isolated using the RNeasy Plus Mini Kit (Qiagen). One microgram of total RNA was transcribed into cDNA using the SuperScript III Reverse Transcriptase (Thermo Fisher Scientific) and preamplified using TaqMan Preamp Master Mix (Thermo Fisher Scientific). Quantitative reverse transcriptase PCR was then performed using the Fluidigm BioMark HD system and dynamic array-integrated fluid circuits (Fluidigm). After thermal cycling, raw Ct (threshold cycles) data were checked for linear amplification. Gene expression was corrected by the mean of housekeeping genes (RLP0, RLP8, and β-actin) using the ΔCt method. Gene expression input for data analysis was ΔCt. The complete list of primers (HPLC purified, Integrated DNA Technologies) used for cDNA amplification by quantitative PCR are described in Table S5.

Recombinant α5β1 plasmid design:

The sequence of the α5 and β1 subunit wild-type ectodomain was incorporated into a CDNA3.1-Hygro(−)-TET template (Invitrogen). The integrin amino acid numbering starts after the signal peptide. Each integrin subunit was fused with either an ACID or a BASE coiled-coil peptide to assist in heterodimer production ⁸⁹. Residues 1 to 954 of the α5 subunit were cloned followed by a HRV-3C protease cleavage recognition site, an ACID coil, and a Strep-Tag II. The residues 1-728 of the β1 subunit were put into the same plasmid backbone filled by a HRV-3C protease cleavage recognition site, a BASE coil, and a 6x histidine tag. The list of primers used for construct generation are described in Table S5.

Purification of integrin α5β1 ectodomain:

Soluble integrin ectodomain (α5 residues 1-954 and β1 residues 1-728) was expressed using ExpiCHO-S cells (Thermo Fisher) in suspension at a density of 6.0 x 10⁶ cells/mL by transient transfection using the ExpiCHO transfection kit. Cultures were co-transfected with plasmids encoding the α5 or β1 subunit in a 2:1 (α:β) ratio with Expifectamine CHO. Expression was according to the manufacturer’s ‘max titer’ protocol. Cells were harvested by centrifugation at 1,000 x g for 15 minutes at 4°C. Phenylmethylsulfonyl fluoride (PMSF) was added to the supernatant at a final concentration of 0.5 mM before undergoing another centrifugation run at 36,000 x g for 45 minutes at 4°C. Clarified supernatant was diluted 1:1 v/v with Histrap binding buffer (20 mM Na₂PO₄, 500 mM NaCl, and 20 mM imidazole, pH 7.4) and applied to 5 mL Histrap FF Crude column (Cytiva) equilibrated with Histrap binding buffer. The column was washed with 1 column volume (CV) of Histrap binding buffer before eluting protein with 2 CV of Histrap Elution Buffer (20 mM Na₂PO₄, 500 mM NaCl, and 500 mM imidazole, pH 7.4). Protein fractions were pooled, concentrated and buffer exchanged into integrin storage buffer (20 mM Tris-Base pH 7.4, 150 mM NaCl, 1 mM CaCl₂, and 1 mM MgCl₂). Protease was added 1:20 w/w (protease:α5β1) to cleave the ACID/BASE coils and purification tags at 4°C overnight with rocking. The cleaved α5β1 integrin was further purified via gel filtration chromatography using a Superdex 200 Increase 10/300 SEC column (Cytiva) equilibrated in integrin storage buffer. Peak fractions of α5β1 were pooled, concentrated, flash-frozen with 10% glycerol (v/v), and stored at −80°C until used for experiments.

Purification of BIIG2 and MINT1526A Fab fragments:

BIIG2 antibody was produced at scale through the Mayo Clinic institutional core facility. Briefly, the production process involved culturing the hybridoma cells in roller bottles containing Iscove’s Modified Dulbecco’s Medium supplemented with gamma globulin-free serum. A single-step Protein G affinity purification was used to isolate antibodies. The eluted antibody fractions were pooled, concentrated, and buffer exchanged into phosphate buffered saline (1x PBS). The concentrated stock solutions of the antibodies were stored at −80 °C until used for experiments. Genentech has kindly provided us with purified MINT1526A antibody.

