The following clinical groups were studied: 1) women with RPL (n=11) and control women with elective termination of pregnancy (n=11). Each RPL sample was paired with gestational age-matched control. Blood samples were collected within the HUN-PER study at the Maternity Private Clinic of Obstetrics and Gynecology (Budapest, Hungary) at the time of surgery. All women were Caucasian. Samples from pregnancies with congenital or chromosomal abnormalities or multiple gestations were not collected. Pregnancies were dated according to ultrasound scans and samples were collected between 6–13 weeks of gestation. Table 1 contains the demographic and clinical data of the two study groups.

RPL was defined as two or more failed clinical pregnancies by the American Society for Reproductive Medicine (21). RPL cases were recruited from patients with a nonviable intrauterine pregnancy detected by ultrasound [gestational sac containing an embryo or fetus without fetal heart activity within the first 12 6/7 weeks of gestation according to the American College of Obstetricians and Gynecologists Practice Bulletin (87)]. We did not pay special concern whether previous failed first-trimester clinical pregnancies were complete/incomplete spontaneous or missed abortions. In the control group, elective termination of pregnancy was performed at the request of pregnant women for non-medical reasons in all cases.

Blood samples were processed immediately after collection. Plasma and serum samples were separated and stored at −80°C in our Perinatal Biobank at the HUN-REN Research Centre for Natural Sciences.

Plasma samples were depleted using a Human-14 Multi Affinity Removal Spin Cartridge (Agilent Technologies) according to the manufacturer’s instructions. After the addition of Biognosys’ Lysis Buffer, digestion was carried out on single filter units (Sartorius Vivacon 500, 30’000 MWCO HY) following a modified FASP protocol (described by the Max Planck Institute of Biochemistry, Martinsried, Germany).

Samples were reduced and alkylated using Biognosys’ Reduction & Alkylation Solution for 30min at 37°C. Subsequently, digestion to peptides was carried out using trypsin (Promega Corporation, Madison, WI, USA) overnight at 37°C and a protein:protease ratio of 1:100.

Peptides were desalted using C18 Midi BioPure columns (NestGroup, Ipswich, MA, USA) according to the manufacturer’s instructions and dried down using a SpeedVac system. Peptides were resuspended in LC solvent A [1% acetonitrile, 0.1% formic acid (FA)] (Sigma-Aldrich, St. Louis, MO, USA). Prior to mass spectrometric analyses, they were spiked with Biognosys’ iRT kit calibration peptides. Peptide concentrations were determined using a UV/VIS Spectrometer (SPECTROstar Nano, BMG Labtech Ortenberg, Germany).

For High-Resolution Mass Liquid Chromatography-Tandem Mass Spectrometry (HRM LC-MS/MS) measurements, 2µg of peptides were injected into an in-house packed C18 column (Dr. Maisch ReproSil Pur, 1.9µm particle size, 120Å pore size; 75µm inner diameter, 50cm length, New Objective) on a Thermo Scientific™ Easy nLC 1200 nano-liquid chromatography system connected to a Thermo Scientific™ Q Exactive™ HF-X mass spectrometer equipped with a standard nano-electrospray source. LC solvents were A: 1% acetonitrile in water with 0.1% FA; B: 15% water in acetonitrile with 0.1% FA (all solvents from Sigma-Aldrich, St. Louis, MO, USA). The nonlinear LC gradient was 1-52% solvent B in 60min followed by 52-90% B in 10sec, 90% B for 10min, 90-1% B in 10sec and 1% B for 5min. A data-independent acquisition (DIA) method with one full-range survey scan and 14 DIA windows was used.

Three selected biomarker proteins were analyzed applying the same study groups (RPL, n=14; control, n=30) including the same samples as described in our previous study (88). Briefly, serum concentrations of CGB, PAPPA, and PGF proteins were measured with immunoassay, using a BRAHMS plus KRYPTOR Analyzer (ThermoFisher Scientific, Waltham, MA, USA) with monoclonal antibodies targeted to different epitopes of the measured analytes. Here, the results of CGB and PAPPA were used for re-analysis.

