Extensive serum biomarker analysis in the prethrombotic state of recurrent spontaneous abortion

Abstract The prethrombotic state (PTS) is a possible cause of recurrent spontaneous abortion (RSA). The aim of this study was to identify serum biomarkers for the detection of RSA with PTS (PSRSA). A Quantibody array 440 was used to screen novel serum‐based biomarkers for PSRSA/NRSA (RSA without PTS). Proteins differentially expressed in PSRSA were analysed using bioinformatics methods and subjected to a customized array and enzyme‐linked immunosorbent assay (ELISA) validation. We used receiver operating characteristic to calculate diagnostic accuracy, and machine learning methods to establish a biomarker model for evaluation of the identified targets. 20 targets were selected for validation using a customized array, and seven targets via ELISA. The decision tree model showed that IL‐24 was the first node and eotaxin‐3 was the second node distinguishing the PSRSA and NRSA groups (an accuracy rate of 100% and an AUC of 1). Epidermal growth factor (EGF) as the node distinguished the PSRSA and NC groups (an accuracy rate of 100% and an AUC of 1). EGF as the node distinguished the NRSA and NC groups (an accuracy rate of 96.5% and an AUC of 0.998). Serum DNAM‐1, BAFF, CNTF, LAG‐3, IL‐24, Eotaxin‐3 and EGF represent a panel of promising diagnostic biomarkers to detect the PSRSA.

thrombophilia, prothrombin 20210A mutation, elevated factor VIII level, and mutations of the gene encoding the enzyme methylenetetrahydrofolate reductase. Acquired PTS is comprised mainly by antiphospholipid antibody syndrome (APS) and various diseases or conditions, such as tumours and long-term immobilization that cause blood hypercoagulability. The general prevalence of PTS in the general population is 3% ~ 8%, and in a study of women with recurrent spontaneous abortion, after excluding other causes, the prevalence of the PTS reached 78%. 7 Hereditary PTS factors affect 3% to 11% of the population. 8 Mitic et al 9 reported that inherited thrombophilia was found in 36% of the study subjects. APS is an acquired autoimmune thrombotic disease, and RSA is one of its clinical classification criteria. 10 Therefore, it is particularly important to study the atypical relationship between PTS and early spontaneous pregnancy loss.
According to the current PTS clinical testing protocols, RSA patients were divided into a PTS group and a non-PTS group.
Anticoagulant therapy was performed on the PTS group. Some patients had significant therapeutic effects, while some patients had insignificant therapeutic effects. It is necessary to find more accurate diagnostic markers to diagnose RSA patients with PTS in order to improve treatment.
Clinically, multi-factor detection using protein chip technology can be used to screen and identify a panel of specific RSA protein markers and, combined with PTS detection technology, provides an important theoretical basis for the screening of RSA caused by PTS.
Previously, Wu et al 11

| Subjects
One sixty-five subjects were recruited from the Department of Traditional Chinese Medicine, Beijing Obstetrics and Gynecology Hospital, Capital Medical University (Beijing, China). All subjects signed informed consent forms before participating in this study.
This study was approved by the Ethics Committee of the Beijing Obstetrics and Gynecology Hospital, Capital Medical University (approval number: 2016-KY-001). None of the subjects in this study (<40 years of age) were in a gestational state. All samples were collected during the subjects' menstrual cycles. Studies have shown that the menstrual cycle will not affect the results of these experiments. 12 14 Currently, commonly used molecular markers include antithrombin III activity (AT-Ⅲ), protein C activity (PC), protein S activity assay (PS), plasminogen activity (PLG), tissue-type plasminogen activator (t-PA), plasminogen activator inhibitor (PAI-1), fibrinogen degradation product (FDP), CD3-CD16 + CD56 + and homocysteine (HCY).
All RSA patients met the diagnostic criteria for RSA and had pregnancy losses prior to 12 weeks. The RSA patients who had one or more atypical PTS diagnostic markers were deemed PSRSA. The RSA patients who did not have one or more atypical PTS diagnostic markers were deemed NRSA. Women with normal pregnant history were deemed NC as the healthy controls, and they did not meet the diagnostic criteria for PTS.
According to the diagnostic criteria for RSA and PTS, the subjects were divided into 58 PSRSA, 55 NRSA and 52 NC subjects.

