Antiphospholipid/antiprotein antibodies, hemostasis-related autoantibodies, and plasma homocysteine as risk factors for a first early pregnancy loss: a matched case-control study

The incidence of pregnancy loss after implantation is high, estimated at 25% to 40%.¹  Spontaneous abortion is the outcome of 14% to 19% of registered pregnancies,^2,3  and the rate of miscarriage in a large population-based study of Chinese women confirmed to be pregnant for the first time reaches 9%.⁴  Very early pregnancy loss during implantation and maintenance of early pregnancy, just before placentation, is mainly due to chromosomal abnormalities, but there is a dearth of knowledge of the causes for later losses.

The maintenance of early pregnancy is inextricably linked with placental growth and differentiation.⁵  One crucial point is the vascular permeability of the placenta maternal side: Women who are carriers of thrombogenic polymorphisms or antiphospholipid antibodies are at higher risk of late fetal loss because of placental thrombosis.^6,7 

It is not until the beginning of the eighth week that undoubted direct communications of maternal uterine spiral arteries with the placental intervillous space can be recognized, mainly by way of narrow gaps in the cytotrophoblastic shell.⁸  Transvaginal ultrasonography has shown that the arterial signals in the yolk sac circulation disappear, and the umbilicoplacental circulation increases between the beginning of the eighth and of the 10th weeks of gestation, indicating that the placenta replaces the yolk sac as an essential source of blood supply to the embryo at that time.⁹  During these 2 crucial weeks, in such an early placental circulation, limited hypercoagulability in the mother may induce pregnancy termination.

Antiphospholipid antibodies are heterogeneous autoantibodies that recognize various combinations of phospholipids, phospholipid-binding proteins, or both. Their association with a clinical syndrome of hypercoagulability defines the “antiphospholipid syndrome,” whose clinical criteria include manifestations of vascular thrombosis and/or complications of pregnancy.¹⁰  In women with no clinical antecedent of vascular thrombosis or fetal death/premature birth, 3 or more unexplained spontaneous abortions before the 10th week of gestation are currently needed to evoke the diagnosis and initiate specific laboratory investigations, including testing for lupus anticoagulant antibodies and for anticardiolipin antibodies. Other antiphospholipid antibodies are not included in the classical laboratory criteria but are under focus: antiphospholipid antibodies directed against phospholipids other than cardiolipin, like phosphatidylethanolamine, or against phospholipid-binding proteins like β2-glycoprotein I or annexin V,¹¹  leading to study antibodies against phospholipid-free hemostasis-related proteins.¹²  Three glycoproteins are known to play a crucial role in the regulation of the hemostatic system: tissue factor, the cell-surface receptor for activated factor VII during coagulation activation; thrombomodulin, the endothelial cell-surface receptor for thrombin during activation of the protein C antithrombotic system; and tissue-type plasminogen activator (t-PA), the main vascular activator of the fibrinolytic system. The functions of tissue factor and of thrombomodulin highly depend on their phospholipid environment.^13,14  Antitissue-type plasminogen activator antibodies have been described in patients with systemic lupus erythematosus¹⁵  and in some primary antiphospholipid syndromes.¹⁶  One may speculate that antibodies against these glycoproteins can modulate the placental thrombotic risk. Moderate hyperhomocysteinemia, a risk factor for thromboembolism,¹⁷  seems to influence the thrombotic tendency in patients with primary¹⁸  or secondary¹⁹  antiphospholipid syndrome and is involved in pregnancy failure.²⁰ 

The objective of the present study was to determine whether women with a first episode of spontaneous pregnancy loss during the eighth and ninth weeks of gestation were more likely to have moderate levels of some hemostasis-related autoantibodies: traditional antiphospholipid antibodies (lupus anticoagulant and anticardiolipin antibodies), nontraditional antiphospholipid antibodies (antiphosphatidylethanolamine, anti-β2-glycoprotein I, and anti-annexin V antibodies), antihemostasis-related key glycoproteins antibodies (antitissue factor, antithrombomodulin, anti-t-PA), or a moderate hyperhomocysteinemia.

Between September 1996 and September 2001, we carried out a matched case-control study at the university hospital of Nîmes, France, and in clinics belonging to the same administrative obstetrical network, which serves about 1 million inhabitants. This study was a part of our Mediterranean Abnormal Pregnancy Study Program called NOHA⁶  (Nîmes Obstetricians and Haematologists study⁶ ).

