Comprehensive Analysis of the Expression and Prognostic Signicance of THBSs in Breast Cancer

Background: Breast cancer is one of the most common tumors for women worldwide. Thrombospondins (THBSs) are reported to play important roles in various cellular processes and are involved in the occurrence and development of human cancers. However, the expression and prognostic value of THBSs family in breast cancer remain unclear. Methods: In this study, we examined the genes and protein expression levels of THBSs and their prognostic value by synthesizing several mainstream databases, including Oncomine, Human Protein Atlas (HPA), UALCAN, and KM Plotter. We also analyzed THBS interaction networks, genetic alterations, functional enrichment, and drug sensitivity with several publicly accessible databases, including GEPIA, GeneMANIA, STRING, cBioPortal, Metascape and NCI-60 database. Results: The results showed that the mRNA expression levels of THBS1, THBS2, THBS3, and THBS5 in breast cancer tissues were signicantly higher than in normal tissues. The mRNA expression levels of THBS4 were different in different subtypes of breast cancer, and the protein expression levels of THBS1, THBS2, and THBS4 in breast cancer tissues were higher than in normal breast tissues. Survival analysis showed that breast cancer patients with high THBS1 gene expression showed worse overall survival (OS), relapse-free survival (RFS), and post-progression survival (PPS), and breast cancer patients with high THBS2 gene expression also showed worse RFS. Conversely, lower THBS3 levels predicted worse RFS, and lower THBS4 levels predicted worse OS, RFS, and distant metastasis-free survival (DMFS). Conclusions: These results suggest that THBSs may be potential biomarkers for breast cancer.

breast cancer is still not clear due to its heterogeneity, and the 5-year survival rate of breast cancer patients needs to be further improved [15]. Therefore, the search for new highly speci c and sensitive biomarkers and new protein families will help clarify the molecular mechanism of breast cancer patients, improve the prognosis of breast cancer patients and individualize patient treatment.
Thrombospondins (THBSs) are extracellular matrix proteins that were rst discovered in human cells and include ve homologous proteins. They are involved in angiogenesis, proliferation, apoptosis, NO-cGMPdependent protein kinase pathway and transforming growth factor-β (TGF-β) activity, embryonic development, wound healing, synaptogenesis and tumorigenesis [16,17]. So far, ve family members have been found, which can be divided into two subfamilies A and B, according to their structure and function. Subtype A includes THBS1 (TSP1) and THBS2 (TSP2). Subtype B includes THBS3 (TSP3), THBS4 (TSP4), and THBS5 (TSP5) or cartilage oligomeric matrix protein (COMP). Type A has thrombospondin type 1 repeat(TSRs), which consist of three prepared motif repeats and bind to transforming growth factor (TGF-β), and is considered to be the structural basis for THBS to inhibit endothelial cell proliferation and induce endothelial cell apoptosis [18]. THBS1 and THBS2 are usually present in speci c developmental stages, tissue remodeling, and speci c physiological processes of injury or in ammation and can bind with TGF-β to affect angiogenesis, while THBS5 is a constituent of the skeletal and cartilage extracellular matrix [19].
Studies have shown differential expression of THBSs in different cancers. However, few studies have focused on the expression and prognostic value of the THBSs family in breast cancer. In this study, we comprehensively examined the transcriptional level of THBSs and investigated the prognostic value of human cancers. In addition, we analyzed the interaction network, genetic alteration, functional enrichment, and drug sensitivity of THBSs.

Oncomine analysis
Oncomine (https://www.oncomine.org) is a publicly accessible online cancer database, containing gene expression array data [51,52].In our study, Oncomine was used to analyze the mRNA expression level of the THBS members in different cancers. We compared the mRNA expression of THBS members in cancer and normal tissues using Student's t-test after set the thresholds as follows: P value: 0.05 and foldchange: 2.
2.2 GEPIA dataset analysis GEPIA (http://gepia.cancer-pku.cn/index.html) is a newly developed analytical tool using a standard processing pipeline and consist of thousands of tumors and normal tissue samples data [53]. We identi ed 100 genes that were similar to each of the THBS family members.

