Comprehensive Proteomic Analysis Reveals Distinct Features and a Diagnostic Biomarker Panel for Early Pregnancy Loss in Histological Subtypes.

In this study, we enrolled 36 patients with EPL in China and categorized them into 4 subclinical groups according to their clinical, morphological, and genetic features ( Table 1 and Supplemental Table S1 ). The mean age of the participants ranged from approximately 28 to 31 years, and the mean gestational age ranged from 46 to 72 days. Higher mean β-human chorionic gonadotropin and Ki-67 positive immunostaining rates were observed in the CHM group than in the other groups ( p < 0.05). Table 1 Clinical characteristics of patients in the discovery proteomics study Characteristic CHM ( n  = 10) PHM ( n  = 6) HA ( n  = 10) NC ( n  = 10) Race/ethnicity  Asian 10 (100%) 6 (100%) 10 (100%) 10 (100%)  Unknown/other 0 (0%) 0 (0%) 0 (0%) 0 (0%) Age at procedure (years) 30.80 ± 9.69 (19–51) 28.17 ± 3.97 (21–32) 29.80 ± 4.32 (22–35) 30.20 ± 5.33 (23–36) Gestational age at tissue collection (days) 61.90 ± 14.08 (50–97) ∗ 72.33 ± 12.11 (58–90) ∗ 58.60 ± 12.96 (45–86) 46.10 ± 6.62 (37–59) β-hCG (10 3 mIU/ml) 127.68 ± 92.43 102.51 ± 81.41 27.2 ± 34.42§ N/A Adverse pregnancy history  Spontaneous abortion 1 (10%) 1 (16.67%) 0 (0%) 0 (0%) Histologic diagnosis  HM (untyped) 2 (20%) 0 (0%) 0 (0%) N/A  CHM 6 (60%) 1 (16.67%) 0 (0%) N/A  PHM 0 (0%) 2 (33.33%) 0 (0%) N/A  HA 2 (20%) 3 (50.00%) 10 (100%) N/A  NC 0 (0%) 0 (0%) 0 (0%) N/A P57 immunostaining  Positive 0 (0%) 6 (100%) 10 (100%) 10 (100%)  Negative 10 (100%) 0 (0%) 0 (0%) 0 (0%) Ki-67 immunostaining (%) 85.40 ± 14.32 51.17 ± 20.89§ 35.00 ± 31.27§ 33.10 ± 22.48§ STR genotyping  Androgenetic diploidy 10 (100%) 0 (0%) 0 (0%) 0 (0%)  Diandric triploidy 0 (0%) 6 (100%) 0 (0%) 0 (0%)  Biparental diploidy 0 (0%) 0 (0%) 10 (100%) 10 (100%) Sex chromosome content  XY 0 (0%) 0 (0%) 5 (50%) 7 (70%)  XX 10 (100%) 0 (0%) 5 (50%) 3 (30%)  XXY 0 (0%) 0 (0%) 0 (0%) 0 (0%)  XYY 0 (0%) 1 (16.67%) 0 (0%) 0 (0%)  XXX 0 (0%) 5 (83.33%) 0 (0%) 0 (0%) Comorbidities  Vaginitis 0 (0%) 0 (0%) 1 (10%) 1 (10%)  Fibroid 0 (0%) 0 (0%) 0 (0%) 1 (10%) Prognosis  GTN 1 (10%) 0 (0%) 0 (0%) 0 (0%) p value < 0.05 for comparisons between CHM or PHM patients and healthy controls∗ and between CHM patients and other groups§. β-hCG, β-human chorionic gonadotropin; CHM, complete hydatidiform mole; GTN, gestational trophoblastic neoplasia; HA, hydropic abortion; HM, hydatidiform mole; NC, normal control; PHM, partial hydatidiform mole; STR, short tandem repeat. Clinical characteristics of patients in the discovery proteomics study p value < 0.05 for comparisons between CHM or PHM patients and healthy controls∗ and between CHM patients and other groups§. β-hCG, β-human chorionic gonadotropin; CHM, complete hydatidiform mole; GTN, gestational trophoblastic neoplasia; HA, hydropic abortion; HM, hydatidiform mole; NC, normal control; PHM, partial hydatidiform mole; STR, short tandem repeat. All patients underwent precise diagnosis through pathological examination, p57 immunohistochemical analysis, and STR genotyping ( Fig. 2 ). However, histopathology revealed CHMs in only 6 patients (60%) and PHMs in 2 patients (33.33%). This result highlights the low diagnostic concordance rate between pathology and STR genotyping in molar pregnancy, emphasizing the urgent need for new effective biomarkers to distinguish EPL subtypes. Fig. 2 Pathological diagnosis, p57 immunohistochemistry, and genotyping analysis of specimens before enrolment in the study. A, typical H&E staining and p57 immunohistochemical expression patterns of the CHM, PHM, HA, and NC groups. CHM is negative for p57 expression in villous cytotrophoblastic and stromal cells. B, short tandem repeat (STR) genotyping analysis. The results depict the purely androgenetic, diandric triploidy, and biparental characteristics corresponding to the CHM, PHM, and HA or NC groups, respectively. CHM, complete hydatidiform mole; HA, hydropic abortion; NC, normal control; PHM, partial hydatidiform mole; POC, product of conception. Pathological diagnosis, p57 immunohistochemistry, and genotyping analysis of specimens before enrolment in the study. A, typical H&E staining and p57 immunohistochemical expression patterns of the CHM, PHM, HA, and NC groups. CHM is negative for p57 expression in villous cytotrophoblastic and stromal cells. B, short tandem repeat (STR) genotyping analysis. The results depict the purely androgenetic, diandric triploidy, and biparental characteristics corresponding to the CHM, PHM, and HA or NC groups, respectively. CHM, complete hydatidiform mole; HA, hydropic abortion; NC, normal control; PHM, partial hydatidiform mole; POC, product of conception. We performed a label-free proteomic analysis of 36 fresh villous tissues (10 CHMs, six PHMs, 10 HAs, and 10 NCs). The quality of the protein extraction met the requirements of the study ( Supplemental Fig. S1 ), and the results indicated good consistency in protein identification and quantification ( Supplemental Fig. S2 , A and B ). A total of 6287 proteins were identified, with 5711 proteins quantified ( Supplemental Fig. S2 C and Supplemental Table S2 ). PCA and hierarchical clustering demonstrated clear separation of the NC, CHM, and PHM or HA groups, but a certain degree of overlap between the PHM and HA groups were observed, which is consistent with the pathological diagnosis status quo ( Fig. 3 , A and B ). Proteomic data and the clinical course of EPL were correlated in a way that encouraged further research on specific proteins that may be used as diagnostic markers. Fig. 3 Proteomic profiling analysis of EPL in the discovery study. A, principal component analysis (PCA) of samples within each subtype; 95% confidence ellipses are shown for the group means. B, the abundance profile of all proteins in the four groups is represented by a heat map. C, the proteomics data were clustered using Mfuzz into significant discrete clusters to illustrate relative expression changes. D, volcano plot indicating DEPs upregulated or downregulated in each group. E, KEGG pathway enrichment analysis of DEPs. F, subcellular distribution of the DEPs in the CHM, PHM, HA, and NC groups. G, steps involved in refining the biomarker candidates to a multiprotein classifier panel. CHM samples (n = 10), PHM samples (n = 6), HA samples (n = 10), and NC samples (n = 10). CHM, complete hydatidiform mole; DEP, differentially expressed protein; EPL, early pregnancy loss; HA, hydropic abortion; KEGG, Kyoto Encyclopedia of Genes and Genomes; NC, normal control; PHM, partial hydatidiform mole; PRM, parallel reaction monitoring. Proteomic profiling analysis of EPL in the discovery study. A, principal component analysis (PCA) of samples within each subtype; 95% confidence ellipses are shown for the group means. B, the abundance profile of all proteins in the four groups is represented by a heat map. C, the proteomics data were clustered using Mfuzz into significant discrete clusters to illustrate relative expression changes. D, volcano plot indicating DEPs upregulated or downregulated in each group. E, KEGG pathway enrichment analysis of DEPs. F, subcellular distribution of the DEPs in the CHM, PHM, HA, and NC groups. G, steps involved in refining the biomarker candidates to a multiprotein classifier panel. CHM samples (n = 10), PHM samples (n = 6), HA samples (n = 10), and NC samples (n = 10). CHM, complete hydatidiform mole; DEP, differentially expressed protein; EPL, early pregnancy loss; HA, hydropic abortion; KEGG, Kyoto Encyclopedia of Genes and Genomes; NC, normal control; PHM, partial hydatidiform mole; PRM, parallel reaction monitoring. The expression patterns of the EPL subtype and NC tissues were subsequently analyzed using Mfuzz ( https://bioconductor.org/packages/release/bioc/html/Mfuzz.html ) ( 26 ). Mfuzz analysis was conducted with default parameters to divide all quantifiable proteins into nine clusters. The results revealed two interesting clusters, cluster 4 and cluster 5, containing 663 and 562 proteins, respectively, the expression of which consistently increased or decreased from CHMs to PHMs, HAs, and NCs. These proteins in the two clusters may serve as indicators of EPL severity ( Fig. 3 C ). Pathway enrichment analysis revealed that the proteins in cluster 4 were enriched mainly in the mitochondrial membrane pathway, and the proteins in cluster 5 were enriched in the blood microparticle pathway. In addition, other clusters illustrated the relative proteomic changes between groups ( Supplemental Fig. S3 and Supplemental Table S3 ). To determine which of the 5711 quantifiable proteins were significantly and specifically upregulated or downregulated in each EPL type, a differential expression analysis (adjusted p value < 0.05) comparing each group to the other three groups ( e.g. , CHMs versus non-CHMs) was applied. The resulting DEPs were considered specifically represented in this group. A total of 1662 DEPs were identified, with 842 (50.66%), 192 (11.55%), 42 (2.53%), and 586 (35.26%) proteins being specific to the CHM, PHM, HA, and NC groups, respectively. Among these DEPs, 602 and 1060 were upregulated and downregulated, respectively ( Fig. 3 D and Supplemental Table S4 , A and B ). We then analyzed the subcellular distribution of the DEPs ( Fig. 3 F and Supplemental Table S4 A ) and found that the DEPs were predominantly distributed in the cytoplasm (447, 26.90%), extracellular space (376, 22.62%), and nucleus (307, 18.47%). A notable portion of the CHM group-specific downregulated DEPs (115, 13.66%) were also annotated as mitochondrial. These findings suggest that finding suggests that the CHM proteome is involved in energy metabolism, which is consistent with previous reports ( 27 , 28 ). Functional enrichment analysis of DEPs revealed that DNA replication (FEN1, MCM3/4/5, PCNA, POLD1, and RFC3/5), the complement and coagulation cascade (A2M, C7/9, CLU, and F10/2), and mismatch repair (MSH2, PCNA, POLD1, and RFC3/5) were overrepresented in proteins that were upregulated in CHMs. In contrast, the peroxisome (ABCD3, ACAA1, ACOX3, ACSL1, and CAT) and chemical carcinogenesis-reactive oxygen species (ROS) (CAT, COX5B/6A1/6B1, EPHX1, and GSTM1) pathways were overrepresented in the proteins with downregulated expression. Notably, unlike the suppression of hypoxia-inducible factor 1 (HIF-1) signaling pathways in patient PHMs and HAs, the HIF-1 signaling pathways were activated in CHMs. Moreover, the expression of collagen family members (COL21A1, COL4A6, and COL6A1/A2/A3) was specifically upregulated in the PHM group. These findings indicate the dysregulation of tissue integrity and the extracellular matrix, which have been reported to be characteristic features of PHM ( 29 ). Furthermore, several important proteins involved in glycolysis and carbon metabolism (ALDH3B2, ALDOA, ALDOC, ENO1, and GPI) were downregulated in PHMs, indicating that metabolic dysfunction is relevant to PHM pathogenesis and development. In contrast, there were fewer DEPs than in the HA group, indicating that the downregulated proteins participate in the pathways of nucleotide sugar biosynthesis (GFPT1 and HK2) and central carbon metabolism in cancer (HK2 and SLC1A5) ( Fig. 3 , D and E and Supplemental Table S5 B ). A multistage refinement workflow, as illustrated in Figure 3 G , was utilized to identify candidate biomarkers for EPL subtypes. First, we focused on placenta- or embryo-related DEPs on the basis of tissue-specific annotation provided by the ProteinAtlas and GeneCards databases, resulting in the inclusion of 599 DEPs ( Supplemental Fig. S4 A and Supplemental Table S5 A ). According to the functional enrichment analysis of these filtered proteins, 150 downregulated DEPs in the CHM group were related to mitochondria ( Supplemental Fig. S4 B and Supplemental Table S5 B ). The overexpressed proteins contributed to the enrichment of pathways related to cell proliferation, cell migration, and paternally expressed imprinted gene coding proteins (MEST, PEG10, CARS1, IGF2, and DLK1), which is consistent with the cell phenotype and epigenetic mechanisms of HMs. Next, we focused on the proteins with a high expression difference ratio (fold change> 4 or < 0.25), especially the top DEPs with high data quality. A final list of 18 proteins ( Supplemental Table S6 B ) was accepted for verification using a PRM method in the same cohort but with the inclusion of three new cases in the PHM group to expand the targeted MS cohort, comprising 10 CHMs, nine PHMs, and 10 HAs, and 10 healthy controls ( Supplemental Table S6 A ). From the 18 protein candidates, 32 peptides containing unique sequences were measured by PRM. The targeted proteomic assay identified 