Multi-omics identification of prognosis-associated proteins in early-stage EC and characterization of the TRAP1low/CAMSAP3low cluster

The brief process of this study was presented in Fig. S1. Clinical characteristics of stage I-II EC patients, including 15 survival and 9 dead patients, were listed in Table S1. The histopathological types of all patients were endometrioid endometrial cancer (EEC). Differentially expressed proteins (DEPs) of three categories, including tumor focal vs. para-cancerous tissues belonging to survival patients, tumor focal vs. para-cancerous tissues belonging to dead patients and tumor focal tissues of survival patients vs. dead patients, were identified (Fig. S2A). By comparing and analyzing the tumor focal tissues of survival and dead patients, the up-DEPs were significantly enriched in the oxidative phosphorylation. When comparing tumor focal and para-cancerous tissues belonging to survival patients, the up-DEPs were enriched in the synthesis of unsaturated fatty acids (Fig. S2B). 34 DEPs were performed LASSO analysis and 13 prognostic proteins were ultimately obtained through cross-matching aboved three categories: AK3, ATF2, NUMA1, CACHD1, ZMPSTE24, TRAP1, COG3, CAMSAP3, COX4I1, PGPEP1, UBL5, HADHA, and SNRPGP15 (Fig. S2C-E). The 13 candidate proteins showed good outcome prediction and diagnostic performance by ROC curves were screened (Fig. 1A and Fig. S2F). Five prognostic proteins, including TRAP1, CAMSAP3, NUMA1, UBL5, COX4I1 (Fig. 1B and Fig. S2G) were engaged in the p53 signaling pathway through GSEA analysis.

ScRNA-seq was carried out for 5 patients with stage IA EEC and 3 adjacent normal endometrial tissues from 3 patients (Table S2). After quality control and filtering, 79,641 high-quality cells were obtained, 5 main known cell types were clustered and identified: epithelial cells, stromal fibroblasts cells, endothelial cells, smooth muscle cells and immune cells (Fig. S3A-C). Through differential expression analysis via scRNA-seq, 3 differentially expressed genes (DEGs) consistent with the proteomic data: NUMA1, TRAP1 and CAMSAP3 (Fig. S3D), and their expression levels were showed in different clusters (Fig. S3E). The proportion of epithelial cells was significantly greater in tumors than in normal samples (Fig. S3F) and previous report has shown that EC originated from unciliated epithelial cells [4]. Therefore, the epithelial cells were firstly classified into 3 prime subclusters: luminal cells, glandular cells and ciliated epithelial cells (Fig. S3G-I). After excluding the ciliated epithelial cells, the epithelial cells were re-clustered, yielding 10 clusters. Then the expression levels of 3 genes were assessed in each cluster and the gene expressions of clusters 6 and 7 were lower than their median (TRAP1, median = 0.096, CAMSAP3, median = 0.040, Nuclear mitotic apparatus protein (NUMA1), median = 0.138) (Fig. S4A and Table S3). By analyzing the unciliated epithelial cells in tumors and para-cancer samples, cluster 7 was predominantly present in normal cells, thus cluster 6 was the focus (Fig. S4B). Moreover, almost all other cells could be classified as glandular cells or luminal cells (Fig. 1C, D and Fig. S4C). The top genes of the cluster 6 included previously reported oncogenes, such as LCN2, SAA1 and TFF3. TRAP1, NUMA1 and CAMSAP3 were highly expressed in glandular cells and luminal cells, but all lowly expressed in cluster 6 (Fig. 1E). Moreover, GSEA analysis of unciliated epithelial cells revealed that the p53 signaling pathway was downregulated in cluster 6 (Fig. 1F) and upregulated in glandular and luminal cells (Fig. S4D). These results were consistent with the GSEA of TRAP1 and CAMSAP3 derived proteomics, except NUMA1. Thus, we defined cluster 6 as the TRAP1low/CAMSAP3low cluster, and then evaluated its basic features in the TME.

Analysis of intercellular interactions with CellChat revealed that epithelial cells interacted most closely with immune cells (Fig. S5A). To identify which immune subclusters interacted intimately with the TRAP1low/CAMSAP3low cluster, the immune subclusters were assigned cell identity, including T cells, B cells, NK cells, and myeloid cells (Fig. S5B-D). The interaction intensity between the TRAP1low/CAMSAP3low cluster and myeloid cells was the greatest (Fig. 1G). The top ligand-receptors included MIF- CD74/CD44, MIF-CD74/CXCR4, MDK-NCL etc. (Fig. 1H). Subsequently, the myeloid cells were divided into three clusters: dendritic cells (DCs), macrophages and monocytes (Fig. S5E-G). The interaction between the TRAP1low/CAMSAP3low cluster and macrophages was the greatest (Fig. S5H), and the top receptor-ligand pairs almost overlapped with those in myeloid cells (Fig. S5I). The results indicated that macrophages are likely to be the main cells that interact with the TRAP1low/CAMSAP3low cluster of myeloid cells, and the major receptor-ligands include MIF-CD74/CXCL4, MIF-CD74/CD44.