Human protein atlas analysis
The Human Protein Atlas (https://www.proteinatlas.org/) is an online tool that contains transcriptome pro ling data and immunohistochemistry pro ling data for more than 8,000 patients and 17 major cancer types, allowing users to directly pro le the protein expression patterns of speci c genes in speci ed tumors [54].In this study, we obtained immunohistochemical images in this tool to directly compare the protein expression of different THBSs family members in human healthy and breast cancer tissues and analyze the THBSs family protein expression patterns in depth.

UALCAN
UALCAN (http://ualcan.path.uab.edu) is an interactive web resource based on level 3 RNA-seq and clinical data of 31 cancer types from TCGA database. It can be used to analyze relative transcriptional expression of potential genes of interest between tumor and normal samples and association of the transcriptional expression with relative clinicopathologic parameters [55].In this study, UALCAN was used to analyze the mRNA expressions of 5 THBSs family members in primary breast tissues and their association with clinicopathologic parameters and nodal metastasis status. Difference of transcriptional expression was compared by students' t test and p < 0.05 was considered as statically signi cant.

Kaplan-Meier plotter analysis
The Kaplan-Meier Plotter (http://www.kmplot.com) is a tool to draw survival plots with gene expression data and survival information from GEO, EGA, and TCGA cancer microarray datasets [56]. We evaluated the relevance of the mRNA expression level of ve THBS proteins to the clinical outcomes (OS, RFS, PPS, and DMFS) of untreated breast cancer patients. This tool automatically calculates the best cutoff value, log-rank p-value, hazard ratio (HR), and 95% con dence intervals (CIs).

GeneMANIA analysis
GeneMANIA (http://www.genemania.org),an online system for network analysis, can be used for predicting and visualizing the protein-protein interaction (PPI) network and gene functional assays [57] and features several bioinformatics methods: physical interaction, gene co-expression, gene co-location, gene enrichment analysis, and website prediction. In our study, GeneMANIA was used to construct the gene networks and predict the functions of THBSs.

STRING analysis
STRING (https://string-db.org/) is a database of known and predicted protein-protein interactions (PPI) [58].Herein, to detect the role of THBS family co-expressed genes, the online database of STRING was applied to analyze associations among the PPI network of THBS family co-expressed genes, and the species were set to Homo sapiens and a combined score of > 0.7 was considered statistically signi cant.
The nodes meant proteins; the edges mean the interaction of proteins and we hide disconnected nodes in the network.
2.8 cBioPortal for cancer genomics analysis cBioPortal (http://www.cbioportal.org) is an online open-access website resource for exploring, visualizing, and analyzing multidimensional cancer genomics data [59].In this study, we analyzed the genomic pro les of 5 THBS family members, which contained mutations, putative copy-number alterations from GISTIC and mRNA Expression z-Scores (RNA Seq V2 RSEM) with a z-score threshold ± 1.8.

Metascape analysis
Metascape (http://metascape.org) is a tool for gene annotation and gene list enrichment analysis [60]. In this study, the pathway and process enrichment of THBSs and similar genes were analyzed in Metascape.

Drug-Gene Interaction Network Analysis
The NCI-60 database, containing data from 60 cancer cell lines, was analyzed by the CELLMINER website (https://discover.nci.nih.gov/cellminer/) [61]. The expression status of THBSs and z-score for cell sensitivity data was downloaded from the website and assessed through Pearson correlation analysis to determine the correlation between THBSs expression and drug sensitivity.