14 protein candidates with a trend consistent with that of the discovery proteomic data ( Supplemental Table S6 , B – D ). The PCA model based on these markers clearly separated the three disease groups ( Fig. 4 A ). Of particular interest was the ability to differentiate PHMs from HAs. The expression profiles of these markers were generated, revealing clear intergroup differences between PHMs and HAs ( Fig. 4 , B and C ). We visualized the protein candidate expression and gene ontology term enrichment analysis results via a Sankey diagram when biomarkers were selected ( Fig. 4 D ). Fig. 4 Summary of targeted proteomics analysis of candidate biomarkers in the validation study. A, principal component analysis (PCA) model based on 18 candidate proteins of all samples illustrating clear separation; 95% confidence ellipses are shown for the group means. B and C, relative quantities of the 14 proteins between CHM ( blue ), PHM ( red ), HA ( green ), and NC ( purple ) are shown. D, a Sankey graphical summary of biological process enrichment and DEPs. E, least absolute shrinkage and selection operator (LASSO) logistic model of the markers in each group. F, receiver operating characteristic (ROC) curve analyses of the diagnostic markers. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. . CHM samples (n = 10), PHM samples (n = 9), HA samples (n = 10), and NC samples (n = 10). CHM, complete hydatidiform mole; DEP, differentially expressed protein; GO, gene ontology; HA, hydropic abortion; ns, nonsignificant; NC, normal control; PHM, partial hydatidiform mole. Summary of targeted proteomics analysis of candidate biomarkers in the validation study. A, principal component analysis (PCA) model based on 18 candidate proteins of all samples illustrating clear separation; 95% confidence ellipses are shown for the group means. B and C, relative quantities of the 14 proteins between CHM ( blue ), PHM ( red ), HA ( green ), and NC ( purple ) are shown. D, a Sankey graphical summary of biological process enrichment and DEPs. E, least absolute shrinkage and selection operator (LASSO) logistic model of the markers in each group. F, receiver operating characteristic (ROC) curve analyses of the diagnostic markers. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. . CHM samples (n = 10), PHM samples (n = 9), HA samples (n = 10), and NC samples (n = 10). CHM, complete hydatidiform mole; DEP, differentially expressed protein; GO, gene ontology; HA, hydropic abortion; ns, nonsignificant; NC, normal control; PHM, partial hydatidiform mole. To examine the diagnostic ability of the markers, we developed diagnostic models using the group LASSO logistic model and ROC analysis. The cut-off value, specificity, sensitivity, and AUC value of the models were evaluated ( Fig. 4 F and Supplemental Table S6 e ). The percentage contributions of the markers to the model in each test were then calculated ( Fig. 4 E ). This process identified four protein markers, DLK1, SPTB/COL21A1, and SAR1A, which represented the CHM, PHM, and NC groups, respectively. The AUCs of DLK1, SPTB/COL21A1, and SAR1A for detecting CHM, PHM or NC were 0.900 [95% confidence interval (CI) 0.806 to 0.994], 0.804 (95% CI 0.631–0.976)/0.885 (95% CI 0.758–1.000), and 0.991 (95% CI 0.970–1.000), respectively ( Fig. 4 F ). First, IHC and immunofluorescence assays were conducted to determine the spatial distribution of the four biomarkers in first-trimester human villous tissues ( Fig. 5 , A and B ). DLK1, an imprinted expressed protein from the paternal allele, was predominantly found in the cytoplasm of villous mesenchymal cells. SPTB exhibited staining primarily within the cytoplasm of cytotrophoblast cells and extravillous trophoblast cells and nucleated red blood cells, with minimal expression in other trophoblast cytoplasmic regions. COL21A1 and SAR1A were widely expressed in the cytoplasm of all trophoblast cell types ( Fig. 5 A ). To further elucidate their specific cellular associations, we assessed the colocalization of these markers with leucocyte antigen-G (HLA-G, a marker for extravillous trophoblast cells), CK7 (a general trophoblast marker), and vimentin (a marker for villous mesenchymal and stromal cells). The results of these colocalization studies are presented in Figure 5 B . Fig. 5 Localization of the four biomarkers in human first-trimester placental tissues. A, H&E and IHC images showing the morphological characteristics and expression of DLK1, SPTB, COL21A1, and SAR1A. Red arrows indicate nucleated red blood cells. Scale bars represent 50 μm. B, immunofluorescence colocalization of DLK1, SPTB, COL21A1, and SAR1A with established trophoblast-specific markers. Vimentin, HLA-G, and CK7 were used as markers of mesenchymal/stromal cells, EVT cells, and trophoblast cells, respectively. Scale bars represent 50 μm. DLK1, delta-like homolog 1; EVT, extravillous trophoblast; HLA-G, human leucocyte antigen-G; IHC, immunohistochemistry. Localization of the four biomarkers in human first-trimester placental tissues. A, H&E and IHC images showing the morphological characteristics and expression of DLK1, SPTB, COL21A1, and SAR1A. Red arrows indicate nucleated red blood cells. Scale bars represent 50 μm. B, immunofluorescence colocalization of DLK1, SPTB, COL21A1, and SAR1A with established trophoblast-specific markers. Vimentin, HLA-G, and CK7 were used as markers of mesenchymal/stromal cells, EVT cells, and trophoblast cells, respectively. Scale bars represent 50 μm. DLK1, delta-like homolog 1; EVT, extravillous trophoblast; HLA-G, human leucocyte antigen-G; IHC, immunohistochemistry. To further extend our study for clinical use, we performed WB analysis and IHC assays in an independent cohort. WB analysis of seven fresh villous tissue samples from each group revealed variability in the expression levels of the markers. Interestingly, an overall trend toward upregulation was observed for DLK1, SPTB/COL21A1, and SAR1A across CHMs, PHMs, and NCs, respectively ( Fig. 6 , A and B ). Another independent cohort of 30 FFPE samples from each group was collected for IHC validation. Stronger staining of DLK1, STPB/COL21A1, and SAR1A was observed in CHMs, PHMs, and NCs, respectively, than in other EPL subtypes ( Fig. 6 C , Supplemental Fig. S5 , and Supplemental Table S7 ), which is consistent with the MS results. The subcellular localization