| Quantibody array
Twenty four PSRSA, 12 NRSA and 12 NC subjects were studied in the biomarker discovery stage. The characteristics, including number of subjects, age of diagnosis, body mass index (BMI) and time of spontaneous abortion, are shown in Table 1. The differences between groups were analysed by Welch's t test, and P values <.05 were considered statistically significant. The Quantibody array QAH-CAA-440 was used to measure the serum protein levels, according to the manufacturer's instructions. It is an array-based multiple ELISA system, which can simultaneously and quantitatively detect the expression levels of 440 proteins and was employed as described previously. 11  Abbreviations: AT-III, antithrombin-Ⅲ ; BMI, body mass index; FDPs, fibrinogen degradation products; HCY, homocysteine; PAI-1, plasminogen activator inhibitor 1; PC, protein C; PLG, plasminogen; PS, protein S; t-PA, tissue plasminogen activator.

| Bioinformatics analysis
All array data analyses were performed using RayBio Analysis Tool software (Q-Analyzer Software for QAH-CAA-440) (https://www. raybi otech.com/produ cts/other -produ cts/softw are/). To elucidate the potential functions of the differentially expressed proteins obtained from the Quantibody array, gene ontology (GO) and pathway analysis were carried out to describe the biological processes, cellular components and molecular functions of these proteins by inputting the gene IDs of the differential proteins into the KOBAS3.0 database (http:// kobas.cbi.pku.edu.cn/index.php). Protein-protein interaction (PPI) analysis was performed using the STRING database (https://strin g-db. org/cgi/input.pl) and the protein IDs to identify node proteins.

| Statistical analysis
Data were presented as means ± SD (standard deviations).
Differences between groups were determined by one-way ANOVA, method was used to assess sensitivity and specificity of potential biomarkers. A decision tree model was built using the R language machine software package C5.0 algorithm. 15

| Study population characteristics
We recruited a total of 165 subjects, 48 subjects in the biomarker discovery stage and all 165 subjects in the biomarker validation stage.  Tables 1 and 2. Table 1 shows that there were no significant differences in the indicators between PSRSA and NRSA. The AT-III and PS indicators were significantly different (P < .05) between PSRSA and NC. The AT-III, PS, and t-PA indicators were significantly different (P < .05) between NRSA and NC. As the sample size increased, the differences in indicators between PSRSA and NRSA appeared. Table 2 shows that the AT-III, PS and HCY indicators had significant differences (P < .05) between PSRSA and NRSA.
The AT-III, PS, t-PA and HCY were significantly different (P < .05) between PSRSA and NC. The AT-III and PS indicators showed significant differences (P < .05) between NRSA and NC. Combining Tables 1 and 2, AT-III and PS were able to distinguish the PSRSA, NRSA and NC groups.
differentially expressed proteins were analysed statistically using one-way ANOVA, followed by multiple comparisons performed with post hoc Bonferroni test between any pair of the PSRSA, NRSA and NC groups. As a result, 18 proteins were found to be differentially expressed between the PSRSA and NRSA groups (Table 3), 20 between the PSRSA and NC groups, and 21 between the NRSA and NC groups (detailed data in Tables S1-S3). Therefore, after Venn diagram analysis, we identified 12 specific PSRSA associated biomarkers ( Figure 1A).
This analysis was confirmed using principal component analy-  Figure 1C, Figures S1B and S2B. The volcano plot shows 18 proteins with differential expression based on an adjusted P value (adj. P. Val) of less than .05. Of these, 11 proteins were up-regulated, and seven proteins were down-regulated ( Figure 1C). To determine whether these DEPs could discriminate patients with PSRSA from the NRSA or NC groups, we performed the heatmap of hierarchical clustering analysis ( Figure 1D, Figures   S1C and S2C). The result showed that most of the PSRSA samples could be separated from the NRSA group to form two major groups, and the two clusters isolated through the different expression of proteins ( Figure 1D).

| Expression characteristics of PSRSA related proteins
There were 18 proteins significantly differentially expressed in PSRSA patients as shown in Table 3 (detailed data of 440 factors   are shown in Table S1-

| CQ20 validation
Using Venn graph intersection analysis to identify PSRSA and NRSA or  (Table 4). 58 PSRSA, 34 NRSA and 25 NC cases were recruited to verify the 20 proteins. Table 4

| Validation array results with ELISA
According to the primary screening and the customized array results, 7 proteins of interest were selected for further validation using ELISA on samples from 165 subjects. As shown in Figure 6, the levels of DNAM-1, IL-24 and CNTF were increased in PSRSA patients, compared with the NRSA and NC groups. The levels of Eotaxin-3, BAFF, LAG-3 and EGF were decreased in the PSRSA group, compared with the NRSA and NC groups. These results were consistent with the array results.

