TFPI Alpha and Beta Regulate mRNAs and microRNAs Involved in Cancer Biology and in the Immune System in Breast Cancer Cells

Emerging evidence indicate a new role of TFPI in cancer biology. We recently reported that both isoforms of TFPI induced apoptosis and inhibited proliferation of cancer cells. The signaling pathway(s) mediating the effects of TFPI is, however, presently still unclear. Our goal was to further investigate the cellular processes affected by TFPI and to get insight into the molecular mechanisms involved in the effects of TFPI, using a global gene expression study approach. TFPIα or TFPIβ cDNA were transfected into SK-BR-3 breast cancer cells for stable overexpression. Global mRNA and microRNA (miRNA) expressions were measured and functional annotation of the differentially expressed genes and miRNAs according to gene ontology terms was conducted. Selected results were validated using qRT-PCR and Western blot. A total of 242 and 801 mRNA transcripts and 120 and 46 miRNAs were differentially expressed in cells overexpressing TFPIα or TFPIβ, respectively. Overexpression of either isoform significantly affected the expression of genes involved in cell development (apoptosis, cell movement, migration, invasion, colony formation, growth, and adhesion) and immune response. Network analyses revealed biological interactions between these genes and implied that several of the genes may be involved in both processes. The expression profiles also correlated significantly with clinical phenotype and outcome. Functional cluster analyses indicated altered activity of the epidermal growth factor receptor, small GTPases, and the NF-κB and JAK/STAT cascades when TFPI was overexpressed, and increased activity of the transcription factors NF-κB and Elk-1 and phospho-Akt levels was observed. Integrated mRNA-miRNA analyses showed that 19% and 32% of the differentially expressed genes in cells overexpressing TFPIα or TFPIβ, respectively, may have been regulated by miRNAs. Overexpression of TFPI in breast cancer cells affected the expression of mRNAs and miRNAs involved in processes facilitating cancer cell growth and immunologic response, possibly by signal transduction involving the EGFR pathway.


Introduction
Tissue factor (TF) pathway inhibitor-1 (TFPI) is a serine protease inhibitor encoded on chromosome 2. Alternative splicing of the TFPI gene results in two main isoforms, TFPIa and TFPIb. The 276 amino acid TFPIa contains three Kunitz protease inhibitor domains and a basic C-terminal end [1]. It is secreted from cells, and found either free in the extracellular compartment or bound to the cell membrane through a yet unidentified glycosylphosphatidylinositol (GPI) anchored protein [2]. TFPIb consists of 223 amino acids, sharing amino acids 1-181 with TFPIa, and thus contains the first two Kunitz protease inhibitor domains. The Cterminal end encodes a GPI anchor attachment site [2], and TFPIb is therefore located exclusively on the cell surface.
Microarrays are widely used for the simultaneous screening of whole genome mRNA expression, giving extensive information about the transcriptome. Functional analysis of expression signa-tures elucidates ongoing cellular and molecular processes. Expression profiles derived from clinically assessed breast tumors also aid in tumor classification and prognostic assessment [3]. MicroRNAs (miRNAs) are short (,22 nt) non-coding RNA fragments that regulate mRNA expression at the post-transcriptional level. Since their discovery in 1993 [4], more than 1500 human miRNAs have been identified according to the miRNA sequence database (http:// www.mirbase.org/, release 18). Through partial sequence complementarity, miRNAs bind to the 39 untranslated region (UTR) of their target mRNAs and facilitate cleavage or degradation of the transcripts [5,6]. Many miRNAs are known to be important in relation to disease, such as cancer, as they regulate genes involved in proliferation, differentiation, and apoptosis [5,7,8]. Altered expression of miRNAs has been observed in human breast cancers, and several of the miRNAs have been shown to regulate tumorigenic processes [9][10][11][12][13].
In general, TFPI is known for its important role in the regulation of TF induced blood coagulation. However, more recent evidence indicates an additional role of TFPI in cancer. Several cancer tissues and cell lines have been shown to express TFPI [14,15], and TFPI treatment has been reported to reduce tumor growth and metastasis in vivo [16,17]. We recently reported that overexpression of TFPIa or TFPIb in SK-BR-3 breast cancer cells resulted in pro-apoptotic and anti-proliferative effects of either isoform in vitro [18], but the mechanism behind the effect of TFPI is poorly understood.
In the present study, the global mRNA expression profiles of SK-BR-3 breast cancer cells stably overexpressing TFPIa or TFPIb were investigated to further understand the cellular processes affected beyond apoptosis and proliferation, and the molecular mechanisms behind these effects. The clinical relevance of the differentially expressed genes was assessed using publicly available, clinically annotated breast cancer expression data. To our knowledge, no reports have described the relationship between TFPI and miRNA expressions. A miRNA screening was therefore conducted to elucidate the possible involvement of these mRNA regulators in mediating the cellular effects of TFPI.