of these four proteins remained unchanged among the subtypes. The ROC-AUCs of DLK1, SPTB, COL21A1, and SAR1A for detecting CHM, PHM, or NC were 0.969 (95% CI 0.936–1.000), 0.801 (95% CI 0.710–0.891), 0.746 (95% CI 0.648–0.845), and 0.879 (95% CI 0.812–0.947), respectively. DLK1 yielded the highest AUC value, with fairly high sensitivity (95.56%) and specificity (96.67%) at the threshold point of 7.5 for distinguishing CHMs from other EPL subtypes. SPTB and COL21A1 served as diagnostic markers for PHM, each with distinct advantages. The SPTB demonstrated high sensitivity, with a rate of 94.44% and a specificity of 50% at an optimal cut-off value of 7.5. In contrast, COL21A1 was characterized by its high specificity, with a rate of 86.67% and a sensitivity of 47.78% at a cut-off value of 3.5. SAR1A had significant diagnostic value for differentiating NC pregnancies from other pathological pregnancies, with a sensitivity of 82.22% and an impressive specificity of 93.33%. Fig. 6 Validation of a four-marker diagnostic panel for EPL in an independent cohort. A, representative WB bands for the expression of DLK1, SPTB, COL21A1, and SAR1A in fresh POC samples of CHM, PHM, HA, and NC. B, quantitative analysis of the WB band intensities for each protein marker, normalized to that of GAPDH as a loading control (n = 28, seven samples per group). C, representative immunohistochemistry (IHC) images of the expression of the four proteins in the independent EPL validation cohort (n = 120, 40 samples each group). Scale bars represent 50 μm. D, semiquantitative staining scores representing the immunohistochemical expression levels of the four proteins in the validation cohort. E, ROC curves depicting the diagnostic performance of the protein biomarkers based on the immunohistochemistry scores. Note: The data are shown as the means ± SDs. Differences were considered significant if p  < 0.05 (∗∗ p  < 0.01, ∗∗∗ p  < 0.001, and ∗∗∗∗ p  < 0.0001). CHM, complete hydatidiform mole; DLK1, delta-like homolog 1; EPL, early pregnancy loss; HA, hydropic abortion; NC, normal control; PHM, partial hydatidiform mole; POC, product of conception; ROC, receiver operating characteristic. Validation of a four-marker diagnostic panel for EPL in an independent cohort. A, representative WB bands for the expression of DLK1, SPTB, COL21A1, and SAR1A in fresh POC samples of CHM, PHM, HA, and NC. B, quantitative analysis of the WB band intensities for each protein marker, normalized to that of GAPDH as a loading control (n = 28, seven samples per group). C, representative immunohistochemistry (IHC) images of the expression of the four proteins in the independent EPL validation cohort (n = 120, 40 samples each group). Scale bars represent 50 μm. D, semiquantitative staining scores representing the immunohistochemical expression levels of the four proteins in the validation cohort. E, ROC curves depicting the diagnostic performance of the protein biomarkers based on the immunohistochemistry scores. Note: The data are shown as the means ± SDs. Differences were considered significant if p  < 0.05 (∗∗ p  < 0.01, ∗∗∗ p  < 0.001, and ∗∗∗∗ p  < 0.0001). CHM, complete hydatidiform mole; DLK1, delta-like homolog 1; EPL, early pregnancy loss; HA, hydropic abortion; NC, normal control; PHM, partial hydatidiform mole; POC, product of conception; ROC, receiver operating characteristic. Due to the lack of suitable HM animal models, the human trophoblast cell lines HTR-8/SVneo and Swan-71 cell lines were selected for subsequent in vitro biological behavior experiments. We evaluated the expression of the four biomarkers in these cell lines using WB and immunofluorescence staining, confirming their presence in both trophoblast cell lines ( Fig. 7 , A and B ). Since we observed that SPTB (β-I-spectrin) and COL21A1 (collagen XXI) are candidate biomarkers for predicting PHM, and previous studies have indicated that β-spectrin may play a role in embryogenesis ( 30 , 31 ), SPTB was chosen for further investigation. We generated SPTB knockdown and SPTB-overexpressing cell lines by transfecting human trophoblasts with siRNA and SPTB plasmids to explore the role of SPTB in this process. The transfection efficiency was assessed by quantitative real-time PCR and WB analysis ( Supplemental Fig. S6 ). As indicated by the results of the EdU assay, the knockdown of SPTB suppressed the proliferative capability of trophoblast cells ( Fig. 7 C ). Silencing SPTB inhibited the migration ability of trophoblast cells in transwell and wound healing assays ( Fig. 7 , D and E ). The overexpression of SPTB had the opposite effect ( Supplemental Fig. S7 , A – C ). Epithelial‒mesenchymal transition contributes to the enhanced migration of trophoblast cells, so we investigated whether SPTB could be involved in this process. The results revealed that SPTB expression was positively correlated with N-cadherin, vimentin, Snail, MMP2, and MMP9 expression ( Fig. 7 , F and G and Supplemental Fig. S7 , E and F ). Fig. 7 SPTB promotes human trophoblast cell proliferation, invasion, and migration in vitro . A and B, basal expression of DLK1, SPTB, COL21A1, and SAR1A in two human trophoblast cell lines determined via WB and immunofluorescence assays. C, an EdU assay was used to verify the change in cell proliferation after induction of SPTB knockdown in HTR-8/SVneo and Swan-71 cells (200×) compared with control (Ctrl) cells. D and E, the invasion and migration capacities of trophoblast cells were evaluated using transwell and wound healing assays. F, the protein expression of EMT markers in HTR-8/SVneo and Swan-71 cells was detected by WB. G, quantification of the WB results for the SPTB and EMT protein markers. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. The data are expressed as the mean ± SD of at least three independent experiments. DLK1, delta-like homolog 1; EMT, epithelial‒mesenchymal transition; WB, Western blot. SPTB promotes human trophoblast cell proliferation, invasion, and migration in vitro . A and B, basal expression of DLK1, SPTB, COL21A1, and SAR1A in two human trophoblast cell lines determined via WB and immunofluorescence assays. C, an EdU assay was used to verify the change in cell proliferation after induction of SPTB knockdown in HTR-8/SVneo and Swan-71 cells (200×) compared with control (Ctrl) cells. D and E, the invasion and migration capacities of trophoblast cells were evaluated using transwell and wound healing assays. F, the protein expression of EMT markers in HTR-8/SVneo and Swan-71 cells was detected by WB. G, quantification of the WB results for the SPTB and EMT protein markers. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. The data are expressed as the mean ± SD of at least three independent experiments. DLK1, delta-like homolog 1; EMT, epithelial‒mesenchymal transition; WB, Western blot. SPTB is a cytoskeletal protein that evolved from α-actins and participates in angiogenic regulation ( 32 ). To evaluate the effects of SPTB on the cytoskeleton network of trophoblasts and the angiogenic potential of endothelial cells, we performed immunofluorescence staining of SPTB with F-actin and tubulin and performed a tube formation assay with human umbilical vein endothelial cells. The results showed that SPTB promoted the tubular formation of endothelial cells ( Fig. 8 A and Supplemental Fig. S7 D ). Immunofluorescence results revealed that SPTB, F-actin and tubulin were colocalized in trophoblast cell lines in vitro to varying degrees ( Fig. 8 B ). Furthermore, WB analysis revealed that SPTB expression was positively correlated with actin- and tubulin-related cytoskeletal proteins ( Fig. 8 , C and D and Supplemental Fig. S7 , G and H ). Together, these results indicate that SPTB facilitates the proliferation, migration, invasion, and tube formation of trophoblast cells in vitro and positively affects the cytoskeletal network. Fig. 8 SPTB maintains trophoblast angiogenesis and the cytoskeleton network in vitro . A, tube formation capacity of human umbilical vein endothelial cells (HUVECs) cultivated for 6 to 8 h in conditioned medium from transfected cells (100×). B, immunofluorescence assay showing the colocalization of SPTB with F-actin and tubulin in HTR-8/SVneo and Swan-71 cells. The scale bar represents 50 μm. C, the expression levels of cytoskeleton-related proteins in the trophoblast cell lines HTR-8/SVneo and Swan-71 were detected by WB after transfection with SPTB siRNA or the control. D, quantification of the WB band intensities for SPTB and cytoskeleton-related proteins. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. The data are expressed as the mean ± SD of at least three independent experiments. WB, Western blot. SPTB maintains trophoblast angiogenesis and the cytoskeleton network in vitro . A, tube formation capacity of human umbilical vein endothelial cells (HUVECs) cultivated for 6 to 8 h in conditioned medium from transfected cells (100×). B, immunofluorescence assay showing the colocalization of SPTB with F-actin and tubulin in HTR-8/SVneo and Swan-71 cells. The scale bar represents 50 μm. C, the expression levels of cytoskeleton-related proteins in the trophoblast cell lines HTR-8/SVneo and Swan-71 were detected by WB after transfection with SPTB siRNA or the control. D, quantification of the WB band intensities for SPTB and cytoskeleton-related proteins. Note: ∗, ∗∗, ∗∗∗, and ∗∗∗∗ represent p values < 0.05, <0.01, <0.001, and <0.0001, respectively. The data are expressed as the mean ± SD of at least three independent experiments. WB, Western blot.

In this study, we describe the first proteomic characterization of EPL according to histological subtype. Our high-dimensional quantitative proteomic analysis revealed numerous differences among CHMs, PHMs, HAs, and normal placentas. We revealed thousands of DEP associations, and their biological relevance was assessed through pathway analyses. CHM is the subtype with the most drastic differences and can be clearly distinguished from the other groups. We also demonstrated the high diagnostic power of four protein classifiers DLK1, SPTB/COL21A1, and SAR1A. Among the DEPs in each group, 63.78% were downregulated, including proteins involved in the peroxisome, ROS, and oxidative phosphorylation pathways in CHMs; glycolysis and carbon metabolism in PHMs, and the biosynthesis of nucleotide sugars in HAs. The upregulated DEPs were primarily involved in DNA replication or mismatch repair and complement and coagulation cascades in CHMs and collagen family members in PHMs. A very limited number of studies have previously examined biomarkers and biological pathways of EPL including HMs. Among these studies, the ROS and HIF-1 pathways, complement and coagulation cascades, and glycolysis carbon metabolism were the most commonly reported pathways ( 33 , 34 , 35 ). The human placenta is classified as hemomonochorial, in which maternal blood directly bathes the fetal trophoblast. During the first trimester, severe hypoxic conditions are present in the placental microenvironment ( 36 ), and the production of potentially harmful ROS is limited to protect the embryo from free radical-mediated teratogenesis ( 37 ). Our pathway analysis revealed that the HIF-1 signaling pathway was suppressed in PHMs and HAs but activated in CHMs ( Fig. 3 E). This effect is probably due to the premature onset of maternal circulation throughout the placenta in most EPL patients ( 38 , 39 ), especially those with PHMs and HAs, resulting in the dysregulation of adaptation to hypoxia and the suppression of HIF-1 signaling pathways. The presence of oxidative stress in CHM has been previously reported ( 40 ), which is consistent with our data. The distinct manifestations of CHM can be attributed to two factors. First, combined with the downregulation of the oxidative phosphorylation pathway, the CHM group appeared to have severe mitochondrial dysfunction. This apparent dysfunction may be related to the poor quality of oocytes because the mitochondria of CHMs are still supplied by maternal sources ( 41 , 42 ). Second, the lower prevalence of normal endovascular trophoblast invasion in CHMs leads to more severe hypoxic conditions and activation of the HIF-1 pathway ( 38 ). Our data also revealed reduced expression of key glucose-related metabolic proteins, such as GPI and HK2, in EPL, especially in PHM and HA. Early placental tissue exhibits high proliferation rates, and the energy needed comes principally from glycolysis ( 43 ). Trophoblast proliferation, differentiation, and invasion are regulated by low oxygen levels through the HIF signaling pathway ( 44 ), and ROS inhibit multiple glycolytic enzymes. Here, we observed the accumulation of ROS and the inhibition of the HIF-1 