An allosteric redox switch in domain V of β₂-glycoprotein I controls membrane binding and anti-domain I autoantibody recognition

Abstract

β₂-glycoprotein I (β₂GPI) is an abundant multidomain plasma protein that plays various roles in the clotting and complement cascades. It is also the main target of antiphospholipid antibodies (aPL) in the acquired coagulopathy known as antiphospholipid syndrome (APS). Previous studies have shown that β₂GPI adopts two interconvertible biochemical conformations, oxidized and reduced, depending on the integrity of the disulfide bonds. However, the precise contribution of the disulfide bonds to β₂GPI structure and function is unknown. Here, we substituted cysteine residues with serine to investigate how the disulfide bonds C32-C60 in domain I (DI) and C288-C326 in domain V (DV) regulate β₂GPI's structure and function. Results of our biophysical and biochemical studies support the hypothesis that the C32-C60 disulfide bond plays a structural role, whereas the disulfide bond C288-C326 is allosteric. We demonstrate that absence of the C288-C326 bond, unlike absence of the C32-C60 bond, diminishes membrane binding without affecting the thermodynamic stability and overall structure of the protein, which remains elongated in solution. We also document that, while absence of the C32-C60 bond directly impairs recognition of β₂GPI by pathogenic anti-DI antibodies, absence of the C288-C326 disulfide bond is sufficient to abolish complex formation in the presence of anionic phospholipids. We conclude that the disulfide bond C288-C326 operates as a molecular switch capable of regulating β₂GPI's physiological functions in a redox-dependent manner. We propose that in APS patients with anti-DI antibodies, selective rupture of the C288-C326 disulfide bond may be a valid strategy to lower the pathogenic potential of aPL.

β₂-glycoprotein I (β₂GPI), also known as apolipoprotein H, is a phospholipid-binding protein that plays multiple roles in the coagulation and complement cascades (1, 2, 3, 4). It is also recognized as the main antigen of antiphospholipid antibody (aPL) in the autoimmune disorder antiphospholipid syndrome (APS) (1). β₂GPI consists of 326 amino acids organized in four canonical complement control protein (CCP) domains (DI-DIV) (5, 6) followed by an aberrant CCP-like domain, DV, which is held responsible for the interaction with a variety of negatively charged surfaces, such as heparin and “activated” plasma membranes (7, 8, 9, 10). Other important structural features of β₂GPI are four N-glycosylations, which account for ∼20% of its molecular weight (11), and 22 conserved cysteines (C), potentially forming 11 disulfide bonds (Fig. 1A).

Figure 1.

Chemical characterization and oligomeric state of the reduced-like variants.A, X-ray crystal structure of oxβ₂GPI (6V09) (10) documenting integrity of the 11 disulfide bonds (red sticks) and the location of the two disulfide bonds in DI targeted in this study (yellow spheres). Also highlighted are other regions of the molecules relevant for this study: the N-terminal HPC4 tag (wheat); the epitope R39-K44, which is recognized by MBB2 (10); residue W53 in DI (orange), the ω-loop (311–317), and W316 in DV, which is not visible in the structure due to disorder. Glycosylations are shown as green sticks. B, SDS-PAGE analysis of the recombinant proteins before (−) and after (+) the treatment with the reducing agent β-mercaptoethanol (β-ME). Samples are 1 = oxβ₂GPI; 2 = β₂GPI C32S/C60S; 3 = β₂GPI C288S/C326S. C, SDS-PAGE analysis of the recombinant proteins before (left) and after (right) removal of the glycosylations. Samples are 1 = oxβ₂GPI; 2 = β₂GPI C32S/C60S; 3 = β₂GPI C288S/C326S. Note how the deglycosylation treatment reduce the difference of electrophoretic mobility between the samples, indicating that β₂GPI C32S/C60S contains more glycans when produced in HEK293 cells. D, reactivity of the recombinant proteins against AF-647 maleimide in the absence (left) and presence (right) of TCEP monitored by SDS-PAGE. Samples are 1 = oxβ₂GPI; 2 = β₂GPI C288S/C326S; 3 and 4 = β₂GPI C32S/C60S. Bands 3 and 4 are two different preparations of β₂GPI C32S/C60S, which document minimal lot-to-lot variability. E, shown are the chromatograms obtained by SEC documenting monomeric proteins and a slightly but appreciable difference between the variant β₂GPI C32S/C60S and the other two samples. The small shift is in agreement with the larger hydrodynamic radius documented by SDS-PAGE in panels B–D.

Even though there is some disagreement regarding the structural architecture of the protein while it circulates in the blood (8, 9, 10, 12, 13, 14), consensus exists on the fact that, in human plasma, β₂GPI adopts two major interconvertible redox states, oxidized (54%) and reduced (46%), depending on the disulfide bonds' integrity (15, 16). In the oxidized form, all 22 cysteine residues are oxidized, forming 11 disulfide bonds (S-S). By contrast, in the reduced form, some of the 11 disulfide bonds are broken. While the exact number of oxidized and reduced species present in human plasma has yet to be defined, pioneering studies led by the group of Krilis have revealed that oxidized β₂GPI is the substrate of several thiol-isomerases and can be reduced by this family of enzymes both in vitro and in vivo (17), in which case the disulfide bond most heavily targeted is C288-C326 in DV, followed by C32-C60 in DI (18).