Demographic and clinical information for both the proteomic and validation cohorts is summarized in Table 1 . In both sets, the mean maternal age was higher in the RPL than in the control group. Women in the RPL group had experienced two to four spontaneous miscarriages, including the current one, while those in the control group did not have any previous miscarriages.

Using high-resolution mass spectrometry with a strict limit of quantification, we identified and quantified 651 proteins across all plasma samples (RPL, n=11; control, n=11), of which 27 were classified as PPE proteins ( Supplementary Table 1 ).

Given the higher average maternal age in the RPL group, we investigated whether maternal age might act as a potential confounding factor influencing the proteomic data. To assess this, we performed a correlation analysis between maternal age and the abundance of all quantified proteins across the entire cohort. After correcting for multiple hypothesis testing using Storey’s q-value method, none of the proteins showed a statistically significant association with maternal age, suggesting that maternal age did not have a detectable impact on the observed proteomic differences.

Between early control (n=7) and late control (n=4) samples ( Figure 1 , Supplementary Table 2 ), 60 DA proteins were identified, among them 11 were PPE proteins (enrichment factor: 4.59, p=4.31×10^-6). 29 proteins were downregulated (one PPE, enrichment non-significant) and 31 upregulated (ten PPE, enrichment factor: 8.08, p=4.11×10^-8).

In the whole gestational age range ( Figure 2 , Table 2 , Supplementary Tables 2, 3 ) 50 DA proteins were identified (top 3 down: PZP, PSG9, CGB; top 3 up: C4BPA, HBA, HBB), 36 were downregulated and 14 were upregulated. Eleven PPE proteins were found among DA proteins, corresponding to an enrichment factor of 5.51 and p=5.91×10^-7. All of these PPE proteins were down-regulated in RPL (enrichment factor: 7.65, p=1.31×10^-8). Significantly enriched GO terms included ‘placental function’ (3/27), ‘oxidative processes’ (4/27), ‘immune function’ (8/27), and ‘blood coagulation’ (1/27) related biological processes.

In early RPL ( Figure 3 , Table 3 , Supplementary Tables 2, 3 ), 40 DA proteins were identified (top 3 down: SHBG, CGB, CGA; top 3 up: C4BPA, SAMP, C4BPB), 32 were downregulated and 8 were upregulated. Among these, eight PPE proteins were found (enrichment factor: 5.01, p=6.91×10^-5). In early RPL, also all PPE proteins were downregulated (enrichment factor: 6.26, p=1.16×10^-5) and the fold changes of these proteins were also the lowest with one exception. Placental (3/14), as well as immune function (7/14) -related biological processes were among the top enriched GO terms.

In late RPL ( Figure 4 , Table 4 , Supplementary Tables 2, 3 ) 90 DA proteins were found (top 3 down: PZP, PAPPA, PSG9; top 3 up: THBS1, ECM, HBB), among them 51 were downregulated and 39 were upregulated. Ten PPE proteins were found (enrichment factor: 2.78, p=1.25×10^-3). In late RPL, similar to early RPL, all of the PPE proteins were downregulated (enrichment factor: 4.91, p=7.36×10^-6), and the fold changes of these proteins were also among the lowest ones with two exceptions. The top enriched GO terms included ‘placental function’ (4/206), ‘oxidative processes’ (8/206), ‘immune function’ (55/206), ‘blood coagulation’ (6/206), ‘angiogenesis’ (3/206), ‘cell migration’ (9/206), as well as ‘blood circulation’ (6/206) -related biological processes.

