Structural basis of human LRG1 recognition by Magacizumab, a humanized monoclonal antibody with therapeutic potential
aStructural Biology and Drug Discovery Group, Centre for Molecular Medicine Norway, Nordic EMBL Partnership, University of Oslo and Oslo University Hospital, 0318 Oslo, Norway, bInstitute of Ophthalmology, University College London, London, United Kingdom, cDepartment of Chemistry, University College London, London, United Kingdom, and dDepartment of Physiology and Biophysics, University of California, Irvine, Irvine, CA 92697, USA
*Correspondence e-mail: email@example.com, firstname.lastname@example.org, email@example.com
The formation of new dysfunctional blood vessels is a crucial stage in the development of various conditions such as macular degeneration, diabetes, cardiovascular disease, neurological disease and inflammatory disorders, as well as during tumor growth, eventually contributing to metastasis. An important factor involved in pathogenic angiogenesis is leucine-rich α-2-glycoprotein 1 (LRG1), the antibody blockade of which has been shown to lead to a reduction in both choroidal neovascularization and tumor growth in mouse models. In this work, the structural interactions between the LRG1 epitope and the Fab fragment of Magacizumab, a humanized function-blocking IgG4 against LRG1, are analysed, determining its specific binding mode and the key residues involved in LRG1 recognition. Based on these structural findings, a series of mutations are suggested that could be introduced into Magacizumab to increase its affinity for LRG1, as well as a model of the entire Fab–LRG1 complex that could enlighten new strategies to enhance affinity, consequently leading towards an even more efficient therapeutic.
Neoangiogenesis is a phenomenon that plays an important role in many diseases, including those affecting the eyes, kidneys and lungs, as well as promoting cancer growth. Vascular dysfunction, whether it arises from a failure of new vessels to stabilize or from existing vessels losing normal function, is driven by various factors that regulate the growth and functional stability of blood vessels (Wang et al., 2013; Carmeliet & Jain, 2000; Potente et al., 2011). While vascular endothelial growth factor (VEGF) is a major player in this context, other secondary factors also contribute to vascular signaling in disease. Such is the case for TGF-β1, TGF-β type II receptor, ALK1, Smad1, Smad5 and Smad8, the regulation of which depends on multiple factors and the presence of additional modulatory proteins. In summary, these pathways involve the interaction of many factors, some as yet unknown, complicating the development of therapies and treatment strategies targeting the previously mentioned diseases.
By studying pathological retinal vascularization in a range of mouse mutants, Wang et al. (2013) found leucine-rich α-2-glycoprotein 1 (LRG1; a glycosylated 50 kDa protein with 347 residues in humans) to be induced nearly exclusively in this vasculature. LRG1 belongs to the leucine-rich repeat family of proteins, many of which are involved in protein–protein interactions and cell-signaling pathways. Further analysis determined that under normal conditions LRG1 is not present in the systemic vasculature and is not involved in developmental angiogenesis, only being present to any extent in the liver and in neutrophils. Thus, the normal function of LRG1 has yet to be elucidated, although there is evidence that it may be involved in wound healing (Gao et al., 2020; Pickert et al., 2009) and also in protection against the pro-apoptotic effects of extracellular cytochrome c (Shirai et al., 2009; Jemmerson, 2021).
When studying disordered vascularization in mice following laser-induced choroidal neovascularization (CNV) and oxygen-induced retinopathy, Wang et al. (2013) further revealed that LRG1 gene deletion or blockade with an anti-LRG1 antibody resulted in a significant reduction in lesion size. These experiments, together with recent reports that LRG1 is involved in tumor growth, tumor metastasis and emphysema (Hisata et al., 2021; Singhal et al., 2021; O'Connor et al., 2021), indicate that LRG1 affects vascular stability and may be a suitable therapeutic target in multiple pathologies associated with aberrant vascularization and vascular destabilization. Accordingly, a program was undertaken to develop a therapeutic antibody that was initiated with the generation of a cohort of more than 100 mouse monoclonal antibodies (mAbs) that specifically bind to human LRG1. The panel of mAbs was derived using conventional hybridoma technology following the immunization of mice with recombinant human LRG1. The entire cohort of mAbs was assayed for therapeutic suitability and reduced to three lead candidates based on their anti-angiogenic activity and affinity, with the anti-angiogenic activity providing a measure of the function-blocking activity of each mAb. Among these three was 15C4 which, due to its specificity and near-irreversible picomolar affinity, was humanized with a view to future testing in clinical trials (Kallenberg et al., 2020). Humanized antibodies generally comprise an Fc region from IgG1, IgG2, IgG3 or IgG4 and a Fab region containing the paratope. The paratope, as well as its surroundings, is crucial for specificity against the target and may be engineered to increase the affinity of the antibody–antigen complex. In the case of 15C4, humanization resulted in a loss of affinity of approximately an order of magnitude, suggesting that affinity maturation may be both desirable and feasible. For the same purpose, obtaining structural information on the binding mechanism between the components will help not only to understand complex formation but also to design new antibodies with improved affinity and/or specificity.