While the mechanisms controlling the redox equilibrium of β₂GPI have yet to be fully understood, similar to other plasma proteins (see Table 1 in Chiu and Hogg (19)), oxidation/reduction cycles may provide a means of regulation of its biological function, with important consequences in APS (20, 21) and other cardiovascular diseases in which patients are subjected to higher levels of oxidative stress (22). For instance, while the oxidized form of β₂GPI was shown to inhibit platelet adhesion to immobilized von Willebrand Factor (vWF) under flow conditions (23), the reduced form was shown to increase adhesion of platelet via platelet glycoprotein Ib alpha (GpIbα) (18). Another functional difference between oxidized and reduced β₂GPI is that reduced β₂GPI protected cells of the vasculature from free radical injury, perhaps working as a chemical scavenger (16, 24). Finally, it was also documented that the levels of oxidized β₂GPI were significantly elevated in APS patients with an active form of the disease (25), raising the hypothesis that oxidized β₂GPI is the preferred target of aPL. In contrast to this hypothesis, studies published by Buchholz et al. (13) while completing this manuscript documented a strong increase in the binding of aPL to reduced β₂GPI that was prepared by thioredoxin reduction followed by 3-(N-maleimido-propionyl) biocytin (MPB) labeling to avoid reoxidation. Hence, it remains unclear under what circumstances aPL prefers oxidized or reduced β₂GPI.

Based on this premise and encouraged by our previous structural analyses of human recombinant β₂GPI (10), we set out to investigate the structural and functional consequences imparted by the selective rupture of the disulfide bonds C32-C60 in DI and C288-C326 in DV of β₂GPI. Differently from previous studies, in which chemical or enzymatic methods were used to reduce disulfide bonds of β₂GPI (13, 17, 26), chemically stable, reduced-like forms of β₂GPI were produced using genetic methods by replacing cysteine residues with serine. The C→S replacement makes the formation of disulfide bonds impossible. At the same time, given that serine differs from cysteine in a single atom—the oxygen of the alcohol replaces the sulfur of the thiol—serine substitutions are likely to preserve the H-bonding network imparted by the thiol groups, which is lost when thiols are reacted with chemicals such as MPB. Hence, replacement of C→S at positions 32 and 60 in DI and 288 and 326 in DV should provide a reasonable experimental model to study the effects caused by the rupture of each disulfide bond.

Results

Removal of C32-C60 and C288-C326 disulfide bonds in β₂GPI does not affect protein expression levels and oligomeric state

The double mutants β₂GPI C32S/C60S and β₂GPI C288S/C326S were expressed in human embryonic kidney cells alongside β₂GPI WT (Fig. 1A). Because β₂GPI WT contains 22 cysteines forming 11 disulfide bonds (10), in this study, we named it oxidized β₂GPI (oxβ₂GPI). Purification of the proteins from the media was performed by immunoaffinity chromatography under mild buffer conditions (20 mM Tris pH 7.4, 145 mM NaCl, TBS) by taking advantage of a calcium-dependent tag located at the N-terminus (10). This strategy was chosen to preserve the structural and chemical integrity of the recombinant proteins. Yields of ∼0.5 mg per liter of cell culture were obtained for all three proteins, documenting similar expression levels.

Immunopurified proteins migrated as a single smeary band at ∼50 kDa under nonreducing and reducing conditions, indicative of a single, heavily glycosylated polypeptide chain (Fig. 1B). Interestingly, the band of β₂GPI C32S/C60S displayed a slightly lower electrophoretic mobility compared with oxβ₂GPI, implying a larger hydrodynamic radius. Since no such difference was observed for β₂GPI C288S/C326S, we attributed this effect to a higher content of glycosylations, which may have resulted from a partial disruption of the DI's secondary structure. In line with this interpretation, the difference between β₂GPI C32S/C60S and oxβ₂GPI almost completely disappeared after 3 h of deglycosylation treatment (Fig. 1C). Thus, unlike C→S substitutions at positions 288 and 326 in DV, substitutions at positions 32 and 60 in DI enhanced glycan incorporation during protein synthesis, possibly as a result of local structural perturbations in DI.

In addition to introducing local structural perturbations, absence of the disulfide bond in DI may result in global disulfide bond instability. This was investigated by reacting β₂GPI C32S/C60S with AlexaFluor-647-C2-maleimide in the absence and presence of the reducing agent tris(2-carboxyethyl)phosphine hydrochloride (TCEP), as performed recently with protein disulfide isomerase (27). oxβ₂GPI and β₂GPI C288S/C326S were used as a control. Regardless of the protein sample, incorporation of the fluorescent dyes was observed almost exclusively in the presence of TCEP (Fig. 1D). This indicates that, in the absence of TCEP, the cysteine residues are incapable of reacting with the maleimide dye, suggesting that they are either oxidized, likely forming chemically stable disulfide bonds, or simply inaccessible to the fluorescent dye, in which case they should be buried, an observation that is consistent with a folded protein structure.

Finally, to elucidate the oligomeric state under native conditions, we performed analytical size-exclusion chromatography (SEC) experiments in 20 mM Tris pH 7.4, 145 mM NaCl, 5 mM CaCl₂. The proteins were prepared at 20 μM, resulting in a concentration ∼2–4 μM in the column due to dilution. This is the concentration of β₂GPI in human plasma (1). The vast majority (>95%) of β₂GPI C32S/C60S and β₂GPI C288S/C326S eluted in single peaks with retention volumes of ∼12.9 ml and ∼13.2 ml, respectively (Fig. 1E). Importantly, the chromatographic profiles and elution volumes were similar to those obtained for oxβ₂GPI (∼13.2 ml). This is consistent with monomeric proteins.