Among DA proteins ( Figure 5 , Supplementary Tables 2, 3 ), 15 were identified in both early and late RPL, including seven PPE proteins (enrichment factor: 4.88, p=3.70×10^-5). 25 proteins (one PPE, non-significant) were found only in early RPL, and 75 (three PPE, non-significant) only in late RPL. Importantly, among DA proteins identified in both early and late RPL, only ‘placental function’-related biological processes (3/4) were enriched.

Principal Component Analysis (PCA) was performed to identify the variance of data in the different sample groups (RPL and control groups) based on all proteins. Principal components 1, 2, and 3 explained 31.7% of the total variance, and little separation between sample groups could be observed (data not shown).

Classification performances were characterized by calculating sensitivity, specificity, and ROC curves for each DA protein in the different gestational age ranges. Scatter plots, box plots, and ROC curves of the best candidate proteins (whole gestational age range: CGA, early RPL: PAEP, late RPL: PAPPA) are presented in Figure 6 , respectively.

The concentration of CGB and PAPPA was also determined using the same immunoassay as in our previous study (88). In concordance with the proteomic findings, similar trends in biomarker levels were observed in the validation cohort. Both CGB and PAPPA showed significantly lower concentrations in the RPL group compared to the control group, reinforcing the proteomic results ( Figure 7 ).

The correlation analysis between immunoassay and mass spectrometry data revealed a good correlation ( Figure 7 ) on the overlapping cohort as the Pearson correlation coefficient (R) was 0.795 for CGB and 0.965 for PAPPA.

1) Sixty DA proteins had gestational age-related concentration changes in controls. 2) Fifty DA proteins were dysregulated in RPL compared to control samples. 3) Forty DA proteins were found in early RPL and ninety in late RPL. 4) PPE proteins were enriched among DA proteins and showed to be the most downregulated during both periods in RPL. 5) Enriched GO terms in early RPL included ‘immune regulation’, ‘complement activation’, and ‘female pregnancy’ while ‘placental function’, ‘oxidative processes’, ‘immune function’, and ‘blood coagulation’ related biological processes were enriched in late RPL among others. 6) Several biomarker candidates for early and late RPL were identified and two were validated with immunoassay.

The fetal support and maintenance of pregnancy by providing nutrition, immunological and hormonal regulation as well as gas and waste exchange, are the most important roles of the placenta (98, 99). It has been known for more than half century that protein products of the placenta get into the maternal circulation, and their circulating concentration changes may well reflect changes in placental functions with gestational age in normal pregnancies as well as dysregulation of placental functions in pregnancy complications. Therefore, placental proteins turned out to be of great significance for the monitoring of placental functions through non-invasive sampling of maternal blood (i.e. liquid biopsy) (100–103).

We have recently defined PPE proteins (90, 91), which play a central role in monitoring placental functions in the maternal blood. Also, our previous results showed that placental dysfunction is often associated with the altered expression of PPE proteins (104–113), therefore, special attention should be paid to these proteins in the prediction and treatment of pregnancy complications such as RPL. Recently, with immunoassay, we examined three of these PPE proteins (free β-hCG, PAPP-A, and PlGF) and confirmed their diagnostic significance in RPL (88). Utilizing the recent developments in the field of proteomics and mass spectrometry that enable the deep proteomic investigation of different tissues (114), including the placenta (115), and the liquid biopsy of these tissues via minimal invasive blood draw (116, 117), here we aimed to explore the proteome-wide changes in maternal plasma in RPL. This current analysis had a special attention to PPE proteins in an attempt to provide relevant information about the developmental status and functional condition of the placenta and to support the early detection and/or prediction of failing placental functions.

The placenta undergoes rapid development in the first weeks of pregnancy (118), however, the lack of its connection to the maternal circulation may hinder placental liquid biopsy at the beginning of the pregnancy (119, 120). Indeed, villous placental vessels and circulation start to develop at around gestational weeks 5–6 (121, 122), while the intervillous circulation is only being established after around the 9^th gestational week. This is a protective mechanism for the developing embryo and placenta to defend their tissues from excessive and harmful oxygen exposure (95, 123). Therefore, placental products enter into the maternal circulation in much larger quantities only after the opening of the intervillous blood space (124).