BIIG2 and MINT1526A antibodies were digested with the Pierce Fab Preparation Kit (Thermo Fisher) using papain and following the manufacturer’s protocol. BIIG2 Fab fragments were separated from the Fc fragments and intact IgG by loading onto a chromatography column packed with 5 ml of CaptureSelect IgG-Fc (Multi-species) Affinity Resin (Thermo Fisher) in PBS using a NGC FPLC instrument (Bio-Rad). The Fab fragment was present in the flow-through. Separation of the MINT1526A Fab fragment from the Fc fragment was achieved using a NAb Protein A spin column following the manufacturer’s protocol. SDS-PAGE gel bands of Fab and Fc samples from both antibodies were digested and analyzed by mass spectrometry (Mayo Clinic Proteomics Core) to verify the identity and separation of the two fragments.

Purification of Fibronectin domains 7-10:

Human fibronectin domains 7-10 (FN7-10) was purified as described previously ¹⁵. FN7-10 was expressed in Escherichia coli BL21 (DE3) in LB media containing ampicillin at 37 °C with shaking. Cells were grown until the OD600 reached 0.6-0.8. The cells were induced using IPTG at 1 mM final concentration and then grown overnight at 18 °C. Cells were pelleted and resuspended in a buffer containing 50 mM Tris-base pH 7.4, 200 mM NaCl, 1% Triton X-100, and a Pierce protease tablet. Cells were then lysed via sonication and clarified by centrifugation at 76,000 x g for 40 minutes at 4°C. Solid (NH₄)₂SO₄ was added to 40% (final w/v) slowly and the solution was placed on a nutator for 30 minutes at 4°C. This solution was then centrifuged again at 76,000 x G for 40 minutes at 4°C and the pellet was resuspended in Hitrap Q Binding Buffer (20 mM Tris-base pH 7.4 and 50 mM NaCl). The protein sample was applied onto a 5 mL Hitrap Q XL column. The column was washed with 5 CV of Hitrap Q Binding Buffer. The protein was eluted in a gradient profile from 50 mM NaCl to 500 mM NaCl over 20 CV. Protein fractions were pooled, concentrated and buffer exchanged into 50 mM Tris-base pH 7.4 and 150 mM NaCl. FN7-10 was further purified via gel filtration chromatography using a Superdex 200 increase 10/300 SEC column (Cytiva) equilibrated in 50 mM Tris-base pH 7.4 and 150 mM NaCl. Peak fractions of FN7-10 were pooled, concentrated, flash frozen with 10% glycerol (v/v), and stored at −80°C until used for experiments.

Biolayer Interferometry:

All binding experiments were performed on an Octet Red96e (Sartorius). MINT1526A was immobilized on Protein A biosensors (Sartorius). These sensors were hydrated using nonactive running buffer (20 mM Tris-Base pH 7.4, 150 mM NaCl, 0.1% BSA, and 0.02% Tween20 with 5mM CaCl₂) for 15 minutes before subsequently being incubated with fresh nonactive running buffer. MINT1526A was diluted with nonactive running buffer to 5 ug/mL and loaded for 300 seconds before a 120 second baseline equilibration in the nonactive running buffer. Association with integrin α5β1 ectodomain was carried out with a two-fold dilution series from 50nM to 1.56nM in nonactive running buffer for 300 seconds before dissociating for 300 seconds. Experiments were repeated with active running buffer (20 mM Tris-Base pH 7.4, 150 mM NaCl, 0.1% BSA, and 0.02% Tween20 with 1mM MnCl₂). MINT1526A binding experiments were performed in triplicate for both nonactive and active conditions.

Binding assays with BIIG2 began with hydrating AR2G biosensors (Sartorius) in water for 15 minutes, followed by an additional 1 minute incubation. The AR2G biosensors were then activated using 20 mM EDC and 10 mM s-NHS for 300 seconds (following manufacturer’s protocol). BIIG2 antibody was diluted with 10 mM sodium acetate pH 5.0 to 5 ug/mL and loaded until a threshold shift of 2nm was achieved. Loading was immediately quenched in 1M ethanolamine for 300 seconds. A baseline equilibration step was performed in nonactive running buffer for 120 seconds. Association with integrin α5β1 ectodomain was performed at a two-fold dilution series from 50nM to 1.56nM in nonactive running buffer for 450 seconds before dissociating for 450 seconds. Experiments were repeated with active running buffer and each buffer condition was repeated for an additional two times.