Structural consequences imparted by the loss of the disulfide bonds

We have recently demonstrated that oxβ₂GPI adopts an elongated conformation under physiological pH (i.e., 7.4) and ionic strength concentrations (i.e., 145 mM NaCl) (10). However, the structures of the reduced form(s) of β₂GPI remain unknown. Recent atomic force microcopy (AFM) analyses of reduced β₂GPI produced by enzymatic reaction indicate that reduced β₂GPI absorbed into mica sheets mostly retains a closed form (65%), with a slight preference for the open form when compared with untreated β₂GPI (35% versus 25%) (13). However, SEC analysis of the proteins in Figure 1E indicates that β₂GPI C32S/C60S and β₂GPI C288S/C326S are similar to oxβ₂GPI and, therefore, should too be elongated in solution. To validate our observations, we performed small-angle X-ray scattering (SAXS) experiments. The scattering profiles and pair distribution functions of the three proteins in TBS buffer are shown in Figure 2A and Figure 2B, respectively. Data were collected in the presence of ∼0.5% (w/w) glycerol, which minimizes radiation damage while preserving proteins' native conformations. From these plots, we calculated radius of gyration (R_g) and maximum linear dimensions (D_max) values, which are shown in Figure 2C. The values of R_g and D_max are similar for oxβ₂GPI and β₂GPI C288S/C326S. These values are consistent with our previously published data for β₂GPI, but not with what AFM analyses would have predicted (13), documenting elongated conformations of oxβ₂GPI and β₂GPI C288S/C326S in solution. Also consistent with the conclusions derived from the SEC experiments in Figure 1E are the SAXS data obtained with the mutant β₂GPI C32S/C60S. In this case, the values of R_g and D_max were slightly larger than that of oxβ₂GPI (Fig. 2C), documenting a more extended conformation of the protein in solution, hence a larger hydrodynamic radius. This difference between the variants is best appreciated in Figure 2D, which shows the ab initio envelopes calculated from the scattering profiles shown in Figure 2A.

Figure 2.

Solution structure of the disulfide bond variants monitored by SAXS. Scattering profiles (A) and normalized pair distribution functions (B) of oxβ₂GPI (black), β₂GPI C32S/C60S (magenta), and β₂GPI C288S/C326S (cyan) obtained in 20 mM Tris pH 7.4, 145 mM NaCl, 0.5% glycerol (w/w) at a concentration of 4 mg/ml. C, radius of gyration (R_g) determined by linear fit from the Guinier plot and D_max calculated for each protein sample. D, surface representation of the ab initio envelopes calculated from the scattering profiles shown in panel A. Note how β₂GPI C32S/C60S is significantly more elongated than oxβ₂GPI and β₂GPI C288S/C326S.

Impact of the disulfide bonds on the stability of β₂GPI

Having established that all three proteins are monomeric and adopt elongated structures in solution, we next investigated their thermodynamic stability by chemical denaturation methods. This is because rupture of disulfide bonds often leads to destabilization of a protein's three-dimensional structure, which, in turn, results in different protein denaturation profiles (28). β₂GPI contains five tryptophan residues, one in every domain. Addition of 7M Gnd-HCl to solutions of oxβ₂GPI, β₂GPI C32S/C60S, and β₂GPI C288S/C326S in TBS buffer produced a significant threefold increase of the intrinsic fluorescence signal (Fig. 3A). This indicates that (1) in TBS buffer, the proteins are folded, and (2) a high concentration of Gnd-HCl causes the proteins to unfold. These spectroscopic signatures are reminiscent of previous data obtained using chemically synthesized DI (29). They are also consistent with the structure of human recombinant β₂GPI shown in Figure 1A. In this structure, residues W53 in DI, W111 in DII, W175 in DIII, and W235 in DIV are buried within the protein core, located in the proximity of a disulfide bond. In contrast, W316 in DV is located in the ω-loop, which is flexible and constitutively exposed to the solvent. Since disulfide bonds quench tryptophan fluorescence by an electron transfer mechanism (30), unfolding of β₂GPI is expected to liberate the tryptophans in DI-DIV, resulting in an increase of the fluorescence signal. In line with this interpretation, we found that the fluorescence intensity of the 280-nm emission spectrum of β₂GPI C32S/C60S was ∼35% higher than that of oxβ₂GPI, whereas it did not change for the mutant β₂GPI C288S/C326S. We speculate that, while the chemical environment of W316 is not perturbed by the mutations C288S/C326S, absence of the disulfide bond C32-C60 results in loss of fluorescence quenching, also facilitated by relaxation of the protein secondary structure.

Figure 3.

Stability of the disulfide bond variants.A, baseline subtracted fluorescence spectra of oxβ₂GPI (black), β₂GPI C32S/C60S (magenta), and β₂GPI C288S/C326S (cyan) (1.0 ml, 0.2 mM in 20 mM Tris pH 7.4, 145 mM NaCl, 0.015% Tween 20 at 25 °C) in the absence (solid lines) and presence (dotted lines) of 7M Gnd-HCl. B, chemical denaturation induced by Gdn-HCl. Protein samples were excited at 280 nm, and the fluorescence intensity was recorded at 350 nm as a function of denaturant concentration. Continuous line represents the best fit of data points to Equation 25 in Eftink (60), yielding a GdnHCl_1/2 and denaturation index (m) values of 3.41 ± 0.10 M and 0.0140 ± 0.0009 for oxβ₂GPI (black), 3.04 ± 0.10 M and 0.0123 ± 0.0010 for β₂GPI C32S/C60S (magenta), and 3.32 ± 0.12 M and 0.0132 ± 0.0009 for β₂GPI C288S/C326S (cyan). Note how the sigmoidal curves for oxβ₂GPI and β₂GPI C288S/C326S are almost identical but different from β₂GPI C32S/C60S.

By taking advantage of the denaturant-dependent increase of fluorescence intensity reported in Figure 3A, concentration-dependent experiments were performed to quantify the impact of the two disulfide bonds on the stability of β₂GPI (Fig. 3B). Sigmoidal curves were obtained for all the constructs, indicative of a cooperative transition between folded and unfolded species (28). Analysis of the fluorescent data with a two-state model (F->U) (29) yielded Gdn-HCl_1/2 values of 3.41 ± 0.10 M for oxβ₂GPI, 3.04 ± 0.10 M for β₂GPI C32S/C60S, and 3.31 ± 0.12 M for β₂GPI C288S/C326S. While the values of Gdn-HCl_1/2 were within experimental error for β₂GPI C288S/C326S and oxβ₂GPI, they were significantly different (0.37 units) from those calculated for the variant β₂GPI C32S/C60S. This documents that, unlike C288-C326, the disulfide bond C32-C60 in DI is important for the thermodynamic stability of β₂GPI: its absence facilitates protein unfolding.

