Background

The tumor microenvironment (TME) consists of heterogeneous components including borders, blood vessels, lymph vessels, extracellular matrix (ECM), immune and/or inflammatory cells, secretomes, coding or non-coding RNAs, small organelles, tumor cells, and surrounding associated or regulatory cells [1]. Among the tumor-associated regulatory cells, stromal cells playing various crucial oncogenic roles in the TME. In the TME, heterogeneous stromal cells are associated with tumor growth, invasion, progression, and metastasis [2, 3]. These tumor-associated stromal cells promote various dysregulated biological functions including extracellular matrix remodeling, cellular migration, neoangiogenesis, and evasion of immunosurveillance through the production of several types of onco-regulators including cytokines, chemokines, matrix metalloproteinases (MMPs), extracellular matrix (ECM), and growth factors [4]. It is recently demonstrated that the tumor-associated stromal cells playing pivotal roles in the resistance of cancer therapy [5]. The oncogenic intrinsic properties of stromal cells substantially regulated the genotype and phenotype of surrounding cancer cells in the TME [6].

In the breast cancer TME, tumor-associated stromal cells are associated with cancer initiation, development, progression, angiogenesis, metastasis, recurrence, and therapeutic resistance [7]. Survival of breast cancer patients is correlated with stromal biology including the reorganization of the extracellular matrix (ECM) to promote cancer invasion and migration, changes in the phenotypes of stromal cell, variability in the stromal gene expression profiles, and changes in cellular signaling cascades to aid surrounding cancer cells [8]. Various parts of stromal compartments have crucial effects on breast cancer TME. The tumor-promoting intrinsic properties of the stroma are associated with the tumorigenesis of breast cancer [7]. Finak et al. reported that the expression of stromal gene signatures is correlated with the clinical outcomes of breast cancer patients [9]. Winslow et al. identified stromal gene signatures that are associated with clinical features in different types of molecular subtypes of breast cancer [10]. Altogether, these studies provide the clue that the stromal cells have substantial onco-regulatory roles in the TME of breast cancer.

Herein, we performed comprehensive bioinformatic analyses to identify molecular alterations in breast tumor stroma versus normal stroma. We identified differentially expressed genes (DEGs), hub genes from the interactions of DEGs, and regulatory transcription factors (TFs), and kinases and pathways associated with the DEGs. We also identified stromal genes having a significant link with recurrence-free survival in breast cancer patients. Moreover, we found certain tumor stromal genes, which were gradually dysregulated through the three different grades of breast cancer, and their dysregulation was associated with poor prognosis in patients.

Materials and methods

Data selection and pre-processing

We systematically searched for the gene expression omnibus (GEO) database using keywords “breast cancer,” “stroma,” and “tumor stroma”. Ultimately, we identified eight datasets: GSE9014 (sample size n = 123) [9], GSE83591 (n = 53) [11], GSE31192 (n = 17) [12], GSE26910 (n = 12) [13], GSE10797 (n = 33) [14], GSE8977 (n = 22) [15], GSE33692 (n = 22) [16], and GSE14548 (n = 34) [17] (Supplementary Table S1). We combined the eight datasets into a single dataset (including stromal data and excluding other data) using the NetworkAnalyst software [18]. The ComBat method was utilized to remove batch effects from the eight datasets [19] (effects of batch removal shown in Supplementary Fig. S1). Each dataset was normalized by base-2 log transformation or quantile normalization. The combined dataset included 240 tumor stroma and 76 normal stroma samples. We also downloaded the dataset of gene expression profiles (RSEM normalized) in TCGA breast cancer cohort (n = 1212) from the genomic data commons (GDC) data portal (https://portal.gdc.cancer.gov/). We further normalized the RSEM gene expression values by base-2 log transformation [20]. In addition, we used the clinical data of a breast tumor stroma cohort (GSE9014) to evaluate the survival time differences in breast cancer patients [9].

Identification of DEGs between breast tumor stroma and normal stroma by a meta-analysis

We employed the R package “limma” to identify the DEGs between BTS and normal stroma [21]. A meta-analysis of the eight datasets was performed using Cochran’s combination test [22]. The false discovery rate (FDR), calculated by the Benjamini–Hochberg method [23], was used to adjust for multiple tests. We selected the all DEGs with a threshold of absolute value with combined effect size (ES) > 0.41 and FDR < 0.05.

Pathway and functional enrichment analysis

We performed pathway enrichment analysis of the set of genes that were differentially expressed between BTS and normal stroma using the GSEA software [24]. The KEGG pathways [25] significantly associated with the set of genes upregulated and downregulated in BTS versus normal stroma were identified, respectively, using a threshold of FDR < 0.05.