| Sensitive and specific analysis of several biomarkers
We used one-way ANOVA, followed by multiple comparisons performed with post hoc Bonferroni test, to analyse the differences  Figure 7B). Using EGF as the node to distinguish PSRSA from the NC groups, the model accuracy rate was 100% (sensitivity 100%, specificity 100%), the Kappa coefficient was 1, the positive predictive value and negative predictive value were 1, and the AUC was 1 ( Figure 7C). Interestingly, when using EGF as the node to distinguish the NRSA and NC groups, the model accuracy rate was 96.5% (sensitivity 100%, specificity 93.94%), the Kappa coefficient was 0.9304, the positive predictive value was 0.9259, the negative predictive value was 1, and the AUC was 0.998 ( Figure 7D), indicating that the EGF indicator distinguished the two groups with the greatest effect.

| D ISCUSS I ON
As one of the most prevalent obstetric complications, recurrent spontaneous abortion affects more than 30% of pregnancies. 16 Although RSA has been the focus of the majority of research, in most cases the cause of RSA is not clear. Recently, the relationship between RSA and the prethrombotic state has become a focus of attention. 17 The prethrombotic state causes decidual vascular fibrinoid necrosis and villus infarction, which affects the material exchange between the foetus and mother, increasing the risk of pregnancy loss. 18

| Epidermal growth factor
Epidermal growth factor is a multifunctional growth factor. Many studies have shown that the survival and invasion capabilities of human trophoblast cells are related to the intercellular signalling of EGF-related peptides. 20 EGF can reduce trophoblast apoptosis induced by exposure to oxidative stress in an in vitro experiment. 21 Yumusak et al 22

| DNAX accessory molecule-1
DNAX accessory molecule-1 (DNAM-1) is also known as platelet / T cell activation antigen 1 (PTS1), 23 also designated as CD226. 24 DNAM-1 is involved in the differentiation of cytotoxic T lymphocytes and anomalous killer cells, 25 is expressed on platelets and is involved in platelet activation and aggregation. 23 It is also involved in intercellular injection and lymphocyte signal transfer and is widely expressed on T cells, NK cells, monocytes and B cells, among other types of leukocytes. 26

| Interleukin-24
Interleukin-24 (IL-24) is a secreted cytokine that belongs to the IL-10 cytokine family. It is also known as melanoma differentiation associ-

| Ciliary neurotrophic factor
Ciliary neurotrophic factor (CNTF) belongs to the IL-6 cytokine family 32 and is involved in a variety of processes from endogenous neuroprotection to energy expenditure regulation in the body. 33 Watanobe et al 34  In our study, there was a significant increase in CNTF expression in PSRSA patients and a decrease in CNTF expression in NRSA patients.
The exact molecular mechanism for these controversial results remains unclear.

| Eotaxin-3
Shinkai et al 36  has been confirmed to enhance EVT function and is closely related to thrombosis. 38 Falcone et al 39 showed that lower levels of serum eotaxin-3 could be used as coronary heart artery independent predictors of future cardiovascular adverse events in patients. In the current study, we found that the serum eotaxin-3 levels in PSRSA patients were significantly decreased compared with the NRSA and NC groups, consistent with previous reports. We hypothesize that low levels of eotaxin-3 cause weakened EVT function and potential cardiovascular problems, which may cause recurrent spontaneous abortion in PSRSA patients.

| B-cell activating factor
B-cell activating factor (BAFF), also known as TALL-1, is an important growth factor for B cells. 40

| CON CLUS IONS
In summary, DNAM-1, IL-24, CNTF, LAG-3, BAFF, Eotaxin-3 and EGF were identified as differentially expressed in PSRSA patients compared with the NRSA and NC groups. The present study had limitations, as the sample size was relatively small. A multi-centre cohort of larger sample sizes of PSRSA and NRSA vs NC is needed to assess the diagnostic power of these seven biomarkers in the prethrombotic state of recurrent spontaneous abortion. Further research will be necessary to explore the related mechanisms between RSA and the prethrombotic state. The combination of protein array technology with the traditional prethrombotic state laboratory indicators and the endometrial blood perfusion assessment, and final follow-ups of F I G U R E 6 ELISA validation results of serum PSRSA biomarkers. The data is shown in the scatter plot with median values. The P value between PSRSA vs NRSA or NC of each protein was obtained from the Mann-Whitney U test analysis. ***P < .0001, **P < .001 and *P < .05 vs NRSA or NC groups pregnancy outcomes will increase the accuracy, specificity and sensitivity of the diagnosis of RSA with prethrombotic state.

ACK N OWLED G EM ENTS
This work was supported by funding from the Natural Science

Foundation of China (81904234) and the Beijing Municipal
Administration of Hospitals' Ascent Plan (DFL20151301). The sponsors of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

CO N FLI C T O F I NTE R E S T
The authors declare that no competing interests exist.