A redox switch in DV controls β₂GPI binding to the membranes

An important physiological role of β₂GPI is to clear necrotic and bacteria cells from the circulation and neutralize cell debris (1), such as polyanions, which may cause cellular damage, inflammation, and may trigger coagulation. To evaluate the functional consequences caused by the loss of the disulfide bonds, binding studies were performed using unfractionated heparin and liposomes made of phosphatidylcholine: cardiolipin (POPC:CL, 80:20% molar ratio) and phosphatidylcholine: phosphatidylserine (POPC:POPS, 80:20% molar ratio), as simplified models for necrotic cells, which are also relevant to the role β₂GPI in APS (31, 32).

Chromatographic studies were used to study the interaction of β₂GPI with heparin. This methodology, although qualitative, has been successfully used in the past and proved useful to identify some of the regions responsible for this interaction (33). Consistent with previous reports (33), oxβ₂GPI bound with moderate strength to the heparin column. It eluted from the column at a value of conductivity equals 29 mS/cm, which corresponds to ∼300 mM NaCl (Fig. 4A). Likewise, elution of β₂GPI C32S/C60S from the heparin column occurred at 28 mS/cm, documenting no perturbation caused by the mutations in DI. More interesting, however, was the effect of the substitutions C288S/C326S. β₂GPI C288S/C326S bound to heparin better than oxβ₂GPI (28 mS/cm versus 33 mS/cm), suggesting that an intact C288-C326 disulfide bond lowers the affinity for heparin.

Figure 4.

Interaction with heparin and plasma membranes.A, heparin chromatographic analyses. Samples solubilized in 20 mM Tris pH 7.4 50 mM NaCl were loaded onto a 1 ml HiTrap Heparin column and eluted with a linear gradient of NaCl, from 50 to 500 mM in 15 min (dotted line). Flow rate was set to 0.8 ml/min and absorbance was set to 280 nm. Note how the peak center of β₂GPI C288S/C326S (cyan) shifts toward the right compared with oxβ₂GPI (black) and β₂GPI C32S/C60S (magenta), documenting higher affinity for heparin. B and C, sensograms obtained by injecting increasing concentrations of protein solutions (from 0.019 to 5 μM) into an L1 chip coated with liposomes made of (B) POPC:CL (80:20%, 100 nm) and (C) POPC:POPS (80:20%, 100 nm). Flow rate was 25 ml/min. Experimental conditions are 20 mM Tris pH 7.4, 145 mM NaCl, 0.1% (w/v) BSA (D and E) RU values at equilibrium plotted versus concentration of β₂GPI (dots) and fit to a simple binding isotherm (line) to extract values of apparent K_d and RU_max reported in panel F. Note how oxβ₂GPI (black) and β₂GPI C32S/C60S (magenta) are similar to each other but different from the variant with the mutations C288S/C326S. F, values of affinity, RU_max, and stoichiometry (S) calculated by SPR. S was calculated using the formula (RU_protein/RU_liposomes)∗(MW_liposome/MW_protein) as described elsewhere (34) and assuming that 50% of the total number of lipids are accessible to the protein.

Next, surface plasmon resonance (SPR) studies were performed to monitor the kinetics of interaction between β₂GPI and immobilized large unilamellar vesicles (LUV) prepared by extrusion. SPR was chosen for two reasons. First, it allows determination of apparent binding affinity constants (K_d), which provide a quantitative description of the system. Second, it provides information about the maximal binding stoichiometry (34), expressed as the number of protein molecules bound at saturation to the outer region of a 100-nm liposome. The sensograms of oxβ₂GPI, β₂GPI C32S/C60S, and β₂GPI C288S/C326S obtained by flowing increasing concentrations of proteins in 20 mM Tris pH 7.4, 145 mM NaCl, 0.1% BSA (w/v) onto a L1 sensor chips immobilized with POPC:CL and POPC:POPS are shown in Figure 4, B and C, respectively. Maximum response obtained at 50 s plotted versus total ligand concentration is shown in Figure 4, D and E The SPR profiles of oxβ₂GPI and β₂GPI C32S/C60S were qualitatively similar to each other and so were the values of K_d and maximal binding stoichiometry (Fig. 4F). In contrast, binding of β₂GPI C288S/C326S to liposomes was reduced compared with oxβ₂GPI. We measured values of K_d,app ∼13 and ∼8 times weaker compared with oxβ₂GPI and a maximum binding stoichiometry of 174 and 66 for POPC:CL and POPC:POPS, respectively. This corresponds to one protein-binding site per ∼50 CL and ∼130 POPS, as compared with one protein-binding site per ∼15 CL and ∼35 POPS for oxβ₂GPI and ∼16 CL and ∼38 POPS for β₂GPI C32S/C60S. The remarkable drop in number of β₂GPI molecules bound at saturation for β₂GPI C288S/C326S translates into less protein bound per square nanometer (nm²). These results are consistent with a model in which the integrity of the disulfide bond C288-C326 in DV regulates binding and accumulation of β₂GPI onto negatively charged plasma membranes.