In this work, we have determined the high-resolution crystal structure of the Fab fragment of Magacizumab (MagaFab), the lead IgG4 humanized from 15C4, in complex with a 15-amino-acid peptide that corresponds to the epitope in LRG1 (Kallenberg et al., 2020). Careful analysis of the atomic interactions allowed us to identify LRG1 residues that play a key role in antibody recognition as well as the surrounding interactions that contribute to binding and strengthen MagaFab–LRG1 complex formation. Lastly, to extract insights into the overall organization of the complex between full-length LRG1 and MagaFab, a homology model of LRG1 was built and structurally aligned with the peptide posed over the paratope.
Magacizumab was generated as a recombinant antibody using a CHO cell-based expression system and purified from the culture medium as described previously (Kallenberg et al., 2020). The antibody was formulated at 75.4 mg ml−1 in digestion buffer (20 mM monobasic sodium phosphate, 10 mM disodium EDTA, 80 mM cysteine–HCl pH 7.2) and reacted with a 1:10(w:w) amount of immobilized papain (Thermo Scientific; 250 µg per milligram of gel) by incubation for 24 h at 37°C whilst shaking (at 1100 rev min−1). Cysteine–HCl was added immediately before Magacizumab digestion. After digestion, the resin was separated from the digest using a filter column and washed with phosphate-buffered saline (PBS) pH 7.0 three times. The digest was combined with the washes, the buffer was exchanged completely for PBS pH 7.4 using diafiltration columns (10 kDa molecular-weight cutoff) and the volume was adjusted to 2 ml. The sample was then applied onto an NAb protein A column (Thermo Scientific). The Fab fraction, Magacizumab Fab (MagaFab), was eluted according to the manufacturer's protocol; the column was washed three times with PBS pH 7.4 and the Fc fraction was eluted with 0.2 M glycine–HCl pH 2.5, which was neutralized with 10% of the volume of 1 M Tris pH 8.5 solution. The MagaFab fraction was combined with the washes and both the Fab and Fc solutions were buffer-exchanged into PBS pH 7.4 using diafiltration columns (10 kDa molecular-weight cutoff). The digests were analyzed by SDS–PAGE and LCMS to reveal the formation of MagaFab. The observed molecular mass was 47 461 Da, while the computed mass was 47 412 Da.
2.2. Crystallization and structure determination of apo MagaFab
Rod-shaped MagaFab crystals of approximately 100 × 50 × 50 µm in size were obtained by sitting-drop crystallization (0.5 µl sample and 0.5 µl well condition) at 23°C using the JCSG screen (Qiagen, Germany). Further optimization of the hit conditions led to larger needle-shaped crystals that grew in 0.2 M NaCl, 0.1 M bis-Tris pH 5.5, 25%(w/v) PEG 3350 at a protein concentration of 5 mg ml−1. The best crystals were cryoprotected using 0.1 M bis-Tris pH 5.5, 35%(w/v) PEG 3350 (Teng & Moffat, 2000). X-ray diffraction data were collected from several crystals at 100 K on the XALOC beamline at ALBA, Spain (Juanhuix et al., 2014). Diffraction images were processed with XDS (Kabsch, 2010) and AIMLESS (Evans, 2011; Winn et al., 2011), resulting in a 2.7 Å resolution P21 data set expected to contain two molecules per asymmetric unit (36.5% solvent content). Molecular replacement was performed with Phaser (McCoy et al., 2007) using the crystal structures of Pembrolizumab (PDB entry 5dk3, CH and CL chains; Scapin et al., 2015), antibody B43 (PDB entry 6al4, VH chain; Teplyakov et al., 2018) and humanized 5c8 Fab (PDB entry 6bjz, VL chain; Henderson et al., 2020) as the highest homology templates for the respective chains. Several rounds of rebuilding and refinement were performed with Coot (Emsley et al., 2010) and Phenix (Liebschner et al., 2019). This apo MagaFab structure was not deposited in the PDB but was used as a molecular-replacement model to determine the structure of MagaFab in complex with the LRG1 epitope peptide.
2.3. Crystallization and structure determination of MagaFab in complex with the epitope peptide
The MagaFab–peptide complex was obtained by co-crystallization at MagaFab:peptide molar ratios of 1:2 and 1:5 using a MagaFab concentration of 12 mg ml−1 in 0.2 M NaCl, 0.1 M bis-Tris pH 5.5, 25%(w/v) PEG 3350. The best crystals were cryoprotected using 0.1 M bis-Tris pH 5.5, 35%(w/v) PEG 3350 (Teng & Moffat, 2000). X-ray diffraction data were collected from several crystals at 100 K on the XALOC beamline at ALBA, Spain (Juanhuix et al., 2014). Diffraction images were processed with XDS (Kabsch, 2010) and AIMLESS (Evans, 2011; Winn et al., 2011), resulting in a 1.7 Å resolution P212121 data set that was expected to contain two MagaFab molecules per asymmetric unit (56.9% solvent content; Table 1). Molecular replacement was performed with Phaser (McCoy et al., 2007) using our previously determined apo MagaFab structure as a template. Several rounds of rebuilding and refinement were performed with Coot (Emsley et al., 2010) and Phenix (Liebschner et al., 2019).