Fig. 1

Multi-omics identification of prognostic proteins in early-stage EC and revealing characteristics of subtypes. A ROC curves of TRAP1, CAMSAP3 and NUMA1 in different groups. (C_D, Cancer tissues at death. N_D, Para-cancerous tissues in death. C_S, Cancer tissues in survival. N_S, Para-cancerous tissues in survival). B TRAP1, CAMSAP3 and NUMA1-related signaling pathways based on GSEA. C t-SNE plot showing clusters of epithelial types after reclustering. D Heatmap showing the top genes in different subpopulations after reclustering. E Violin plot showing the mean expression of TRAP1, CAMSAP3 and NUMA1 in the subgroups after regrouping. F GSEA analysis of the TRAP1low/CAMSAP3low cluster. G CellChat plot of TRAP1low/CAMSAP3low cluster and immune subpopulations. H Receptor-ligand pairs chords of TRAP1low/CAMSAP3low cluster and immune cells

Validation of prognostic proteins and their ability to predict mortality risk in EC patients

IHC was used to verify the expression of prognostic proteins in different samples. TRAP1 and CAMSAP3 expression was greater in the para-cancerous group than in the cancer group with different prognoses. The expression of TRAP1 and CAMSAP3 in tumor tissues of dead patients was lower than survival patients. However, the expression of NUMA1 did not change significantly in cancer and para-cancerous tissues of survival patients and in cancer tissues with different prognoses (Fig. 2A). The Youden index (Youden index = specificity + sensitivity − 1) was calculated according to the immune score of prognostic proteins, and the value corresponding to the maximum Youden index was selected as the cut-off value (TRAP1:5.5, CAMSAP3: 9.8, NUMA1: 6.95). The expression of TRAP1 and CAMSAP3 was consistent with the proteomic analysis, but the expression of NUMA1 did not match.

To further confirm the correlation between prognostic proteins expression and EC, and validate whether they can be used as independent prognostic proteins, we analyzed TCGA cohorts for external validation. The data were divided into early (stage I-II), advanced (stage III-IV) and adjacent tissues (normal). Differential expression comparison and survival analysis were performed (Fig. 2B). Then we selected the patients who died from EC (n = 57) and matched 110 surviving patients for analysis by propensity matching score (N = 167) in the TCGA database. The results showed that patients with low-expression scores which was calculated for the TRAP1 and CAMSAP3 signatures in endometrial lesions had significantly lower survival rates across the selected cohorts, which was consistent with our results (p < 0.05). The same results can be observed in patients with stage I-II EC and patients without TP53 mutation (Fig. S7 A, B). After combined analysis of TRAP1 and CAMSAP3 with TP53 mutation, it can be seen that the 5-year survival rate of patients decreased rapidly, suggesting the rapid progression of the disease (Fig. 2C). And the 5-year survival rate of patients significantly decreased, suggesting the worst prognosis of I-II EC patients (p < 0.05, Fig. S7B). A clinical predictive model was also constructed with the same data. The AUC for TP53 combined with TRAP1 and CAMSAP3 was 0.84, which was significantly (p < 0.001) greater than that for TP53 (AUC = 0.72) (Fig. 2D and Table S4). The results confirmed that the prognostic proteins we selected can be used for further research in the future.

Fig. 2

Verification of prognostic proteins in clinical. A Representative images of immunohistochemical staining for TRAP1, CAMSAP3 and NUMA1 in cancer tissues and normal tissues of EC patients with different prognosis and statistical box plot of prognostic protein expression in different groups. (*p < 0.05; **p < 0.01; ***p < 0.001.) B The expression of TRAP1 and CAMSAP3 was validated in early (Stage I-II, n = 386), advanced (Stage III-IV, n = 153) stages of EC as well as in normal patients (n = 35) through the TCGA database. (Kruskal-Wallis test, p-values are noted). The expression level of each gene was visualized by log2(TPM + 1). Box plots showed the median (center), 25–75 percentile (box), and lower whisker = smallest observation greater than or equal to lower hinge − 1.5 * IQR (interquartile range); upper whisker = largest observation less than or equal to upper hinge + 1.5 * IQR. C Combination with TRAP1, CAMSAP3 and TP53 mutation status predicted a disease outcome across TCGA EC cohorts. The median was set as the cut-off value to stratify EC patients into low-expression and high-expression groups. Survival curves were visualized by Kaplan–Meier method. D ROC curves of the predictive models were constructed by combining TP53, TRAP1 and CAMSAP3 from the TCGA database. The AUC for each model was shown