Identification of TFs, protein kinases, and master transcriptional regulators (MTRs) that are significantly associated with the DEGs

To link gene expression signatures to upstream cell signaling networks, we used the eXpression2Kinases algorithm [26] to identify the upstream TFs and kinases that regulate the DEGs. In the eXpression2Kinases algorithm, we used a threshold of hypergeometric P value ≤ 0.05 for identifying upstream TFs and kinases. Besides, we utilized the Cytoscape plug-in iRegulon [27] to identify the MTRs for the upregulated and downregulated DEGs, with a threshold of the minimum normalized enrichment score (NES) > 3.0 which corresponds to an approximate FDR between 3 and 9%.

Protein–protein interactions (PPIs)

We constructed PPI networks of the DEGs using STRING (version v11 [28]). We input all DEGs into STRING. The rank of genes was identified by the Cytoscape plugin cytoHubba [29]. Hub nodes were identified using a threshold of medium interaction score ≥ 0.40 and we selected the degree of interaction ≥ 25 for identifying most interacted genes in the PPI. We identified hub genes, protein kinases, oncogenes, and tumor suppressor genes (TSGs) by comparing the hub nodes with TFs, protein kinases, oncogenes, and TSGs obtained from GSEA (https://www.gsea-msigdb.org/gsea/index.jsp). The online tool “Calculate and draw custom Venn diagrams” (http://bioinformatics.psb.ugent.be/webtools/Venn/) was used to identify common TF encoding genes, protein kinase encoding genes, oncogenes, and TSGs between different groups. We visualized the PPI networks using Cytoscape (version 3.6.1) [30].

Evaluation of immune scores, stromal scores, and tumor purity in stromal breast cancer subtypes

We utilized the “ESTIMATE” R package to calculate immune score representing the enrichment levels of immune cells, stromal score representing content of stromal cells, and tumor purity for each breast tumor sample [31] in (GSE9014 [9]). We compared immune scores, stromal scores, and tumor purity between the patients without disease recurrence and the patients with disease recurrence. We consider the Wilcoxon sum rank test P value ≤ 0.05 for identifying significant difference between both groups.

Quantification of the enrichment levels of immune and stromal signatures

We used the single-sample gene-set enrichment analysis (ssGSEA) to quantify the enrichment scores of immune and stromal signatures in tumors based on the expression levels of their marker genes [32]. We defined the ratio of immune signatures in a tumor sample as the ratio of the average expression levels of their marker genes. The immune and stromal signatures analyzed included B cells, CD8 + T cells, CD4 + regulatory T cells, macrophages, neutrophil, natural killer (NK) cells, tumor-infiltrating lymphocytes (TILs), regulatory T cells (Tregs), cytolytic activity, T cell activation, T cell exhaustion, T follicular helper cells (Tfh), M2 macrophages, tumor-associated macrophage (TAM), myeloid-derived suppressor cell (MDSC), endothelial cell, and cancer-associated fibroblasts (CAFs). Their marker genes are shown in Supplementary Table S2.

Survival analysis

We used the clinical data of a BTS cohort (GSE9014) which involved 53 breast cancer patients with clinical information available [9] (Supplementary Table S3) for survival analysis. We compared the recurrence-free survival (RFS) between breast cancer patients classified based on gene expression levels (expression levels > median versus expression levels < median). Kaplan–Meier survival curves were used to show the survival time differences, and the log-rank test was utilized to evaluate the significance of survival time differences between both groups. We used the function “survfit” in the R package “survival” to perform survival analysis and the function “coxph” in the R package “survival” for the univariate and multivariable Cox regression analyses [33].

Identification of DEGs between breast cancer patients with different tumor grades, clinical outcomes, and survival prognosis

In the GSE9014 database [9], we identified the DEGs between the breast cancer patients without and with disease recurrence and the DEGs among the breast cancer patients with different grades (grade I, grade II, and grade III) (Student’s t test, P < 0.05). We then identified the common genes between both groups of DEGs. We further analyzed the association of the expression of these common genes with the RFS of breast cancer patients. To identify the DEGs among the breast cancer patients with different grades, we utilized the R package “multcomp” [34].

Statistical and computational analysis

We used the two-tailed Student’s t test to compare two classes of normally distributed data, including gene expression levels and the ratios of immune signatures, and the one-tailed Mann–Whitney U test to compare two classes of data that were not normally distributed, including immune scores, stromal scores, tumor purity, and ssGSEA scores. The FDR evaluated by the Benjamini–Hochberg method [23] was used to adjust for multiple tests. We used the R package “ggplot2” to visualize the plots. For multiple probes of a single gene, we averaged the expression values of all probes into a single value by NetworkAnalyst [18]. The online tool “Calculate and draw custom Venn diagrams” (http://bioinformatics.psb.ugent.be/webtools/Venn/) was used to identify common genes between the different groups.