To account for this physiological change in placental circulation during the first trimester, we investigated the proteomic changes before and after the supposed start of the intervillous circulation while considering differences in maternal environmental, metabolic, and immunological characteristics. In control pregnancies, our study identified key DA proteins between gestational weeks 6–9 and 9–13, shedding light on the physiological adaptations during early pregnancy. There were 29 proteins with elevated concentrations in gestational weeks 6–9 compared to 9–13. These included FN1 (fibronectin 1) (125), ECM1 (extracellular matrix protein 1) (126), and COMP (cartilage oligomeric matrix protein) (127), which are all associated with extracellular matrix organization, cell adhesion, and tissue remodeling. Their dropping concentration with gestational age suggests a shift from initial implantation and rapid tissue expansion toward the established placental milieu. Other proteins with decreasing concentration over gestational age included PF4 (platelet factor 4), which is a key regulator of vascular homeostasis (128, 129), highlighting the high demands for vascular remodeling during the early phase of placentation. Furthermore, IGFBP2 (insulin-like growth factor-binding protein 2), which regulates the bioavailability of insulin-like growth factors crucial for cellular proliferation and differentiation (130), had a decreasing concentration with advancing gestational age. Several platelet-related proteins (e.g., PF4, PF4V1, PPBP alias CXCL7, THBS1, and VWF) were also found to decrease over gestational age. This may be due to a decrease in mean platelet count gradually throughout gestation; otherwise, upon activation, maternal platelets would provide a source of proinflammatory mediators in the intervillous space of the placenta (129).

Conversely, 31 proteins demonstrated increased concentrations in gestational weeks 9–13 compared to 6–9. These included PZP (pregnancy zone protein) which supports the transition towards immune tolerance (131, 132), as well as PAPPA which increases the bioavailability of insulin-like growth factors, which are crucial for fetal growth and placental development. Also, ADAM12 (a disintegrin and metalloproteinase domain-containing protein 12) is a well-known regulator of trophoblast invasion and extracellular matrix (ECM) remodeling (133). Proteins such as PRG2 (proteoglycan 2, pro eosinophil major basic protein), PSG1, PSG3, and PSG9 (pregnancy-specific beta-1 glycoprotein 1, 3, and 9) are key immunomodulators that help to establish maternal immune tolerance towards the fetus (134). Many of these proteins are also well-known to have increasing concentrations in maternal blood during this timeframe of pregnancy which validates our study (112, 116). These findings provide insights into the dynamic proteomic changes occurring in early pregnancy. The elevated concentrations of ECM-related proteins (FN1, ECM1, COMP) and vascular regulators (PF4) in the earlier phase suggest the importance of ECM reorganization, tissue remodeling, and vascular homeostasis essential for implantation and initial placental formation. ECM proteins not only provide structural scaffolding for invading trophoblasts but also participate in immune regulation by modulating leukocyte adhesion and migration. FN1, for instance, can interact with integrins and Toll-like receptors on maternal immune cells, influencing macrophage activation and cytokine secretion (135).

In contrast, the increased abundance of proteins after week 9, involved in immunomodulation (PZP, PSG1, PSG9, PRG2) and placental development (PAPPA, ADAM12), aligns with the physiological progression toward immune tolerance and increased placental growth.

PZP, a pregnancy-associated protease inhibitor, has been shown to suppress T-cell proliferation and promote a tolerogenic environment by forming complexes with glycodelin-A, and leading to reduced IL-2 production (131).

Pregnancy-specific glycoproteins (PSG1, PSG3, PSG9) are produced by the syncytiotrophoblast (136) and act on maternal monocytes, dendritic cells, and NK cells to induce anti-inflammatory cytokines (e.g., IL-10) and activate TGF-β1 and TGF-β2 (134, 137), which together promote the expansion of regulatory T cells (Tregs) and dampen cytotoxic responses (138).