Cell cultures and stable cell lines
The human mammary adenocarcinoma SK-BR-3 cells (ATCC HTB-30, Manassas, VA, USA) were grown in RPMI1640 containing phenol red and 2 mM L-glutamine (Lonza, Viviere, Belgium), supplemented with 10% heat inactivated FBS (Lonza). Cells were cultured at 37uC in an incubator with a humidified atmosphere and 5% CO2.
Stable cell lines with TFPI upregulated were established as previously described [18]. In short, cDNAs encoding TFPIa or TFPIb were cloned into the pcDNA3.1/V5/His-pTOPO vector (Invitrogen Life Technology, Carlsbad, CA) and transfected into SK-BR-3 wild type cells using Lipofectamine 2000. Stable clones were selected using neomycin. The stable cell line pools with TFPIa or TFPIb upregulated were named pTOPO-TFPIa and pTOPO-TFPIb, respectively. To minimize non-specific effects of the stable overexpression, wild type cells transfected with an empty vector pTOPO were used as controls.

RNA isolation
Total RNA was isolated from the cells using the mirVANA RNA isolation kit (Ambion Life Technologies) according to the manufacturer's instruction. The quantity and quality of the isolated RNA were measured using the NanoDropH ND-1000 UV-Vis Spectrophotometer (NanoDrop Technologies, Wilmington, DE) and Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA), respectively.

Microarray analyses and statistical analysis of data
Microarray analyses were performed using the Affymetrix GeneChip Human Gene 1.0 ST Arrays (Affymetrix, Santa Clara, CA) which contains approximately 28,000 gene transcripts. 150 ng of total RNA was subjected to GeneChip HT One-Cycle cDNA Synthesis Kit and GeneChip HT IVT Labeling Kit, following the manufacturer's protocol for whole genome gene expression analysis (Affymetrix). Biotinylated and fragmented single stranded cDNAs were hybridized to the GeneChips. The arrays were washed and stained using FS-450 fluidics station (Affymetrix). Signal intensities were detected by Hewlett Packard Gene Array Scanner 3000 7G (Hewlett Packard, Palo Alto, CA, USA). Three chips with samples from three independent RNA isolations were run for each cell line.
The scanned images were processed using the AGCC (Affymetrix GeneChip Command Console) software and the CEL files were imported into the Partek Genomics Suite software (Partek, Inc. MO, USA). The Robust Multichip Analysis (RMA) algorithm was applied for generation of signal values and normalization. Probe sets with maximal signal values of less than 5 across all arrays were removed to filter for low and non-expressed genes, reducing the number of mRNA transcripts to 25,492. For expression comparisons of different groups, profiles were compared using a 1-way ANOVA model. The results were expressed as fold changes (FC), i.e., ratios of mean signal values from cells with TFPI upregulated and empty vector control cells. Gene lists were generated with the criteria of false discovery rate (FDR) 10% and a FC of $|2|. The data has been deposited in the NCBI Gene Expression Omnibus (GEO) database with accession number GSE30037 in compliance to MIAME guidelines (http://www.ncbi.nlm.nih.gov/geo/).