Results

Identification of DEGs between BTS and normal stroma

We identified 1058 DEGs between BTS and normal stroma. Among these DEGs, 782 were upregulated (Supplementary Table S4) and 276 were downregulated (Supplementary Table S5) in BTS. The top 25 upregulated (the highest ES) genes included COL10A1, SULF1, INHBA, NOX4, COMP, COL11A1, RAB31, IFI30, COL8A1, CTSB, LRRC15, SDC1, WISP1, LAMP5, LEF1, ASPN, MSR1, MNDA, SLAMF8, UNC5B, SLA, TYROBP, C3AR1, ITGAX, and COL8A2 (Table 1). Among them, the upregulation of COL11A1 and IFI30 was associated with a worse prognosis in breast cancer patients (Fig. 1A–C). In addition, the top 25 downregulated (the lowest ES) genes included FIGF, SPRY2, DLK1, SFRP1, TGFBR3, HLF, CD36, GPC3, LIFR, CAPN6, RELN, AKR1C3, CAV1, PLP1, MATN2, SDPR, SOCS2, ITM2A, LDB2, SYNM, EGFR, NACA, NOVA1, SPTBN1, and SEMA3G in the BTS (Table 1). Among these genes, the downregulation of SPRY2, CAV1, SOCS2, ITM2A, LDB2, and NACA in BTS was associated with worse prognosis of breast cancer patients (Fig. 1A, D–I).

Table 1 Top 25 upregulated and top 25 downregulated genes in breast tumor stroma
Full size table
Fig. 1
figure 1

Expression levels of eight prognostic genes in BTS and their associations with survival prognosis in breast cancer. A Expression of eight prognostic genes in BTS versus normal stroma. We investigated the survival of top 25 upregulated and top 25 downregulated genes. The upregulation of COL11A1 and IFI30 (upregulated in BTS) is associated with a worse prognosis (B, C). The downregulation of SPRY2, CAV1, SOCS2, ITM2A, LDB2, and NACA (downregulated in BTS) is associated with a worse prognosis (D–I)

Full size image

Identifications of pathways significantly associated with the breast tumor stromal DEGs

GSEA [24] identified 82 KEGG pathways [25] significantly associated with the DEGs upregulated in BTS (Fig. 2A and Supplementary Table S6). Among them, the top 20 (the lowest FDR) pathways are displayed in Fig. 2A. These pathways were mainly involved in immune, stromal signatures, including cytokine–cytokine receptor interaction, Toll-like receptor signaling, antigen processing and presentation, chemokine signaling, T cell receptor signaling, B cell receptor signaling, natural killer cell-mediated cytotoxicity, leukocyte transendothelial migration, hematopoietic cell lineage, complement and coagulation cascades, Fc gamma R-mediated phagocytosis, Fc epsilon RI signaling pathway, NOD-like receptor signaling, Jak-STAT signaling pathway, cytosolic DNA-sensing, RIG-I-like receptor signaling, cell adhesion molecules (CAMs), focal adhesion, ECM–receptor interaction, regulation of actin cytoskeleton, adherens junction, tight junction, and gap junction. Moreover, many cancer-associated pathways were included in the 82 pathways, including MAPK signaling, TGF-beta, VEGF signaling, calcium signaling, mTOR signaling, and apoptosis. Besides, we identified 67 KEGG pathways [25] associated with the DEGs downregulated in BTS. The top 20 (the lowest FDR) pathways are displayed in Fig. 2B and Supplementary Table S7. The downregulated pathways are mainly associated with metabolism (propanoate metabolism, tryptophan metabolism, valine, leucine and isoleucine degradation, lysine degradation, beta-alanine metabolism, limonene and pinene degradation, arachidonic acid metabolism, butanoate metabolism, fatty acid metabolism, histidine metabolism, metabolism of xenobiotics by cytochrome P450, drug metabolism-cytochrome P450, pyruvate metabolism, glycerolipid metabolism, arginine and proline metabolism, steroid hormone biosynthesis, retinol metabolism, ascorbate and aldarate metabolism, linoleic acid metabolism, glycine, serine and threonine metabolism, alanine, aspartate and glutamate metabolism, ether lipid metabolism, glycolysis/gluconeogenesis, etc.), cancers (pathways in cancer, small cell lung cancer, endometrial cancer, non-small cell lung cancer, bladder cancer, thyroid cancer, pancreatic cancer, renal cell carcinoma, prostate cancer, acute myeloid leukemia, colorectal cancer, etc.), and cellular signaling and development (ribosome, PPAR signaling pathway, neurotrophin signaling pathway, insulin signaling pathway, spliceosome, p53 signaling pathway, Wnt signaling pathway, etc.). Altogether, our pathway analysis underlines that the stromal gene signatures are associated with the alteration of pathways that regulating tumor immunity, cellular signaling, metabolism, and cancers.

