Evidence for heterogeneity of the obstetric antiphospholipid syndrome: thrombosis can be critical for antiphospholipid-induced pregnancy loss

Authors

  • V. POINDRON,

    1. National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, Strasbourg
    Search for more papers by this author
    • These authors contributed equally to this paper.

  • R. BERAT,

    1. Molecular Imaging and NanoBioTechnology, IECB, CBMN UMR-5248 CNRS Université de Bordeaux-1 ENITAB, Avenue des Facultés, Talence
    Search for more papers by this author
    • These authors contributed equally to this paper.

  • A. M. KNAPP,

    1. National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, Strasbourg
    Search for more papers by this author
  • F. TOTI,

    1. U. 770 INSERM, Hôpital de Bicêtre; Université Paris-Sud-11, Faculté de Médecine, Le Kremlin-Bicêtre
    2. Université de Strasbourg, Faculté de Médecine, Institut d’Hématologie & Immunologie, Strasbourg
    Search for more papers by this author
  • F. ZOBAIRI,

    1. U. 770 INSERM, Hôpital de Bicêtre; Université Paris-Sud-11, Faculté de Médecine, Le Kremlin-Bicêtre
    Search for more papers by this author
  • A. S. KORGANOW,

    1. National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, Strasbourg
    Search for more papers by this author
  • M. P. CHENARD,

    1. Pathology Department, Hôpital de Hautepierre, Hôpitaux Universitaires de Strasbourg, Strasbourg, France
    Search for more papers by this author
  • C. GOUNOU,

    1. Molecular Imaging and NanoBioTechnology, IECB, CBMN UMR-5248 CNRS Université de Bordeaux-1 ENITAB, Avenue des Facultés, Talence
    Search for more papers by this author
  • J. L. PASQUALI,

    1. National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, Strasbourg
    Search for more papers by this author
  • A. BRISSON,

    1. Molecular Imaging and NanoBioTechnology, IECB, CBMN UMR-5248 CNRS Université de Bordeaux-1 ENITAB, Avenue des Facultés, Talence
    Search for more papers by this author
  • T. MARTIN

    1. National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, Strasbourg
    Search for more papers by this author

Vincent Poindron, National Referral Center for Systemic Autoimmune Diseases, Clinical Immunology Department, Nouvel Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Université de Strasbourg and CNRS UPR 9021, 67091 Strasbourg, France.
Tel.: +33 3 69 55 12 70; fax: +33 3 69 55 18 35.
E-mail: vincent.poindron@chru-strasbourg.fr

Abstract

Summary. Background: Antiphospholipid antibodies are associated with thrombosis and repeated pregnancy losses during the antiphospholipid syndrome. Several experimental findings indicate that purified antiphospholipid antibodies are directly responsible for inflammation-induced pregnancy losses, or for disruption of the annexin A5 shield at the trophoblastic interface. We previously showed that passive transfer of CIC15, a monoclonal antiphospholipid antibody binding to cardiolipin and annexin A5 that was isolated from a patient with primary antiphospholipid syndrome, induces fetal resorption in pregnant mice. Objectives: To investigate the mechanisms of CIC15-induced pregnancy loss. Methods/results: We show that CIC15 induces fetal loss through a new mechanism that is probably related to procoagulant activity. The time course is different from those of previously described models, and histologic analysis shows that the placentas are devoid of any sign of inflammation but display some signs of thrombotic events. Despite these differences, the CIC15 and ‘inflammatory’ models share some similarities: lack of FcγRI/III dependency, and the efficacy of heparin in preventing fetal losses. However, this latter observation is here mostly attributable to anticoagulation rather than complement inhibition, because fondaparinux sodium and hirudin show similar efficiency. In vitro, CIC15 enhances cardiolipin-induced thrombin generation. Finally, using a combination of surface-sensitive methods, we show that, although it binds complexes of cardiolipin–annexin A5, CIC15 is not able to disrupt the two-dimensional ordered arrays of annexin A5. Conclusions: This human monoclonal antibody is responsible for pregnancy loss through a new mechanism involving thrombosis. This mechanism adds to the heterogeneity of the obstetric antiphospholipid syndrome.

Introduction

Antiphospholipid antibodies (APLs) are autoantibodies associated with arterial and/or venous thrombosis as well as various complications during pregnancy that characterize the antiphospholipid syndrome (APLS) [1,2]. The heterogeneity of APLs is now well established. First, APLs recognize different anionic phospholipids, such as cardiolipin (CL), phosphatidylserine (PS), and phosphatidylethanolamine. Second, the protein dependency (so-called cofactors) for the phospholipid binding differs between autoantibodies [3–5]. Even in a single patient with APLS, clonal analysis revealed significant diversity of the recognized antigens [6]. Third, the molecular structures of the paratope appear to be highly different from one monoclonal APL to another, even in healthy individuals [6,7]. Fourth, the mechanisms of the pathogenicity of APLs may also be diverse [8–15].

From a clinical perspective, the variety of APL-associated pregnancy disorders, ranging from early pregnancy loss to severe HELLP syndrome, also suggests that different APLs may be responsible for different pathologic events. The mechanisms for APL-associated pregnancy loss include the disruption of a two-dimensional (2D) shield of annexin A5 (Anx5) proteins, which has been proposed to protect the placenta from abnormal coagulation [16], and an inflammatory process. In a model of passive transfer of APLs to pregnant mice that reproduces human pathology [17,18], the effector mechanism involves neutrophil infiltration into the placenta, and complement activation with no evidence of thrombosis despite efficient prophylaxis with heparin [19].