MagaFab 1 (chains A and B), most of MagaFab 2 (chains C and D) and both peptide molecules (chains E and F) fit very well into the electron density (Supplementary Fig. S1), showing excellent geometry and no outliers (as assessed by MolProbity; Chen et al., 2010). However, a region of about 60 residues located after Ser25 of chain D (corresponding to the VH chain of MagaFab 2) displayed significant Fo − Fc positive density at contouring levels below 5σ along the main chain and side chains. In general, this positive difference density displays the overall features and shape of several side chains shifted between 2 and 3 Å in defined directions, as well as a thickening of the main chain. The magnitude of these features varies in range depending on the area, consistent with a pseudo-rigid-body displacement and slight torsion of the whole region (Supplementary Fig. S2a). Different molecular-replacement and refinement strategies were performed to minimize this positive difference density. Being unable to obtain electron-density maps that lacked the positive density in this region with a single conformation for chain D, we explored the possibility of differential crystal contacts and a global double conformation affecting not only isolated residues but a substantial area of the VH region of molecule 2 (chain D). The best structure, with an R factor of 0.182 and an Rfree of 0.206 (Table 1), was obtained by building two fragments in a double conformation (Ser25–Val37 and Ile48–Ser85) and by refining XYZ in both reciprocal and real space, translation–libration–screw rotation (TLS) factors, occupancies and individual B factors without applying noncrystallographic symmetry (NCS) restraints (Usón et al., 1999; Supplementary Figs. S2b and S2c). The final occupancies were 0.47:0.53 for the first fragment and 0.43:0.57 for the second fragment. Although scattered Fo − Fc positive density was still visible in some minor exposed areas, building a double conformation of these two regions decreased both R factors by about 2% compared with the same strategy performed with a single conformation at an occupancy of 1.00 as a control. Composite omit maps and refinements of partial models were performed to minimize bias during the whole process.
In summary, MagaFab 1 (chains A and B), as well as its respective peptide (chain E), do not suffer from differential crystal contacts, hence the single conformation combined with the high quality of both the maps and model in these regions allowed us to describe and analyze both the overall MagaFab structure and its interactions with the LRG1 epitope at the reliability expected for a 1.7 Å resolution study.
Three-dimensional homology models of mouse hybridoma 15C4 variable domains were generated with SWISS-MODEL (Waterhouse et al., 2018). The best model of the murine hybridoma 15C4 VH chain, according to the GMQE (global model quality estimation), was built using the pyroglutamate-amyloid-β-specific Fab c#24 heavy chain (PDB entry 5myx, chain B) as a template (Piechotta et al., 2017), returning a GMQE of 0.97, a sequence identity of 83.9%, a sequence similarity of 90.7% and a coverage of 97.0%. The VL chain of the murine hybridoma 15C4 Fab was generated using the VL fragment of Mus musculus IgG as determined in the crystal structure of the anion-exchanger domain of human erythrocyte band 3 (PDB entry 4yzf, chain F; Arakawa et al., 2015) as a template. In this case the GMQE was 0.99, with 89.2% sequence identity and 95.5% sequence similarity with 100% coverage.
A three-dimensional homology model of LRG1 was built with SWISS-MODEL using NetrinG ligand 2 (NGL2; PDB entry 3zyi, chain A) as a template (Seiradake et al., 2011), returning a GMQE of 0.59, a sequence identity of 22.1%, a sequence similarity of 39.9% and a coverage of 86.0%. Additional models were built by threading with RaptorX (Peng & Xu, 2011; Xu et al., 2021). Up to five models were predicted, with estimated Cα r.m.s.d.s ranging between 2.02 and 2.13 Å.
The crystal structure of MagaFab in complex with the LRG1 epitope shows the 15-residue peptide (Pep; GNKLQVLGKDLLLPQ) bound at its paratope along the VL and VH chains of MagaFab, forming a partially folded structure consisting of a short 310-helix (8-GKDL-11) flanked by loop regions without specific secondary structure. In addition to this arrangement, a symmetry-related peptide molecule (Pep*) as well as a symmetry-related MagaFab molecule are located near the main peptide, establishing additional contacts that strengthen crystal packing (Fig. 1, Supplementary Fig. S3 and Table 2).
The strongest interaction between MagaFab and the peptide is a pair of hydrogen bonds between the respective side-chain amides of Gln5Pep and Gln50VH, supported by a water-mediated hydrogen bond to the carbonyl group of Ile100VH (Fig. 2a). This arrangement is flanked on one side by Trp33VH, which, in addition to a possible stacking effect, is likely to stabilize the main chain of Lys3Pep. This lysine is of special importance due to its Coulombic interactions with Asp55VH and Asp57VH. In addition to these key interactions, the N-terminal Gln1Pep is located between Tyr52VH, the carbonyl group of Gly31VH, the C-terminal hydroxyl group of Gln15Pep* and a structural water molecule, while Asn2Pep interacts with Asp10Pep* from the symmetry-related peptide, completing the stabilization of this region (Fig. 2b).