Fig. 2
figure 2

KEGG pathways are significantly associated with the upregulated and downregulated genes in BTS versus normal stroma identified by GSEA [24]. A Top 20 pathways significantly associated with the DEGs upregulated in BTS. B Top 20 pathways significantly associated with the DEGs downregulated in BTS. FDR: false discovery rate

Full size image

Upstream TFs, kinases, and MTRs regulating the DEGs

The eXpression2Kinases algorithm identified 20 upstream TFs playing a significant regulatory role toward the DEGs (Supplementary Table S8). These TFs included IRF8, NFE2L2, TP63, RUNX1, SPI1, SMAD4, TRIM28, GATA2, AR, SUZ12, EGR1, KLF4, GATA1, RELA, TCF3, PPARD, RCOR1, TP53, SALL4, and NANOG. Interestingly, among the 20 upstream TFs, the genes encoding IRF8, PPARD, and RUNX1 were significantly upregulated in BTS and the gene encoding KLF4 was significantly downregulated in BTS (Supplementary Fig. S2A). IRF8 is a tumor suppressor involved in the regulation of the signaling of breast cancer cells [35]. Dysregulation of NFE2L2 is correlated with poor outcomes in breast cancer patients [36]. In the harsh metabolic conditions of the TME, PPARD promotes the survival of breast cancer cells [37]. In triple-negative breast cancer (TNBC), the expression of RUNX1 is correlated with the poor survival prognosis [38]. One of the isoforms of KLF4 is associated with the carcinogenesis of breast cancer [39]. Besides, we identified 116 upstream protein kinases, including CDK1, LYN, CSNK2A1, MAPK3, RPS6KA1, PRKACA, and TGFBR2 (Supplementary Table S9). Among these kinases, the genes encoding DYRK2, LYN, ERBB2, RPS6KA1, and PRKACA were significantly upregulated in BTS and the gene encoding PKD1, RPS6KA5, EGFR, and FOXO3 were significantly downregulated in BTS (Supplementary Fig. S2B). CSNK2A1 expression levels are significantly higher in a basal subtype of breast cancer [40].

MAPKs are associated with the downstream oncogenic signaling pathways in breast tumorigenesis [41]. Furthermore, we identified 13 MTRs, including IRF8, ETV7, STAT2, SPI1, IRF1, RELA, STAT1, IKZF1, RUNX1, IRF7, NFKB1, NKX3-2, and BATF, which were involved in the regulation of the genes upregulated in BTS (Fig. 3A and Supplementary Table S10). We also identified 11 MTRs targeting the genes downregulated in BTS, including HLF, FOXP2, JUND, NANOS1, RBBP9, FOS, TAF1, FOSL1, JUN, HAND1, and FOXJ3 (Fig. 3B and Supplementary Table S10). Interestingly, the genes encoding the MTRs RUNX1, IRF7, ETV7, STAT2, and IRF8 were upregulated in BTS (Fig. 3A), and the genes encoding the MTRs HLF and FOXJ3 were downregulated in BTS (Fig. 3B). Altogether, these results indicate that a number of TFs and protein kinases play significant roles in regulating the breast cancer stromal gene signatures and are associated with the pathogenesis of breast cancer.

Fig. 3
figure 3figure 3

Regulatory networks of the master transcriptional regulators (MTRs) and their targeted differentially expressed genes (DEGs) between BTS and normal stroma. A Regulatory network of the MTRs and their targeted upregulated genes in BTS. B Regulatory network of the MTRs and their targeted downregulated genes in BTS. In the center, green color octagon indicates MTRs, and purple color oval indicates DEGs