We have previously shown that a human monoclonal APL, CIC15, originating from a patient with primary APLS who experienced multiple early abortions, induced fetal resorption when passively transferred to pregnant mice [20]. CIC15 binds to CL and to Anx5, as well as to CL-Anx5 complexes, the affinity for the latter beeing significantly higher than those for CL or Anx5 alone. Here, we show that CIC15 is responsible for fetal resorption in pregnant mice through a mechanism that is linked neither to inflammation nor to its ability to disrupt an Anx5 2D network. However evidence of thromboses in placental arteries and the protective effects of hirudin and fondaparinux strongly suggest that CIC15-induced pregnancy loss results from fetal ischemia as a consequence of arterial thrombosis.

Patient, materials and methods

Patient and monoclonal APLs

The patient CIC suffering from primary APLS and the production of monoclonal APLs have been previously described [20]. Briefly, the patient’s APL-expressing B cells were sorted by flow cytometry with CL-containing labeled vesicles. After cDNA preparation, each pair of the rearranged variable region genes for the light and heavy chains of the single sorted cells was amplified and cloned into a recombinant baculovirus, allowing the expression of human IgG. Monoclonal APLs were produced by infecting insect cells (sf9) [6]. CIC3 and CIC15 belong to this set of monoclonal APLs, and therefore originate from two different patients’ B cells. After purification of the monoclonal APLs on protein A–sepharose columns, they were tested with a passive transfer model in pregnant mice. CIC15 was shown to be pathogenic, with a fetal resorption rate close to 35%, but CIC3 was not. CIC15 displays three mutations in complementarity determining region 1 (CDR1) of the κ-chain [20]. The germline antibody (GL) that was obtained by reversing these mutations by directed mutagenesis to the germinal configuration is not pathogenic.

Model of monoclonal IgG transfer into pregnant mice

BALB/c females were mated with males previously isolated, and the presence of a vaginal plug defined mating. Monoclonal APLs (CIC3, CIC15, and GL) or normal polyclonal human IgG were injected intravenously (100 μg per injection) in the tails of mice at days 1 and 2 postcoitum, or intraperitoneally (300 μg per injection) either on days 1 and 2 or on days 8 and 12 postcoitum, depending on the experiment. Mice receiving early injections (days 1 and 2) were killed on day 10, and mice receiving late injections (days 8 and 12) were killed on day 15. Fetal resorption was defined as at least 50% reduction of the normal fetus weight and volume on macroscopic examination. Resorption rates were compared by use of Fisher’s test. Mice with FcγRI/RIII knockout on a BALB/c background (Takonic, Hudson, NY, USA) and BALB/c mice were bred in our own facilities. The INSERM ethics committee on animal experimentation approved all of these experiments.

The efficiency of intravenous injections was tested by measuring serum levels of human IgG by ELISA, as previously described [20]. To control the efficiency of intraperitoneal injections, serum levels of human IgG were measured 3 h after injections by use of the same ELISA. Some BALB/c mice were mated and killed on day 5 for histologic and immunofluorescence studies. Gravid uteri were frozen and stored at − 80 °C.

Anticoagulation of CIC15-injected pregnant mice

BALB/c mice, previously infused with the pathogenic CIC15, received anticoagulation treatment with subcutaneous injections of low molecular weight heparin (tinzaparin). Twelve IU once daily was shown to provide efficient anti-factor Xa activity (∼ 0.7 IU mL−1; measured 4 h after the second dose,) and was used for the anticoagulation. Pregnant mice treated intravenously with CIC15 (days 1 and 2 postcoitum) received tinzaparin from day 0 to day 8 of pregnancy, and were killed on day 10. In other experiments, several mice were treated with hirudin (8 μg, subcutaneous injection twice daily) or with fondaparinux sodium (10 μg, subcutaneous injection daily) administered for the same time period as tinzaparin. Doses of anticoagulants were calculated on the basis of the manufacturer’s suggestions for murine studies and from the literature [19]. Anticoagulation was monitored by activated cephalin time (ACT) for hirudin and by anti-FXa activity in tinzaparin-treated and fondaparinux-treated mice. ACT ratios in mice treated with hirudin ranged from 3.5 to 5. Anti-FXa activity was measured with a colorimetric method (anti-FXa STA-Rotachrom; Stago, Asniére/Seine, France); anticoagulation was confirmed for values ranging from 0.6 to 1.2 IU mL−1.

Histology and immunofluorescence assays

Five-day-old and 10-day-old embryos from mice with early injections of CIC15 (days 1 and 2) were fixed in 4% paraformaldehyde and embedded in paraffin. Four-micrometer section samples were stained with hematoxylin and eosin. For indirect immunofluorescence assay, embryos were frozen. Cryostats were fixed with acetone at 4 °C, saturated with a murine monoclonal IgAκ (Pharmingen, San Diego, CA, USA) at 10 μg mL−1 for 30 min at room temperature, and washed with phosphate-buffered saline (PBS)–Tween. Samples were incubated with monoclonal APLs (4 μg mL−1) for 45 min at room temperature, and washed with PBS–Tween. CIC3, a non-pathogenic monoclonal APL, was used as a control. A fluorescein isothiocyanate-coupled (Fab)′2 anti-human IgG (Immunotech, Marseille, France) was used to reveal monoclonal APL fixation (7.5 μg mL−1, 30 min at room temperature).