While the binding of the Gly1Pep–Gln5Pep portion of the peptide mainly relies on a strong network of polar interactions with residues from the VH chain and water molecules, the second portion (Val6Pep–Gln15Pep) mainly establishes hydrophobic interactions between Leu7Pep and Val103VH, as well as stacking interactions of the short 310-helix in Pep with Phe36VL and Trp96VL (Fig. 2c). Another key structural water, located between Tyr54VL and the carbonyl group of Thr102VH, is involved in the stabilization of the peptide by its proximity to the carbonyl group of Val6Pep. Finally, a network of interactions between Lys9-Asp10-Leu11/Gln15Pep and the symmetry-related neighbors (Gly1Pep*, Asn2Pep*, Ser206VL* and Thr101VH*) completes the interactions of this region (Fig. 2d). While these interactions are the strongest, the tight packing of the different molecules in the crystals leads to a significant contact surface area between the main peptide and a symmetry-related molecule (352.0 Å2), similar to the area between the main peptide and each of the MagaFab chains (APep–VH = 356.0 Å2, APep–VL = 338.2 Å2; Table 2). The contact areas between the main peptide and the symmetry-related MagaFab chains are smaller (APep–VH* = 187.1 Å2, APep–VL* = 14.2 Å2), but definitely contribute to crystal packing.
From a therapeutic point of view, increasing the affinity of Magacizumab for LRG1 would be expected to provide certain benefits, since this would permit a potential reduction in dosage and frequency of administration to patients. Affinity maturation of an anti-VEGF Fab led to the development of ranibizumab (Lucentis), a therapeutic widely used in eye disease that has approximately a tenfold higher affinity for VEGF than the parental Fab (Chen et al., 1999). The crystal structure of MagaFab in complex with its LRG1 epitope described in this work has allowed us to analyze and explore potential mutations that could similarly lead to a significant increase in affinity. Some key features that can be extracted from this analysis are that Gln50VH is probably the main residue involved in the recognition of the antigen by strongly interacting with Gln5Pep, while Asp55VH and Asp57VH are required to stabilize Lys3Pep. The remaining interactions are either nonpolar or are mediated by structural waters.
While mutations of the key residues (Gln50VH, Asp55VH and Asp57VH) could cause negative effects in the paratope, leading to loss of affinity, mutations of residues showing weaker or water-mediated interactions could result in new and stronger direct contacts with the epitope, hence increasing the stability of the complex. Such is the case for Trp33VH. Located between the bidentate Gln50VH–Gln5Pep interaction and the Lys3Pep side chain, it forms a hydrogen bond to the carbonyl group of the lysine. However, two important structural waters (W46 and W297) are located next to this position. Mutation of Trp33VH to an arginine could place the guanidino group in the same position as these waters, presumably increasing the affinity of the peptide main chain (Supplementary Fig. S4a). However, due to the proximity to the bidentate Gln–Gln interaction, an arginine in this position could potentially destabilize or prevent this strong interaction due to steric hindrance, causing the opposite of the desired effect.
Another interesting candidate is Gly31VH since it is located next to Gly1-Asn2Pep and is exposed to the solvent. Therefore, a mutation of this residue is less prone to interfere with other key residues. A short polar residue such an aspartate or a serine might be sufficient to create an interaction with Asn2Pep and would not alter the current main-chain geometry, therefore keeping the φ/ψ values in the favored region of the nonglycine Ramachandran plot (Supplementary Fig. S4b).
On the nonpolar section of the peptide, mutation of Met98VL to a leucine might strengthen the hydrophobic interactions between Leu7Pep and Val103VH (Supplementary Fig. S4c). A similar effect might be obtained by mutating Thr31VL to a leucine or an isoleucine, placing a nonpolar chain between Leu11Pep and Leu12Pep to potentiate hydrophobic interactions with one of these side chains (Supplementary Fig. S4d). However, due to the exposed location of both Thr31VL and the peptide 310-helix segment, the current spatial arrangement (4.4 Å between the Thr31VL hydroxyl and Leu11Pep carbonyl groups) could differ in the MagaFab–LRG1 complex, where the distances between these residues might differ significantly.
To explore the differences in affinity between the mouse hybridoma 15C4 mAb and MagaFab, three-dimensional homology models of 15C4 mAb variable domains were built with SWISS-MODEL (Waterhouse et al., 2018) using the pyroglutamate-amyloid-β-specific Fab c#24 heavy chain (PDB entry 5myx, chain B; Piechotta et al., 2017) and the VL fragment of M. musculus IgG from the crystal structure of the anion-exchanger domain of human erythrocyte band 3 (PDB entry 4yzf, chain F; Arakawa et al., 2015) as the most similar templates for the VH and VL chains, respectively. These models were later aligned with the corresponding domains of the MagaFab structure (Cα r.m.s.d.s of 0.36 Å for FabVH and 15C4VH and 0.38 Å for FabVL and 15C4VL).
