Oncomine is a publicly available database (https://www.oncomine.org). It consists online cancer microarray data of 715 datasets, 86,733 cancer and normal tissue samples (15). The Oncomine database was used to illustrate the transcriptional level of CD47 in ovarian cancer. The thresholds were set as follow: P-value<1e-4, fold change >2, and gene rank in the top 10%.

The GEPIA database (http://gepia.cancer-pku.cn/) consists of RNA sequencing expression data of 9,736 tumors and 8,587 normal samples derived from the TCGA and GTEx database (16). CD47 expression level in different cancer types versus normal tissues were processed using GEPIA. P<0.05 was considered to be statistically significant.

CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/home.jsp) is the first dedicated database to analyze distinct functional states of different cancer cells at single-cell level (13). It involves 14 functional states of 41900 cancer single cells in 25 types of different cancers. Cellular functional states include angiogenesis, apoptosis, cell cycle, differentiation, DNA damage, DNA repair, Epithelial-Mesenchymal Transition (EMT), hypoxia, inflammation, invasion, metastasis, proliferation, quiescence and stemness. The CancerSEA database was used to analyze CD47 function in 14 different states using cancer single cell data.

Ualcan (http://ualcan.path.uab.edu/) is a comprehensive web portal for analyzing cancer OMICS data (17). In this study, expression of CD47 in ovarian cancer was analyzed in Ualcan based on individual cancer stages, tumor grades, patient’s race, patient’s age and TP53 mutation status. The expression level of CD47 was normalized as transcript per million reads. P<0.05 was considered statistically significant.

HPA (https://www.proteinatlas.org/) is an open access protein database (18). It is composed of proteomic data based on 26941 antibodies targeting 17165 unique proteins. CD47 protein expression level between ovarian cancer and normal ovary was verified using immunohistochemistry (IHC).

The KM plotter (http://kmplot.com) database evaluates survival of different genes in 21 cancer types including breast (n=6234), ovarian (n=2190), lung (n=3452) and gastric (n=1440) cancer (19). Prognostic values including PFS (progression-free survival) and OS (overall survival) of CD47 were evaluated in ovarian cancers. Four different probes were used to evaluate CD47 related PFS and OS in ovarian cancer. Hazard ratios (HRs) with 95% confidence intervals and logrank P-value were determined.

TISCH collected data from Gene Expression Omnibus (GEO) (20) and ArrayExpress (21) to formulate its scRNA-seq atlas (14). TISCH includes 79 databases and 2045746 cells from both tumor patients and healthy donors. The datasets were uniformly processed to enable clarifying components of the TME at both single-cell and annotated cluster levels. In this work, we used datasets derived from TISCH to decipher the TME heterogeneity between primary and metastatic ovarian tumor sites at single cell level.

GeneMANIA (http://genemania.org/) is a flexible website for predicting protein-protein interactions based on gene functions (22). It indexes 2, 813 interactive functional association networks which includes 660, 443, 499 interactions derived from 9 organisms. In this work, we adopted GeneMANIA to construct the network of CD47 interactive proteins based on physical interactions, co-expression, predicted, co-localization, pathway, genetic interactions, and shared protein domains.

TIMER (https://cistrome.shinyapps.io/timer/) is a database for comprehensive analysis of tumor-infiltrating immune cells (23). TIMER database consists of 10897 samples of 32 cancer types from the TCGA database to evaluate immune infiltrates abundance. CD47 expression correlation with six immune infiltrates (B cells, CD4+ T cells, CD8+ T cells, Neutrophils, Macrophages and Dendritic cells) were estimated using TIMER. The gene expression level was assessed using log2 TPM. CD47 was also evaluated with different gene markers of B cell, T cell, CD8+ T cell, Dendritic cell, macrophage type1 cells (M1), macrophage type2 cells (M2), Th1 cell, Th2 cell, regulatory T cells (Treg), monocyte, tumor-associated macrophages (TAM) and T cell exhaustion.