Competitive binding assays were carried out where α5β1 was bound to MINT1526A or BIIG2 Fabs or FN domains 7-10 fragment in a 1:2 molar ratio 20 minutes prior to assay running in either active or nonactive buffer. MINT1526A or BIIG2 antibodies or FN 7-10 used as immobilized ligands were loaded onto their respective biosensors (Protein A for MINT1526A and AR2G for BIIG2 and FN 7-10) at 5ug/mL. Experimental conditions and association and dissociation step times follow as described above.

Crystallization of BIIG2:

BIIG2 Fab (7.14 mg/mL in PBS + 0.02% sodium azide) was diluted in CIEX binding buffer (25 mM Bis-Tris-HCl, pH 6.5) and loaded onto a pre-equilibrated 1 mL HiTrap SP HP chromatography column (Cytiva), connected to an ÄKTA Purifier 10 chromatography system (Cytiva). The protein was eluted with a 0-100% gradient over 100 mL of CIEX elution buffer (25 mM Bis-Tris-HCl, 1 M NaCl, pH 6.5). The main protein component eluted as a highly symmetric peak around 12% v/v Buffer B (approximately 120 mM NaCl). Fractions containing BIIG2 Fab protein were pooled and concentrated to 10 mg/mL for crystallization studies.

Preliminary screening of BIIG2 Fab crystallization conditions was performed with pre-formulated commercial screens (JCSG+, Morpheus, PACT Premier, Molecular Dimensions/CalibreScientific Inc.) ^90–92. Sitting-drop vapor diffusion crystallization experiments were set up on UVXPO Two-Lens Plates (SwissCI) by mixing BIIG2 Fab at 10 mg/mL and the crystallization solutions in a 1:1 volume ratio (300 nL + 300 nL) using an Oryx4 crystallization robot (Douglas Instruments Ltd.) Plates were stored at 20°C in a RI-182 crystal plate hotel (Formulatrix), and monitored at regular time intervals for the formation of crystals. Initial crystals grew from several conditions from the JCSG+, PACT Premier and Morpheus formulations. Two sets of 48 optimized crystallization solutions were designed, combining the chemical components from several PACT Premier hits. A new sitting-drop vapor diffusion experiment was set up by mixing the optimized formulations with CIEX-purified BIIG2 Fab, which was diluted with type I water (Milli-Q) to a protein concentration of 7.6 mg/mL.

Crystallization drops were set up on an MRC Maxi PS (SwissCI) crystallization plate, mixing the diluted protein solution and the crystallization cocktails in a 1:1 v/v ratio (300 nL + 300 nL). Diffraction-quality crystals grew from several conditions. Crystals with the most promising morphologies were harvested in nylon loops or lytholoops, cryoprotected (if appropriate), flash-cooled in a nitrogen cryostream, and sent for synchrotron data collection in a liquid nitrogen-cooled CX 100 dry shipper (Taylor Wharton). The crystal that provided the highest-diffracting data set grew from 20% w/v PEG 3350, 0.2 M sodium chloride, 2.5% v/v DMSO; cryoprotectant solution was prepared by complementing the crystallization solution from the reservoir with glycerol, to a final concentration of 20% v/v.

X-ray data collection and refinement:

Synchrotron data collection on BIIG2 Fab crystals was performed at the BioMAX beamline, MAX IV, Lund, Sweden (at 100 K, 0.9537 Å). Data were initially integrated and scaled with autoPROC ⁹³, and ranked by their diffraction resolution cut-off, based on the assessment of the Pearson’s correlation coefficient (CC_1/2 > 10%) ⁹⁴.The highest-resolution data set (1.4 Å), collected on a crystal belonging to space group P2₁2₁2₁, was chosen for structure determination. Many of the programs and tools used belong to the CCP4 software suite ⁹⁵. Data were integrated and scaled with DIALS ⁹⁶; merging and truncation were carried out with AIMLESS, a component of the CCP4 software suite ⁹⁵. A PDB-restricted BLAST ^97,98 homology search identified a suitable search model for molecular replacement, the Fab fragment of a rat monoclonal antibody (PDB ID: 1C5D; VL=80.5 % seq. id.; VH = 79.1% seq. id.). Ligands and water molecules were manually deleted, and alternative conformations removed using PDBSET ⁹⁵. Amino acids were truncated to the last conserved atom using the CCP4 tool CHAINSAW ⁹⁹. The truncated model was segmented into four separate components, the heavy and light chains and their constant and variable regions, which were used as separate search models in Phaser ¹⁰⁰. The phased data set contained a single copy of the Fab heterodimer in the asymmetric unit. Search model bias was mitigated by running an extensive initial round (60 cycles) of maximum-likelihood refinement using REFMAC5 ⁹⁵. Two cycles of real-space model rebuilding were carried out in Coot ¹⁰¹, each followed by maximum-likelihood isotropic refinement with REFMAC5 ⁹⁵. Missing atoms from the protein main chain were automatically added with BUCCANEER ⁹⁵, followed by several rounds of manual building with Coot ¹⁰¹, alternating with maximum-likelihood isotropic Single-wavelength anomalous diffraction (SAD) refinement with REFMAC5 ⁹⁵. The final refinement step included TLS parameterization ¹⁰².