Prothrombinase activity in suspensions containing CL or PS liposomes in the presence of CIC15 and Anx5

Prothrombinase activity was measured in a liposome suspension composed of 30% CL, 58.3% 1,2-dioleoyl-sn-glycero-3-phosphocholine (PC) and 11.6% cholesterol prepared by the dialysis method and initially used for the sorting of CIC15 B cells [6]. Control liposomes (PS liposomes) consisted of vesicles containing 33% PS and 66% PC (mol/mol) used at 0.7 mm final concentration. The prothrombinase assay was performed at 37 °C with slight modifications of a previously described multiwell assay [20,21]. Briefly, CIC15 or GL was incubated with 7.5 mm CL liposomes for 15 min before 1 h of incubation in the presence of 0.1 μg mL−1 Anx5, a concentration enabling > 80% inhibition of thrombin generation by 0.7 mm PS liposomes. FV (ADF, America Diagnostica Inc, Stanford, CA, USA), FXa (Biogenic Perols, Biogenic SA, Mauguiot, France), FII (Hyphen Biomed, Andresy, France), and CaCl2, at final respective concentrations of 220 pm, 53 pm, 1 μm, and 1 mm, were then added. FII conversion to thrombin was revealed after 15 min by the cleavage of a specific chromogenic substrate (0.3 mm final concentration; PNAPEP, Hyphen Biomed, Andresy, France), using a microplate spectrophotometric reader equipped with kinetics software.

In vitro studies of Anx5 binding and 2D organization on lipid membranes

PC, PS and CL were purchased from Avanti Polar Lipids (Albaster, AL, USA). Fetal bovine serum was from Gibco (Invitrogen, Life technologies SAS, Villebon/Yvette, France). All other chemicals were from Sigma Aldrich, (Saint-Louis, MO, USA). Water was purified with a RiOs system (Millipore, Molsheim, France). Recombinant Anx5 was produced as previously described [22].

The following buffer solutions were prepared: buffer A contained 10 mm Hepes, 150 mm NaCl, and 2 mm NaN3 (pH 7.4), supplemented or not supplemented with either 2 mm CaCl2 or 2 mm EGTA, as indicated in the text; and buffer B contained 10 mm Hepes, 15 mm NaCl, 2 mm CaCl2, and 2 mm NaN3 (pH 7.4). Muscovite mica disks 12 mm in diameter were purchased from Metafix (Montdidier, France). Quartz crystal microbalance with dissipation monitoring (QCM-D) sensor crystals (5 MHz), coated with 50 nm of silicon oxide, were from Q-Sense (Gothenburg, Sweden).

Preparation of small unilamellar lipid vesicles

PC and CL lipids were dissolved in chloroform, mixed in the desired amounts (PC/CL, 8 : 1 molar ratio), dried in a rotary evaporator, resuspended in buffer A at 2.5 mg mL−1 final concentration, and vortexed to give a suspension of multilamellar lipid vesicles (MLVs). To form small unilamellar vesicles (SUVs), the MLV suspension was first homogenized by five cycles of freeze–thawing and vortexing, and then sonicated with a tip-sonicator (Model-250; Branson, Danbury, CT, USA) operated in pulse mode at an 18% duty cycle for 30 min over ice. The sample was centrifuged for 10 min at 16 000 × g in an Eppendorf centrifuge, in order to remove titanium particles. SUVs were stored at 4 °C until use. CL was used for these experiments because CIC15 does not bind to PS [20].

Formation of supported lipid bilayers (SLBs)

SLBs were formed by depositing PC/CL (8 : 1) SUVs, at 0.1 mg mL−1 total lipid concentration in buffer A supplemented with 2 mm CaCl2, over silica-coated quartz crystals or mica disks used in QCM-D and atomic force microscopy (AFM) experiments, respectively [23,24]. For QCM-D experiments, SLBs were formed within the measurement chamber. For AFM experiments, SUVs were deposited over the mica disks outside the AFM cell; the excess of vesicles was removed after 20 min of incubation by exchanging the volume of the chamber 10 times with buffer A containing 2 mm CaCl2.

QCM-D

QCM-D measurements were performed with a Q-Sense D300 system equipped with a QAFC 302 axial flow chamber (Q-Sense), as described in detail elsewhere [23]. In brief, the adsorption of matter at the surface of a sensor crystal induces changes in the resonance frequency, F. The frequency change, ΔF, is linearly related to the adsorbed mass, according to the Sauerbrey equation [25], m = − C × ΔF, with C = 17.7 ng cm−2 Hz−1 [25]. The QCM-D method allows measurement of the dissipation changes associated with deposition of adsorbates. For the sake of simplicity, the dissipation changes are not included in the presented figures.