Full size image

Identification of prognostic hub genes in breast tumor stroma

To identify the hub genes of the DEGs in BTS, we input all the DEGs into the STRING tool [28]. We identified 233 hub genes (degree ≥ 25), including 194 upregulated and 39 downregulated genes in BTS (Supplementary Table S11 and Supplementary Fig. S3). Finally, we displayed the top 50 hub genes (EGFR, TLR4, ITGAM, IL10, TLR2, CD86, IL1B, MMP9, ITGB2, TLR8, TLR7, ITGAX, MYC, CXCL10, TYROBP, CXCR4, IRF8, TLR3, CASP3, CTLA4, CSF1R, PLEK, LCP2, CD80, C3AR1, MYD88, IL10RA, PIK3R1, CYBB, SYK, SELL, FCGR2A, CXCL9, CCR7, CCR1, LYN, IRF7, CXCL1, PTGS2, RAC2, ERBB2, FCER1G, ISG15, HCK, CXCR3, CD4, IL7, FCGR2B, COL1A1, and OASL) in Fig. 4A. We found that the upregulation of hub genes MMP9, FCER1G, CD86, ITGAM, TLR2, and COL1A1 (upregulated DEGs in BTS) was significantly associated with poor RFS (Fig. 5A–F). These data indicate that the dysregulation of breast cancer stromal hub genes is likely to be associated with poor prognosis in breast cancer patients.

Fig. 4
figure 4

Protein–protein interactions of prognostic hub genes in BTS. A Protein–protein interaction network of top 50 hub genes. The hub genes in yellow are associated with poor prognosis in breast cancer. B Two prognostic hub oncogenes (IL21R and COL1A1) interact with other hub genes. C Interactions of two prognostic protein kinases encoding genes (PRKACA and CSK) with other hub genes. D Prognostic growth factor encoding gene PLAU is a hub gene that interacts with other stromal hub genes

Full size image
Fig. 5
figure 5

Protein–protein interaction network analysis identifies prognostic hub genes in BTS. A–F High expression of the top hub genes is associated with poor prognosis in breast cancer. F–G The elevated expression of oncogenic hub genes COL1A1 and IL21R is associated with poor prognosis in breast cancer. H–I The elevated expression of protein kinases encoding genes (PRKACA and CSK) is correlated with poor prognosis in breast cancer. J The elevated expression of growth factor encoding gene PLAU is linked with poor prognosis in breast cancer. The survival analysis is performed in the breast cancer dataset GSE9014

Full size image

Hub oncogenes, protein kinases encoding genes, and cytokines and growth factor-encoding genes are associated with poor survival prognosis in BTS

We identified the hub genes belonging to four gene families, including oncogenes, genes encoding protein kinases, genes encoding cytokines and growth factors, and tumor suppressor genes (Supplementary Fig. S3). We found 17 oncogenic hub genes, including CD74, CIITA, CLTC, COL1A1, ERBB2, FCGR2B, IL21R, MUC1, SYK (upregulated in BTS), and EGFR, EPS15, FOXO3, MET, MYC, PPARG, RPL22, ZBTB16 (downregulated in BTS). Also, we found 11 protein kinases encoding genes (CSF1R, CSK, EIF2AK2, HCK, LYN, PRKACA, RNASEL, SYK, and ERBB2 (upregulated in BTS) and EGFR and MET (downregulated in BTS), 20 cytokines and growth factors encoding genes (CCL11, CCL7, CMTM6, CXCL10, CXCL11, CXCL9, IL10, IL16, IL1B, IL1RN, IL7, OSM, PLAU, PMCH, and TNFSF4 (upregulated in BTS) and CAT, CCL27, CXCL1, CXCL2, and CXCL3 (downregulated in BTS), and 2 tumor suppressors genes (upregulated TNFAIP3 and downregulated PIK3R1).

We investigated the association of these hub genes with survival prognosis. Besides, we investigated the specific gene-family-centric PPI of prognostic hub oncogenes (IL21R and COL1A1) (Fig. 4B), hub protein kinase genes (PRKACA and CSK) (Fig. 4C), and hub cytokines-and-growth-factor genes (PLAU) (Fig. 4D) with other stromal hub genes. We revealed that these families of genes were interacted with other stromal hub genes (Fig. 4B–D), indicating their regulatory roles in the TME of breast cancer. Survival analysis revealed that the upregulation of two oncogenes (IL21R and COL1A1), two protein kinase genes (PRKACA and CSK), and a cytokine and growth factor gene (PLAU) is associated with shorter RFS in breast cancer patients (Fig. 5F–J). Altogether, these results indicate that the dysregulation of many tumor stroma-derived gene signatures is associated with unfavorable clinical outcomes in breast cancer patients.