All residues involved in peptide recognition and their surroundings are conserved between MagaFab and the hybridoma 15C4 domains (Figs. 3a and 3b). The differences are mostly scattered across exposed areas of MagaFab that do not interact with the epitope. However, there is a difference at Leu50Fab-VL, a highly conserved residue amongst human immunoglobulins that is a phenylalanine in the original murine fragment. This Leu/Phe interacts with Val103Fab-VH, which is located at the end of a flexible loop (Arg98Fab-VH–Tyr107Fab-VH) that is involved in interaction with the peptide (Val103VH–Leu7Pep) (Fig. 3c). In the 15C4 VH homology model this loop is oriented differently to in our MagaFab crystal structure but, considering that it was modeled independently of the VL chain and the peptide, one cannot assume that this conformation might be due to the Leu50Phe substitution; rather, it might be due to an orientation generated during the modeling process or the original orientation in the template used by SWISS-MODEL. As mentioned before, the template selected by SWISS-MODEL as the most similar was the pyroglutamate-amyloid-β-specific Fab c#24 heavy chain (PDB entry 5myx, chain B; Piechotta et al., 2017), the corresponding loop of which, although three residues shorter (Arg98–Tyr104), shows the orientation predicted in the 15C4 VH homology model. Nevertheless, the Leu50Phe substitution points to a region that could be critical in stabilizing the VH loop that, at the same time, interacts with the peptide.
In the absence of a three-dimensional structure of LRG1, analysis of its sequence and comparison with similar proteins with structures that are known can be utilized to construct a homology model of LRG1, which in turn can be used to predict the structure of its complex with Magacizumab. Analysis of the LRG1 sequence with PSIPRED (Buchan & Jones, 2019), BlastP (Altschul et al., 1990), PROSITE (Sigrist et al., 2013) and XtalPred (Slabinski et al., 2007) suggests the presence of a 35-residue N-terminal signal peptide formed by a 17-amino-acid disordered region (Met1–Pro17) followed by a 16-residue helix (His18–Ala34). Up to eight leucine-rich repeats (LRRs) are predicted between residues 93 and 282, while the C-terminal domain is identified as a leucine-rich-repeat C-terminal domain (LRRCT).
The three-dimensional structure of LRG1 was modeled with SWISS-MODEL (Waterhouse et al., 2018) using NetrinG ligand 2 (NGL2) as the template (PDB entry 3zyi, chain A; Seiradake et al., 2011; 24.7% sequence identity and 32.0% sequence similarity with 86.0% coverage). Usually, it is thought that the structure of a protein can be predicted with some accuracy when its sequence identity compared with the template is greater than 30% (Xiang, 2006). For this reason, additional models will be discussed later. NGL2 is a type I transmembrane protein composed of a nine-leucine-rich-repeat domain (NGL2LRR), an immunoglobulin-like domain (NGL2Ig), a glycosylated region, a transmembrane helix and a cytoplasmic domain (Seiradake et al., 2011). The resulting LRG1 model reveals a canonical leucine-rich-repeat domain (LRR domain) composed of nine leucine-rich repeats (LRRs) containing the MagaFab epitope (at LRR7), capped by N- and C-terminal domains (Fig. 4a). The N-terminal domain is similar to the N-terminal cap in NGL2LRR but lacks the disulfide bridge present in NGL2LRR between Cys50 and Cys61 (Cys43 and Ile54 in LRG1). However, LRG1 shows a second cysteine close to Ile54 (Cys56), which could establish a disulfide bridge with Cys43 in a slightly different fold to that predicted by SWISS-MODEL. The LRG1 C-terminal domain was modeled very similarly to the C-terminal cap of NGL2LRR, containing two helices and a disulfide bridge (Cys303–Cys329) equivalent to the Cys304–Cys329 bridge in NGL2LRR.
Also noticeable is the fact that, as mentioned before, BlastP (Altschul et al., 1990) and PROSITE (Sigrist et al., 2013) sequence analyses predict only eight LRRs instead of the nine LRRs modeled by SWISS-MODEL. While these programs identify the first LRR beginning at residue Lys93, homology modeling predicts an additional LRR between Thr70 and Ser92, corresponding to the first LRR of NGL2LRR. Therefore, although the consensus LRR motif LXXLXLXXN/CXL is not found between residues 70 and 92, the abundance and location of leucine residues is similar enough to model this section as an LRR (Supplementary Fig. S5).
Sequence alignment of LRG1 and NGL2LRR (excluding the NGL2Ig domain) performed with EMBOSS-Needle resulted in an identity of 23.6% and a similarity of 43.0%. Since the identity was below 30%, additional LRG1 models were built with RaptorX (Peng & Xu, 2011; Xu et al., 2021), a program that uses threading instead of homology as a method for structure prediction. Five additional models were predicted with RaptorX using the LRG1 sequence without the signal peptide as input. The estimated Cα r.m.s.d. of these LRG1 models ranged from 2.02 to 2.13 Å and they showed very similar folding of all domains compared with the homology model calculated by SWISS-MODEL (Fig. 4b). Interestingly, although Cys43 and Cys56 do not form a disulfide bridge in these threading models, they are predicted to be close enough (7.2 Å between Cα atoms) and could potentially establish such a bond in vivo, while the homology model showed a slightly longer separation between these cysteines (10.0 Å between Cα atoms), preventing a bridge. The cysteines located in the C-terminal domain (Cys303–Cys329) that form a disulfide bridge in the homology model are also predicted in the same position by threading.