We identified 1709 consecutive women, aged 19 to 31 years, hospitalized with a medically confirmed diagnosis of spontaneous abortion between the beginning of the eighth and the end of the ninth weeks of pregnancy (10th and 11th weeks of amenorrhea). Those women with a blighted ovum on ultrasound examination (9%) or with chromosome abnormality in the conceptus karyotype (42%) were excluded, leading to study the potential inclusion of 923 women. These women, known to have a positive pregnancy test, had medical symptoms related to abortion, which had occurred during the 10th and 11th weeks of amenorrhea. Transvaginal ultrasound was performed before management to confirm spontaneous abortion (no heartbeat detection), but the embryonic size was not systematically evaluated (total length and biparietal diameter measured in 811 patients).

As abortion itself durably influences the levels of antiphospholipid antibodies,¹¹  the control patients were consecutive women hospitalized for a provoked abortion performed between the beginning of the eighth and the end of the ninth weeks of pregnancy.

We assessed eligibility and performed clinical examinations on admission to hospital, but inclusion, informed consent, and blood collection took place at least 6 months (6-12 months) after the acute event.

Case and control patients who reported previous spontaneous pregnancy loss, previously known systemic disease, previous venous or arterial thrombosis, or a family history of thromboembolism were not eligible for the study. We excluded participants who were taking vitamin supplements, including supplemental folic acid, or drugs that affect homocysteine metabolism. Case and control patients were systematically tested for polymorphisms of factor V Leiden (Gly1691Ala) and factor II (Gly20210Ala), and positive individuals were not eligible: In putative control patients, the prevalence of the factor V and factor II polymorphism was identical to that we had previously found, in our region, in 1000 healthy individuals (1.6% and 1.9%, respectively). In putative patients, the prevalence of the 2 polymorphisms was significantly higher (4.7% and 3.9%, respectively).

Of 923 patients, 180 were not eligible: 33 had personal or familial history of thrombosis, 10 had a systemic disease (rheumatoid arthritis, 1; chronic hepatitis C virus infection, 3; HIV infection, 4; diabetes mellitus, 2), 48 patients took drugs, 41 were heterozygous for the factor V Leiden polymorphism, 34 for the factor II, 21 had no matched control, and 19 did not adhere to study procedures. Thus, there were 743 case patients. Of control patients, the proportion of women with a systemic disease, who took drugs, with no matched case patient, or who did not adhere to study procedures was similar to that in case patients, but the proportion who had personal or familial history of thrombosis was significantly lower, because of the lower prevalence of factor V and factor II polymorphisms.

Each case patient was matched according to age (years), total number of pregnancies, and time elapsed (months) since abortion had occurred. Table 1 gives the characteristics of case and control patients.

Data collection procedures were identical for case and control patients. The study was approved by the ethics committee of our institution, and informed consent was obtained from all participants.

Samples. All blood samples were obtained by clean venipuncture performed between 8:00 and 10:00 am after overnight fasting. Samples for antiphospholipid antibodies were collected in 1/10 volume of CTAD anticoagulant-antiplatelet mixture (0.109 M trisodium citrate, pH 5.4; theophylline; 3.7 mM adenosine; 0.198 mM dipyridamole; Diatube H; Becton Dickinson, Rungis, France)²¹  after systematic rejection of the first 2 mL of blood. Following a double centrifugation at 4000g for 20 minutes, aliquots of platelet-poor plasma were immediately stored at -80°C until use and secondarily used nonfiltered.

Two hundred healthy individuals (102 women, 98 men; mean age, 45 years; range, 18-82 years) were recruited in a Public Health Centre during a systematic medical check-up. Plasma from each healthy individual and the obtained healthy pooled plasma were divided into aliquots and frozen at -80°C.

Assays. Total plasma homocysteine concentrations were measured by high-performance liquid chromatography (HPLC) with the use of a commercially available reagent kit and analytical cartridge from Bio-Rad Laboratories GmbH (München, Germany). Intra-individual variability is less than 4.5% in young healthy adults. In this assay, as an indication, normal ranges were calculated by using the healthy individual plasmas and a cut-off corresponding to the 99th percentile of the obtained normal values were less than 14.7 μM.