We used TCGA OV RNA-Seq data to find DEGs between CD47 high and low expression groups with R package DESeq2. There are 376 patients in the TCGA OV RNA-Seq dataset. We used quartile method to divide patients into CD47 high and low group. We sorted descending CD47 TPM expression value in all patients, and chose top 94 patients as CD47 high group, and bottom 96 patients as CD47 low group. The genes with P value less than 0.05 and log fold change larger than 1 or lower than -1 were considered significant. They were showed in a volcano plot with R package ggplot2. These different up-expression genes were enrichment in KEGG pathways with KOBAS web tool (http://kobas.cbi.pku.edu.cn/), P value less than 0.05 was considered significant in statistic. Besides, we also analyzed pathways in GSEA software about CD47 high expression group.

Results generated using CancerSEA was displayed with correlation and P-value. GEPIA and Ualcan results were calculated using P-value. KM plotter and log-rank tests were utilized to represent the survival curves. Results displayed using Oncomine was processed with P-values, fold changes and ranks. The correlation of CD47 with different immune infiltration levels was represented using P-value. The DEGs analyses were produced using R software (version 3.6.3). P<0.05 was considered statistically significant (*P<0.05 **P<0.01, ***P<0.001).

Oncomine database was used to probe into CD47 differential expression at pan-cancer level. The results showed high CD47 expression in head and neck cancer, kidney cancer, myeloma, ovarian cancer, pancreatic cancer and sarcoma ( Figure 1A ). Meanwhile, GEPIA database was used to further validate the results. CD47 was highly expressed in bladder urothelial carcinoma (BLCA), BRCA, CESC, cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD) and ovarian cancer ( Figure 1B ). The data was also evaluated using TIMER database, which showed significant difference (P<0.05) in BRCA (1093 tumor vs. 112 normal), CHOL, COAD (457 tumor vs. 41 normal), GBM, HNSC, LUAD (515 tumor vs. 59 normal), LUSC, PRAD, STAD, THCA and UCEC (545 tumor vs. 35 normal) ( Figure 1C ). The results showed that CD47 played different roles in distinct types of cancer. Although TIMER only investigated CD47 expression in 303 ovarian tumor samples but not in adjacent normal tissues, CD47 expression level in ovarian cancer was relatively high.

To explore how CD47 might affect different cancers, we used CancerSEA single-cell database to analyze the correlation of CD47 with 14 distinct functions in 14 types of cancer. As shown in Supplementary Figure 1 , correlation of CD47 was analyzed using an interactive bubble chart. The upper bar represents the correlation of CD47 with 14 distinct functional states of all single-cell datasets. Expression of CD47 showed positive correlation with angiogenesis, apoptosis, cell cycle, differentiation, EMT, hypoxia, inflammation, metastasis, proliferation and quiescence. The data showed that CD47 might play a tumor activator role in several types of cancer.

Functional relevance analysis showed that CD47 expression was positively correlated with metastasis in ovarian carcinoma ( Figure 2A ); with DNA repair in renal cell carcinoma (RCC) ( Supplementary Figure 2A ); with inflammation, cell cycle, DNA repair, and proliferation in breast cancer (BRCA) ( Supplementary Figure 2B ); with invasion, quiescence, proliferation, and angiogenesis in prostate cancer (PC) ( Supplementary Figure 2C ); with metastasis, invasion and EMT in head and neck cancer ( Supplementary Figure 2D ); with invasion, EMT, hypoxia and metastasis in glioblastoma (GBM) ( Supplementary Figure 2E ); with metastasis, cell cycle, proliferation and EMT in glioma ( Supplementary Figure 2F ); with inflammation, differentiation, quiescence and hypoxia in astrocytoma (AST) ( Supplementary Figure 2G ); with metastasis, EMT, inflammation and apoptosis in high-grade glioma (HGG) ( Supplementary Figure 2H ); with stemness, inflammation, differentiation and metastasis in Oligodendroglioma (ODG) ( Supplementary Figure 2I ); with no significant correlation in lung adenocarcinoma (LUAD) ( Supplementary Figure 2J ); with quiescence, inflammation, proliferation and differentiation in non-small cell lung cancer (NSCLC) ( Supplementary Figure 2K ); with metastasis, quiescence, apoptosis and inflammation in melanoma (MEL) ( Supplementary Figure 2L ).