Binding of pathogenic anti-DI antibodies to β₂GPI is reduced by loss of the disulfide bond C32-C60

Antibodies against β₂GPI (anti-β₂GPI) are considered a hallmark of APS and are believed to play a key role in the disease pathogenesis (35). Among them, the ones targeting DI are considered highly pathogenic (36, 37). We, therefore, evaluated how the mutations C32S/C60S and C288S/C326S might affect anti-β₂GPI recognition. We performed a series of ELISA assays in which β₂GPI was immobilized onto a hydrophilic plastic plate and patients' plasma was used as a source of autoantibodies. Plasma samples from 18 APS patients with lupus anticoagulant (LA) and also positive for anticardiolipin (aCL) and anti-β₂GPI antibodies, so called “triple positive,” were chosen for these studies. This is because, in these patients, most anti-β₂GPI antibodies are believed to target DI (38). The results of the ELISA assays are shown in Figure 5A. They document that anti-β₂GPI from triple-positive APS patients, while reacting equally well against immobilized oxβ₂GPI and β₂GPI C288S/C326S, reacted very poorly toward immobilized β₂GPI C32S/C60S. Since the difference between oxβ₂GPI and β₂GPI C288S/C326S was not statistically significant (p > 0.05), we concluded that loss of the disulfide bond C32-C60, but not C288-C326, impairs autoantibody recognition toward immobilized β₂GPI.

Figure 5.

Interaction of β₂GPI with antiphospholipid antibodies.A, reactivity of aPL from 18 triple-positive APS patients monitored by ELISA. Each dot represents a patient, and the value of each dot is the average of two independent determinations. Statistical analysis between the three groups was performed using ANOVA (∗∗∗ p < 0.001, ∗∗∗∗ p < 0.0001). B, binding of MBB2 to immobilized oxβ₂GPI and variants followed by ELISA, using the same protocol for panel A. Dose-dependent curves were globally fit to a Hill equation. Best-fit values of EC₅₀ are reported in panel E. C, interaction between immobilized MBB2 and soluble β₂GPI followed by SPR. D, maximum response at each concertation was plotted against total concentration of soluble β₂GPI. Dose-dependent curves were globally fit to a simple binding equation. Best-fit values of K_d are reported in panel E. Reproducibility was tested by performing each binding experiment two times, by randomizing the order of the injections and immobilizing the same antibody in the two different flow channels. E, values of affinity calculated by ELISA (EC₅₀) and SPR (K_d,app).

A possible explanation for the data in Figure 5A, which is consistent with the biophysical structural studies in Figure 3, is that loss of the disulfide bond C32-C60 results in partial unfolding of DI. This, in turn, may alter the conformation of the loop G40-R43, which is considered the immunodominant epitope of anti-DI antibodies (39). Since patients' autoantibodies are heterogenous, we tested this hypothesis by performing binding studies using the monoclonal antibody MBB2. Previous biochemical studies have shown that MBB2 recognizes the epitope R39-K44 in DI (10) and is pathogenic in vivo (40). To allow direct comparison, binding studies were conducted by ELISA, using the same protocol developed for the experiments shown in Figure 5A. We found that soluble MBB2 reacted very strongly toward immobilized oxβ₂GPI and β₂GPI C288S/C326S, reaching saturation at a concentration of 1 μg/ml, which corresponds to a molar concentration of 12.5 nM (Fig. 5B). By contrast, soluble MBB2 reacted very poorly toward immobilized β₂GPI C32S/C60S, indicating impaired binding. Global fitting of the dose-dependent curves using a Hill equation was used to quantify the differences between the variants. This method yielded EC₅₀ values of 0.92 ± 0.1 nM for oxβ₂GPI, 146 ± 20 nM for β₂GPI C32S/C60S, and 0.77 ± 0.1 nM for β₂GPI C288S/C326S, documenting a drop ∼150-fold when the disulfide bond C32-C60 in DI is broken.

Although unlikely due to the conservative nature of the C→S substitutions, the loss of affinity of MBB2 for β₂GPI C32S/C60S may arise from unfolding and/or a preferential orientation upon the interaction with the protein to the plastic surface, resulting in masking of the epitope R39-K44, which is recognized by MBB2. To rule out these possibilities and, at the same time, eliminate the complexity introduced by multiple washing/binding steps associated with ELISA, we performed SPR experiments. MBB2 was immobilized to the chip's surface and β₂GPI was kept in the fluid phase. In agreement with our previously published studies (10), we found that oxβ₂GPI interacted with MBB2 with a K_d = 5.3 ± 0.3 μM (Fig. 5C, top panel), characterized by fast association and dissociation rate constants. As for the disulfide bond variants, we found that the binding profile of β₂GPI C288S/C326S was identical to oxβ₂GPI (Fig. 5C, bottom panel). In contrast, the binding of β₂GPI C32S/C60S to immobilized MBB2 was approximately 5-fold reduced compared with oxβ₂GPI. Loss of binding was clearly visible by inspecting the sensograms shown in Figure 5C, middle panel, and the dose-dependent plots are displayed in Figure 5D. This data is qualitatively consistent with the results obtained by ELISA and supports a model whereby anti-β₂GPI antibodies in triple-positive APS patients target an epitope in DI, which is compromised, yet not disrupted, when the disulfide bond C32-C60 is broken. Differences between the values of affinities measured by ELISA and SPR for MBB2 were expected from previous studies (10, 40), implying that immobilization of the antigen onto the plastic plate results in optimal antigen recognition. Whether conformational changes are responsible for this phenomenon is certainly a possibility. However, such conformational changes must be different from opening of the structure caused by the interaction of circular β₂GPI with the plastic plate inferred by previous studies (12, 13, 37) since the recombinant proteins used in this study are already elongated.