Lupus anticoagulants (LAs) were detected with screening assays and were first identified by mixing studies using the healthy pooled plasma, then confirmed by neutralization procedures according to the recommendations of the Subcommittee on Lupus Anticoagulant/Antiphospholipid Antibody of the Scientific and Standardisation Committee of the International Society on Thrombosis and Haemostasis.^22,23  All normal values had been defined from the statistical analysis (1st-99th percentiles) performed on data obtained from the healthy individual plasmas. In a given sample, a positive LA conclusion was given if screening, identification, and neutralization were all positive. Three commercially available screening assays were used: a lupus anticoagulant-sensitive activated partial thromboplastin time (aPTT; PTT LA; Diagnostica Stago, Asnières, France), a diluted Russell viper venom time (dRVVT; Bioclot LA; Biopool, Umea, Sweden), a tissue thromboplastin inhibition test (TTIT) using a single 1:500 thromboplastin dilution (Neoplastin; CI Plus; Diagnostica Stago). For each prolonged time that was found, the corresponding clotting time of a 1:1 mixture of sample and healthy pooled plasma was performed. Samples with a ratio of mixture clotting time to healthy pool plasma clotting time more than 1.2 for the aPTT, or/and more than 1.15 for the dRVVT or/and more than 1.15 for the TTIT indicated the presence of an inhibitor activity. In such cases, lupuslike anticoagulant was confirmed, in the case of positive aPTT by a commercially available hexagonal (II) phase neutralization assay (Staclot LA; Diagnostica Stago) and, in the case of positive dRVVT, by high-phospholipid concentrations using washed, ionophore-treated platelets. They were prepared as follows: Citrated healthy blood was centrifuged at 150g for 15 minutes at room temperature. The platelet-rich plasma supernatant was centrifuged at 1500g for 10 minutes at room temperature, and the platelets were washed twice by suspension and centrifugation in a buffer consisting of 0.15 M NaCl, 0.02 M Tris (tris(hydroxymethyl)aminomethane), 0.001 M EDTA (ethylenediaminetetraacetic acid), 0.005 M glucose, pH 7.5. The twice-washed platelets were resuspended at a concentration of 8 × 10⁸/mL in the same buffer, but without EDTA (buffer A). Activation was achieved by adding to 2 mL of platelet suspension 1 μL of 5 mM solution of calcium ionophore A23187 in absolute ethanol (giving a final concentration of 2.5 μM) and incubating for 5 minutes at room temperature. Buffer A, instead of platelet suspension, was treated the same way. The ionophore-treated platelets or buffer A was frozen in aliquots at -80°C for future use. Finally, the confirmation study was performed by using 50 μL ionophore-treated platelets or buffer A added to 100 μL plasma and to 100 μL dRVVT reagent: The ionophore-treated platelets had been diluted in buffer A to obtain a dRVVT of 25 to 30 seconds when performed on the healthy pooled plasma. The normal values of the ratio of time in presence of activated platelets to time in presence of buffer obtained with the healthy individual plasmas were 0.92 to 1.23 (1st-99th percentiles). A ratio lower than 0.92 was considered to indicate the presence of a lupus anticoagulant.

Plasma anticardiolipin immunoglobulin G (IgG) and IgM antibodies were measured according to an adaptation of the in-house method that has been previously published by Reber et al.²⁴  In brief, plates (Immunoplate Maxisorp; Nunc, Roskilde, Denmark) were coated with cardiolipin (50 μg/mL in methanol; Sigma, St Louis, MO) or just with methanol (nonspecific binding wells) and incubated overnight at 4°C. Blocking and dilution solutions were phosphate-buffered saline (PBS) with 10% fetal calf serum. Blocking step was performed by incubation for 2 hours at room temperature. Plasma diluted 1:100 was tested with coated and noncoated plates. The anticardiolipin reactivity was assessed by using alkaline phosphatase-conjugated affinity-purified goat antihuman (Fc fragment-specific) IgG or IgM antibodies (Jackson Immunoresearch Lab, West Grove, PA). Results were expressed in δ optical density (OD; the absorbency of noncoated wells was subtracted from that of coated wells). A calibration curve was performed with standards from the Antiphospholipid Standardization Laboratory (University of Louisville, Louisville, KY). The anticardiolipin antibody (aCL) content of samples was determined from a log/log plot of δ OD versus IgG antiphospholipid (GPL) or IgM antiphospholipid (MPL) units. In this assay, normal ranges were calculated by using the healthy individual plasmas, and a cut-off corresponding to the 99th percentile of the obtained normal values were IgG-aCL less than 15.2 GPL units and IgM-aCL less than 12.7 MPL units.