Analysis of clinical breast cancer datasets
The publicly available breast cancer datasets GSE6532, GSE4922, and GSE7390 were downloaded from GEO at NCBI and merged. Expression data generated from untreated patients using the Affymetrix HG-U133A arrays were selected, and profiles from samples not containing the clinical variables age, size, grade, estrogen receptor status, relapse-free survival time, and relapsefree survival were excluded, resulting in 545 samples. The datasets were RMA normalized individually using the R/Bioconductor package affy, and each probe median centered before merging. The probe IDs of the 242 and 801 differentially expressed genes following overexpression of TFPIa or TFPIb, respectively, were matched to probe IDs identical to the IDs used in the clinical datasets using netaffx, giving 262 and 773 probes available for comparison. Only these probes were extracted from the merged dataset and used in the analysis. Associations between the differentially expressed genes following overexpression of either isoform of TFPI and the clinical gene profiles were evaluated using the globaltest package in R. In each test of a clinical variable, all other clinical variables were controlled for. The resulting p-values were adjusted for multiple testing with Hommel's method in R.
The correlation between the expression of TFPI isoforms in the normalized datasets and the clinical variables were investigated using a non-parametric Spearman's rho correlation test in R. Pvalues were corrected using the Holm-Bonferroni method.
MicroRNA profiling and target gene prediction miRNA quantification was conducted using the Taqman Human MicroRNA Array Card A v2.0 (Applied Biosystems Life Technologies) following the manufacturer's instructions. The array measures the expression of 377 highly characterized miRNAs and 4 controls simultaneously. In short, 350 ng total RNA was reversely transcribed into cDNA using the Megaplex Human RT primers pool A and TaqMan MicroRNA RT kit (Applied Biosystems Life Technologies). After addition of TaqMan Gene Expression Master mix (Applied Biosystems Life Technologies), samples were loaded on the arrays and run on the ABI PRISM 7900 HT Sequence Detection System (Applied Biosystems Life Technologies) using the low density array format. Results were normalized against the endogenous control U6 snRNA and changes in relative miRNA expression were calculated using the comparative Ct method and expressed as FC of empty vector pTOPO control. miRNAs with a FC of $|2| were considered differentially expressed.
For the integrated miRNA and mRNA analysis, the lists of differentially expressed miRNAs were loaded into the Partek Genomics Suite software (Partek) already containing the analyzed microarray data. The software connects to the targetscan 5.1 database (http://www.targetscan.org/) for identification of predicted mRNAs possible regulated by the differentially expressed miRNAs.

Western blot analysis
For Western blot analysis, cells were harvested and lysed as described previously [18]. The cell lysates were incubated at 97uC for 5 minutes before separated on a SDS-polyacrylamide gel (Bio-Rad Laboratories, Hercules, CA). The membrane was incubated over night at 4uC with the primary human antibodies antiphospho-Akt, anti-Akt (#9271 and #9272, Cell Signaling Technology, Denver MA), anti-EGFR (#44-798G, Invitrogen Life Technologies) and anti-a-tubulin (#T5168, Sigma-Aldrich, St. Louis, MO) and for 1 hour at room temperature with the appropriate horseradish peroxidase (HRP) -conjugated secondary antibody. HRP signals were developed using the ECL Western blotting detection system (GE Healthcare, Fairfield, CT). Bands were quantified using the image processing and analysis program ImageJ version 1.4.3.67 (Rasband W; National Institute of Health, Bethesda, MD, http://rsbweb.nih.gov/ij/index.html).

Luciferase analysis
Pathway analysis was performed using Cignal Finder TM 10-Pathway Reporter Arrays (SuperArray Biosciences, Fredrick, MD) according to the manufacturer's instruction. In short, SK-BR-3 cells (3610 4 ) transiently overexpressing TFPIa or TFPIb or the empty vector pTOPO as a control were seeded into 96-well arrays containing luciferase reporters to common signal transduction pathways and transfection agent. After 48 h, cells were lysed and luciferase activity measured using a Wallac Victor 1420 plate-reader (Perkin Elmer, Waltham, MA). The intensity of the firefly luciferase was normalized to renilla luciferase, and the firefly/renilla ratios of cells overexpressing TFPI and empty vector control cells were divided to determine the relative luciferase activity.