Interplay between disulfide bonds and anti-β₂GPI/β₂GPI complex formation in the presence of anionic phospholipids

Even though our binding assays in Figure 5 document that autoantibodies bind to DI and therefore are not directly affected by the structural integrity of the disulfide bond C288-C326, in vivo, pathogenic anti-β₂GPI/β₂GPI complexes often form on the surface of activated or dying cells exposing negatively charged phospholipids and very rarely in solution (40, 41). We, therefore, set out to elucidate the interplay between the integrity of the disulfide bonds and anti-β₂GPI/β₂GPI complex formation in the presence of anionic phospholipids. Similar experiments were recently performed by our group to uncover functional differences between type I and type II antiphosphatidylserine/prothrombin antibodies (aPS/PT) in APS patients (42). To mimic a physiologically relevant scenario, we incubated a fixed concentration of β₂GPI (i.e., 2 μM, the concentration of β₂GPI in human plasma) with increasing concentrations of MBB2 (from 0.01 to 2 μM) in 20 mM Tris pH 7.4, 145 mM NaCl, 5 mM CaCl₂, and 0.1% BSA (w/v) and then injected this mixture into a surface coated with POPC:CL. The concentrations of MBB2 were chosen according to previous literature to cover a reasonable range of anti-β₂GPI concentrations found in APS patients (43). Figure 6A shows the sensograms obtained for the three protein samples injected into the same phospholipid surface. The sensograms are interpreted as accumulation of β₂GPI onto the lipids that is amplified by the presence of MBB2. In the presence of MBB2, we observed maximum response for oxβ₂GPI, followed by β₂GPI C32S/C60S. A weak response was measured for β₂GPI C288S/C326S even at the highest concentration of MBB2 tested, which produces saturation of the signal for oxβ₂GPI. Since MBB2 does not bind to the phospholipids, this data provides evidence that multimerization of oxβ₂GPI and β₂GPI C32S/C60S onto the membranes is stimulated by MBB2. It also shows that binding of β₂GPI C288S/C326S to the membranes cannot be rescued by addition of MBB2.

Figure 6.

Interplay between anti-DI antibodies and phospholipid binding.A, sensograms of oxβ₂GPI and mutants obtained by injecting 2 μM of protein alone (dotted line) or protein that was complexed with increasing concentrations of MBB2 (from 0.01 to 2 μM) into an L1 chip coated with POPC:CL (80:20). Flow rate was 25 μl/min. Running buffer was 20 mM Tris pH 7.4, 145 mM NaCl, 5 mM CaCl₂, 0.1% (w/v) BSA. B, shown is the response at each concentration of MBB2 obtained after 110 s subtracted for the protein alone sample versus the concentration of MBB2. Solid lines were obtained used the phenomenological equation y = (R₀ + R_max∗(x/EC₅₀))/(1+(x/EC₅₀)) to measure the concentration of MBB2, which provides 50% of the maximum effect (R_max = 675 units). C, schematic model summarizing the interplay between oxidative modifications, phospholipid binding, and MBB2 binding.

A nonlinear fit of the response versus the concentration of MBB2, shown in Figure 6B, was used to quantify the difference between the protein samples. This analysis yielded values of EC_50, interpreted as the concentration of MBB2, which provides 50% of the maximum effect, of ∼0.2 μM for oxβ₂GPI, ∼2 μM for β₂GPI C32S/C60S, and >90 μM for β₂GPI C288S/C326S, underscoring the remarkable difference between oxβ₂GPI and β₂GPI C288S/C326S and documenting that the presence of anionic phospholipids enhances MBB2 binding of ∼100-fold for oxβ₂GPI. Of note, accumulation of β₂GPI-MBB2 complexes started at concentrations of MBB2 that were significantly lower (>100-fold) than the values of K_d calculated for the proteins in solution, implying that binding of β₂GPI to the lipids precedes antibody binding and that formation of a soluble complex is not necessary to explain the kinetics of binding. Hence, we conclude that oxidized β₂GPI bound to anionic membrane is the preferred target of MBB2 and perhaps other anti-DI antibodies.

Discussion

In this work, we performed biophysical and biochemical studies to characterize how loss of the disulfide bonds C32-C60 and C288-C326 in the plasma glycoprotein β₂GPI perturbs some of its physiological functions, which are also important for aPL binding. Our results lend support to the hypothesis that the disulfide bond C288-C326 acts as a redox switch that regulates membrane binding and, therefore, indirectly controls the formation of immune complexes under conditions most relevant to APS pathology (Fig. 6C). Furthermore, we surprisingly found that loss of the disulfide bond C288-C326 stimulates heparin binding (Fig. 4A). Since (auto)activation of coagulation factor XII is accelerated by heparin (44, 45), enhanced binding of reduced β₂GPI to polyanions such as heparin, and perhaps polyphosphates (46), could help explain its mild anticoagulant activity in the bloodstream (47). Based on this evidence, mechanisms to preserve the reduced form of β₂GPI in the circulation or stimulate the reduction of the disulfide bond C288-C326 may find utility in the treatment of APS patients. In addition to highlighting the functional role of the disulfide bond C288-C326, our results are also informative because (1) they consolidate previous observations regarding the epitope specificity of anti-β₂GPI antibodies in triple-positive APS patients, who carry the greatest risk of developing thrombosis (10, 38, 39, 48); and (2) they stress the importance of oxidative posttranslational modifications as a highly relevant biological mechanism to fine-tune the function of extracellular proteins involved in hemostasis (49, 50, 51, 52, 53, 54).

Differently from previous studies (13, 17, 26), which relied on chemical and enzymatic reduction of disulfide bonds, we generated two variants in which the cysteine residues 32 and 60 in DI and 288 and 326 in DV were substituted to serine. β₂GPI WT, which represents the oxidized monomeric form of β₂GPI (10), was used as a control. We opted for this protein engineering strategy because it overcomes many challenges associated with the production, purification, and storage of reduced protein species generated by chemical and enzymatic reactions without, however, introducing significant chemical perturbations to the system.

An initial finding documented by SEC (Fig. 1E) and SAXS (Fig. 2) made possible by this strategy was that the absence of either disulfide bond did not perturb the oligomerization state, nor did it affect the overall structure of β₂GPI, which remained elongated in solution. Since β₂GPI is believed to adopt “closed” and “open” forms (1), this information helps us to rule out a potential link between looping of the molecule and integrity of the disulfide bonds. Whether other mechanisms controlling circularization of β₂GPI exist, such as surface-induced oligomerization or reshuffling of disulfide bonds mediated by thiol-isomerases, remains a possibility. Thus far, however, we have no evidence that monomeric β₂GPI made recombinantly in mammalian cells or purified from plasma using the perchloric acid method can adopt a circular form in solution, as proposed by other research groups (12, 13).