The mRNA and protein expression of THBSs in breast cancer
We rst used the Oncomine database to analyze transcription levels of THBSs in various cancer types and the corresponding normal tissues. The results showed that there were 465, 447, 436, 434, and 438 unique assays for THBS1, THBS2, THBS3, THBS4, and THBS5, respectively. In tumor tissue, THBS1 was signi cantly increased in 40 datasets and decreased in 33 datasets. THBS2 expression is elevated in tumor tissues compared to normal tissues, especially in breast cancer, colorectal cancer, gastric cancer, head and neck cancer, lung cancer, and pancreatic cancer. For THBS3 and THBS4, the expression increased in 8 and 25 datasets, and decreased in 9 and 15 datasets, respectively. In addition, high expression of THBS5 was observed in 64 datasets, while reduced levels were found in four datasets.
Oncomine analysis showed a signi cant increase in the THBS1 gene levels in breast cancer tissues. According to the TCGA Breast Dataset, compared to normal breast tissue, THBS1 is upregulated in invasive ductal and lobular carcinoma, mixed lobular and ductal breast carcinoma and male breast carcinoma. According to the Radvanyi Breast Dataset [62], the expression of THBS1 is upregulated in invasive ductal breast carcinoma (IDC). The transcriptional levels of THBS2 were signi cantly higher in eight data sets and signi cantly lower in two datasets. In the TCGA Breast Dataset, THBS2 is upregulated in invasive breast carcinoma, invasive lobular breast carcinoma, and IDC.  Fig. 1 and Table 1.
We also compared THBS transcriptional levels in breast cancer and normal tissue using UALCAN analysis (Fig. 2). We found that THBS2, THBS3, and THBS5 were upregulated in tumor tissues, and THBS1 and THBS4 were not signi cantly different between tumor and normal tissues. In addition, the relationship between THBS gene levels and breast cancer tumor stage was also analyzed. The results showed that the expression of THBS1, THBS2, THBS3, and THBS5 were correlated with tumor stage (Fig. 3). Finally, we evaluated the association between THBSs and the lymph node metastatic status. The results showed that THBS1, 2, 3, and 5 were associated with lymph node metastatic state, while THBS4 was not associated with lymph node metastatic state (Fig. 4).
To further explore the expression of THBS proteins in breast cancer, we analyzed the immunohistochemical staining images in Human Protein Atlas (HPA). The results showed that THBS1, THBS2, and THBS3 were not expressed in normal breast tissue, while the protein expression levels of THBS1, THBS2, and THBS3 in breast cancer tissue were high, middle, and not detected, respectively. The THBS4 protein levels are low in normal breast tissue and medium in breast cancer tissue (Fig. 5). Unfortunately, we did not nd immunohistochemical images of THBS5 in breast cancer and normal breast tissue in the HPA. In conclusion, we found that the protein expressions of THBS1, THBS2 and THBS4 in breast cancer tissues were higher than in normal breast tissues.

Prognostic values of THBSs in breast cancer
Using KM plotter, we found that high THBS1 and THBS2 levels were associated with worse RFS (HR = 1.13, P = 0.006; HR = 1.13, P = 0.018) and that high expression of THBS3 and THBS4 was associated with better RFS ( Table 3. In addition, we analyzed the prognostic value of THBSs according to the ER/PR/HER2 status in Table 4. The results showed that high THBS1 expression was associated with worse RFS (

Correlation analyses of THBSs in breast cancer
We used GeneMANIA to analyze the relationship of THBSs at the gene level (Fig. 6A). Results showed that all THBSs have shared protein domains, and physical interactions were found between THBS1 and THBS2, THBS1 and THBS3, and THBS3 and THBS4. Relationships have been found in the co-expression  of THBS1 and THBS2, THBS2 and THBS5, THBS3 and THBS5, and THBS4 and THBS5. In addition, the  relationships of THBS1 and THBS2, THBS2 and THBS3, THBS3 and THBS4, and THBS4 and THBS5 were also found in website predict.
We identi ed THBS interactions at the protein expression level using STRING (Fig. 6B). THBS2 interacts with THBS1 and THBS3 based on protein homology, co-expression data, experimentally determined data, and textmining. In addition, the relationship between THBS1 and THBS2 was only noticed on gene cooccurrence.

THBSs genetic alteration in breast cancer
We analyzed the alteration of the THBS genes in breast cancer using cBioPortal.