Finally, the recent public release of human proteome predictions by AlphaFold (Jumper et al., 2021) allowed us to compare our LRG1 homology and threading models with a novel ab initio model based on the primary amino-acid sequence and deep learning. In this study, a very high per-residue confidence score (pLDDT > 90) is obtained for the entire structure, excluding the first 35 residues that correspond to the signal peptide (which were absent in our SWISS-MODEL homology model and the threading predictions). The folding of all domains is very similar to the homology and threading models and predicts disulfide bridges in both the N- and C-terminal domains (Fig. 4b).
The similarity between the homology, threading and ab initio models suggests that these predictions might be relatively accurate since different methods returned comparable models. However, the crystal structure of LRG1 will be required to perform further analysis, since homology, threading and ab initio models generally are not sufficiently reliable for reaching deeper conclusions.
To generate a model of whole LRG1 in complex with MagaFab, we manually superposed the epitope of the SWISS-MODEL homology model of LRG1 with the peptide from our crystal structure (Fig. 5a). The peptide crystallized in complex with MagaFab corresponds to the bold section of LRG1 LRR7 (214-LERLHLEGNKLQVLGKDLLLPQPD-237); therefore, an alignment was performed based on the location of this repeat and was manually adjusted to fit the orientation of the key residue Gln225 (Gln5Pep) and the surrounding main chain (Fig. 5b). While Gln225 and Val226 of the LRG1 homology model align well with Gln5 and Val6 of the peptide from the crystal structure, a 180° flip of the main chain (and, as a consequence, of the side chain) is observed for Leu227 (Fig. 5b). In the MagaFab–peptide complex crystal structure, Leu227 (Leu7Pep) appears to play an important role by interacting with Val103VH, but this interaction is not suggested by the homology model. Attempts to superpose the Cα atoms of Leu227 and Leu7Pep, in addition to Gln225 and Val226, resulted in clashes of LRR7 and LRR8 with a MagaFab loop (Ala25–Phe36). However, since this MagaFab loop is exposed to the solvent it is likely to be flexible, possibly allowing the binding of LRG1 without clashes and maybe even assisting in the recognition of LRR7 (Fig. 5a). In addition, the crystal contacts between the peptide molecules in the asymmetric unit described above might have favored peptide conformations different from the structure of the complex in solution. Crystal structures of either LRG1 or the LRG1–MagaFab complex would allow a more complete description of the interactions leading to LRG1 recognition.
In this work, we have determined the three-dimensional structure of the MagaFab fragment in complex with a 15-amino-acid peptide corresponding to the epitope of LRG1. A key Gln–Gln bidentate hydrogen bond and the presence of charged residues in the surroundings comprise the main interactions between MagaFab and LRG1. This binding is assisted by secondary residues whose weaker interactions could be strengthened by the mutations suggested herein, potentially leading to an increase in the affinity of the MagaFab–LRG1 complex. In addition, computational homology and threading models of LRG1 were generated and analyzed to suggest a model of the MagaFab–LRG1 complex.
The following references are cited in the supporting information for this article: Madeira et al. (2019) and Waterhouse et al. (2009).
We thank ALBA, Barcelona, Spain for providing synchrotron-radiation facilities. We are especially grateful to the staff of beamline BL13 (XALOC) for their assistance. Author contributions were as follows. JGF designed experiments, performed crystallization and determined three-dimensional structures, built computational models, performed data analysis and wrote the manuscript with input from all authors. HL contributed to data processing, model refinement and interpretation. FJ prepared MagaFab under the guidance of VC. All authors contributed to the study design and to writing the manuscript.
We thank the Worldwide Cancer Research for funding the research of Javier Gutiérrez-Fernández (grant No. 101944001). We thank the Wellcome Trust for providing a PhD studentship to Faiza Javaid (grant No. 203859/Z/16/Z) and Diabetes UK for support to Giulia De Rossi.
Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. (1990). J. Mol. Biol. 215, 403–410. CrossRef CAS PubMed Web of Science Google Scholar
Arakawa, T., Kobayashi-Yurugi, T., Alguel, Y., Iwanari, H., Hatae, H., Iwata, M., Abe, Y., Hino, T., Ikeda-Suno, C., Kuma, H., Kang, D., Murata, T., Hamakubo, T., Cameron, A. D., Kobayashi, T., Hamasaki, N. & Iwata, S. (2015). Science, 350, 680–684. Web of Science CrossRef CAS PubMed Google Scholar
Buchan, D. W. A. & Jones, D. T. (2019). Nucleic Acids Res. 47, W402–W407. Web of Science CrossRef CAS PubMed Google Scholar
Carmeliet, P. & Jain, R. K. (2000). Nature, 407, 249–257. Web of Science CrossRef PubMed CAS Google Scholar
Chen, V. B., Arendall, W. B., Headd, J. J., Keedy, D. A., Immormino, R. M., Kapral, G. J., Murray, L. W., Richardson, J. S. & Richardson, D. C. (2010). Acta Cryst. D66, 12–21. Web of Science CrossRef CAS IUCr Journals Google Scholar
Chen, Y., Wiesmann, C., Fuh, G., Li, B., Christinger, H. W., McKay, P., de Vos, A. M. & Lowman, H. B. (1999). J. Mol. Biol. 293, 865–881. Web of Science CrossRef PubMed CAS Google Scholar
Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. (2010). Acta Cryst. D66, 486–501. Web of Science CrossRef CAS IUCr Journals Google Scholar
Evans, P. R. (2011). Acta Cryst. D67, 282–292. Web of Science CrossRef CAS IUCr Journals Google Scholar
Gao, Y., Xie, Z., Ho, C., Wang, J., Li, Q., Zhang, Y. & Zhou, J. (2020). J. Investig. Dermatol. 140, 455–464. Web of Science CrossRef CAS PubMed Google Scholar
Henderson, J. N., Simmons, C. R., Fahmi, N. E., Jeffs, J. W., Borges, C. R. & Mills, J. H. (2020). Biochemistry, 59, 3401–3410. Web of Science CrossRef CAS PubMed Google Scholar
Hisata, S., Racanelli, A. C., Kermani, P., Schreiner, R., Houghton, S., Palikuqi, B., Kunar, B., Zhou, A., McConn, K., Capili, A., Redmond, D., Nolan, D. J., Ginsberg, M., Ding, B., Martinez, F. J., Scandura, J. M., Cloonan, S. M., Rafii, S. & Choi, A. M. K. (2021). J. Exp. Med. 218, e20200938. Web of Science CrossRef PubMed Google Scholar
Jemmerson, R. (2021). Front. Cell Dev. Biol. 9, 744908. Web of Science CrossRef PubMed Google Scholar
Juanhuix, J., Gil-Ortiz, F., Cuní, G., Colldelram, C., Nicolás, J., Lidón, J., Boter, E., Ruget, C., Ferrer, S. & Benach, J. (2014). J. Synchrotron Rad. 21, 679–689. Web of Science CrossRef CAS IUCr Journals Google Scholar
Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., Back, T., Petersen, S., Reiman, D., Clancy, E., Zielinski, M., Steinegger, M., Pacholska, M., Berghammer, T., Bodenstein, S., Silver, D., Vinyals, O., Senior, A. W., Kavukcuoglu, K., Kohli, P. & Hassabis, D. (2021). Nature, 596, 583–589. Web of Science CrossRef CAS PubMed Google Scholar
Kabsch, W. (2010). Acta Cryst. D66, 125–132. Web of Science CrossRef CAS IUCr Journals Google Scholar
Kallenberg, D., Tripathi, V., Javaid, F., Pilotti, C., George, J., Davis, S., Blackburn, J. W., O'Connor, M., Dowsett, L., Bowers, C. E., Liyanage, S., Gourlaouen, M., Hoeh, A., Mota, F., Selwood, D., Bainbridge, J., Chudasama, V., Greenwood, J. & Moss, S. E. (2020). bioRxiv, 2020.07.25.218149. Google Scholar
Krissinel, E. & Henrick, K. (2007). J. Mol. Biol. 372, 774–797. Web of Science CrossRef PubMed CAS Google Scholar
Liebschner, D., Afonine, P. V., Baker, M. L., Bunkóczi, G., Chen, V. B., Croll, T. I., Hintze, B., Hung, L.-W., Jain, S., McCoy, A. J., Moriarty, N. W., Oeffner, R. D., Poon, B. K., Prisant, M. G., Read, R. J., Richardson, J. S., Richardson, D. C., Sammito, M. D., Sobolev, O. V., Stockwell, D. H., Terwilliger, T. C., Urzhumtsev, A. G., Videau, L. L., Williams, C. J. & Adams, P. D. (2019). Acta Cryst. D75, 861–877. Web of Science CrossRef IUCr Journals Google Scholar
Madeira, F., Park, Y. M., Lee, J., Buso, N., Gur, T., Madhusoodanan, N., Basutkar, P., Tivey, A. R. N., Potter, S. C., Finn, R. D. & Lopez, R. (2019). Nucleic Acids Res. 47, W636–W641. Web of Science CrossRef CAS PubMed Google Scholar
McCoy, A. J., Grosse-Kunstleve, R. W., Adams, P. D., Winn, M. D., Storoni, L. C. & Read, R. J. (2007). J. Appl. Cryst. 40, 658–674. Web of Science CrossRef CAS IUCr Journals Google Scholar