Comparisons of immune and stromal signatures between breast cancer patients with good and bad clinical outcomes

We found that stromal scores were lower in the breast cancer patients with bad clinical outcomes (Wilcoxon sum rank test, P ≤ 0.05) (Fig. 6A). In contrast, tumor purity was higher in breast cancer patients with bad clinical outcomes (Fig. 6A). Interestingly, the enrichment scores (ssGSEA scores) of CD8 + T cells (P = 0.007), TILs (P = 0.03), and endothelial cells (P = 0.05) were lower in breast cancer patients with bad clinical outcomes than in those with good clinical outcomes (Fig. 6B). In contrast, MDSCs (P = 0.05) is more highly enriched in the breast cancer patients with bad clinical outcomes (Fig. 6B). The ratios of CD8 + /CD4 + regulatory T cells were lower in the breast cancer patients with bad clinical outcomes (Student’s t test, P = 4.5 × 10–05) (Fig. 6C). These results indicate that increased immune-promoting signatures are associated with better prognosis in breast cancer, while increased immunosuppressive signatures are associated with worse prognosis. This is consistent with the findings of previous studies [42,43,44,45,46].

Fig. 6
figure 6

Comparisons of immune and stromal signatures between breast cancer patients with good and bad clinical outcomes. A Comparisons of stromal scores and tumor purity between the bad and good clinical outcome groups of breast cancer patients. B The enrichment scores (ssGSEA scores) of CD8 + T cells, TILs, and endothelial cells are lower in the bad clinical outcome group. The enrichment scores of MDSCs is higher in the bad clinical outcome. C The ratios of CD8 + /CD4 + regulatory T cells are lower in the bad clinical outcome group

Full size image

Stromal gene signatures significantly altered with the grades, clinical outcomes, and survival prognosis

We found 1955 DEGs among three grades (grade I, II, and III) of breast cancers (F-test, P < 0.05) (Fig. 7 and Supplementary Table S12). Besides, we found 1471 DEGs between the good clinical outcome (patients without disease recurrence) and bad clinical outcome (patients with disease recurrence) groups (Student’s t test, P < 0.05) (Fig. 7 and Supplementary Table S13). There were 124 common genes between both groups of DEGs (Fig. 7 and Supplementary Table S14). Furthermore, we found 20 of the 124 genes (MCM4, SPECC1, IMPA2, AGO2, COL14A1, ESR1, SLIT2, IGF1, CH25H, PRR5L, ABCA6, CEP126, IGDCC4, LHFP, MFAP3, PCSK5, RAB37, RBMS3, SETBP1, and TSPAN11) whose expression had a significant association with RFS prognosis (Fig. 7 and Supplementary Table S14).

Fig. 7
figure 7

20 genes which are significantly and gradually upregulated (MCM4, SPECC1, IMPA2, and AGO2) or downregulated (COL14A1, ESR1, SLIT2, IGF1, CH25H, PRR5L, ABCA6, CEP126, IGDCC4, LHFP, MFAP3, PCSK5, RAB37, RBMS3, SETBP1, and TSPAN11) through grade I, II, and III of breast cancers, deregulated in the bad clinical outcome group, and associated with poorer survival prognosis

Full size image

MCM4, SPECC1, IMPA2, and AGO2 were gradually upregulated through grade I, II, and III of breast cancers, and their elevated expression was associated with the worse clinical outcomes RFS (Fig. 8A). In contrast, COL14A1, ESR1, SLIT2, IGF1, CH25H, PRR5L, ABCA6, CEP126, IGDCC4, LHFP, MFAP3, PCSK5, RAB37, RBMS3, SETBP1, and TSPAN11 were gradually downregulated through grade I, II, and III of breast cancers and their reduced expression were associated with worse clinical outcomes and RFS (4 genes are shown in Fig. 8B and 12 genes in Supplementary Fig. S4).

Fig. 8
figure 8figure 8

Stromal genes significantly altered among the grades (grade I, II, and III), clinical outcomes, and their association with survival prognosis. A Gradually upregulated genes among three grades and their association with bad clinical outcome. B Gradually downregulated genes among three grades and their association with bad clinical outcome (Only 4 genes are shown)

Full size image

The univariate Cox regression analyses identified 23 genes (out of the 38 prognostic genes) were significant prognostic factors. These genes included COL11A1, CAV1, ITM2A, LDB2, CD86, TLR2, COL1A1, SPECC1, IMPA2, AGO2, COL14A1, SLIT2, IGF1, CH25H, PRR5L, ABCA6, CEP126, IGDCC4, LHFP, MFAP3, RAB37, SETBP1, and TSPAN11 (Supplementary Fig. S5). We further performed the multivariate Cox regression analysis with the expression levels of the 23 genes, breast cancer subtypes (including Her2 status, ER status, and PR status), and age being the predictor variables. We found that 19 prognostic genes (COL11A1, ITM2A, LDB2, CD86, TLR2, COL1A1, IMPA2, AGO2, COL14A1, SLIT2, CH25H, PRR5L, CEP126, IGDCC4, LHFP, MFAP3, RAB37, SETBP1, and TSPAN11), the three subtypes, and age were significant prognostic factors (Supplementary Fig. S6).