Global changes in mRNA expression
The stable overexpression of the two isoforms of TFPI in SK-BR-3 breast cancer cells have previously been reported [18] and are summarized in Table 1. TFPIb is exclusively located on the cell surface and an increase in protein expression was therefore only detected in the cell medium after PI-PLC treatment, which cleaves the GPI anchor and releases TFPIb from the surface. The effect of the upregulation of TFPIa or TFPIb on the global mRNA expression was investigated using microarray technology. After correcting for multiple testing, only genes with a FC of $|2| and a P-value of ..05 were considered differentially expressed. A total of 242 transcripts were found to be differentially expressed in SK-BR-3 cells overexpressing TFPIa compared to empty vector control cells. 223 of these were upregulated (top ten: IFI44, IFI44L, IFIT1, MX1, SAMD9, DDX60, IFIT3, OAS2, IFIT2, PARP9), while 19 were downregulated (top ten: ASCL1, RET, SBCB, SLC4A10, PODXL, SLC6A14, FREM2, IGF1R, LXN, RN5S9; Table S1). In comparison, 801 transcripts were differentially expressed in SK-BR-3 cells with TFPIb overexpressed. Of these, 546 (top ten: GPNMB, WNT5A, DOCK11, SCIN, PLS3, KRT4, VGLL1, LDHB, GJA5, SCL15A1) and 175 (top ten: ASCL1, NCRNA00052, SLITRK6, PCDH10, NPY1R, NCAM2, PGR, C6orf141, SGCG, RET) were up-or downregulated, respectively, compared to control cells (Table S2). 106 of the differentially expressed mRNAs were identified in both cell lines overexpressing TFPI. Overexpression of TFPIa or TFPIb in SK-BR-3 breast cancer cells, displayed as a summary of previously reported data [18]. mRNA levels were quantified using qRT-PCR. Protein levels were measured in cell medium using total TFPI ELISA.
Results are presented relatively as mean fold change (FC) of control (n = 36 SD). * Relative increase in TFPI antigen measured in cell medium after PI-PLC treatment. doi:10.1371/journal.pone.0047184.t001

Gene ontology categories
The results generated from the microarray experiments were loaded into the Ingenuity Pathway Analysis software (Ingenuity Systems, www.ingenuity.com) for functional analysis. 183 and 669 of the transcript IDs that were differentially expressed in cells overexpressing TFPIa or TFPIb, respectively, were mapped in the ingenuity analysis. Of these, 158 and 534 were eligible for functional analysis. The ten most significant biological functions affected are listed in Table 2 (for complete lists, see Table S3). Eight of the ten gene ontology (GO) categories, Cell Death, Cellular Development, Cellular Movement, Cellular Growth and Proliferation, Cell-To-Cell Signaling and Interaction, Cellular Function and Maintenance, Cell Morphology, and Cellular Assembly and Organization, were affected when either isoform was overexpressed. Within these categories, lower GO terms such as apoptosis, cell movement, migration, invasion, colony formation, proliferation, and adhesion were affected. The ten most significant diseases are listed in Table 2 (For complete lists, see Table S3). Four of these, Infectious Disease, Cancer, Inflammatory Disease, and Reproductive System Disease, were identified in both cell lines. Within the cancer category, lower GO terms such as tumorigenesis and metastasis were affected. We then investigated the differentially expressed genes associated with the lower GO terms. Several genes were found to be associated with more than one category and some seemed to play key roles. The key genes included AQP3, ASCL1, BAK1, CAPN2, CASP4, CDK12, CTSC, CYP1B1, DOCK11, EGFR, EPAS1, FERMT2, GLRX, IFIH1, IGF1R, ITGB6, LGALS3BP, MED1, MYDD88, NAV1, NT5E, NTN4, OLR1, PDGFD, PODXL, RET, SCIN, SGCG, SCL15A1, ST6GAL1, STAT3, SWAP70, TAP1, TIMP2, TPD52, TRIM29, VCL, and WNT5A when either isoform of TFPI was overexpressed. The differentially expressed genes associated with Inflammatory disease were also investigated and in addition to the underlined genes listed above, FREM2, KIAA1217, OSBPL3, PSMB8, RIN2, SLC41A2, UNC93B1, and WBSCR17 were identified in both cell lines.