AFM

AFM experiments were performed in liquid with a Nanoscope IV-Multimode (Veeco, Dourdan, France), equipped with a J-scanner (120 μm). Oxide-sharpened silicon nitride cantilevers with nominal spring constant of 0.06 N m−1 (Digital Instruments, Santa Barbara, CA, USA) were exposed to UV/ozone (BHK, Ontario, CA, USA) for 10 min prior to use [23]. The SLB-covered mica substrates were installed in the contact mode fluid cell equipped with an O-ring, and sample solution or buffer was injected with a syringe. Images were recorded in contact mode at scanning rates of 4–6 Hz and a scanning angle of 0°. Images were flattened.

Results

CIC15 is an IgG human monoclonal anti-CL antibody originating from a patient with primary APLS [6]. We have previously shown that CIC15 binds CL, Anx5 and CL–Anx5 complexes, and is pathogenic during passive transfer to pregnant mice [20].

Differences from the inflammatory model of fetal resorption

The analysis of the time course of the effects of the passive transfer of APLs points to a striking difference between our results and the inflammatory model. In the latter, transfer of pathogenic polyclonal APLs, or of the monoclonal Mab 519, was performed late (on days 8 and 12 of pregnancy), but was still able to induce close to 40% of fetal resorption [17,18]. As shown in Fig. 1, the intraperitoneal injections of CIC15 into pregnant mice did achieve a similar rate of fetal resorption (35%) when transfers were performed at days 1 and 2 of pregnancy, whereas they were completely inefficient when performed at days 8 and 12 of pregnancy (P = 0.007). Similar results were obtained with intravenous injections (not shown).

Figure 1.

 CIC15 induces fetal loss only when injected early during pregnancy. Rates of fetal resorption after intraperitoneal injections of normal human IgG (HuIgG), CIC3, germline antibody (GL) and CIC15 mAb at days 1 and 2 postcoitum; pregnant BALB/c mice were killed at day 10. CIC15 was also injected at days 8 and 12 (designated ‘late CIC15’) in another group of mice, which were killed at day 15 (n = 6–8 mice per group). Horizontal bars indicate mean values; *CIC15 vs. HuIgG, CIC3 and GL, P < 0.01; **CIC15 vs. ‘late CIC15’, P = 0.007.

The absence of polynuclear neutrophil infiltration at the sites of fetal resorption on day 10 of pregnancy after CIC15 treatment is another important discrepancy (Fig. 2). However, we observed some tissue necrosis in the fetuses, and several signs of decidual arterial thrombosis. Considering the time course of the pathogenic effect of CIC15, we searched for evidence of neutrophil infiltration at day 5 of pregnancy, and, again, no inflammation was observed.

Figure 2.

 Conventional uterine histology after CIC15 treatment. Mice (n = 5–8) were treated at days 1 and 2 postcoitum and killed at day 10 if not otherwise stated. Mice were treated with CIC15 (B–F) or CIC3 as control (A). (A) Normal embryo (E) with normal trophoblast (T) and placenta (P), × 40. (B) Necrotic embryo (NE) without embryo or trophoblast neutrophil infiltration, × 40. (C) Advanced necrosis (N) without neutrophil infiltration. (D) Thrombosis (Th) between necrotic embryo and trophoblast. NRC indicates nucleated red cells originating from the embryo, × 100. (E) Necrosis as shown by calcium deposit, × 100. (F) Conventional uterine histology after CIC15 injection and killing at day 5 of pregnancy shows the early stage of necrotic embryo (N) without decidual (D) neutrophil infiltration, × 200.

CIC15 polyreactivity precludes the identification of the antigenic target involved in fetal loss

Our attempt to localize the structure targeted by CIC15 in vivo by the use of direct immunofluorescence remained unsuccessfull, owing to the presence of normal human IgGs in the decidual tissues of mice injected with normal human polyclonal IgGs (Fig. 3A,B); this probably reflects human IgG binding to murine Fcγ receptors. Indirect immunofluorescent assay on tissues from CIC15-treated mice killed at day 5 of pregnancy revealed diffuse tissue staining with strong nuclear fluorescence (Fig. 3E). This is most likely attributable to the polyreactivity of CIC15, which precludes any possibility of identifying a specific target involved in the pathogenesis. Indeed, like its germline counterpart (GL), which was shown to be non-pathogenic [20], CIC15 displays some polyreactivity, including antinuclear activity on Hep2 cells (Fig. 3F) and anti-DNA activity [20].

Figure 3.

 CIC15 polyreactivity precludes the identification of its specific antigenic target in vivo. (A, B) Direct immunofluorescence of uteri of mice injected on days 1 and 2 postcoitum with (A) CIC15 or (B) polyvalent human IgG, × 80. (C–E) Indirect immunofluorescence assay on day 5 gravid uterus. (C) Decidua, hematoxylin and eosin, × 100. (D) Indirect immunofluorescence with CIC3 shows no staining. (E) Indirect immunofluorecence of CIC15 shows diffuse tissue staining with strong nuclear fluorescence. (F) Antinuclear activity of CIC15 on Hep2 cells.