For all the following solid-phase assays, because of the absence of available standardized calibration, absorbencies were measured every 120-second period for 12 minutes, using a Genesis software-controlled iEMS microplate reader (Labsystems OY, Helsinki, Finland) and a dual-wavelength reading type. Concentrations of the alkaline phosphatase substrate had been sufficient to preserve the linearity of OD variations during the development of reactions. δ OD values (the absorbency of the coated wells minus that of noncoated wells) were used for the results. Results of assays were given as the calculated slopes of regressions describing the increase of δ OD values (Staview 5 software; Abacus Concepts, Berkeley, CA) (units, OD × [120 sec]^-1).

IgM and IgG antibodies directed against the zwitterionic phospholipid, phosphatidylethanolamine, were detected using a modified version of the enzyme-linked immunosorbent assay (ELISA) described by Berard et al.²⁵  Plates (Immuno Plate Maxisorp) were coated with egg phosphatidylethanolamine (1.5 μg/well; Sigma) diluted in methanol/chloroform (4:1), or with methanol/chloroform alone (nonspecific binding wells) and incubated overnight at 4°C. Blocking and dilution solutions were PBS with 10% adult bovine plasma. Blocking step was performed by incubation for 1 hour at room temperature. Plasma diluted 1:100 was tested with coated and noncoated wells. The antiphosphatidylethanolamine antibody (aPE) reactivity was assessed by using alkaline phosphatase-conjugated affinity-purified goat antihuman (Fc fragment-specific) IgG or IgM. Normal ranges calculated by using the healthy individual plasma (cut-off, 99th percentile of the obtained healthy values) were antiphosphatidylethanolamine IgG less than 0.415 OD × [120 sec]^-1 and antiphosphatidylethanolamine IgM less than 0.510 OD × [120 sec]^-1.

IgG and IgM antibodies against various phospholipid-dependent cofactors, namely phospholipid-free anti-β2-glycoprotein I antibodies and anti-annexin V antibodies were tested as previously described.¹¹  Human β2-glycoprotein I was isolated from freshly frozen citrated plasma according to Horbach et al,²⁶  and human annexin V was isolated from human placentas according to Romisch and Heimburger.²⁷  Micro ELISA plates (Immuno Plate Maxisorp) were coated with the purified human protein at 10 μg/mL in 0.05 M phosphate buffer at pH 7.50, or with the buffer alone (nonspecific binding wells) overnight at room temperature. Plates were then saturated with PBS containing 10% bovine serum albumin (BSA), and plasma was assayed at a 1:101 dilution in PBS containing 10% BSA and 2% Tween 20. The reactivity was assessed by using alkaline phosphatase-conjugated affinity-purified goat antihuman (Fc fragment specific) IgG or IgM antibodies (Jackson Immunoresearch Lab). δ A₄₀₅ at 20°C values (the absorbency of the coated wells minus that of noncoated wells) were used for the results. In these assays, normal ranges were calculated by using the healthy individual plasma, and cut-off corresponding to the 99th percentile was anti-β2-glycoprotein I antibodies, IgG less than 0.264 OD × [120 sec]^-1 and IgM less than 0.296 OD × [120 sec]^-1; anti-annexin V antibodies, IgG less than 0.348 OD × [120 sec.]^-1 and IgM less than 0.298 OD × [120 sec]^-1.

IgG and IgM antibodies against tissue factor and against thrombomodulin were tested using home-made assays. Both glycoproteins were extracted from human placenta as described,^28-30  followed by an immunoaffinity chromatography step using an antitissue factor antibody (American Diagnostica, Greenwich, CT) or an antithrombomodulin antibody (Diagnostica Stago, Asnières, France). The same procedure as the one described for anti-β2-glycoprotein I antibodies and anti-annexin V antibodies was then applied, the concentration of the purified human protein in the coating buffer always being at 10 μg/mL. Normal ranges were calculated by using the healthy individual plasma, and cut-off corresponding to the 99th percentile was antitissue factor antibodies, IgG less than 0.184 OD × [120 sec]^-1 and IgM less than 0.205 OD × [120 sec]^-1; antithrombomodulin antibodies, IgG less than 0.169 OD × [120 sec]^-1 and IgM less than 0.193 OD × [120 sec]^-1.