Network analysis
To investigate the biological interaction between the differentially expressed genes we generated networks using the Ingenuity Pathway Analysis software. Of the 183 and 669 transcript IDs mapped in the ingenuity analyses, 161 and 560 were network eligible. To express the probability that the genes in a network are actually connected, a p-value is calculated and displayed as the -log value (z-score). All the interactions were species specific. Overexpression of TFPIa resulted in ten networks with a z-score of .15 (p,1.0e 215 , Table S4). The networks included genes involved in the immune system, cell death, protein modification, gene expression, cell communication, cell movement and cancer. Networks one to nine overlapped and several of the genes were connected to more than one network. The most significant network associated with both biological function and disease contained genes involved in Inflammatory Response, Infection Mechanism, and Cell Death. Two genes in this network, BAK1 and TNFSF10, were found to be associated with both immune response and cell death ( Figure S1). Overexpression of TFPIb resulted in 16 networks with a z-score of .15 (Table S4). These networks included genes involved in metabolism, growth, immune system, cardiovascular system, cell movement, cell development, cell cycle, gene expression, cancer, cellular organization and cell death. Seven of these networks (network 1, 2, 7-9, 13 and 14) overlapped, and there were some interactions between the networks. The most significant network associated with both biological function and disease contained genes involved in Cellular Growth and Proliferation, Inflammatory Disease, and Respiratory Disease. Five genes, EGFR, NCOA3, PRKAR2B, SAT1 and STAT3 were associated with both inflammatory disorder and cellular growth in this network ( Figure S2).

Signal transduction mechanisms
To explore the candidate signal transduction mechanisms possibly mediating the effects of TFPI, we investigated the functional clusters among the differentially expressed genes using eGOn v2.0 (http://www.genetools.microarray.ntnu.no/egon/ index.php). The Affymetrix IDs were converted to Genebank accession numbers using DAVID v6.7 (http://david.abcc.ncifcrf. gov/) and of the 242 and 801 transcripts with Affymetrix IDs, 222 and 753 were converted to Genebank accession numbers and loaded into eGOn. 207 and 687 transcripts were mapped to unigene clusters and 156 and 554 of these were associated with GO terms when TFPIa or TFPIb was overexpressed, respectively. The list of all the expressed genes contained 25,492 transcripts IDs, of which 24,371 were converted to Unigene IDs and uploaded. 23,332 were mapped to unigene clusters and 13,061 were associated with GO terms. Using the Master-Target test, the differentially expressed genes identified in cells overexpressing TFPIa or TFPIb were compared to all the expressed genes in the microarray analysis to distinguish GO categories significantly over-represented. Results showed a significant enrichment of genes involved in the NFkB (BST2, LGALS9, NFKBIA, STAT1, TNFSF10, TRIM38) and JAK/STAT (NMI, STAT1, STAT2, STAT3) signaling pathway when TFPIa was overexpressed (Table 3). In cells overexpressing TFPIb, a general effect on the intracellular signaling and receptor tyrosine kinase activity (EGFR, EPHA4, EHPA7, ERBB2, FGFR2, RET, TYRO3) was observed (Table 3). Epidermal growth factor receptor (EGFR) activity (EGFR, ERBB2) and small GTPase signaling (ARHGDIB, DIAPH3, DOCK3, DOCK4, DOCK9, DOCK11, ELMO1, EXPH5, GBP1, GBP3, GDI2, MX1, MX2, NRAS, PLCE1, RIN2, RND1) were affected when either isoform was overexpressed. The luciferase reporter system cignal finder and western blotting were used to further identify the signaling molecules that were affected in cells overexpressing TFPI. Results showed that overexpression of either isoform of TFPI led to increased activity of the NF-kB and Elk-1 transcription factors, but not AP1 ( Figure 1A). An increase in phospho-Akt was also observed in SK-BR-3 cells after overexpression of TFPIa or TFPIb ( Figure 1B).