Partial similarities with the inflammatory model of fetal resorption

In the inflammatory model, APL pathogenicity is independent of the receptors FcγRI and FcγRIII [18]. The CIC15 pathogenic effect is also independent of the presence of these receptors. Indeed, BALB/c FcRγ-chain deficient pregnant mice, when injected with CIC15, expressed similar levels of fetal resorption as wild-type animals (fetal resorption rates: BALB/c FcR knockout, 0.323 ± 0.14 [n = 5]; BALB/c, 0.32 ± 0.17). We were unable to evaluate the involvement of C5a receptor (C5aR) in the abortive effect of CIC15, because pregnant C57/Bl6 mice were resistant to CIC15 pathogenicity; C5aR knockout pregnant mice, being on the resistant background, were also unaffected by CIC15 (not shown).

As we observed signs of placental thrombosis in CIC15-injected pregnant mice, we investigated whether heparin, which is efficient in the treatment of pregnant women with APLS [26], was also able to prevent fetal loss after passive transfer of CIC15. Mice were treated subcutaneously with 12 U of tinzaparin daily from postcoital days 0 to 8, and killed at day 10. Mice were anticoagulated, as shown by anti-FXa activity between 0.6 and 1.2 IU mL−1. As shown in Fig. 4A, mice treated with tinzaparin were completely protected from fetal injury induced by CIC15 (P < 0.0001). The inflammatory model suggests that the preventive effect of heparin is not attributable to its anticoagulation properties, but rather to its capacity to inhibit complement activation [19]. To test this hypothesis, we treated mice with two other anticoagulants that have no effect on complement activation [19]. Hirudin is a direct inhibitor of thrombin, whereas fondaparinux is a specific inhibitor of FXa. Figure 4A shows that both inhibitors protected mice from CIC15-induced pregnancy loss. There was no significant difference between the three anticoagulants. These results, together with the histology data showing arterial thromboses, strongly suggest that CIC15-mediated fetal injury is mainly the consequence of a prothombotic effect.

Figure 4.

 Thrombosis is critical for the effect of CIC15. (A) Anticoagulants protect mice from CIC15-induced pregnancy loss. Pregnant mice were treated with CIC15 and CIC3 as a control. Several CIC15-injected mice were also treated with tinzaparin (TINZA), fondaparinux (FDX), or hirudin (HIR) (n = 6 mice per group). *CIC15 vs. CIC15 + tinzaparin, P < 0.0001; *CIC15 vs. CIC15 + fondaparinux, P = 0.005; *CIC15 vs. CIC15 + hirudin, P = 0.007; **no significant difference between the three anticoagulants. (B) A prothrombinase assay was performed with CL-containing or PS-containing liposomes as a source of phospholipid-dependent thrombin generation. To evaluate the effect of CIC15 on annexin A5 (Anx5) anticoagulant properties, Anx5 was added (Anx5+, 0.1 μg mL−1) or not added (Anx5−) 15 min after CIC15 (0–20 μg mL−1 final concentration), and conversion of prothrombin to thrombin was measured at 405 nm with a chromogenic substrate in the presence of FXa, FVa, and CaCl2, after an additional 1 h of incubation (n = 3). Data are represented as mean optical density (OD) min–1 values ± two standard deviations.

To obtain insights into the mechanisms by which CIC15 may induce thrombosis, we investigated its effect on in vitro PL-dependent coagulation reactions. The effect of CIC15 on procoagulant activities mediated by CL was assessed by using CL liposomes in a modified functional prothrombinase assay. Figure 4B shows that CIC15 had no effect on thrombin generation by control liposomes containing PS, even when applied at concentrations able to promote 50% inhibition, as observed with other APLs from the patient (20 μg mL−1) [20]. In contrast, thrombin generation promoted by CL liposomes was enhanced in the presence of CIC15, by up to eight times the basal activity. Interestingly, the ability of Anx5 to inhibit thrombin generation by control liposomes remained stable (80%) and was even enhanced by CIC15 (96% inhibition at 21 μg mL−1) in the presence of CL liposomes (Fig. 4B). GL had no effect in these assays (not shown). It is noteworthy that CIC15 was not able to induce tissue factor expression on neutrophils or on monocytes when injected into pregnant mice, and no tissue factor-expressing microparticles could be detected in the plasma samples from their peripheral blood (data not shown).

CIC15 does not disrupt the 2D ordered networks of Anx5

Anx5 has potent anticoagulant activity in vitro, owing to its high affinity for anionic phospholipids and its capacity to displace coagulation factors from phospholipid surfaces [27]. Rand et al. [16,28] have proposed that APLs induce abortion by disrupting a putative 2D shield composed of Anx5, which covers the surface of syncytiotrophoblasts. As CIC15 binds Anx5, we investigated whether CIC15 was able to disrupt the 2D ordered networks formed by Anx5 [21,28].

Formation of 2D assemblies of Anx5 on CL-containing SLBs, as determined by QCM-D and AFM.  The influence of CIC15 on the binding and 2D organization of Anx5 on SLBs was investigated by means of QCM-D and AFM. First, we verified that Anx5 forms 2D ordered networks on CL-containing SLBs, as previously established on PS-containing SLBs (Fig. 5) [22,29,30]. Figure 5B shows the results of a typical QCM-D experiment: the formation of an SLB by deposition of PC/CL (8 : 1) SUVs on a SiO2-coated sensor (phase I), the Ca2+-dependent adsorption of Anx5 to the SLB surface (phase II), and the release of Anx5 induced by exposure to EGTA, a Ca2+-chelating agent (phase III). Figure 5C–F shows typical AFM images of a PC/CL SLB on mica (Fig. 5C), 2D self-assembled domains of Anx5 formed over an SLB before complete coverage of the surface (Fig. 5D), and enlarged views of an Anx5 domain revealing its 2D crystalline structure (Fig. 5E,F). The observed structures are identical to those obtained on PS-containing SLBs [22,29], establishing that the adsorption and 2D self-organization behavior of Anx5 is the same on CL-containing and PS-containing SLBs.

