Breast cancer is the most commonly diagnosed cancer and leading cause of cancer-related deaths in women. In the US, it is the second-most common cause for cancer-related death in women, just behind lung cancer, with the expectation that 231,840 new cases will be diagnosed with 40,290 deaths in 2015 [1]. While breast cancer is typically referred to as a single disease, human breast tumors comprise heterogeneous and diverse groups, with patients in the same stage of disease varying in morphologies, treatments, treatment responses and overall outcomes [2]. With the advent of gene expression profiling technologies, researchers have been able to dissect the genetic and phenotypic variability among tumors and differentiate breast cancer into four molecular subtypes based on the presence or absence of the estrogen and/or progesterone hormone receptors (HR) and overexpression of the human epidermal growth factor 2 (HER2) protein: luminal A (HR+/HER2-), luminal B (HR+/HER2+), HER2-enriched (HR-/HER2+) and basal-like (HR-/HER2-) [2–5]. These groups are determined through the analysis of biological markers, which can provide diagnostic, prognostic and therapeutic response information about a certain cancer and are important in the early detection, diagnosis and treatment to improve patient outcome [6, 7]. Of the one million breast cancer cases annually diagnosed around the world, approximately 15–20%, or 170,000, of the cases will be of the Triple-Negative Breast Cancer (TNBC) subgroup [8–10]. Similar to breast cancers as a general group, TNBCs exhibit a disparity among racial groups, with premenopausal African and African American women demonstrating higher rates of diagnosis. Younger women, as well as Hispanic and non-Hispanic women of lower socioeconomic statuses, are also more frequently diagnosed with aggressive TNBCs [1, 9, 10]. Other risk factors include increased parity, younger age at first pregnancy, shorter period of breast feeding and higher hip-to-waist ratio [8].

Despite the widespread use of standard chemotherapy such as Paclitaxel (Taxol) or the combination of taxanes and genotoxic drugs, TNBCs lack the appropriate targets for the commonly used targeted breast cancer therapies, conferring an aggressive phenotype and poorer survival rate to the disease [8–12]. For example, Tamoxifen, which was originally used to treat all breast cancers, is now known to be effective against tumors expressing hormone receptors (ERs and PRs), while Trastuzumab therapy is used to treat patients presenting an over-amplification of HER2 [13]. Due to the lack of targeted therapies, TNBC patients have a poorer prognosis with more frequent relapse, distant recurrence and higher proliferation rates than other subtypes of breast cancer patients [8, 10–12].

Currently, many researchers are analyzing the dysfunctional pathways unique to TNBC in order to identify possible gene targets and develop drug therapies [14–17]. Although a couple of drugs are currently in undergoing clinical trials, the biology behind TNBC is still largely unknown. It is known that the TNBC represents distinct heterogeneity which complicates clinical treatment strategies. Further classification of TNBC may help in achieving better clinical outcome through. Currently, TNBC can be separated into distinct subtypes with gene expression profiling. Six subtypes have been reported with unique gene expression and ontologies: basal-like 1 (BL1), basal-like 2 (BL2), immunomodulatory (IM), mesenchymal (M), mesenchymal stem-like (MSL) and luminal androgen receptor (LAR) [18]. Masuda et al. [19] determined seven subtypes. In this study, we explored the pathways that are upregulated and downregulated in TNBC with respect to normal breast tissue samples. We hypothesized these up- and down-regulated pathways represent two opposing effects (Yin and Yang) that determine the cancer outcome [20–22]. These Yin and Yang pathways could help identify potential therapeutic targets for TNBC. They can be also used to build pathway classifiers in which the Yin and Yang pathways present a strong contrast pathway profile together. The TNBC subtypes classified by Yin and Yang pathways would aid in the personalized therapy for TNBC.

A number of the top pathways shown by GSEA to be upregulated in TNBC play a variety of roles in the mitotic cell cycle, cell division, and specific chromosomal processes. Of these pathways, the FOXM1, which is the top Yin pathway in TNBC but not in other breast cancer subtypes (i.e luminal, HER2 enriched), is listed as the most significant with a FDR of 0 (Additional file 1: Table S1). The FOXM1 includes Nek2, which is ranked first among all the genes from the gene sets characterized by GSEA (data not shown). Nek2, a member of the serine-threonine kinase family, is a cell cycle dependent protein kinase that has been shown to be upregulated in cancers such as lymphoma, cholangiocarcinoma, breast, prostate and cervical. Nek2 functions in the regulation of mitotic spindle formation, chromosome segregation, cell division, carcinogenesis, and the tumorigenic growth of breast cancer [27, 28]. It is especially known to play a role in the mitotic progression of cells where it prompts the separation of the centrosomes by centering itself on the centrosome and establishing a bipolar spindle [27]. This is noteworthy as chromosome instability is considered a common defect in cancer cells which may arise from malfunctions in cell division and the unequal separation of chromosomes to their respective daughter cells during mitosis [29].

PPARα is the top listed TNBC Yang pathway but is a pathway shared with the other breast cancer subtypes (Additional file 2: Table S2). Some of the key players in the PPARα pathway are the nuclear receptors from the family of peroxisome proliferator activator receptors (PPARs). They generally control cellular proliferation and differentiation, glucose and lipid metabolism, as well as adipocyte differentiation [30, 31]. PPARα ligands have been shown to induce cell cycle arrest at the G₁ phase of the cell cycle to prompt the differentiation of liposarcoma and colon, prostate and breast cancer cells, conferring a less malignant phenotype to the cells. The induction of apoptosis through the PPARα pathway in the cells was accompanied by the activation of the NF-κB pathway, which functions in the inflammatory response, innate and adaptive immunity, and prevention of cells undergoing apoptosis following DNA damage [31, 32].