signaling pathway in PHMs and HAs. Thus, we hypothesized that decreased HIF-1 leads to decreased glycolysis and carbon metabolism in the EPL. Consequently, the shortage of energy production compromises embryo development, but the outer part of the conceptus, such as the trophoblast, may survive because of the surrounding maternal nutrients, causing a phenotype resembling a HM ( 34 ). After the discovery of several placenta-specific EPL-associated proteins, this study aimed to develop an early diagnostic biomarker panel for EPL subtypes. In this study, 14 of the DEPs were verified by PRM assays, demonstrating their ability to effectively categorize patients with different EPL subtypes. Among these candidates, four markers (DLK1, SPTB, COL21A1, and SAR1A) were ultimately selected by LASSO logistic regression. DLK1 is a paternally imprinted but maternally expressed gene. As indicated by our data, DLK1 was significantly upregulated in HM, especially in CHM. In the etiology of HMs, overrepresentation of the paternal genome leads to global alterations in imprinted gene expression in the molar trophoblast ( 13 ), so this finding was expected. The developmentally important DLK1-DIO3 imprinted domain on human chromosome 14 has been reported to be negatively associated with fetal growth ( 45 , 46 ). According to the parental conflict hypothesis ( 47 ), the paternal genome tends to maximize resources for its progeny, resulting in the hyperplastic nature of HM ( 48 ). Type XXI collagen, which is encoded by the COL21A1 gene, belongs to the FACIT collagen family ( 49 ). Notably, for the first time, we report the overexpression of collagen family proteins in PHMs. Collagen is an important component of the extracellular matrix in mammals, and trophoblasts secrete various types of collagen, including type IV, VI, and XXI collagen ( 50 , 51 ). Pregnancy-associated hormones and hypoxia-induced upregulation of HIF-1α could promote collagen deposition and secretion ( 52 , 53 , 54 ). Collagen overexpression facilitates trophoblast invasion, induces immune tolerance and promotes angiogenesis at the maternal-fetal interface ( 55 , 56 , 57 ). To our knowledge, the function of COL21A1 in human reproduction has not been reported by other researchers thus far. Therefore, the role of COL21A in embryogenesis requires further investigation. In our study, SPTB was selected for further in vitro functional exploration. On the one hand, SPTB was found to be upregulated in the PHM group, and has the potential to be used as a biomarker to correct the high misdiagnosis rate of PHM. On the other hand, the spectrin family has been reported to be related to embryonic development. This gene, encoding the β-I spectrin subunit, was initially discovered in erythrocytes and subsequently confirmed to be expressed in the membrane skeleton of all metazoan cells ( 32 ). In a Drosophila model, β-spectrin mutations were lethal during early development, and in mice, β-II-spectrin–depleted embryos underwent only two rounds of cell division and failed to compact ( 30 , 58 ). Interestingly, spectrin was found predominantly in the form of a narrow condensed subplasmalemmal band ( 59 , 60 ), colocalizing with the subcortical maternal complex (SCMC). Mutations in genes encoding SCMC lead to adverse reproductive outcomes in humans, including HMs ( 61 , 62 ). Thus, the interaction and colocalization between SPTB and SCMC suggest that they may have a subtle connection to the pathogenesis of EPL. In our in vitro study, proliferation, invasion, migration, and angiogenesis were positively related to SPTB expression in trophoblast cell lines. Recent studies have indicated that β-spectrin regulates cytoskeleton and cell dynamics by interacting with cadherin and catenin, which is consistent with our results ( 60 , 63 , 64 ). Ras-related GTPase 1A (SAR1A) was upregulated in the NC group and has not been previously reported to be involved in human reproduction. SAR1A plays a key role in controlling vehicle transport between the endoplasmic reticulum and Golgi ( 65 , 66 , 67 ) and might affect early pregnancy through autophagosome biogenesis ( 68 , 69 , 70 ). Our observations of extensive proteomic alterations and their connection to embryonic development may help to explain the pathophysiology of EPL. Future investigations on EPL should focus on identifying upstream regulators. We recognize several limitations of this study. First, historical incidence rates of HMs have been reported to be approximately 1 per 1000 pregnancies in China ( 71 ). This low incidence, coupled with the scarcity of fresh samples, particularly PHMs, constrained the size of our MS cohorts. The resulting limited and imbalanced sample sizes may introduce statistical biases. The focus of our study on a Chinese patient population may not fully represent the global population, which inevitably introduces the possibility of bias toward particular demographics. Additionally, the absence of a public database for HMs precludes external validation of our findings, limiting our study to proteomic analysis. Future research will benefit from multiomics approaches and multicenter collaborations to expand the study of EPL histological subtypes. While the four-protein signature we identified shows promise as a novel diagnostic tool and initial efficacy in an independent cohort, further extensive validation is essential for clinical implementation. Large-scale, multicenter trials will be instrumental in confirming the robustness of the signature and in developing standardized assays for clinical translation. In summary, our study revealed dysregulated proteins and pathways in different histological subtypes of EPL. We also developed a diagnostic signature composed of DLK1, SPTB, COL21A1, and SAR1A, which showed high diagnostic power for patients with EPL. Additionally, we conducted further investigations of SPTB as a biomarker for PHMs, and explored its functions in trophoblast cell lines to elucidate its role in the pathogenesis of PHM. Moreover, a rich pool of DEPs was identified in this study, including potentially important DEPs that were not included in our final diagnostic panel. Nonetheless, these additional DEPs should not be underestimated, and further exploration of these potential factors is warranted for the development of diagnostic, prognostic, and even intervention strategies.