Figure 5.

 Adsorption and two-dimensional (2D) organization of annexin A5 (Anx5) on 1,2-dioleoyl-sn-glycero-3-phosphocholine (PC)/cardiolipin (CL) supported lipid bilayers (SLBs). (A) Scheme of the 2D self-assembly of Anx5 on a negatively charged membrane in the presence of calcium ions. Soluble monomeric Anx5 molecules bind to the membrane surface, form trimers, and then form 2D crystals of trimers assembled with p6 symmetry. (B) Adsorption of Anx5 on a PC/CL (8 : 1) SLB, as determined by quartz crystal microbalance with dissipation monitoring. Variation of the adsorbed mass (ng cm−2) vs. time: phase I corresponds to the formation of an SLB by deposition of 100 μg mL−1 PC/CL small unilamellar vesicles in buffer A supplemented with 2 mm CaCl2 (arrow 1); phase II corresponds to the adsorption of Anx5, after injection (arrow 2) of 20 μg mL−1 Anx5 in buffer A supplemented with 2 mm CaCl2; Anx5 binding is saturated at about 280 ng cm−2 and is quasi-irreversible, as indicated by the absence of Anx5 displacement upon rinsing with a protein-free buffer (arrowhead). Phase III represents the immediate and complete release of Anx5 triggered by rinsing with an EGTA-containing buffer (arrow 3). (C) Atomic force microscopy (AFM) image of a PC/CL SLB formed on mica. A color-coded height table is shown in the inset (this table is valid for all AFM images presented here). (D–F) AFM images of 2D crystalline domains of Anx5 formed on a PC/CL SLB. (D) Image recorded before saturation of the surface. (E, F) images revealing the 2D crystalline structure of Anx5 domains, together with the presence of holes. (F) Scheme of the a 2D crystal of membrane-bound Anx5, revealing the presence of large open spaces around the six-fold symmetry centers (three six-fold symmetry centers are indicated by red hexagons). These open spaces, ∼ 9 nm in diameter, either stay empty, in which case holes are detected in AFM images, or accommodate additional trimers; two Anx5 trimers (colored in blue) have been incorporated into the 2D crystal, with arbitrary orientations. A hexagonal unit cell is indicated, with 17.7-nm unit cell side.

A special comment concerns the presence of dark structures, referred to hereafter as holes, within the Anx5 2D crystalline domains (Fig. 5E,F). The Anx5 crystals are composed of Anx5 trimers arranged at the vertices of hexagons with p6 symmetry, as shown in Fig. 5G [28]. The central area of each hexagon is large enough to accommodate another Anx5 trimer, the presence or absence of which depends only on local conditions (for more details, see [29,31]. In conclusion, the holes observed on AFM images correspond to positions where the central trimers are absent. They do not correspond to crystalline defectsm and therefore they cannot be interpreted as perturbations of the crystalline organization.

Interaction between CIC15, Anx5, and SLBs.  We first studied the interaction between CIC15 and PC/CL SLBs in the absence of Anx5. As shown in Fig. 6A, CIC15 bound to CL-containing SLBs. In saturating conditions, an adsorbed mass of 177 ± 20 ng cm−2 (n = 2) of CIC15 covered the lipid surface. The influence of the lipid and ionic compositions on CIC15 binding was investigated. We found that the presence of negatively charged phospholipids was required for CIC15 binding, in agreement with previous ELISA results [20], and that CIC15 binding was Ca2+-independent (data not shown). Surprisingly, we also found that CIC15 bound more strongly and more stably at low ionic strength, with saturation at 530 ng cm−2 in buffer B containing 15 mm NaCl (Fig. 6B). Therefore, for the follow-up of this study, the behavior of CIC15 was investigated at both high (150 mm, buffer A) and low (15 mm, buffer B) NaCl concentrations. AFM images of SLB-bound CIC15 monolayers were extremely noisy, revealing the presence of some material deposited on the SLB surface (Fig. 6C). The lack of molecular detail in such AFM images is explained by the high flexibility of IgG molecules.

Figure 6.