Analysis of the dysregulated stromal genes in TCGA BRCA cohort

We further analyzed the expression of the stromal prognostic upregulated and downregulated genes in the TCGA breast invasive carcinoma (BRCA) cohort. Interestingly, we found that the stromal upregulated prognostic genes COL11A1 and IFI30 were significantly upregulated in BRCA than in healthy tissue (Fig. 9A). Also, the downregulated stromal prognostic genes SPRY2, CAV1, SOCS2, ITM2A, LDB2, and NACA were significantly downregulated in BRCA (Fig. 9A). Moreover, we compared the expression levels of stromal prognostic hub genes (COL1A1, MMP9, CD86, FCER1G, ITGAM, and TLR2), oncogenes (COL1A1 and IL21R), protein kinases encoding genes (PRKACA and CSK), and chemokines and growth factors encoding gene (PLAU) between TCGA BRCA and healthy tissue. We found that the expression levels of COL1A1, MMP9, CD86, FCER1G, ITGAM, IL21R, CSK, and PLAU genes were consistently upregulated in BRCA (Fig. 9B). Only PRKACA and TLR2 were slightly downregulated in BRCA. Finally, we investigated the expression levels of prognostic stromal genes whose expression levels gradually altered among the three grades of tumor stroma and also altered between breast cancer patients with bad and good clinical outcomes. Interestingly, we found that MCM4, SPECC1, and AGO2 genes were upregulated, and ABCA6, CEP126, CH25H, COL14A1, IGDCC4, IGF1, MFAP3, PCSK5, RBMS3, SLIT2, TSPAN11, LHFP, and SETBP1 were downregulated in BRCA (Table 2). These results confirmed that the stromal transcriptomes contribute to breast tissue carcinogenesis.

Fig. 9
figure 9

Comparisons of the expression levels of stromal dysregulated prognostic genes between TCGA BRCA and healthy tissue. A Stromal upregulated prognostic genes (COL11A1 and IFI30) are upregulated in BRCA, and downregulated prognostic genes (SPRY2, CAV1, SOCS2, ITM2A, LDB2, and NACA) are downregulated in BRCA. B The stromal hub genes (MMP9, COL1A1, CD86, FCER1G, ITGAM, and TLR2), oncogenes (COL1A1 and IL21R), protein kinases encoding gene (CSK), and chemokines and growth factors encoding gene (PLAU) are significantly upregulated in BRCA. The Student’s t test P values and fold change are shown

Full size image
Table 2 Comparisons of the expression levels of the genes between TCGA BRCA and healthy tissue, whose expression levels gradually altered among breast cancers with three different grades and between breast cancers with bad and good clinical outcomes
Full size table

Discussion

Since stromal cells control various types of tumor phenotypes, including tumor growth, invasion, progression, metastasis, and angiogenesis [2, 3], identification of novel molecular features in BTS is significant. In this study, by analyzing a combined dataset composed of eight breast tumor stromal transcriptomic datasets, we identified the deregulated stromal gene signatures and their associated cellular signaling pathways and PPIs, as well as their associations with antitumor immunosuppression, poor clinical outcomes, and tumor progression in breast cancer (Fig. 10). Our meta-analysis identified 782 upregulated and 276 downregulated stromal genes in BTS versus normal stroma. The previous studies have shown that COL10A1, COL11A1, NOX4, and COL8A1 were upregulated in the TME of breast cancer, and their upregulation was associated with the progression of aggressive breast cancers [17]. Overexpression of COL11A1 is associated with worse clinical outcomes, including overall survival and disease-free survival [47]. SULF1, elevated with the second-highest ES in the BTS, is associated with the remodeling of extracellular matrix during the progression of breast cancer [48]. Elevated expression of COMP was found in the epithelial and stromal cells of invasive breast carcinomas [49]. Another bioinformatics study revealed that IFI30, INHBA, and CTSB were upregulated in breast cancer [50]. A previous study showed that FIGF, SFRP1, and SPRY2 were consistently downregulated in multifocal invasive lobular breast tumors [51]. In breast cancer reactive stroma, ITM2A downregulation is associated with shorter survival of patients [13]. Matrix-producing stromal LDB2 has prognostic value in breast cancer [52]. Among the downregulated genes in BTS, many have been associated with breast cancer onset, invasion, progression, and metastasis [53,54,55,56,57,58]. It was also reported that the expression levels of DLK1 and CD36 were downregulated in breast cancer [53]. The type III TGF-beta receptor (TGFBR3), a tumor suppressor gene, is associated with breast cancer progression and metastasis [54]. Silencing of GPC3 expression is associated with the growth, invasion, and metastasis of MCF-7 human breast cancer cells [55]. Altogether, many of the aberrantly expressed genes in BTS have been associated with breast cancer pathogenesis and carcinogenesis.