Clinical relevance of the differentially expressed genes
To further evaluate the involvement of TFPI in cancer progression, the clinical importance of the differentially expressed genes was assessed. Our data were compared to a merged dataset of publicly available, clinically annotated breast cancer expression profiles, comprised of 545 untreated patients. Results showed that the differentially expressed genes were significantly associated with tumor grade and ER status when either isoform of TFPI was overexpressed. Furthermore, overexpression of TFPIb resulted in an expression signature that significantly correlated with relapsefree survival time and occurrence (Table 4).
We also investigated the correlation between the expression of TFPI isoforms and clinical variables in the merged datasets. The results showed that the expression of two probes identifying TFPIa (209676_at and 213258_at) correlated significantly with tumor size (p = .035 and .008, respectively).
We investigated the possible interactions between the miRNA and mRNA expression results. Using the Partek software, the miRNA and mRNA data were compared to identify whether any of the predicted miRNA gene targets were actually differentially expressed at the mRNA level. The results showed that predicted targets of 83 miRNAs were oppositely expressed at the mRNA level in cells overexpressing TFPIa, possibly regulating 46 different mRNAs (Table S7). In cells overexpressing TFPIb, predicted targets of 39 miRNAs were differentially expressed in the microarray analysis, possibly regulating 252 different mRNAs (Table S8). 26 miRNAs were identified in both cell lines overexpressing TFPI, with predicted target genes oppositely regulated. 16 of these had the same predicted targets oppositely expressed in both cell lines, which included ACSL1, FERMT2, KIAA1217, KRT80, NAV1, NT5E, NTN4, PDGFD, RNF213, STAT3, SWAP70, TPD52, and WNT5A ( Table 5). All of these predicted targets except two (RNF213 and KRT80) were among the key genes identified in the functional analysis above.

qRT-PCR and Western blot validation
To validate the results obtained from the microarray and miRNA array analyses, qRT-PCR was performed on selected genes using Taqman single assays. 16 mRNAs, which were either among the most differentially expressed genes or identified as a key gene in the functional and/or cluster analyses, and 13 miRNAs, which were identified in the functional analysis or had predicted mRNA targets differentially expressed, were selected. All of the 16 mRNAs and 11 of the 13 miRNAs selected for verification were validated (Figure 3). The two miRNAs that were not validated, miR-103 and miR-652, had expression values just beneath the fold change limit of |2|. The correlation between the array and qRT-PCR values was significant for both mRNA (r = 0.9, P,.0001) and miRNA (r = 0.8, P,.0001) measurements. The fold change values were generally higher when measured by qRT-PCR than in arrays, which is consistent with earlier reports [19]. Increased mRNA expression of EGFR was observed in both cell lines overexpressing TFPIa or TFPIb, and it was identified as a key molecule in the functional and network analyses, in connection to both Cancer and Inflammation. EGFR protein levels were validated in the cell lysates by Western blotting, and the results showed elevated EGFR protein expression in both cell lines overexpressing TFPI (Figure 4). The mRNA and protein levels corresponded as higher EGFR levels were measured in cells overexpressing TFPIb than in cells overexpressing TFPIa by either method.