IgG and IgM antibodies against t-PA specifically bound to a solid-phase fibrin surface were detected as previously described by Anglés-Cano and Sultan.³¹  Briefly, solid-phase fibrin was prepared in polyvinyl-chloride microtitration plates as published.³²  Purified t-PA (200 IU/mL; Biopool, Umea Sweden) was then bound to the fibrin surfaces (incubation overnight at 4°C). Excess t-PA was removed by intensive washing and the fibrin-t-PA surface was used for testing. Plasma diluted 1/10 in diluent buffer (0.05 M PO₄, pH 7.4, 4% BSA, 0.08 M NaCl) was incubated for 2 hours at 37°C in fibrin-bound t-PA wells. The plates were then washed 3 times with assay buffer (0.05 M PO₄, pH 7.4, 2 mg/mL bovine albumin, 0.08 M NaCl, 0.01% Tween 20, 0.01% Thymerosal), and bound immunoglobulins were detected by using alkaline phosphatase-conjugated affinity-purified goat antihuman IgG or IgM antibodies. δ A₄₀₅ at 20°C values (the absorbency of the t-PA-containing wells minus that of noncontaining wells) were used for the results. This procedure mimics the physiologic fibrin/t-PA interaction and permits the unveiling of epitopes appearing after the conformational change of t-PA when bound to fibrin, thus allowing the detection of autoantibodies against fibrin-induced t-PA neoepitopes.³³  Normal ranges were calculated by using the healthy individual plasmas, and cut-off corresponding to the 99th percentile was IgG less than 0.248 OD × [120 sec]^-1 and IgM less than 0.263 OD × [120 sec]^-1.

Variability control. Because of the absence of standards in ELISAs other than anticardiolipin, day-to-day variability of the various assays had to be controlled. In the predevelopment phase of a given solid-phase assay, we tested each healthy individual plasma in a same run. For a given type of solid-phase assay, we thereafter used the slope value obtained with the healthy plasma giving the highest activity in this assay, and with the healthy pooled plasma, as references. They were run on each plate, leading, for a given type of ELISA, to corresponding “mean slope value of the healthy pooled plasma” and “mean slope value of the healthy highest activity.” We then calculated, for each plate, 2 test-specific, plate-associated variability ratios defined as (plate-associated slope value of the healthy pooled plasma) to (mean slope value of the healthy pooled plasma) and (plate-associated slope value of the healthy highest activity) to (mean slope value of the healthy highest activity). For a given plate, a 2% difference between these 2 variability ratios led us to discard the results and to test again. Finally, individual results obtained in a given plate were systematically divided by the mean value of the 2 corresponding variability ratios.

Each plasma sample was tested in triplicate, and the obtained median values were used for statistical analysis.

First, we examined means and standard deviations of all biomarkers in case and control patients. To examine associations between biomarkers, we computed Spearman correlation coefficients (separately in case and control patients). Then, we categorized each continuous variable into 4 levels, corresponding to percentiles 1 to 80, 81 to 95, 96 to 99, and 100, among control patients. All comparisons of case and control patients were conducted by means of methods for matched pairs: McNemar test and odds ratio for positive lupus anticoagulant, paired Mann-Whitney test for mean values of continuous variables, paired t test for the 95% confidence interval on the mean difference between case and control patients, and conditional logistic regression for odds ratios corresponding to the 3 upper levels of the 4-level categorical variables (percentiles 81-95, 96-99, and 100, compared with percentiles 1-80). The latter models were repeated with each covariate coded as a continuous variable, to obtain a P value for linear trend across the 4 levels. Conditional logistic regression was also used to derive a multivariate model predicting spontaneous miscarriage (ie, case status), from all available autoantibodies and homocysteine levels. On the basis of the final multivariate logistic regression model, we computed for each participant a propensity score for having a spontaneous miscarriage (this score corresponds to each participant's covariate levels, applied to the regression equation, and is expressed on a log-odds scale), and we used the proportion of pairs in which the case patient had a higher score than the control patient as an indication for the model's goodness-of-fit (this is an equivalent of the area under the receiver operating characteristic curve, applied to paired data). All P values less than .05 were considered to be statistically significant.

We enrolled 743 case-control pairs of women, matched for age, total number of pregnancies, and time elapsed since the occurrence of the abortion. No significant difference between case and control patients was found in terms of body mass index, cigarette smoking, and oral contraceptive use (Table 1). The patients and control subjects were similar in terms of educational level, occupation, and social class (data not shown). Women were primary aborters (pregnancy loss during the first pregnancy) or secondary aborters (first pregnancy loss during the second or third pregnancy); after classification of the women accordingly to the presence/absence of a previous successful pregnancy, our final analysis could not detect any significant heterogeneity in the results. No significant difference could be evidenced between women with loss of a first pregnancy and women with 2 successful pregnancies who lost the 3rd one.