ADAM12 plays a dual role, facilitating trophoblast invasion through ECM remodeling while also influencing the availability of cytokines and growth factors within the decidua hence contributes to the pathogenesis of tissue inflammation and TH1 differentiation (139). Altogether, these proteins help in reprogramming the maternal immune system, essential for maintaining maternal-fetal tolerance as the intervillous circulation opens and the fetal-derived placenta becomes more exposed to maternal immunity.

In early RPL cases, we identified the predominant changes in the immune-related proteome. Several proteins involved in immune tolerance and complement regulation were downregulated in early RPL, including C4BPA (complement component 4 binding protein alpha) and C4BPB (complement component 4 binding protein beta), which play crucial roles in modulating complement activation and preventing excessive immune responses at the maternal-fetal interface (140). Additionally, PAEP (progestagen-associated endometrial protein, also known as glycodelin) (141), a critical modulator of maternal immune tolerance that suppresses T-cell activation and supports embryo implantation (142, 143), showed decreased levels in early RPL cases. In addition, GO enrichment analysis further highlighted the disrupted biological processes in early RPL, with enriched terms predominantly linked to ‘immune regulation’, ‘complement activation’, and ‘female pregnancy’. Key terms included ‘multi-organism processes’ (reflecting maternal-fetal interactions), ‘regulation of opsonization’, and ‘negative regulation of complement activation’. These findings may reflect impaired regulation of immune interactions and tolerance mechanisms at the maternal-fetal interface in RPL.

The proteomic profile in late RPL revealed an increased number of DA proteins associated with angiogenesis and inflammatory responses. ANGPTL3 (angiopoietin-related protein 3), a regulator of lipid metabolism (144), has been implicated in modulating angiogenesis (145) and may contribute to altered placental vascularization. PF4, a chemokine released from activated platelets, is involved in immune cell recruitment and inflammatory signaling (146, 147), and its elevation may reflect heightened immune activation or vascular stress in the decidua. In addition, the expression level of the PZP, a known immunomodulator that neutralizes immune activation (148) was downregulated, similar to anti-microbial peptides, important components of the innate immune system at the maternal-fetal interface (149). Similarly, the intensity of the fetal growth regulator PAPPA (150, 151), was also reduced, suggesting an impairment in placental development function. Furthermore, increased levels of hemoglobin subunits (HBA, HBB, HBD) in late RPL compared may result from hyperviscosity/hemoconcentration, which subsequently increases the risk of miscarriage (152, 153). Interestingly, in these studies, low concentration of hemoglobin in the preconception or early pregnancy period, resulting in anemia, also increased the risk of miscarriage (152, 153).

GO analysis in late RPL revealed enriched terms including ‘immune response’, ‘neutrophil degranulation’, ‘platelet activation’, and ‘inflammatory response’, further emphasizing an excessive and dysregulated immune activation, which may contribute to trophoblast stress and placental dysfunction. The presence of oxidative stress-related terms such as ‘cellular oxidant detoxification’, ‘response to reactive oxygen species’, and ‘positive regulation of reactive oxygen species metabolic process’ further supports the notion that placental oxidative stress (123) is a major contributing factor to late RPL. These findings collectively suggest that late RPL is characterized by a pathological environment marked by impaired angiogenesis, dysregulated immune responses, substantial oxidative stress of the trophoblast and placental dysfunction.

Among the shared characteristics, we found consistently low levels of CGA across both early and late RPL, indicating a persistent defect in placental function. The pregnancy-specific glycoproteins, such as PSG1, PSG2, PSG4, PSG9, and PSG11, which are essential for immune tolerance by inducing anti-inflammatory cytokines like IL-10 and promoting trophoblast survival (134, 154, 155), were also significantly reduced in both RPL groups. Additionally, we also observed reduced levels of sex hormone-binding globulin (SHBG) in both RPL groups, a key regulator of sex steroid bioavailability. Low SHBG level has been previously associated with early pregnancy loss (67, 156) and not only reflects hormonal or metabolic dysregulation affecting endometrial receptivity and placental development, but may also alter SHBG-mediated estradiol signaling in lymphocytes, potentially disrupting maternal immune adaptation during early pregnancy (157).