This study was approved by the Institutional Review Board of the First Affiliated Hospital of Zhejiang University (No. 2020IIT339), and written consent was obtained from the patients. All procedures were conducted in accordance with the Declaration of Helsinki. POC samples, including villi or molar tissues and maternal decidua, were obtained in the first trimester by therapeutic surgical dilation and curettage. For MS analysis and WB analysis, a total of 67 fresh POC tissues (17 CHMs, 16 PHMs, 17 HAs, and 17 NCs) were prospectively collected from patients at the Department of Obstetrics and Gynecology of the First Affiliated Hospital, Zhejiang University, from March 2019 to July 2023. The fresh tissue samples were rinsed with precooled sterile saline, examined under a stereomicroscope to separate the villi and decidua, and then dissected into aliquots and stored at −80 °C until use. Another corresponding aliquot was formalin-fixed and paraffin-embedded (FFPE) at room temperature for subsequent microscopic examination and IHC. Blood or buccal swab samples were also collected from these patients or their partners for comparative genotyping and stored at −80 °C until use ( 23 ). For tissue immunofluorescence and IHC analysis, 120 FFPE tissue samples (30 CHMs, 30 PHMs, 30 HAs, and 30 NCs) were retrieved from various collaborators’ pathology laboratories in Zhejiang Province. All POC samples were initially diagnosed by histological examination, p57 analysis, Ki-67 immunohistochemical analysis, and molecular genotyping with short tandem repeat (STR) markers (see the Supplementary materials and methods). According to the diagnostic criteria for HMs ( 24 ) and the clinical consensus of three experienced pathologists (X. Z., H. Z., B. H.), each sample was reclassified as CHM, PHM, HA, or NC for subsequent discovery and validation cohorts. Medical history, pathological results, ultrasound performance, and biochemical data were obtained at enrolment ( Supplemental Table S1 ). The purpose of this study was to identify diagnostic biomarkers through comprehensive proteomic analysis of POC tissues from EPL patients. The workflow is schematically summarized in Figure 1 , and includes four phases: (1) discovery MS and bioinformatic analyses; (2) targeted MS and biomarker selection; (3) analysis by IHC and WB; and (4) in vitro validation using cell lines. Fig. 1 Over all experimental design and schematic of the proteomic analyses of EPL. High-resolution MS-based and targeted proteomic profiling for subtype proteomic analysis, diagnostic signature construction, and validation. CHM, complete hydatidiform mole; EPL, early pregnancy loss; HA, hydropic abortion; MS, mass spectrometry; NC, normal control; PHM, partial hydatidiform mole. Over all experimental design and schematic of the proteomic analyses of EPL. High-resolution MS-based and targeted proteomic profiling for subtype proteomic analysis, diagnostic signature construction, and validation. CHM, complete hydatidiform mole; EPL, early pregnancy loss; HA, hydropic abortion; MS, mass spectrometry; NC, normal control; PHM, partial hydatidiform mole. Phase 1 included 36 frozen fresh samples (10 CHMs, 6 PHMs, 10 HAs, and 10 NCs). A label-free proteomics workflow was used to process these 36 samples as biological replicates for quantitative proteomic analysis. The resulting data were processed using the Proteome Discoverer search engine (v 2.4.1.15, Thermo Fisher Scientific; https://thermo.flexnetoperations.com/control/thmo/login?nextURL=%2Fcontrol%2Fthmo%2Fdownload%3Felement%3D11324157 ). Proteins identified in all replicates with “unique peptides” >1 and a <1% false discovery rate (FDR) at the peptide-spectrum match (PSM), peptide, and protein levels were considered. Due to the scarcity of fresh samples of HMs (especially PHMs), we initially enrolled 10 pairs of samples but six PHMs for MS discovery in phase 1. During the study, three new PHM cases were collected, diagnosed, and included to expand the cohort for targeted MS analysis. Therefore, 39 frozen fresh tissues (10 CHMs, 9 PHMs, 10 HAs, and 10 NCs) were included in phase 2. We selected 18 proteins for parallel reaction monitoring (PRM) verification. Lists of all peptides targeted in the PRM analyses are provided in Supplemental Table S6 . The resulting MS data were processed using the MaxQuant search engine (v.1.6.15.0; https://www.maxquant.org/ ) and Skyline (v. 21.1; https://skyline.ms/project/home/software/skyline/begin.view ), and the protein expression, LASSO algorithm, and ROC analysis were used for biomarker selection. In phase 3, we validated the expression of four candidate proteins (DLK1, SPTB, COL21A1, and SAR1A) in a larger independent cohort using WB analysis of 28 frozen fresh tissue samples and IHC analysis of 120 FFPE samples, with equal samples across each group. In the last phase, we explored one of the biomarkers, SPTB, and its biological functions in vitro using two human trophoblast cell lines. Proteins were extracted from villous tissue samples from patients and digested according to the standard proteomic sample preparation. The sample was ground into the cell powder with liquid nitrogen and lysed with buffer (1% Triton X-100, 1% protease inhibitor cocktail), followed by sonication (Scientz). The protein concentration was then measured. For digestion, each protein was added to the concentration of 20% (m/v) trichloroacetic acid, then vortexed to mix, and incubated for 2 h at 4 °C. The precipitated protein was collected and then washed within 200 mM triethylamine bicarbonate and ultrasonically dispersed. Trypsin was added at 1:50 trypsin-to-protein mass ratio for the first digestion overnight. The sample was reduced with 5 mM DTT for 60 min at 37 °C and alkylated with 11 mM iodoacetamide for 45 min at room temperature in darkness. Finally, the peptides were desalted by Strata X SPE column. Peptides were dissolved in solvent A (0.1% formic acid, 2% acetonitrile in water) and loaded onto a reversed-phase analytical column (25 cm length, 100 μm i.d.) for separation via an EASY-nLC 1200 UPLC system (Thermo Fisher Scientific) with a gradient from 6% to 80% solvent B (0.1% formic acid, 90% acetonitrile in water). The separated peptides were analysed using an Orbitrap Exploris 480 instrument (coupled with the FAIMS Pro interface) with a nanoelectrospray ion source. The electrospray voltage applied was 2300 V. The FAIMS compensation voltage was set to −45 V and −65 V. Precursors and fragments were analyzed using the Orbitrap detector. The full MS scan resolution was set to 60,000 for a scan range of 400 to 1200  m/z . The tandem mass spectrometry (MS/MS) scan was fixed first at 110  m/z at a resolution of 15,000, with the TurboTMT set as off. The automatic gain control was set at 100%, with an intensity threshold of 5E4 ions/s and a maximum injection time of “Auto.” The resulting MS data were processed using the Thermo Proteome Discoverer search engine (v 2.4.1.15). Tandem mass spectra were searched against the human SwissProt database (20,387 entries) concatenated with reverse decoy and contaminant databases. Trypsin (Full) was set as the cleavage enzyme, allowing up to 2 missed cleavages. The minimum peptide length was 6, with a maximum of 3 modifications per peptide. The mass errors were 10 ppm for