However, CD47 is negatively correlated with invasion (R=-0.58; P<0.001) and DNA repair (R=-0.28; P<0.05) in ovarian cancer ( Figure 2A ); with differentiation (R=-0.19; P<0.01) in colorectal cancer (CRC) ( Figure 2B ). These data indicated that CD47 might play different functional roles in ovarian cancer and CRC. Therefore, in this study, we aimed to clarify functional role of CD47 in ovarian cancer.

First, we used the HPA database to confirm CD47 expression in ovarian cancer and normal ovary tissue. CD47 expression level in ovarian cancer was significantly higher than that of normal ovary tissue ( Figures 3A, B ). CD47 expression level was notably higher in ovarian cancer compared to normal ovarian tissues (P<0.05; num (T)=426; num (N)=88) ( Figure 3C ). Two independent datasets derived from oncomine showed that CD47 was highly expressed in ovarian cancer in Yoshihara Ovarian (P=3.66e-7) and Lu Ovarian (P=3.09e-8) ( Figures 3D, E ). GEPIA and Oncomine results were consistent with the HPA database.

We next investigated CD47 expression based on cancer stages using GEPIA database. Expression level of CD47 in ovarian cancer showed no significant difference among different stages ( Figure 3F ). Ualcan database demonstrated no correlation of CD47 expression with cancer stages, tumor grade, patient’s race, patients’ age or TP53 mutation status ( Figures 3G–K ). These results suggest that CD47 may serve as a diagnostic tumor marker in ovarian cancer regardless of different pathological parameters.

The KM plotter was used as an indicator for evaluating prognostic value of CD47 expression based on Affymetrix microarrays. CD47 mRNA level correlated with overall survival (OS) and progression-free survival (PFS) in ovarian cancer patients. We found that CD47 mRNA expression level was positively correlated with worse OS among all ovarian cancer patients based on four different arrays (HR=1.16 (1.02-1.33), logrank P=0.025 for 211075_s_at; HR=1.18 (1.01-1.37), logrank P=0.033 for 213857_s_at; HR=1.33 (1.09-1.63), logrank P=0.0058 for 226016_at and HR = 1.23 (1-1.51), logrank P = 0.048 for 227259_at ( Figures 4A–D ). We found CD47 mRNA expression level reversely correlated with PFS of ovarian cancer patients with HR=1.29 (1.11-1.49), logrank P=0.00059 for 211075_s_at; HR=1.29 (1.11-1.49), logrank P=0.00078 for 213857_s_at; HR=1.51 (1.25-1.83), logrank P=1.7e-05 for 226016_at and HR=1.36 (1.13-1.65), logrank P=0.0013 for 227259_at ( Figures 4E–H ). The data suggested CD47 to be a potential biomarker for predicting ovarian cancer prognosis.

We used two datasets (OV_GSE115007 and OV_GSE118828) of the TISCH database to evaluate CD47 expression in TME-related immune cells. As to different immune cell types, for instance, plasma cells, dendritic cells and Mono/Macro cells of OV_GSE115007, myofibroblasts, fibroblasts and endothelial cells of OV_GSE118828, the CD47 expression level was relatively low compared to expression level in malignant cells of OV_GSE118828 dataset ( Supplementary Figure 3A ). In OV_GSE118828 dataset, CD47 expression level remains the highest in malignant cells, suggesting high CD47 expression in malignant ovarian cancer cells ( Supplementary Figure 3A ). As to other components of the TME, relatively higher CD47 expression level was observed in CD4 T conventional cells and mono/macro cells of OV_GSE118828 dataset. Violin plot showed the same trend of CD47 expression in the ovarian cancer cell microenvironment ( Supplementary Figure 3B ). In OV_GSE115007, only three cell types were found, including plasma cells, dendritic cells and Mono/Macro cells, while DC cells exhibited the most abundant cell counts (n=3415) ( Figures 5A, B ). And all of the three cell types exhibited CD47 expression at a lower level compared to malignant cells in OV_GSE118828 dataset. While in OV_GSE118828 dataset, six cell types were found, with the highest number of 793 to malignant cells ( Figures 5C, D ). Because it was reported that the OV_GSE115007 contains only tumor ascites from primary ovarian cancer patients and OV_GSE118828 dataset contains both primary and metastatic tumor sites (24, 25), we got the fact that the cell components and distribution in primary ovarian cancer ascites and metastatic ovarian cancers are quite distinct. Figures 5E, F represented the distribution of various immune cells related to Figures 5B, D . These results suggest that CD47 expression level was quite different in distinct cell types with the highest in malignant ovarian cancer cells and the cell components of primary and metastatic patients are different, which might be the source of ovarian cancer microenvironment heterogeneity.