Functional enrichment analysis of THBSs in breast cancer
To investigate the function of THBSs and their similar genes, we analyzed the GO and KEGG pathways using Metascape. First, we explored GEPIA to identify THBS-like genes in breast cancer tissues; we used the rst 100 similar genes for each THBS gene (Additional table 1). The results showed that the rst 20 GO pathways were enriched, namely collagen-containing extracellular matrix, vasculature development, collagen binding, collagen bril organization, cell-substrate adhesion, integrin binding, positive regulation of cell migration, response to transforming growth factor beta, basement membrane, anchoring junction, response to wounding, skeletal system development, heart development, negative regulation of cell proliferation, endothelium development, extracellular matrix disassembly, actin lament-based process, calcium ion binding, circulatory system process, and platelet alpha granule (Fig. 8A). The top 10 KEGG pathways were: ECM-receptor interaction, protein digestion, and absorption, proteoglycans in cancer, hypertrophic cardiomyopathy (HCM), TGF-beta signaling pathway, regulation of actin cytoskeleton, AGE-RAGE signaling pathway in diabetic complications, leukocyte transendothelial migration, Wnt signaling pathway, and hedgehog signaling pathway (Fig. 8B). Then, we carried out enrichment term network analysis of THBS genes and similar genes (Fig. 8C).

Drug sensitivity analysis
We evaluated the effect of THBSs on drug sensitivity using the NCI-60 database, which contributes to increase the precision of the treatments. Drug sensitivity was measured by a z-score, with a higher score indicating that the cells were more sensitive to drug therapy (Fig. 9). It is worth noting that the high expression of THBS1, THBS2, THBS3, and THBS4 is associated with the resistance of different cell lines to several chemotherapeutic drugs. Unfortunately, we failed to nd data on the effect of THBS5 on drug sensitivity from the NCI-60 database. In addition, we noticed that different genes had similar associations with the same drug. For example, THBS1 and THBS3 are associated with reduced cell sensitivity to vinblastine, eribulin mesilate, and actinomycin D.