 Formation of complexes between CIC15, 1,2-dioleoyl-sn-glycero-3-phosphocholine (PC)/cardiolipin (CL) supported lipid bilayers (SLBs) and annexin A5 (Anx5). (A, B) Quartz crystal microbalance with dissipation monitoring (QCM-D) responses of the adsorption of CIC15 to PC/CL (8 : 1) SLBs; 100 μg mL−1 CIC15 was injected (arrow 1) at 150 mm NaCl in buffer A (A) or 15 mm NaCl in buffer B (B), respectively. The phase of SLB formation, which precedes the addition of CIC15, is not represented here. Arrowheads indicate rinsing with protein-free buffer solutions. (C) Atomic force microscopy (AFM) image of a monolayer of CIC15 covering a PC/CL (8 : 1) SLB at 15 mm NaCl. The homogeneous white signal indicates that the whole SLB surface is covered by a disordered monolayer of CIC15, the thickness of which is significantly higher than that of an Anx5 monolayer. (D–G) Adsorption and organization of CIC15 on a two-dimensional (2D) monolayer of membrane-bound Anx5. (D, E) QCM-D responses of the adsorption of CIC15 to a saturating layer of Anx5, at 150 mm NaCl in buffer A (D) and at 15 mm NaCl in buffer B (E). In (D), arrows 1–3 indicate (1) injection of 20 μg mL−1 Anx5 in buffer A supplemented with 2 mm CaCl2, (2) injection of 100 μg mL−1 CIC15 in the same buffer, and (3) rinsing with an EGTA-containing buffer. Arrowheads indicate rinsing with protein-free buffer solutions. In (E), arrows 1–5 indicate (1) injection of 20 μg mL−1 Anx5 in buffer A supplemented with 2 mm CaCl2, (2) exchange of buffer A (150 mm NaCl) with buffer B (15 mm NaCl) supplemented with 2 mm CaCl2, (3) injection of 100 μg mL−1 CIC15 in the same buffer, (4) exchange of buffer B with buffer A supplemented with 2 mm CaCl2, and (5) rinsing with an EGTA-containing buffer. (F, G) AFM images of CIC15 molecules bound to 2D crystalline domains of Anx5. CIC15 molecules appear as white dots in AFM images, because they are protruding above the Anx5 layer. The concentration of CIC15 used in this AFM experiment was undersaturating, in order to reveal both the presence of adsorbed CIC15 molecules and the preserved crystallinity of the underlying Anx5 layer.

The formation of ternary complexes between CIC15, Anx5 and PC/CL SLBs was then investigated, in two symmetric conditions: (i) interaction of CIC15 with a preformed Anx5 monolayer; and (ii) interaction of Anx5 with a preformed CIC15 monolayer. We found that CIC15 did not bind to a saturating monolayer of Anx5 at 150 mm NaCl (Fig. 6D). At 15 mm NaCl, the addition of CIC15 to a saturating Anx5 monolayer induced a mass increase of about 220 ng cm−2 (Fig. 6E). Upon rinsing with 150 mm NaCl–buffer A, the mass of the adsorbed layer decreased by 200 ng cm−2, stabilizing at 250 ng cm−2. This, together with the fact that rinsing with buffer A supplemented with 2 mm EGTA induced the release of about 250 ng cm−2, demonstrates that adsorption of CIC15 had not displaced Anx5, or only to a negligible extent. These results indicate, in addition, that a large amount of CIC15 (∼ 220 ng cm−2) can coexist with a saturating monolayer of Anx5. On AFM images, the 2D crystalline domains of Anx5 were conserved after addition of CIC15 (Fig. 6F), but their entire surface was covered with white dots that corresponded to stably bound CIC15 molecules (Fig. 6F,G). At high magnification (Fig. 6G), the antibodies were clearly seen protruding on top of the Anx5 crystals, but the crystalline organization was not disturbed at all by the presence of CIC15.

When Anx5 was added to a preformed CIC15 monolayer at 150 mm NaCl, Anx5 was able to entirely displace CIC15 and to form 2D crystals on the SLB surface (data not shown). At 15 mm NaCl, the addition of Anx5 to a saturating layer of SLB-bound CIC15 resulted in the partial displacement of CIC15 (data not shown). The system at equilibrium consisted of a mixture of 280 ng cm−2 CIC15 and 250 ng cm−2 Anx5, values that are close to those obtained when CIC15 was added to a preformed monolayer of Anx5. AFM images were also very similar to those presented in Fig. 6F,G, showing CIC15 antibodies associated with the Anx5 crystalline domains, but no hint of disruption of the crystalline structure.

Finally, as β2-glycoprotein-I has been proposed to act as a cofactor of APL in APLS [14,16], we investigated whether the addition of β2-glycoprotein-I as well as of components present in blood serum or plasma influenced the CIC15–Anx5–SLB interaction. The addition of β2-glycoprotein-I, serum or plasma samples had no effect on the CIC15–Anx5–SLB interaction (data not shown).

This ensemble of results indicates that: (i) CIC15 and Anx5 can coexist on negatively charged lipid membranes; (ii) the order in which CIC15 and Anx5 are added has no influence on the end result; and (iii) CIC15 is able to integrate into 2D crystalline domains of Anx5, without inducing any type of disorder detectable by AFM.

Discussion

This study focused on CIC15, a human monoclonal APL that induces abortions when passively transferred to pregnant mice. We have shown that CIC15 acts via a previously unrecognized mechanism of pregnancy loss during APLS.

Mice provide a relevant and widely used model for the study of human APL obstetric pathogenicity. Indeed, human and murine placentas share many features that are most probably relevant to APL-induced fetal loss [32,33]. Both humans and mice have hemochorial placentas. In both species, the uterine mucosa is transformed into a specialized tissue known as decidua, and Anx5-binding sites are present on the surfaces of trophoblasts.