An additional finding made possible by our reagents was the discovery that two disulfide bonds play very distinct structural roles in the context of protein folding. While absence of C32-C60 led to profound conformational changes in DI, which ultimately facilitated protein unfolding (Fig. 3), loss of C288-C326 imparted structural alterations that were well tolerated by the protein fold, with no measurable effects on the thermodynamic stability of β₂GPI. Based on this evidence, and according to the classification recently proposed by Chiu el al. (19), we conclude that the disulfide bond C32-C60 plays a “structural” role, whereas the disulfide bond C288-C326 is “allosteric.” Given that the topology of the disulfide bond C32-C60 is conserved among CCP domains (6, 8, 9, 10), our results may also be relevant to other complement factors, such as complement factor H, which, similar to β₂GPI, acts at the interface of coagulation and complement cascades and whose detrimental function is linked to many human diseases, including APS (55).

Beyond the new structural insights, an important aspect of this study is that it provides an improved understanding of the role of the different reduced forms of β₂GPI in APS. In the context of aPL recognition, our biochemical data indicates that it is the disulfide bond C288-C326, and not C32-C60, that most heavily regulates protein's function. This is because the integrity of C288-C326 controls binding of β₂GPI to the plasma membranes (Fig. 4), and in turn, binding to the membranes is necessary for complex formation. Indeed, even though pathogenic antibodies in APS patients target DI of β₂GPI, loss of binding affinity due to the absence of the disulfide bond C32-C60 (Fig. 5) is not enough to prevent complex formation. Importantly, this scenario is consistent with a model of anti-β₂GPI/β₂GPI recognition originally proposed in the 1990s by several groups (see Table 2 of Giles et al. (56)) and recently extended by us (10), envisioning that the preferential binding of anti-DI antibodies to phospholipid-bound-β₂GPI arises from the ability of the preexisting elongated form to accumulate on the membranes. It also helps explain why increased levels of oxidative stress, which are believed to favor the oxidation the disulfide bond C288-C326 in circulating β₂GPI, may worsen disease in APS patients (20, 25). Interestingly, hydroxychloroquine (HQ) prevents formation of immune complexes by interfering with binding of β₂GPI to anionic phospholipids (57). Whether rupture of the disulfide bond C288-C326 contributes to the protective effects seen with HQ remains unknown.

Although less important for aPL recognition, the data obtained for the mutant β₂GPI C32S/C60S also merits attention, especially in the context of aPL generation. Since a relationship exists between misfolding of β₂GPI and autoantibody production (58), we speculate that reduction of the disulfide bond C32-C60 may facilitate the generation of aPL by favoring protein unfolding. Future studies facilitated by these reagents may help clarify whether APS belongs to the growing realm of protein folding diseases.

Experimental procedures

Protein expression, purification, and reagents

Human β₂GPI wild-type (residues 1–326, referred in this work as to oxβ₂GPI) modified to include an epitope for the HPC4 antibody at the N-terminal was cloned into a pDEST40 expression vector using the Gateway cloning technology (Life Technologies). The mutations C32S, C60S, C288S, C326S were generated using the Quickchange Lighting kit (Agilent) and appropriate primers (Integrated DNA Technologies). After sequencing, the recombinant proteins were expressed in ExpiHEK293 cells (Thermo Fisher Scientific) and purified by affinity chromatography, as described previously (10). The purity of each protein was verified by 4–12% SDS-PAGE (NuPAGE, Invitrogen) in the absence and presence of 2-Mercaptoethanol as a reducing agent. Proteins were aliquoted and stored at –80 °C until used. The proprietary monoclonal antibody MBB2 (patent number US2016/0152696A1) was produced and validated as described before (40). Protein concentrations were determined by reading at 280 nm with molar extinction coefficients adjusted according to the amino acid sequence. A value of 49,735 M⁻¹ cm⁻¹ was used for oxβ₂GPI, whereas values of 49,610 M⁻¹ cm⁻¹ were used for β₂GPI C32S/C60S and β₂GPI C288S/C326S. 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS), and bovine heart cardiolipin sodium salt (CL) were from Avanti. All other chemicals were purchased from Sigma-Aldrich.

Deglycosylation experiments

oxβ₂GPI and variants (80 μl, 0.5 mg/ml) were mixed with 8 μl of deglycosylation buffer II (New England Biolabs) and incubated at 75 °C for 15 min to ensure denaturation of the protein samples. After 5 min at room temperature (20–22 °C), the solution was mixed with 5 μl of enzyme. The reaction was left for 3 h at 37 °C before quenching with 5× SDS-containing loading buffer (30 μl). Samples (40 μl/sample ∼10 μg) at t = 0 and t = 3 h were analyzed side by side on a 4–12% SDS-PAGE (NuPAGE, Invitrogen). The gel was stained with SimplyBlue SafeStain following the manufacturers' protocols.

Fluorescent labeling

oxβ₂GPI and variants (50 μl, 5 μM) in Tris 20 mM, 145 mM NaCl, pH 7.4 were reacted with AlexaFluor 647-C2-maleimide (40 μM) (Thermo Fisher Scientific) in the absence or presence of TCEP (5 mM) for 2 h at room temperature (20–22 °C) in the dark. After 2 h, reactions were quenched with 50 mM DTT in SDS-containing loading buffer and run on a 4–12% SDS-PAGE (NuPAGE, Invitrogen). The gel was imaged on a Typhoon Gel and Blot Imaging Systems (Cytiva).