After correcting inaccuracies in the main protein chain, first and second-coordination sphere water molecules were added to the model, followed by alternative main and side-chain conformations. Water molecules were initially placed using Coot:Find waters and then manually inspected for several criteria, including distances from hydrogen-bond donors/acceptors and quality of the electron density. Occupancies for different side chain conformations were assigned during the course of a combined real-space/maximum-likelihood refinement round through phenix.refine ^103,104. The model was completed by adding ligands, including part of the N-linked glycosyl chain visible on N60 and glycerol, which had been used as cryoprotectant. Well-defined saccharide units from the glycan core region were manually positioned in the difference density map and linked by editing the coordinates file. Glutamine H1, the N-terminal amino acid of the Fab heavy chain, was modeled as pyroglutamate. Some of the disulfide bridges were found to be partly reduced (e.g., C23-C88, chain L), due to minor radiation damage; the relevant cysteine residues were modeled as partly reduced, when appropriate. K199 showed residual density on the terminal amino group of its side chain and was modeled as methyllysine. His H100C showed residual density suggesting post-translational modification, and was modeled as 5-hydroxyproline. As this residue is not included in the REFMAC dictionary, a ligand definition file was created with AceDRG ¹⁰⁵. 5-hydroxyproline (PubChem CID: 14325405) coordinates were generated with Coot LIDIA ¹⁰⁶, starting from an isomeric SMILES string obtained from the PubChem server ¹⁰⁷. The AceDRG output for the new ligand (HY5) was edited, renaming and renumbering atoms to match the standard proline numbering; occupancy refinement suggested an 86% post-translational modification rate for His H100C. Further model improvements were assessed with the PDB_REDO server ¹⁰⁸, while the diffraction limit was validated by running paired refinement using PAIREF ¹⁰⁹, accessed through the CCP4i2 GUI ¹¹⁰. The overall quality of the model was validated through the PDB Validation server ¹¹¹. The percentages of amino acid residues occupying the favored, allowed and outlier regions in the Ramachandran plot are 98/2/0%. The quality of the carbohydrate chain was evaluated by Privateer ¹¹², a CCP4 tool accessible through the CCP4i2 interface ¹¹⁰. σ_A-weighted electron density maps were generated using the FFT tool, a component of the CCP4 software package ⁹⁵. Figures were generated using PyMOL (Schrödinger Inc.), assigning secondary structure elements as defined by the SST web server ¹¹³. Position numbering in the final model was adjusted to match the Kabat scheme ¹¹⁴, before depositing the refined coordinates and structure factors in the PDB ¹¹⁵, with PDB ID: 8R38.

Negative-stain EM sample preparation, data acquisition and processing:

The α5β1-Fab complexes were formed using a 1:2 molar ratio of α5β1 to antibody Fab, incubated at room temperature for 20 minutes, and diluted to a final concentration of 7.5 ug/mL in 20 mM Tris-base pH 7.4, 150 mM NaCl supplemented with either 5 mM CaCl₂ or 1 mM MnCl₂. For experiments determining what integrin conformations MINT1526A binds, we used the commercially available conformation-inducing antibody, 9EG7 IgG (BD Biosciences cat# 553715). Integrin α5β1 ectodomain was buffer-exchanged into nonactivating conditions (5 mM Ca²⁺, 20 mM Tris-base pH 7.4, 150 mM NaCl) for 10 minutes with rocking at room temperature. 9EG7 IgG was added to integrin α5β1 at a 1:1 molar ratio and incubated for 10 minutes at room temperature with rocking. 9EG7 binding locks α5β1 into the extended state ⁴³, then MINT1526A or BIIG2 Fab was added at a 1:1 molar ratio to integrin α5β1 ectodomain and incubated for an additional 10 minutes with rocking at room temperature. Experiments were repeated in activating buffer (1 mM Mn²⁺, 20 mM Tris-base pH 7.4, 150 mM NaCl).