Discussion
THBSs, which are overexpressed in various human cancers, can be involved in a series of cellular processes, such as angiogenesis in the tumor microenvironment [70]. However, little is known about the expression and prognosis role of THBSs in breast cancer. Therefore, our study comprehensively analyzed the transcriptional level and prognostic value of THBSs in breast cancer.
The results showed that the mRNA levels of THBS1, THBS2, THBS3, and THBS5 in breast cancer tissues were higher than those in normal tissues, while the mRNA levels of THBS4 were different in different subtypes of breast cancer. The protein expression levels of THBS1, THBS2, and THBS4 in breast cancer tissues were higher than those in normal breast tissues. It has been reported that THBS1 overexpression promotes melanoma invasion and a malignant phenotype by inducing epithelial-mesenchymal transformation of tumor cells [71]. THBS1 is highly expressed in invasive ductal carcinoma of the breast and promotes lymph node metastasis, and THBS-1 potentially could be a predictive marker for metastasis [72]. In our study, we found that high THBS1 expression was signi cantly associated with worse OS, RFS, and DMFS. In addition, high THBS1 expression was signi cantly associated with worse RFS in the basal subtypes. In the Lumina A subtypes, there was a signi cant correlation with worse PPS.
Zhang et al. demonstrated that THBS1 mRNA expression was the highest in HER2 subtype and the lowest in Luminal B subtype, and taxol resistance gene 1 and thrombospondin 1 expression may vary according to the molecular subtypes of breast cancer [25]. In addition, the tRNA-derived fragment tRF-17-79MP9PP can reduce invasion and migration of breast cancer cells via the THBS1 /TGF-β1/Smad3 axis and THBS1 as a downstream target of tRF-17, and reduction of THBS1 expression also partially recovered the effects of tRF-17 inhibition breast cancer cell viability, invasion, and migration [73]. Recent studies have shown that THBS2 is a target of miR-20a, and THBS2 knockdown could eliminate the antiproliferation, pro-apoptotic and anti-autophagy effects mediated by miR-20a inhibitors in cervical cancer cells [74]. We found that high THBS2 expression was associated with worse RFS and that high THBS2 expression was signi cantly associated with worse OS, RFS, and PPS in the Lumina B subtype and with worse DMFS and better RFS in the HER2 + subtype. CD36 can mediate the N-terminal recombinant fragment of THBS2 to activate endothelial cell apoptosis and inhibit the growth and metastasis of breast cancer [75]. Schips et al. found that THBS3 enhances traumatic cardiomyopathy through intracellular integrin inhibition and myo lm instability, and transgene-mediated overexpression of α7β1D integrin in the heart ameliorates the predisposing disease effects of THBS3 by augmenting sarcolemmal stability [76]. In our report, patients with high THBS3 expression showed better RFS. In addition, high THBS3 expression predicted better RFS of the Lumina A and Lumina B subtypes, and worse DMFS of the HER2 + subtypes. Recently, Hou et al. found that THBS4 silencing regulates the cancer stem cell-like properties of prostate cancer by blocking the PI3K/Akt pathway, and the overexpression of THBS4 promoted selfrenewal and proliferation, curbed the apoptosis of prostate cancer stem cells, and enhanced the in vivo tumorigenicity [77]. Patients with elevated THBS4 gene levels showed better OS, RFS, and DMFS. In the basal subtype, elevated THBS4 gene levels were associated with worse OS and DMFS; High expression of THBS4 in Lumina A subtype is associated with better OS, RFS, and PPS. In our study, the mRNA expression levels of THBS4 varied across different types of breast cancer, with high expression in invasive breast cancer and low expression in ductal and medullary breast cancer. Previous studies have shown that the mRNA level of THBS4 in breast cancer is variable, usually the highest in tumors with rich interstitial content (ILC, ER positive, low grade IDC, Luminal A, and normal-like subtypes) [78]. This is consistent with our study results, and THBS4 expression levels in ILC and IDC are not different; thus, THBS4 cannot be used as a biomarker to distinguish ILC and IDC [78]. Studies have shown that RvD1 inhibits the dry characteristics of CAFs-induced EMT and HCC cells by inhibiting the secretion of COMP [79]. In the HER2 + subtype, high expression of THBS5 is associated with worse RFS. Studies have shown that high expression of COMP in breast cancer cells is associated with a worse survival rate and reduced relapsed-free survival rate, moreover, due to the upregulation of matrix metalloprotease-9, cells with high expression of COMP-expressing cells had a more invasive phenotype. Finally, in vitro experiments showed that compared with control cells, Comp-expressing cells showed better survival and higher protein synthesis rates when treated with brefeldin A [50].
In this study, GeneMANIA and STRING analysis showed that co-expression was only observed at the gene level for THBS1 and THBS2, THBS2 and THBS5, THBS3 and THBS5, and THBS4 and THBS5. The gene co-expression relationship was only observed at the protein expression levels for THBS1 and THBS2. Drug sensitivity analysis showed that THBS1 and THBS3 were associated with reduced cell sensitivity to vinblastine, eribulin mesylate, and actinomycin D.
There are some limitations to our study. First, all the data analyzed were obtained from different online databases, which may lead to background heterogeneity; further studies with larger sample sizes are needed to con rm our ndings. Second, we did not conduct experiments to verify the results obtained from the bioinformatics analysis. Finally, we did not explore the molecular mechanisms of different THBSs in breast cancer. Subsequently, further in vitro and in vivo studies should be conducted to con rm our results.

Conclusions
The gene and protein expression of THBSs in breast cancer and their prognostic signi cance were veri ed in this study. In addition, we analyzed THBSs considering their co-expression and interaction networks, genetic alteration, enrichment pathways, and drug sensitivity. The results showed that, compared with normal tissue, the expression levels of THBS1, THBS2, THBS3, and THBS5 were signi cantly upregulated in breast cancer tissue. In contrast, the expression levels of THBS4 were different in different subtypes of breast cancer, and the protein expression of THBS1, THBS2, and THBS4 proteins was higher in breast cancer tissue. Survival analysis revealed that breast cancer patients with high expression of THBS1 and low expression of THBS4 showed worse OS and RFS, and high expression of THBS2 and low expression of THBS3 were associated with worse RFS. In conclusion, THBSs may be novel prognostic biomarkers for breast cancer. Our ndings will provide valuable clues to understanding the pathogenesis and progression of breast cancer in humans and contribute to developing of more effective clinical therapies in the future.       The scatter plot indicates the correlation between THBSs expression and drug sensitivity (the z-score of the CellMiner interface) for the Pearson correlation test using NCI-60 cell line data.