Data obtained, in case patients, by transvaginal ultrasonography before abortion management were available in 622 patients (83.7%). A biparietal diameter ranging between the 10th and the 90th percentile for 10 weeks of amenorrhea (11-18 mm) or for 11 weeks of amenorrhea (14-21 mm) was evidenced in 579 patients (77.9%). A total length between 28 and 42 mm, thus, compatible with 10 or 11 weeks of amenorrhea, was found in 556 patients (74.8%).

Measurements of autoantibody levels and of homocystinemia in case patients were not substantially correlated with corresponding measurements in control patients (all Spearman correlation coefficients between -0.09 and +0.05). However, in both case and control patients, we identified variables (anti-β2-glycoprotein I IgM, anti-annexin V IgM, antitissue-type plasminogen activator IgG, antitissue factor IgM and IgG, antithrombomodulin IgM and IgG) that were moderately to highly correlated (Spearman correlation coefficient 0.49-0.95 in case patients, 0.68-0.96 in control patients; Table 2). Patients with antiphospholipid antibody syndromes are usually characterized by a number of related autoantibodies. Because of this, it may be difficult to identify their separate effects.

Mean values of biomarkers were higher in case patients than in control patients, for about half of the variables considered (Table 3). This finding suggests a shift in the whole distribution and not only an increased risk at the upper end of the distribution.

These observations were confirmed by an analysis in which each predictor was split into 4 categories: percentiles 1 to 80, 81 to 95, 96 to 99, and 100 (Tables 4, 5, and 6). Eight variables display a statistically significant gradual increase in risk of miscarriage with an increasing level of the biomarker: among traditional antiphospholipid antibodies, lupus anticoagulant, and anticardiolipin IgM; among nontraditional antiphospholipid antibodies, antiphosphatidylethanolamine IgM, anti-β2-glycoprotein I IgG, and anti-annexin V IgG; among antihemostasis-related key glycoproteins antibodies, anti-t-PA IgG, and antithrombomodulin IgG; and finally homocystinemia.

The multivariate analysis identified 6 independent predictors of first spontaneous miscarriage during the eighth and ninth weeks of pregnancy: a positive lupus anticoagulant activity, and high levels of anticardiolipin IgM, anti-phosphatidylethanolamine IgM, anti-annexin V IgG, anti-t-PA IgG, and homocysteine (Table 7).

Most of lupus anticoagulant activities were positive with the aPTT-based assay and the dRVVT-based assay (control patients, 8 of 10; case patients, 42 of 48). It was, thus, impossible, in our population of women, to evidence a specific risk associated with a positivity for 1 of these 2 types of assay systems. The TTIT-based system was disappointing, being rarely positive (control patients, 1; case patients, 4), always in association with 1 of the 2 other assays. We also looked whether it was possible to find a higher risk depending on the degree of positivity. The prolongation ratio of (mixture clotting time) to (healthy pool plasma clotting time) was used as an indicator, but the values, in control and case patients, are roughly similar, with only, on the right side of the distributions, 3 case patients and no control patients having ratios higher than 2. We, thus, could not evidence any clear gradation of risk associated with an increasing degree of positivity.

For each of the 5 continuous variables, the adjusted odds ratio increases as the variable value increases. The strongest risk factor is anti-t-PA IgG, the 100th percentile being associated with a 20-fold higher risk than percentiles 1 to 80, the 2 intermediate categories being associated with an 8-fold increased risk. There was a clear gradation of increasing risk associated with the 4 categories of homocystinemia. For the 3 other variables, the highest category of values is associated with a 3- to 4-fold increased risk than the lowest one.

A prediction score on the basis of this multivariate model was computed in case and control patients, and the propensity scores in case and control patients are shown in Figure 1. The predicted risk of miscarriage is higher in the case patient than in the control patient in 577 pairs (77.7%). In these 577 case patients, 536 (92.9%) had transvaginal ultrasound data compatible with a 10 to 11 weeks' gestation failure.

Finally, for each of the 6 independent predictors, we studied 20 case patients and 20 control patients 3 months after initial testing, including 10 case patients and control patients with an initial test in a category of values associated with a significant risk. This second test was performed before the end of the 12th month after miscarriage, because our threshold values had been defined for a blood sample taken before this date. In control patients, this second blood sample was not associated with any move to another category of results, except in 2 women with an increased homocysteine level, in a context of severe food restriction for weight loss. Four patients had a significant variation of at least one of the antibodies; however, this variation was associated with an ongoing transient epidemic viral infection, and a new evaluation, 1 month later, gave results in agreement with the initial ones. Two other patients had an enhancement of their homocysteine level that was secondary to the intake of drugs affecting homocysteine metabolism. These are not insubstantial changes. These changes are, however, ones that can be explained by the clinical context, pointing to the necessary careful knowledge and control of pre-analytical conditions for an adequate use of these tests.