O'Connor, M. N., Kallenberg, D. M., Camilli, C., Pilotti, C., Dritsoula, A., Jackstadt, R., Bowers, C. E., Watson, H. A., Alatsatianos, M., Ohme, J., Dowsett, L., George, J., Blackburn, J. W. D., Wang, X., Singhal, M., Augustin, H. G., Ager, A., Sansom, O. J., Moss, S. E. & Greenwood, J. (2021). Med, 2, 1231–1252. PubMed Google Scholar
Peng, J. & Xu, J. (2011). Proteins, 79, 161–171. Web of Science CrossRef CAS PubMed Google Scholar
Pickert, G., Neufert, C., Leppkes, M., Zheng, Y., Wittkopf, N., Warntjen, M., Lehr, H. A., Hirth, S., Weigmann, B., Wirtz, S., Ouyang, W., Neurath, M. F. & Becker, C. (2009). J. Exp. Med. 206, 1465–1472. Web of Science CrossRef PubMed CAS Google Scholar
Piechotta, A., Parthier, C., Kleinschmidt, M., Gnoth, K., Pillot, T., Lues, I., Demuth, H. U., Schilling, S., Rahfeld, J. U. & Stubbs, M. T. (2017). J. Biol. Chem. 292, 12713–12724. Web of Science CrossRef CAS PubMed Google Scholar
Potente, M., Gerhardt, H. & Carmeliet, P. (2011). Cell, 146, 873–887. Web of Science CrossRef CAS PubMed Google Scholar
Scapin, G., Yang, X., Prosise, W. W., McCoy, M., Reichert, P., Johnston, J. M., Kashi, R. S. & Strickland, C. (2015). Nat. Struct. Mol. Biol. 22, 953–958. Web of Science CrossRef CAS PubMed Google Scholar
Seiradake, E., Coles, C. H., Perestenko, P. V., Harlos, K., McIlhinney, R. A., Aricescu, A. R. & Jones, E. Y. (2011). EMBO J. 30, 4479–4488. Web of Science CrossRef CAS PubMed Google Scholar
Shirai, R., Hirano, F., Ohkura, N., Ikeda, K. & Inoue, S. (2009). Biochem. Biophys. Res. Commun. 382, 776–779. Web of Science CrossRef PubMed CAS Google Scholar
Sigrist, C. J., de Castro, E., Cerutti, L., Cuche, B. A., Hulo, N., Bridge, A., Bougueleret, L. & Xenarios, I. (2013). Nucleic Acids Res. 41, D344–D347. Web of Science CrossRef CAS PubMed Google Scholar
Singhal, M., Gengenbacher, N., Abdul Pari, A. A., Kamiyama, M., Hai, L., Kuhn, B. J., Kallenberg, D. M., Kulkarni, S. R., Camilli, C., Preuss, S. F., Leuchs, B., Mogler, C., Espinet, E., Besemfelder, E., Heide, D., Heikenwalder, M., Sprick, M. R., Trumpp, A., Krijgsveld, J., Schlesner, M., Hu, J., Moss, S. E., Greenwood, J. & Augustin, H. G. (2021). Sci. Transl. Med. 13, eabe6805. Web of Science CrossRef PubMed Google Scholar
Slabinski, L., Jaroszewski, L., Rychlewski, L., Wilson, I. A., Lesley, S. A. & Godzik, A. (2007). Bioinformatics, 23, 3403–3405. Web of Science CrossRef PubMed CAS Google Scholar
Teng, T. & Moffat, K. (2000). J. Synchrotron Rad. 7, 313–317. Web of Science CrossRef CAS IUCr Journals Google Scholar
Teplyakov, A., Obmolova, G., Luo, J. & Gilliland, G. L. (2018). Proteins, 86, 495–500. Web of Science CrossRef CAS PubMed Google Scholar
Tickle, I. J., Laskowski, R. A. & Moss, D. S. (2000). Acta Cryst. D56, 442–450. Web of Science CrossRef CAS IUCr Journals Google Scholar
Usón, I., Pohl, E., Schneider, T. R., Dauter, Z., Schmidt, A., Fritz, H. J. & Sheldrick, G. M. (1999). Acta Cryst. D55, 1158–1167. Web of Science CrossRef IUCr Journals Google Scholar
Wang, X., Abraham, S., McKenzie, J. A. G., Jeffs, N., Swire, M., Tripathi, V. B., Luhmann, U. F. O., Lange, C. A. K., Zhai, Z., Arthur, H. M., Bainbridge, J., Moss, S. E. & Greenwood, J. (2013). Nature, 499, 306–311. Web of Science CrossRef CAS PubMed Google Scholar
Waterhouse, A., Bertoni, M., Bienert, S., Studer, G., Tauriello, G., Gumienny, R., Heer, F. T., de Beer, T. A. P., Rempfer, C., Bordoli, L., Lepore, R. & Schwede, T. (2018). Nucleic Acids Res. 46, W296–W303. Web of Science CrossRef CAS PubMed Google Scholar
Waterhouse, A. M., Procter, J. B., Martin, D. M., Clamp, M. & Barton, G. J. (2009). Bioinformatics, 25, 1189–1191. Web of Science CrossRef PubMed CAS Google Scholar
Winn, M. D., Ballard, C. C., Cowtan, K. D., Dodson, E. J., Emsley, P., Evans, P. R., Keegan, R. M., Krissinel, E. B., Leslie, A. G. W., McCoy, A., McNicholas, S. J., Murshudov, G. N., Pannu, N. S., Potterton, E. A., Powell, H. R., Read, R. J., Vagin, A. & Wilson, K. S. (2011). Acta Cryst. D67, 235–242. Web of Science CrossRef CAS IUCr Journals Google Scholar
Xiang, Z. (2006). Curr. Protein Pept. Sci. 7, 217–227. Web of Science CrossRef PubMed CAS Google Scholar
Xu, J., McPartlon, M. & Li, J. (2021). Nat. Mach. Intell. 3, 601–609. Web of Science CrossRef PubMed Google Scholar
This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.