For all experiments, 3 μL of protein sample was added to a glow-discharged (PELCO easiGlow; 30 seconds; 15 mA) carbon-coated copper grid (homemade carbon layer). The specimens were stained with 3 μL of 2% (w/v) uranyl formate solution as previously described ¹¹⁶.

Data were acquired using a Thermo Fisher Scientific Talos L120C transmission electron microscope operating at 120kV and recorded on a 4k x 4k Thermo Fisher Scientific Ceta camera at a nominal magnification of 92,000x with a pixel size of 1.58 Å/px. Leginon ¹¹⁷ was used to collect 252 (apo α5β1 nonactivating), 253 (apo α5β1 activating), 385 (α5β1:BIIG2 nonactivating), 399 (α5β1:BIIG2 activating), 422 (α5β1:MINT1526A nonactivating), 482 (α5β1:MINT1526A activating), 368 (α5β1:9EG7:BIIG2 nonactivating), 373 (α5β1:9EG7:BIIG2 activating), 366 (α5β1:9EG7:MINT1526A nonactivating) and 395 (α5β1:9EG7:MINT1526A activating) micrographs with a nominal range of 1.7-2.0 μm under focus and a dose of approximately 50 e⁻/Ų. The contrast transfer function of each micrograph was estimated using GCTF ¹¹⁸, and particles were automatically picked unbiasedly using Gautomatch ¹¹⁸ (htttp://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch). Particles were extracted at 352 pixels and binned to 128 pixels using RELION3 ¹¹⁹ before being imported into cryoSPARC ¹²⁰. Particles were subjected to multiple rounds of reference-free 2D alignment and classification to remove false-positive particle images. The final particle count contributed to 2D class averages were 26,057 (apo α5β1, nonactivating), 31,786 (apo α5β1, activating), 18,200 (α5β1:BIIG2, nonactivating), 16,797 (α5β1:BIIG2, activating), 9,742 (α5β1:MINT1526A, nonactivating), 41,387 (α5β1:MINT1526A, activating), 21,329 (α5β1:9EG7:BIIG2 nonactivating), 25502 (α5β1:9EG7:BIIG2 activating), 12,149 (α5β1:9EG7:MINT1526A nonactivating) and 48,765 (α5β1:9EG7:MINT1526A activating).

Cryo-EM sample preparation:

α5β1-BIIG2/MINT1526A Fab complexes were formed using a 1:2 α5β1 to Fab molar ratio in nonactivating buffer (20 mM Tris pH 7.4, 150 mM NaCl, 5 mM CaCl₂) at room temperature with rocking for 20 minutes. nonactivating conditions were used to drive the equilibrium toward a more homogenous sample since we had already established that the epitope binding site was nowhere near the relevant cation coordination sites. These protein complexes were subjected to size exclusion chromatography using a Superdex200 Increase 10/300 GL (Cytiva) equilibrated with nonactivating buffer. Fractions of α5β1:BIIG2/MINT1526A Fab were pooled and concentrated to 0.15 mg/mL before 3 uL of the protein sample was placed on a glow discharged (PELCO easiGlow; 30 seconds; 15 mA) R1.2/1.3 Au 300-mesh grid (QUANTIFOIL), blotted for 8 seconds with a blot-force of 1, and plunge-frozen immediately in liquid ethane using a Thermo Fisher Scientific Vitrobot Mark IV at 4° C and 100% humidity.

Cryo-EM Data collection and processing:

Two datasets (one each for α5β1:BIIG2 Fab and α5β1:MINT1526A Fab) were acquired on a Thermo Fisher Scientific Glacios cryo-transmission electron microscope operating at 200kV and recorded with a Gatan K3 Direct Detection Camera. Automated data collected was carried out using the SerialEM software ¹²¹. 99 frame movies were recorded in super-resolution mode with a super-resolution pixel size of 0.561 Å/px at a nominal magnification of 36,000x. Both datasets were collected in a single session with a nominal defocus range of 1.4–1.8 μm under focus and a dose of approximately 50e⁻/Ų. 3656 micrographs were collected of the α5β1:BIIG2 Fab complex and 3079 micrographs of α5β1:MINT1526A Fab.