Discussion
We recently reported that stable overexpression of TFPIa or TFPIb induced death receptor activated apoptosis and inhibited proliferation of SK-BR-3 breast cancer cells [18]. The molecular basis of the anti-tumor effects are, however, still unclear. To further investigate the cellular processes affected by TFPI and the possible molecular mechanisms behind the effects observed we characterized the global mRNA and miRNA expression profiles of these SK-BR-3 cell lines. To our knowledge, this is the first report on mRNA and miRNA profiles in cancer cells overexpressing TFPI. A total of 242 and 801 mRNAs were differentially expressed in cells overexpressing TFPIa or TFPIb, respectively. Notably, TFPI was not among the upregulated genes. This was because the primers that were used to amplify TFPI cDNA only recognized the coding regions, thus leaving out the 5'-and 3'-UTR. The overexpressed TFPIa and TFPIb were therefore not detected by the probes on the array that recognized these UTR sequences of the TFPI mRNA transcripts. When the PLIER algorithm, which The differentially expressed mRNAs were loaded into eGOn for functional annotation, and the Master-Target test was used to identify significantly over-represented gene ontology categories (by comparing the number of genes associated with a gene ontology category to the total number of genes in the array associated with that category). doi:10.1371/journal.pone.0047184.t003 leaves out probes with low signal intensities, was used to calculate expression differences, TFPI was upregulated 8.9 fold in cells overexpressing TFPIa and 11.9 fold in cells overexpressing TFPIb.
In line with our previous findings [18], several of the differentially expressed genes were found to be involved in apoptosis and proliferation. Furthermore, genes involved in cellular processes such as adhesion, migration, invasion and colony formation were differentially expressed, indicating additional growth-related effects of either isoform of TFPI in the cancer cells. This finding is in line with the increased metastatic growth we recently observed in MDA-MB-231 and Sum102 breast cancer cells after downregulation of TFPI [20], and provides possible candidate genes responsible for these functional effects. In addition to cancer cell development, the functional and network analyses revealed that many of the differentially expressed genes were also involved in the immune response. An association between cancer progression and the immune system is recognized  Three publicly available, clinically annotated breast cancer datasets (GSE6532, GSE4922, and GSE7390) were downloaded from the Gene Expression Omnibus (GEO) database at NCBI and merged. Associations between the differentially expressed genes and clinical variables were evaluated using the globaltest package in R. doi:10.1371/journal.pone.0047184.t004 [21] and chronic inflammation increases the risk of cancer development. Inflammatory molecules such as chemokines have been shown to regulate tumor development [22], and death receptor signaling may activate inflammation in addition to apoptosis [23]. Moreover, Van Den Boogaard and colleagues have previously linked coagulation and inflammation as they reported of an anti-inflammatory effect of recombinant (r) TFPIa in mice [24], supporting our findings.
Overexpression of either isoform of TFPI resulted in expression profiles that associated significantly with tumor grade and ER status in a breast cancer patient material. This indicates a possibly therapeutic potential of TFPI that could affect tumor grade and perhaps patient outcome. However, TFPIa and TFPIb expression in the patient material was not associated with tumor grade or ER status illustrating TFPI as a poor predictive marker for these variables. The association between TFPI and ER status is also intriguing as anti-estrogen therapy is important in breast cancer treatment [25]. Overexpression of TFPIa or TFPIb resulted in FC values of 21.6 and 24.5, respectively, for the ER gene. Preliminary results from our laboratory indicate that estrogen decreases TFPI expression in ER positive breast cancer cells, which could contribute to the estrogen-driven growth of such tumors. It is clear that tumor cells differs considerably from cell lines in the environment that they are growing, and the generalizability of results obtained from cell lines should be taken under consideration. However, studies comparing breast tumors with established tumor derived breast cancer cell lines have concluded that the cell lines mirror the primary tumors with high fidelity regarding subtype and genomic features [26,27].
In a previous study, 16 genes were reported to be differentially expressed in HUVECs after treatment with rTFPIa [28]. None of these were among the 242 genes differentially expressed in cells overexpressing TFPIa. The reason for this discrepancy could be differences in the signal transduction mechanisms mediated by rTFPIa compared to endogenously expressed TFPI or differences between the cell types. The authors suggested that binding of rTFPIa to the very low density lipoprotein receptor (VLDLR) resulted in the downstream gene regulation. In contrast, our results indicated a possible involvement of the EGFR signal transduction pathway. This signaling pathway is initiated by growth factors such as EGF and TGFa and activates signal transduction through small GTPases, MAP kinase cascades (MAPKs), such as ERK, p38, and JNK, and the NF-kB, PI3K/Akt, and JAK/STAT pathways among others. Although EGFR signaling is usually   associated with increased cell survival and growth [29], reports have shown that EGFR stimulation can have the opposite effect [30][31][32][33][34][35], particularly when at high concentrations [36][37][38]. The SK-BR-3 cells used in this study have a high endogenous expression of the EGFR. In addition to increased EGFR expression, we observed an increase in NF-kB [18] and Elk-1 transcription factor activity, but not AP1 activity, in cells overexpressing TFPIa or TFPIb. Elk-1 and AP1 are downstream transcription factors activated by the MAPK signaling cascade. They are mainly regulated by the MAPKs ERK and JNK, respectively, thus indicating involvement of the MAPK ERK but not JNK. Moreover, increased levels of phosphorylated Akt were observed in cells overexpressing TFPI. These findings further indicate involvement of the EGFR pathway. rTFPIa has previously been shown to affect MAPK/ERK signaling, although this was in HUVECs and a decrease in signaling was seen [39]. miRNAs play an important role in cancer development as they can alter the expression of many tumor suppressor-and oncogenes [5]. Overexpression of either isoform of TFPI resulted in altered expression of several miRNAs, and the integrated miRNA-mRNA analysis showed that up to 32% of the differentially expressed mRNAs may have been regulated by miRNAs. Of the 26 miRNAs that had predicted target genes oppositely regulated when either isoform was overexpressed, 21 (miR-15a, miR-21, miR-23b, miR-29c, miR-30c, miR-93, miR-98, miR-101, miR-135a, miR-135b, miR-148b, miR-193b, miR-200a, miR-200b, miR-203, miR-205, miR-218, miR-339, miR-365, miR-429, and miR-454) have previously been linked to breast cancer [40][41][42][43][44][45][46][47][48][49]. Among the predicted target genes identified as differentially expressed in cells overexpressing TFPIa or TFPIb were several of the key genes, which were involved in apoptosis (ACSL1, STAT3, SWAP70, TPD52, and WNT5A), proliferation (ACSL1, STAT3, PDGFD, TPD52, and WNT5A), cell movement (ACSL1, FERMT2, NAV1, NT5E, PDGFD, STAT3, SWAP70, and WNT5A), cancer (ACSL1, FERMT2, NT5E, NTN4, PDGFD, TPD52, STAT3, and WNT5A) and inflammation (ACSL1, KIAA1217, PDGFD, TPD52, STAT3, SWAP70, and WNT5A).
Interestingly, only a portion of the differentially expressed mRNAs and miRNAs were identical after overexpression of TFPIa or TFPIb. It therefore seems likely that these genes were responsible for the effects observed in both cell lines related to cancer cell growth. However, additional mechanisms and processes were also affected, which were unique for each of the isoforms of TFPI. Compared to cells overexpressing TFPIa, more genes were differentially expressed in cells overexpressing TFPIb, and processes such as Carbohydrate metabolism, Free Radical Scavenging, Lipid Metabolism, Cellular Response to Therapeutics, Drug Metabolism, Vitamin and Mineral Metabolism, and Amino Acid Metabolism were significantly affected (Table S3). This may indicate a specific effect of TFPIb on the general metabolism of the cells. Furthermore, the association between the TFPIb expression profile and relapse-free survival could indicate additional therapeutic potential of this isoform.
In summary, we here report the mRNA and miRNA expression profiles of SK-BR-3 breast cancer cells overexpressing TFPIa or TFPIb. The differentially expressed genes provide a genetic insight into the pro-apoptotic and anti-proliferative effects previously observed in these cells [18]. In addition, our findings show that increased expression of either isoform of TFPI affected the expression of genes involved in the immune response and development of cancer, linking these processes. Moreover, the mRNA expression signatures correlated significantly with tumor grade and ER status, showing clinical relevance of the TFPI induced gene expression. The molecular mechanisms mediating the effects of TFPI are presently unknown, but functional analysis and Western blot analysis indicated that the EGFR signaling pathway could be involved.