When we input all Yang pathway genes into Ingenuity Pathway Analysis system (IPA), again the top one is the PPARα/RXRα pathway with a p-value of 1.95 × 10⁵³. The PPARα/RXRα pathway functions in both the cytoplasm and nucleus of cells. Retinoid X receptors (RXRs) are nuclear receptors that form heterodimers with retinoic acid receptors (RARs), which are ligand-regulated transcription factors, to control cell growth and survival. Retinoic acid binds to RARs to regulate processes such as development and cell proliferation, differentiation and apoptosis [33]. In the PPARα/RXRα pathway, PPARα and RXRα form a heterodimer which then binds to DNA to regulate gene transcription. From the IPA output, genes are then transcribed that function in fatty acid oxidation, lipoprotein metabolism, and anti-inflammation. There has been evidence that therapies combining PPARα and RXRα ligands in the treatment of breast cancer are effective [34]. Recently, there has been interest in the treatment of cancers using RAR and RXR modulators as it has been shown that the use of RAR modulation to treat acute promyelocytic leukemia has been successful. Therefore, the use of selective receptor modulators may help address the limitations of some drugs [35]. Selective agonist retinoids were studied in vitro to determine their effects on the proliferation and apoptosis of human breast cancer cells. As the PPARα/RXRα IPA pathway was constructed from the list of downregulated genes, it is possible that induction or amplification of PPARα/RXRα within TNBC cells may provide a better treatment for the disease.

Gene expression profiling has been used to separate TNBC into six subtypes with unique gene expression and ontologies: basal-like 1 (BL1), basal-like 2 (BL2), immunomodulatory (IM), mesenchymal (M), mesenchymal stem-like (MSL) and luminal androgen receptor (LAR) [18]. It was found that the EGFR, VEGFR and FGFR gene products were particularly amplified in TNBCs and serve as putative targets for drug therapies [18]. Although initially it was unclear as to the clinical significance of these subtypes, Masuda et al. [19] determined that a seven subtype classification, which includes an unstable (UNS) subtype, has the potential to aid in the development of innovative personalized medicine regimes for TNBC patients. More recently, though, Burstein et al. [36] analyzed the prognosis of TNBC subtypes and separated the disease into four groups: luminal androgen receptor (LAR), mesenchymal (MES), basal-like immunosuppressed (BLIS) and basal-like immune activated (BLIA) subtypes, with the worst prognosis conferred to BLIS and the most favourable to BLIA. Potential targets included androgen receptor and cell surface mucin (MUC1) for LAR, growth factor receptors such as platelet-derived growth factor (PDGF) receptor A for MES, immunosuppressing molecule (VTCN1) for BLIS and stat signal transduction molecules and cytokines for BLIA [36]. In this study, we used the pathway score profiles of the Yin and Yang pathways as a classifier for TNBC. The 6 subtypes of TNBC generated by our approach showed different pathway patterns and distinct clinical outcomes. We compared our 16-contrasting pathway classifier to the previous 7-subtype classifier using the same validation data [18]. We found that these two classifiers resulted in different classifications (Additional file 3: Figure S4). This is expected since we used the same pathway but different scores to differentiate subtypes while previous methods used gene expression profiling for clustering.

A different YMR signature model has demonstrated significance in stratifying TNBC into high- and low-risk groups though the cohort size is small. Due to the high level of molecular and clinical heterogeneity of TNBC, this range of significance suggested that the YMR built from the Yin Yang pathway genes or FOXM1, PPARα pathway genes has potential significance in some subgroups of TNBC. However, currently TNBC data are mostly collected from patients who underwent chemotherapy, which may disturb the prognosis detection we encountered in this study.

The limitation of this study is the validation of prognostic model of FOXM1 and PPARα pathways. In contrast to previous studies that purposely selected prognostic genes or pathways; we identified important pathways in TNBC tumor compared to normal and then tested their prognostic significance. We validated the 2-pathway prognostic model using the METABRIC data set. We attempted to validate our 2-pathway YMR model in other data sets (GSE28812, GSE25066), however although a similar pattern was found it did not achieve statistical significance. Therefore this is a limitation of our study. The reasons for this are unclear, although different treatments and the frequency of treated versus untreated cases in the cohorts may underlie the different results obtained. We must cautiously interpret the data where patients underwent therapy because therapy can alter prognosis or we were testing the treatment benefit. There is also a limitation in finding large sample size of TNBC without therapy treatment for our validation.

Through the use of GSEA we explored the regulatory signaling pathways in TNBCs. The upregulated FOXM1 pathway and downregulated PPARα pathways were found to be the most significant in TNBC. Therefore, simultaneously targeting these two opposing pathways potentially could provide novel treatments options for some TNBC patients. The pathways can also be used as classifiers to subtype TNBC further for prognosis. The resulting TNBC subtypes exhibit different clinical outcomes, which supports the utility of our approach. This is a primary study using contrasting pathways for TNBC subtyping. Further study will focus on prognosis and treatment prediction signatures for each of these subgroups using more data sets.