Our matched case-control study identified, in women, some risk factors for a first episode of spontaneous abortion at the very early moment when the placenta replaces the yolk sac as an essential source of blood supply to the embryo. These risk factors are acquired ones and are members of a family of markers generally associated with hypercoagulability: 2 commonly detected subgroups of antiphospholipid antibodies (ie, lupus anticoagulant antibodies and anticardiolipin IgM antibodies) plus 2 antiphospholipid antibodies currently not included in the laboratory criteria for the antiphospholipid syndrome, antiphosphatidylethanolamine IgM antibodies and anti-annexin V IgG antibodies, plus antitissue-type plasminogen activator IgG antibodies and high homocysteine levels. The associations were concentration dependent (except for lupus anticoagulant, measured as a dichotomous variable), and the matched odds ratios were moderate to high (between 2.6 and 19.5). These arguments, along with the biologic plausibility of hypercoagulability-associated pregnancy loss, suggest that these associations may be causal and that some biologic risk factors for a first pregnancy loss during and after the 10th week of pregnancy may also be effective for a first pregnancy loss during the eighth and ninth weeks of pregnancy.

A main criticism should be the degree of credibility with which we have studied pregnancies that really failed at 8 and 9 weeks (10-11 weeks' gestation). Miscarriages can come to medical attention days or weeks after the embryo has died. However, we have not studied patients with blighted ovum (ie, with an absence of embryo) died several weeks earlier. Moreover, the available transvaginal ultrasonographic measurements are concordant with a 10 to 11 weeks' gestation failure in roughly 93% of the patients, with an earlier failure in a maximum of 8.9% of them, which cannot explain the differences in the case-control pairs (higher risk in 78% of all case patients). This massive orientation of the case-control pairs toward a higher risk in patients cannot be explained, too, by the fact that ultrasonographic data are unavailable in 16.3% of the patients.

As for all case-control studies, the main limitation of our study is that the causality of the observed associations cannot be established. It is possible that in some instances, the elevation of biomarkers is the consequence, rather that the cause, of spontaneous miscarriage. We attempted to avoid this problem by measuring all biomarkers 6 to 12 months after the event, but formally this possibility cannot be excluded. A prospective study, in which biomarkers would be measured before conception, would resolve this issue but would be logistically difficult to perform. Furthermore, which of the observed elevation in biomarkers represent a causal pathogenic mechanism, and which merely accompany such a mechanism, cannot be determined. It is worthless that the 6 independent predictors are not correlated, whether in case patients or in control patients, so that we have not so much identified a single syndrome as 6 independent risk factors. This identification for instance may reveal different constitutional profiles of predisposition to autoimmunization in patients. Which of the 6 risk factors will prove to represent a clinically important causal mechanism remains to be seen.

There is a substantial overlap between the case patients and control patients, even with regard to those antibodies found statistically significant. This overlap may lead to confusion if these assays are introduced as clinical tests. We propose, in the field of autoantibodies, a new attitude: not to define positive and negative values, but the presence of values in different risk-associated categories. However, on a practical point of view, this proposal implies that clinical laboratories can detect normal values as precisely as abnormal values and can discriminate values close to multiple threshold values, even in the normal range. This proposal also implies that the variability between tests results performed using different finally commercially available kits can be controlled. This control has to be proven, and our results need to be confirmed first in routine conditions.

In the previous NOHA⁴  case-control study,¹¹  performed in women with at least 3 episodes of pregnancy loss from the eighth week of pregnancy, anti-β2-glycoprotein I IgG was an independent risk factor but is not one in the present study. It may be that part of the information given by anti-β2-glycoprotein I IgG is modified by anti-t-PA IgG. During in vivo activation of fibrinolysis, β2-glycoprotein I is cleaved, leading to the formation of molecular forms with a much lower affinity for negatively charged phospholipids³⁴ : It may be that autoimmunization against t-PA modifies the equilibrium between cleaved and native forms of β2-glycoprotein I and thus modifies autoimmunization against β2-glycoprotein I. Anticardiolipin IgM was not an independent risk factor in the NOHA⁴  study but is significant in the present one. The populations under focus are different. The current work is methodologically different, leading us to identify increased risks associated with values of biomarkers thus far considered as normal. Methodologically, our anticardiolipin assay uses fetal calf serum and not adult bovine serum as the source of β2-glycoprotein I, a point that is known to influence the results of the assay.³⁵  However, the fact that IgM anticardiolipin results were significant markers but IgG was not is contradictory to most findings in women with recurrent miscarriages; our findings must not be generalized to the recurrent miscarriage population.