However, other findings indicate that immune dysfunction in RPL follows a gestational age-dependent pattern. Early RPL appears to be associated with maternal immune maladaptation characterized by defective complement regulation and insufficient maternal immune tolerance, whereas late RPL shows stronger associations with dysregulation of placental immune function, vascular remodeling, and inflammation. Despite these differences, the persistently low levels of CGA and PSGs across both gestational age groups suggest that placental dysfunction remains a central feature of RPL throughout pregnancy. In addition, in late RPL we found signs of reduced trophoblast invasion and inadequate spiral artery remodeling, which were implicated already in miscarriages (158). The consequently altered placental circulation characterized by turbulent jets damage the villous structure that can also be seen with ultrasound (159). These rheological and mechanical changes also have indirect molecular effects (160), as evidenced by oxidative stress triggered pathological processes (90). Our findings are also consistent with oxidative stress-related molecular changes in the placenta, as reflected in the maternal circulation.

In summary, our results highlight distinct shifts in the maternal immune proteome before and after the establishment of placental circulation. The distinct and overlapping enrichment patterns across gestational ages underscore the dynamic interplay between immune adaptation, placental function, and vascular remodeling during pregnancy, offering insight into the molecular pathways underlying early and late RPL. While these proteomics results offer valuable biological insights and identify potential diagnostic biomarkers, also validated by immunoassay in a small sample, they are exploratory in nature and require further independent validation in larger, external cohorts to confirm their robustness and clinical relevance.

This study has several strengths, including the use of next-generation proteomics with stringent clinical definitions and homogenous patient groups, enabling the identification of gestational age-specific molecular pathways in RPL. Robust methodologies, such as depletion of abundant proteins, mass spectrometry with stringent FDR controls, and validation through immunoassay correlation, enhance the reliability of findings. Indeed, the correlation coefficient between the mass spectrometry and immunoassay results for PAPPA was outstandingly high, while for CGB it was slightly lower, however, even this is entirely sufficient and consistent with the literature (161).

The limitations of the study include the timing of sample collection at the point of miscarriage, which prevents assessing the predictive value of biomarkers before clinical symptoms. The small sample size which inherently reduces statistical power and increases the risk of both type I (false positive) and type II (false negative) errors, and the use of elective termination pregnancies as controls may limit the generalizability of results, while other potential confounders, such as environmental or lifestyle factors, were not addressed. While our use of multiple testing correction, and cross-validation methods aimed to mitigate these risks, the possibility of overfitting and spurious associations cannot be fully excluded. Additionally, the RPL population in our study primarily consisted of women with two or three prior losses, which represents the lower end of the clinical spectrum. It remains to be investigated whether proteomic alterations are more pronounced or distinct in cases with higher-order losses, which may reflect more severe or recurrent pathological processes.

Since one potential confounder that could influence protein levels is maternal age, which was higher in the RPL group, we investigated the potential effect of this confounder by looking into the data focusing on age-related changes in the proteome. It has been described that aging-related changes in the plasma proteome generally occur over much longer timeframes (i.e., decades) and typically result in modest fold changes (162). For instance, in the case of GDF15 - showing the strongest known association between protein abundance and aging from 1301 proteins - the log₂ fold change per year of age (β = 0.018) corresponds to only ~1.065-fold change over five years (163). This is minimal compared to the ≥1.5-fold changes observed for DA proteins in our study. Furthermore, correlation analysis between maternal age and proteomic data revealed no significant associations after correcting for multiple comparisons. These findings suggest that maternal age is unlikely to be the primary driver of the observed group differences.