To further clarify ovarian cancer microenvironment and how CD47 might affect ovarian cancer heterogeneity and prognosis, we investigated whether CD47 was related to immune infiltration. TIMER database was used to estimate immune infiltration levels in ovarian cancer. CD47 expression was positively correlated with CD8+ T cell infiltration (r=0.102, p=2.48e-02), CD4+ T cell (r=0.107, p=1.90e-02), Neutrophil (r=0.246, p=4.50e-08) and dendritic cell (r=0.162, p=3.71e-04), negatively correlated with tumor purity (r=-0.17, p=1.73e-04). No correlation was observed with B cell (r=0.074, p=1.04e-01) and macrophage (r=0.076, p=9.81e-02) ( Figure 6A ). In addition, we evaluated CD47 expression with other immune cells markers. The results showed that CD47 was not correlated with B cell ( Figure 6B ), CD8+ T cell ( Figure 6D ), dendritic cell ( Figure 6E ) and Th2 cell ( Figure 6I ), but correlated with other immune cells of T cell ( Figure 6C ), M1 cell ( Figure 6F ), M2 cell ( Figure 6G ), Th1 cell ( Figure 6H ), Treg cell ( Figure 6J ).

To illustrate whether CD47 is correlated with other immune infiltrating cells, we also analyzed its correlation with monocyte and TAM cells. CD47 expression was positively correlated with monocyte and TAM infiltration in ovarian cancer ( Supplementary Figures 4A, B ). We then evaluated correlation of T cell exhaustion biomarkers PD1 (PDCD1), CTLA4, LAG3 and HAVCR2 with the expression of CD47 in ovarian cancer ( Supplementary Figure 4C ). The results showed that CD47 was positively correlated with T cell exhaustion. In summary, the above results suggested that CD47 expression might affect the TME through regulating immune cell infiltration.

We analyzed the interactive networks of CD47 with its interactive genes using the GeneMANIA database ( Supplementary Figure 5 ). The circle represented CD47 gene located at center of the interaction network. CD47 gene was surrounded by 20 circles representing genes showing close relation based on physical interactions, co-expression, predicted, co-localization, pathway, genetic interactions and shared protein domains. The top five genes included the signal regulatory protein gamma (SIRPG), SIRPα, src kinase associated phosphoprotein2 (SKAP2), protein tyrosine kinase 2 beta (PTK2B) and the thrombosondin1 (THBS1) gene. The five genes displayed greatest correlation with the CD47 gene. Further analysis showed that these 20 closely correlated genes were related with functions of cell adhesion mediator activity, leukocyte migration, cell-substrate adhesion, regulation of reactive oxygen species metabolic process, leukocyte cell-cell adhesion, cellular extravasation and sulfur compound binding.

To explore the association of CD47 high expression with immune regulation, we performed KEGG and GSEA analysis. First, we found CD47 high expression associated DEGs in ovarian cancer ( Figure 7A ). Then, KEGG pathway enrichment was performed to explore the biological processes related to CD47 high expression in ovarian cancer. Biological processes were mainly involved in cytokine-cytokine receptor interaction, IL-17 signaling pathway, TNF signaling pathway, chemokine signaling pathway, NF-kappa B signaling pathway, NOD-like receptor signaling pathway, PI3K-Akt signaling pathway and Toll-like receptor signaling pathway ( Figure 7B ). According to GSEA analysis, high CD47 expression was mainly involved in the process of toll-like receptor signaling pathway (NES=2.05, p-value=0.0), NOD-like receptor signaling pathway (NES=2.01, p-value=0.0), chemokine-signaling pathway (NES=1.90, p-value=0.0) and cytokine-cytokine receptor interaction pathways (NES=1.86, p-value=0.0) ( Figure 7C ). Accordingly, these above data suggest that CD47 high expression is closely linked to ovarian cancer cell immune cell infiltration and thus might affect TME and induce ovarian cancer heterogeneity.