precursor ions and 0.02 Da for fragment ions. Carbamidomethylation of Cys was fixed, whereas oxidation of Met, acetylation of the protein N terminus, Met loss (M), and Met loss + acetyl (M) were variable modifications. The FDRs of the proteins, peptides and PSMs were adjusted to <1%. Another aliquot of the same samples as those in the discovery cohort was taken, and three new PHM samples were added for targeted proteomics analysis. PRM methods were developed according to Tier 3 level analyses ( 25 ). PRM analysis was subsequently developed and applied to verify that the differentially expressed peptides combined with significant candidate proteins. Each protein was quantified using two unique peptides ( Supplemental Table S6 D ). The selection of the unique peptides was based on the results of the sample data dependent acquisition pilot experiment combined with the search results from MaxQuant 1.6.15.0. Prioritizing those with high scores, appropriate lengths, and strong signals, the m/z ratios and charges were confirmed and the peptides were used further for PRM analysis. PRM was performed on an EASY-nLC 1000 UPLC system (Thermo Fisher Scientific). The separated peptides were analyzed using a Q Exactive Plus instrument with a nanoelectrospray ion source. There was a full scan window followed by, a selective scan of precursor m/z listed in supplemental Table S6 C . The full MS scan resolution was set to 70,000 for a scan range of 360 to 1138  m/z . The fragments were detected using an Orbitrap at a resolution of 17,500. Higher energy collisional dissociation fragmentation was performed at a normalized collision energy of 27%, with a maximum injection time of 50 ms for MS and 210 ms for MS/MS. The isolation window was set at 1.6  m/z ( Supplemental Table S6 C ). The resulting MS/MS data were processed using the MaxQuant search engine (v.1.6.15.0) for database searching to construct spectral libraries for targeted peptides. Tandem mass spectra were searched against against the human SwissProt database (20,387 entries) concatenated with reverse decoy and contaminant databases. The mass tolerance for precursor ions was set as 20 ppm in first search and 4.5 ppm in main search, and the mass tolerance for fragment ions was set as 20 ppm. FDR of protein, peptide, and PSM was adjusted to <1%. The resulting MS data were processed using Skyline (v. 21.1) by considering the peak areas of the peptide fragment ions. The peptide settings were as follows: the enzyme was set to trypsin [KR/P], and the maximum number of missed cleavages was set to 0. The peptide length was set as 7 to 25 amino acid residues, and cysteine alkylation was set to a fixed modification. The transition parameters were as follows: the parent ion charge was set to 2 or 3, the daughter ion charge was set to 1, and the ion type was set to b or y. The selection of fragment ions ranged from third to last, and the tolerance for mass error in ion matching was set to 0.02 Da. Paraffin 4-μm sections were cut from the tissue blocks and subjected to H&E and IHC staining. The markers used in this study included p57, Ki-67, DLK1, SPTB, COL21A1, and SAR1A. The antibodies used are listed in Supplemental Table S8 . Immunostaining was performed with a universal two-step detection kit (ZsgbBio) following the manufacturer's instructions. The histological slides were independently and blindly evaluated by three pathologists (X. Z., H. Z., B. H.). The examiners were instructed to reach a consensus on discordant cases. The extent of biomarker antibody staining on every slide was semiquantitatively assessed, and both the staining intensity and area were assessed ( Supplemental Table S7 ). Examples of the intensity grading are depicted in Supplemental Fig. S5 . Equal amounts of protein were resolved by 4 to 12% SDS‒PAGE and then transferred onto polyvinylidene fluoride membranes. The membrane was blocked with 5% skim milk for 2 h at room temperature. Following overnight incubation with primary antibodies, anti-horseradish peroxidase–conjugated anti-rabbit/mouse IgG (H + L) was used (Proteintech). The immunoblots were imaged with an enhanced chemiluminescence Western blotting kit (Biosharp). Two human trophoblast cell lines, HTR-8/SVneo and Swan-71, were purchased from the American Type Culture Collection and the Cell Bank of China, respectively. HTR-8/SVneo and Swan 71 cells were maintained in RPMI-1640 medium (Gibco) and high glucose-Dulbecco's modified Eagle's medium, respectively. Medium containing 10% fetal bovine serum (Gibco) and 1% penicillin‒streptomycin (Gibco) was added to the cells before placing the cells in an incubator at 37 °C with 5% CO 2 . Detailed in vitro biological function experiments see the Supplementary materials and methods. The relative quantitative value of protein (R) was log2-transformed, and then the transformed R values were compared between groups using the DESeq2 package (version 1.40.2; https://bioconductor.org/packages/release/bioc/html/DESeq2.html ). Adjusted p values were computed using the Benjamini–Hochberg method. Proteins with a |log2Foldchange|>0.5 and adjusted p value less than 0.05 were defined as DEPs. The list of significant DEPs was further curated from the Protein Atlas and GeneCards databases and the literature to identify interesting or placenta-related DEPs. For the selection of the most useful prediction candidates, the least absolute shrinkage and LASSO methods were used. The LASSO logistic models were fitted using the R package “glmnet.” For visualizations, features with a nonzero coefficient value in at least half of the tests were selected. Contributions to the models were calculated on the basis of percentage absolute values. The area under the receiver operating characteristic curve (AUC) was used to evaluate the performance of the prediction model. Principal component analysis (PCA), enrichment-based clustering, Mfuzz expression pattern clustering ( 26 ), gene ontology analysis, and Kyoto Encyclopedia of Genes and Genomes pathway annotation were performed for proteome profiling, and the details are presented in the Supplementary materials and methods. Different visualization plots were generated using R (version 3.6.3) packages and GraphPad Prism software (version 9.0.2; https://www.graphpad.com/ ). Figure 1 was created with Biorender.com . The data are presented as the means ± SDs. The comparison of significance was evaluated by a two-sided Student’s t test for two groups and a one-way ANOVA test with Tukey’s post hoc test for three or more groups. AUC-ROC curves were calculated to evaluate the diagnostic performance of the LASSO model and the immunostaining scores of the biomarkers. Statistical analyses were performed using SPSS (version 19; https://www.ibm.com/spss ), Microsoft Excel ( https://www.microsoft.com/zh-cn/microsoft-365/excel ), and GraphPad Prism software (version 9.0.2) unless otherwise specified. A p value < 0.05 was considered statistically significant.

The authors declare no competing interests.

The label-free proteomics data are deposited in the PRIDE ( PXD055182 ). Username: [email protected] . Password: Wx1fzrESybaC. The PRM proteomics data are deposited in the PRIDE ( PXD053739 ). Username: [email protected] . Password: o24GGQL3Np2b.

This article contains supplemental data ( 72 , 73 ).