Size exclusion and heparin chromatography analyses

Immunopurified proteins were concentrated to 2 mg/ml by ultrafiltration using a MWCO of 10 kDa. After centrifugation (10,000g for 2 min), 100 μl (200 μg) of solution was loaded into a Superdex 200 HR 5/10 (Cytiva) at a flow rate of 0.5 ml/min that was equilibrated with Tris 20 mM, 145 mM NaCl at pH 7.4. Fifty microliters (100 μg) of the same 2 mg/ml solution was mixed with 150 μl of MilliQ water and loaded onto a HiTrap Heparin HP (1 ml). The column was eluted with a linear gradient of NaCl (50–500 mM in 15 min) at a flow rate of 0.8 ml/min. Absorbance was monitored at 280 nm using an ÄKTApurifier system (Cytiva).

Small-angle X-ray scattering (SAXS) measurements

SAXS data were collected at beamline 12-ID-B of the Advanced Photon Source at Argonne National Laboratory at concentration of 4 mg/ml in the absence and presence of glycerol and analyzed using the software ATSAS (59). Briefly, data sets of each protein sample and corresponding buffer were loaded into PRIMUS to obtain a baseline subtracted data set (∗.dat). Baseline subtracted data sets were then analyzed using existing built-in applications, specifically radius of gyration (R_g) and pair distribution functions. The range used to calculate R_g was determined automatically by the software and carefully inspected to ensure linearity. The low-resolution envelopes were produced using DAMMIF (q up to 0.3 Å⁻¹) by directly fitting the reciprocal space-scattering profile. Twenty models were generated for every calculation and then aligned and averaged using DAMAVER. Output files were replotted in Datagraph and Pymol for visualization.

Fluorescence spectroscopy

Proteins were solubilized in 20 mM Tris pH 7.4, 145 mM NaCl (TBS) at a concentration of 200 nM. Tween 20 was added at a concentration of 0.015% (v/v) to minimize protein adsorption to the quartz cuvette. Since a small but significant signal variation of the fluorescent signal (10–15%) was detected over time even after addition of Tween 20, probably due to protein absorption to the quartz cuvette, all fluorescence spectra were collected exactly after 1 min from transferring the protein solutions into the quartz cuvette. For the chemical denaturation experiments, protein samples were prepared by mixing TBS-Tween buffer with a stock solution of 7.6 M Gnd-HCl prepared in 20 mM Tris pH 7.4, 145 mM NaCl into 1.5 ml plastic tubes to a final volume of 900 μl, followed by the addition 100 μl of 4 μM protein solution in TBS-T. Samples were incubated for at least 1 h at room temperature to allow the system to reach equilibrium. This was verified by incubating some protein samples overnight. Fluorescence spectra were recorded on a FluoroMax instrument equipped with water bath for temperature control. Protein samples were excited at 280 nm and the fluorescence intensity was recorded between 295 and 450 nm, using 5/10 nm as excitation and emission slits, respectively. The intensity at 350 nm corrected for baseline was then graphed as a function of denaturant concentration and fit to Equation 25 in ref. (60). Data reported in the manuscript represents the average of two independent experiments.

Surface plasmon resonance (SPR) experiments

Liposomes composed of 100% POPC, 80:20% (molar ratio) POPC:CL and 80:20% (molar ratio) POPC:POPS were prepared by extrusion using 100 nm polycarbonate membranes (Avanti) in TBS at a final concentration of 1 mM. They were kept at 4 °C and used within 7 days. Binding affinities for liposomes were measured using an L1 sensor chip in which liposomes were immobilized at 1600 ± 50 RU, as done before (10, 42). Titrations were performed by injecting increasing concentrations (0–5 μM) of β₂GPI and variants in running buffer (20 mM Tris pH 7.4, 150 mM NaCl, 0.1% w/w BSA) at a flow rate of 25 μl/min at 25 °C. Binding affinities for MBB2 were measured using CM7 sensor chip in which MBB2 was immobilized at 6000 RU using NHS/EDC chemistry. Titrations were performed by injecting increasing concentrations (0–20 μM) of β₂GPI and variants in running buffer (20 mM Tris pH 7.4, 145 mM NaCl, 0.01% w/w Tween20) at a flow rate of 25 μl/min at 25 °C. All experiments were carried out using on a BIAcore-S200 instrument (GE-Healthcare). The dissociation constants (K_d) were obtained as a fitting parameter by plotting the value of the response units (RU) at the steady state for each concentration using Origin Pro 2015. Data reported in the manuscript represents the average of three independent experiments.

ELISA assays

One-hundred microliter of a 10 μg/ml of oxβ₂GPI and variants solubilized in 0.1 M sodium bicarbonate pH 9.6 were added to a Nunc MAXISORP plate and incubated overnight at 4 °C. After washing three times with 200 μl/well of 20 mM Tris pH 7.4, 145 mM NaCl, 5 mM CaCl₂, Tween 20 0.02% (TBS-T), wells were blocked with 200 μl of 20 mM Tris pH 7.4, 145 mM NaCl, 5 mM CaCl₂, 1% BSA (TBS-B) for 60 min at room temperature. One-hundred microliters of plasma (1:60 v/v dilution) or MBB2 prepared in TBS-B was next added to each well and incubated for 60 min at room temperature. Plates were then washed three times TBS-T and then 100 μl of 1:10,000 dilution of alkaline phosphatase conjugated anti-human IgG (γ-chain specific) antibody (Sigma Aldrich) was added for 60 min at room temperature. Plates were washed three times with TBS-T and then incubated with 100 μl of 3,3′,5,5′-tetramethylbenzidine (TMB) liquid substrate (Sigma Aldrich). After 30 min, the colorimetric reaction was quenched with 100 μl of TMB-stop solution. The optical density at 405 nm was recorded using a SPARK microplate reader (TECAN). Statistical analysis between the groups was performed using ANOVA in Prism 9.0.

Data availability

All data are contained in the manuscript. Deidentified plasma samples were provided by Prof Vittorio Pengo.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Edited by Joseph Jez