The dose fractionated super-resolution image stacks were motion corrected and binned 2x2 by Fourier cropping using MotionCor2 ¹²². Motion corrected images were then processed using CTFFIND4 within the cryoSPARC wrapper ^120,123. 2,425,530 (α5β1:BIIG2 Fab) and 1,101,393 (α5β1:MINT1526A Fab) particles were initially picked using the unbiased blob picker in cryoSPARC. These particles were subjected to multiple rounds of reference-free 2D alignment and classification in cryoSPARC. Next, particles were polished in Relion ^124,125 and final 3D classification and refinement steps were done in cryoSPARC. The final particle count for α5β1:BIIG2 Fab is 371,066. The final particle count for α5β1:MINT1526A Fab is 342,608. Images showing EM maps were generated using UCSF ChimeraX ^126–128. A detailed processing schematic is in Figure S10.

CryoEM model building:

The closed α5β1 headpiece (PDB: 3VI3) and our solved BIIG2 Fab structure were used as initial models for the α5β1:BIIG2 Fab complex. The half bent α5β1 ectodomain (PDB ID: 7NXD) was used as an initial model for α5β1:MINT1526A Fab complex. MINT1526A was modeled de novo using peptide sequences derived from mass spectrometry. Initial models were fitted into their respective cryoEM density using UCSF ChimeraX ^126–128 and manually adjusted in COOT ¹⁰¹. Glycosylations were manually added using COOT. Models (including glycans) were refined using Phenix and ISOLDE ^104,129,130. Modeling was aided by using sharpened and unsharpened cryoEM maps. BIIG2 and MINT1526A were numbered according to the Kabat numbering scheme ¹³¹. CryoEM maps and model statistics are found in Table S3.

Molecular dynamics simulations:

All-atom molecular dynamics simulations were performed using the OpenMM 7.7.0 package ¹³² using AMBER ff19SB parameters and topologies. The initial structure for simulations were constructed from the α5β1:BIIG2 or α5β1:MINT1526A cryoEM models. To reduce the system size, only the headpiece domains of α5β1 and variable domains of BIIG2 and MINT Fabs were simulated. N- and C- terminal residues were capped with acetyl and methyl amide groups, respectively. Protonation states for titratable residues were modeled based on a pH of 7.4. The systems were solvated in a truncated octahedron with the minimal distance between the solute and the box boundary set to 8Å and 150 mM NaCl. Calcium ions resolved in the cryoEM model were retained. The system was parameterized using the AMBER19ffSB ¹³³ force field combined with the OPC water model and 12-6-4 Li Merz ion parameters ¹³⁴. As a control, an α5β1 without Fab bound, referred to as apo α5β1, was modeled and parameterized similarly.

The systems were first energy minimized for 20,000 steps followed by a gradual heating to 298K. During minimization and heating, protein backbone atoms were harmonically restrained by a 5 kcal/mol*Ų force constant. Production simulations were performed using periodic boundary conditions, hydrogen mass repartitioning ¹³⁵, and a Langevin integrator with an integration timestep of 4 fs and collision rate of √2 ps⁻¹. Pressure was maintained at 1 atm using a MonteCarlo barostat with an update frequency of 100 steps. Nonbonded interactions were calculated with a distance cutoff of 10Å. Trajectory snapshots were saved every 100 ps during production simulations. A total of 7 replicates, 2μs each in length, were simulated for α5β1 systems.

Trajectories were processed using in-house scripts utilizing the CPPTRAJ ¹³⁶, MDTraj 1.9.7 ¹³⁷, GetContacts (https://getcontacts.github.io/), numpy-1.67.4 ¹³⁸ and matplotlib-3.8.4 packages ¹³⁹. Hydrogen bond frequencies were calculated using a 3.5 Å distance cutoff between the donor-hydrogen and acceptor atoms. The first 50 ns of production simulations were not included in analysis. Trajectories were visualized using VMD 1.9.2 ¹⁴⁰.