A previous case-control study has suggested that women with a first spontaneous abortion are unlikely to have either lupus anticoagulant or IgG anticardiolipin antibodies.³⁶  This work was not focused on the survival of a very early pregnancy. IgM anticardiolipin antibodies were not studied. The percentage of patients with a positive LA activity was close to ours (5.1% versus 6.5%), but the percentage of positive control patients was significantly more important than ours (3.8% versus 1.3%).

Moderate or high levels of antiphospholipid antibodies and antiphospholipid-binding protein antibodies are generally necessary to be linked to clinical manifestations, and IgM antibodies present at low levels are usually not associated with thrombotic events.³⁷  A causal link may, thus, be questioned. Autoantibodies could be a consequence of the mechanism of pregnancy loss in patients, rather than a cause. T-cell tolerance to self-proteins by peptide presentation is limited³⁸ ; abnormally high concentrations of endogenous annexin V or t-PA-derived self-peptides may induce autoimmunization, for instance in the situation of a local inflammatory process, accompanied by cellular and hemostasis activation.

These positive results may, however, be due to the vulnerability of the early placentation stage. Anticardiolipin IgM, antiphosphatidylethanolamine IgM, anti-annexin V IgG, or anti-t-PA IgG may interfere with the various cells and molecular mechanisms involved in normal local hemostasis.^39-41  Another possible explanation is the induction of a defective placentation.^42,43  t-PA may play a key role in the early stages of placentation⁴⁴ : Anti-t-PA antibodies may limit trophoblast invasiveness; an impaired t-PA reactivity has previously been described in unexplained early recurrent abortion.⁴⁵ 

The current findings concerning homocysteine evidence a significant risk associated with traditionally normal homocysteine plasma concentrations, recalling what we have previously described for venous thromboembolism.¹⁷  The question of whether homocysteine concentrations are causally related to early miscarriage remains debatable. Published data only support hyperhomocysteinemia as a risk factor for recurrent early pregnancy loss,²⁰  but the pathophysiology of this relationship is basically not elucidated. Folate levels, which can modulate homocystinemia, have also been incriminated⁴⁶ : The inability of low-dose folic acid supplements to reduce the risk of miscarriage in Chinese women⁴  argues against the hypothesis that spontaneous miscarriage is caused by a weak disequilibrium of the homocysteine remethylation cycle.

We thank all the study participants, patients, control patients, and healthy volunteers who agreed to join us in this long-distance running adventure. We thank E. Cardi and H. Bres for technical assistance, Margaret Manson for editorial assistance, and Prof M. Ramuz and Prof J. P. Bali for their encouragement. We also thank the numerous current and past obstetricians and gynecologists who agreed to contribute to our Mediterranean Abnormal Pregnancy Study Program: S. Balara, M. P. Le Gac, M. Lévy, E. Ranque, J. Leonard, M. Schimpf, B. Vermeulen, N. Abecassis-Bouenal, A. Castel, C. Dumontier-Da Silva, C. Ferrer, M. C. Hoffer-Pinel, S. Kussel, C. Roure, O. Rousseau, G. Masson, C. Courtieu, P. Rudel, J. L. Ter Schiphorst, J. Vignal, H. Coulondre, R. Delpon de Vaux, D. Dupaigne, B. Durieu, C. Gerbino, G. Masson, G. Rouanet, J. L. Alliez, J. L. Alteirac, G. Bensakoun, E. Bergez, E. Bolzinger, and J. Campillo.

J-C.G., P.M., and M.D. were responsible for study design and organization, data collection, interpretation of data, and writing the report; J-P.D. and P.F-P. designed the study; T-V.P. did the statistical analysis, interpreted data, and wrote the study; I.Q., H.D., and M.H. were responsible for patients, interpreted data, and wrote the report; E.M., J-C.B., and G.L-L. were responsible for collection and organization of plasma and for genetic analysis, homocysteine analysis, and plasma testing; S.R-N. and M-L.T. were responsible for patients.