Fig. 10
figure 10

The overall flow of the study that identifying key genes and pathways in the BTS

Full size image

Based on these dysregulated genes, we identified pathways upregulated and downregulated in BTS. The upregulated pathways were mainly involved in immune signatures, stromal signatures, oncogenic signatures, and metabolism. Many of these pathways were involved in cancer initiation, progression, angiogenesis, and metastasis in breast cancer [59,60,61]. For example, the group of cytokines and their receptors are associated with breast cancer growth and progression [59]. The ECM–receptor interaction pathway, a major stromal signaling pathway, is involved in breast cancer development [62]. The downregulated pathways were mainly associated with the alteration of metabolisms. In fact, deregulated cellular metabolisms in glucose, amino acid, and other nutrients, are major hallmarks of cancers [63]. The ribosome pathway is one of the major targeting pathways in cancer therapeutic [64]. As a tumor-suppressive signaling, the p53 signaling pathway is a crucial target in cancer therapy [65]. These results indicate that the transcriptional signatures of BTS are associated with the alteration of numerous cancer-associated pathways.

We identified 13 and 11 MTRs which regulate the upregulated stromal DEGs and downregulated stromal DEGs, respectively. Previous studies have shown that these MTRs are associated with breast carcinogenesis [13, 66,67,68,69]. For example, ETV7 plays a substantial role in the oncogenesis of breast tissue [67]. In nodal positive breast cancer patients, another MTR, STAT2 is associated with RFS [66]. RELA is elevated in the CAFs derived from Her2 + breast cancer tissue [68]. Another study confirmed the altered expression of JUND, FOS, and JUN in breast cancer tissue [69]. These results indicate that these TFs may control the BTS-associated transcriptions in the TME of breast cancer.

We also identified key hub genes, especially oncogenes, protein kinases encoding genes, cytokines and growth factors encoding genes in BTS based on the PPI network analysis. PPI networks indicate cellular signaling, communication, and crosstalk between the cells [70]. These hub genes included adverse prognostic factors, such as MMP9, FCER1G, CD86, ITGAM, TLR2, COL1A1, IL21R, PRKACA, CSK, PLAU, MYC, and RNASEL. In luminal A breast cancers, the matrix metalloproteinase gene MMP9 is associated with poor clinical outcomes [10]. In addition, a previous study have shown that ITGAM and TLR2 act as hub genes in breast cancer [71]. IL21R is highly expressed in breast cancer cells and is associated with proliferation, invasion, and migration of breast cancer cells [72]. The elevated levels of COL1A1 are linked with shorter survival and chemotherapy resistance [73]. Moody et al. reported that PRKACA mediated the therapy resistance in breast cancer [74]. Another prognostic stromal protein kinase, CSK, is associated with the cellular growth of hormone-independent breast cancer tissue [75]. In metastatic breast cancer, elevated PLAU is associated with poor survival prognosis [76]. However, we found that the upregulation of MYC (an oncogene downregulated in BTS) and downregulation of RNASEL (a protein kinase gene downregulated in BTS) were associated with worse survival in breast cancer patients. These data suggest that the deregulation of key molecules in PPI networks associated with stromal signatures contributes to the aggressive TME compartment in breast cancer.

The BTS-specific molecules and their associated pathways and interaction networks are potential prognostic biomarkers and therapeutic targets for breast cancer. For example, the upregulation of COL11A1 and IFI30 could indicates a worse prognosis, while the downregulation of SPRY2, CAV1, SOCS2, ITM2A, LDB2, and NACA could indicates a better prognosis in breast cancer. The small molecule inhibitors targeting the kinases upregulated in BTS, such as CSF1R, CSK, EIF2AK2, HCK, LYN, PRKACA, RNASEL, SYK, and ERBB2, could be effective in controlling breast tumor progression.

This study has several limitations. First, the findings were obtained by the bioinformatics analysis but lack of experimental validation. Second, the number of cancer samples with clinical data, such as survival time, tumor grade, and cancer subtypes, is limited in this study. Finally, we analyzed mRNA expression profiles which are not necessarily the same as protein expression profiles, due to some factors affecting the translation from mRNA level to protein level, such as post-translational modification. Therefore, to translate our findings into clinical applications, further experimental and clinical validation would be necessary.

Conclusions

Our data provide pivotal molecular insights into breast tumor stroma characterization, which may have substantial effects on the stroma-based treatment recommendations for breast cancer patients.