Supporting Information
Figure S1 Network analysis of differentially expressed genes in cells overexpressing TFPIa compared to empty vector control cells. The most significant network (z score = 38, Fisher's exact test) which integrated both molecular function and disease is shown. Upregulated genes are red, downregulated genes are green. Genes with no color are not differentially expressed with the criteria set (FDR 5%, p-value ,0.05 and fold change |2|). Gray lines show biological interaction between genes as identified by the software (whole lines = direct interaction, dotted lines = indirect interaction, arrows = act on). Blue lines show genes involved in inflammatory response, pink lines show genes involved in cell death. (TIF) Figure S2 Network analysis of differentially expressed genes in cells overexpressing TFPIb compared to empty vector control cells. The most significant network (z score = 37, Fisher's exact test) which integrated both molecular function and disease is shown. Upregulated genes are red, downregulated genes are green. Genes with no color are not differentially expressed with the criteria set (FDR 5%, p-value ,0.05 and fold change |2|). Gray lines show biological interaction between genes as identified by the software (whole lines = direct interaction, dotted lines = indirect interaction, arrows = act on). Blue lines show genes involved in inflammatory disease, pink lines show genes involved in cellular growth and proliferation. (TIF) Table S1 Differentially expressed mRNAs following increased expression of TFPIa in SK-BR-3 cells.