The pathogenic effect of CIC15 is different from that reported by Salmon et al. with polyclonal APLs. In this ‘inflammatory’ model, APLs bind to an unknown target, activate complement, and induce FcγRI–III-independent inflammation. We found that CIC15 effects are also independent of the presence of the proinflammatory FcγRI–III, but the involvement of the complement could not be directly assessed, because complement-deficient mice with a BALB/c background are not available. However, no inflammatory cell infiltration could be detected in the decidual or fetal tissues.

Another unexpected level of complexity is exemplified by the sensitivity of CIC15 pathogenic effects to the genetic background (C57/B16 are resistant to CIC15 pathogenic effects). Our results are reminiscent of the genetic influences on the end-stage effector phase of experimental arthritis in the K/BxN model [34]. The distinction between BALB/c (responder) and C57/Bl6 (resistant) is interesting, and could provide insights into the mechanisms of CIC15-induced abortion. However, exploring the genetic heterogeneity in the response to CIC15 is difficult: BALB/c and C57/Bl6 are genetically distant strains, and the pathogenic effect of CIC15 in BALB/c mice does not have very high penetrance.

It is generally assumed that polyreactivity of autoantibodies is a hallmark of natural, unmutated and innocent antibodies. However, some exceptions challenge this idea: hemolytic anemia [35,36], lupus glomerulonephritis [37] and arthritis [38] have been shown to be linked to some pathogenic polyreactive autoantibodies. CIC15 variable regions contain three mutations clustered in the Vκ CDR1. These mutations were shown to be responsible for Anx5 binding, Anx5-enhanced anti-CL reactivity, and pathogenicity [20]. However, these mutations did not abolish the polyreactivity of the germline counterpart of CIC15. Unfortunately, this polyreactivity of CIC15 precludes our attempts to identify the target self-antigen(s) linked to the pathogenicity. Despite this caveat, our data strongly support the idea that CIC15 induces fetal ischemia secondary to thrombosis. First, CIC15 has a procoagulant effect in vitro when assayed with liposomes containing CL. Second, the abortive effect of CIC15 can be prevented by heparin, but also by fondaparinux sodium and hirudin, which have no effect on complement activation. Third, signs of placental thrombosis were present.

Nevertheless, the mechanism by which CIC15 induces placental thrombosis remains elusive. Our in vitro data support the hypothesis that CIC15 may favor the lateral organization of functional CL domains with high net anionic charge, which in turn could favor local coagulation. On the other hand, CIC15 is unable to inhibit the anticoagulant properties of Anx5 (Fig. 4B) or to disrupt an Anx5 shield. We used SLB membrane models and two complementary techniques, QCM-D and AFM, to study the interaction between CIC15, CL-containing SLBs and Anx5 at the molecular level. Our results confirm and extend those from previous ELISA experiments, showing that CIC15 recognizes both CL and Anx5 [20]. Higher affinity of CIC15 for Anx5–CL–SLB complexes is found at low NaCl concentrations; the possible physiologic relevance of this observation is as yet unclear. The major finding of this in vitro analysis is that CIC15 molecules coexist with, but do not disturb, the 2D crystalline organization of Anx5. The behavior of CIC15 thus differs significantly from that of other APLs and from that of anti-Anx5 autoantibodies, whose prothrombotic activity has been proposed to result from their property of disrupting the Anx5 2D ordered organization [16,28,39]. In fact, the role of Anx5 in pregnancy remains unclear, as Anx5-deficient mutant mice are normally fertile [40]. It is tempting to think that binding to Anx5 is important for CIC15 pathogenesis, because GL does not induce abortions. However, the mutations in CIC15 also increase its affinity for CL [20], which could be the clue to pathogenicity. If so, the question is how CIC15 gains access to CL in vivo. CL lipids, like PS, are essentially located intracellularly, CL being mainly present in mitochondrial membranes. However, it was recently shown that CL is present in the plasma as a normal component of lipoproteins [41]. Probably more relevant to the obstetric manifestations of APLS, it has been proposed that PS becomes exposed on the extracellular side during processes of plasma membrane rupture, which occur frequently in many cell types under mechanical stress [42]. There are numerous cases of membrane rupture in syncytiotrophoblasts, related to maternal flux into the intervillous space. In addition, throughout normal pregnancy, there is deportation of syncytial fragments into the maternal circulation [43]. We therefore propose that CL may become exposed to the extracellular medium during membrane rupture events that occur in the placenta.

We and others have previously shown that APLs are extremely heterogeneous in terms of molecular structure, recognized antigens (phospholipids and cofactors), and capacity to induce pathology. We have also shown that pathogenic APLs may represent a minority of this large and complex population. The data presented herein increase the level of complexity by showing that APLs may also be heterogeneous in terms of the mechanisms of pathogenicity, even in one aspect of APLS, namely pregnancy loss. The elucidation of the different mechanisms could lead to different treatments if we are able to detect ‘the needles in the haystack’.

Addendum

Study conception and design: A. Brisson (QCM and AFM studies), T. Martin, J. L. Pasquali; acquisition of data: R. Berat, M. P. Chenard (histopathology), C. Gounou, A. M. Knapp, V. Poindron, and F. Zobairi (prothrombinase assays); analysis and interpretation of data: A. Brisson, A. S. Korganow, T. Martin, J. L. Pasquali, and F. Toti (coagulation assays).

Disclosure of Conflict of Interests

The authors state that they have no conflict of interest.

Ancillary