Coactivator Or Corepressor To Facilitate

Published Date: 02 Nov 2017

Transforming growth factor-beta (TGFβ) is a member of a large family of secreted polypeptides including activins, bone morphogenetic proteins (BMPs), growth and differentiation factors (GDFs), nodal, and anti-müllerian hormones [1]. TGFβ signaling controls a multitude of biological processes, ranging from embryonic development to adult tissue homeostasis. At the cellular level, TGFβ regulates stem cell self-renewal, cell proliferation, differentiation, apoptosis, immortalization, adhesion and migration [2-5]. TGFβ signal transduction begins with ligand binding to cognate transmembrane type II receptor, a constitutively active Ser/Thr kinase receptor. This allows type II receptor to recruit and phosphorylate the type I Ser/Thr kinase receptor. The intracellular receptor-regulated R-Smads are directly phosphorylated by the activated type I receptor kinase. Phosphorylation of R-Smads results in subsequent heterotrimerization with the common partner Co-Smad [6-8]. The activated Smad complexes are then translocated to the nucleus and associated with various transcription coactivator or corepressor to facilitate DNA binding and gene transcriptional activation (Figure 1.1) [9].

1.1.1 TGFβ ligands

The three prototypic TGFβ isoforms (TGFβ1, β2, and β3) are synthesized as dimeric pro-protein (pro-TGFβ) which is formed by three conserved disulfide bonds between six cysteine residues, a structure motif so-called "cysteine knot" [10-12]. The dimeric pro-TGFβ, also known as a latency-associated protein (LAP), is cleaved by furin convertase to form a small latent TGFβ complex (SLC). The SLC consists of the mature TGFβ and LAP that binds to the large latent TGFβ-binding protein (LTBP) to form a large latent complex (LLC) [13, 14]. TGFβ is secreted in the form of either LLC or SLC. LTBP and its bound latent TGFβ are primarily stored in the extracellular matrix (ECM) through transglutaminase-induced crosslinking of the N-terminal domain of LTBP to ECM proteins such as fibronectin (Figure 1.2) [15]. This process is critical for activating latent TGFβ, as previous study showed that LTBP1 antibody and transglutaminase inhibitors block the activation of latent TGFβ [15, 16].

Because the receptor-interacting epitopes of mature TGFβ are shielded by LAP, the LLC cannot interact with TGFβ receptors [11]. In order to release the bioactive TGFβ ligand, a series of proteolytic process is required to convert the pro-TGFβ to active TGFβ by directly targeting LTBP and LAP. Several proteases, including plasmin, mast-cell chymase and thrombin, release LLC from the ECM. For example, BMP1-like metalloprotease cleaves two sites of LTBP1 to liberate the LLC [17]. Thrombospondin-1 binds to LAP and prevents the interactions between LAP and mature TGFβ, therefore leading to an activation of mature TGFβ [18]. It was also shown that the pro-TGFβ is activated and released by binding ανβ6 and ανβ8 integrins to the LAP [19].

There are also structurally diverse ligand-interacting proteins, called ligand traps, which prevent TGFβ ligands to access to their receptors. For example, BMP ligand traps include noggin, chordin, gremlin, follistatin, DAN/cerberus, and Bmper, in which follistatin and DAN/cerberus also traps activins and nodals individually [3, 20, 21]. These ligand traps play an important role in TGFβ-regulated cell responses from embryonic stem cell renewal, differentiation to cancer cell metastasis.

1.1.2 Ligand binding and receptor activation

In human genome, there are two functional classes of receptors including seven type I (ALK1-7) and five type II (ActRIIa, ActRIIB, BMPRII, AMHRII, TGFBR2) receptors [22]. The receptor domain is consisted of a Cys-rich extracellular, a single pass transmembrane, and an intracellular Ser/Thr kinase domain [23-26]. In the absence of ligands, type I and type II receptors exist as homodimers at the cell surface. Upon ligand binding, the homodimers of type II and type I receptors form a heterotetrameric complex where the constitutively active type II receptor phosphorylates type I receptor at the conserved GS-domain [27, 28].

For TGFβ and activin ligands, they interact more effectively with the type II receptor TGFBR2 and ActRII/ ActRIIB [29, 30]. BMP2 and BMP4 have a high affinity binding to the type I receptors BMPRIA/ALK3 and BMPRIB/ALK6 rather than the type II receptor BMPRII [31]. For ligand TGFβ2 or BMP7, both type II and type I receptors are required for efficient binding [32, 33]. The specific interaction between ligands and receptors divides the pathway into the TGFβ-activin-nodal signaling through ALK4, ALK5 and ALK7 and the BMP-GDF signaling through ALK1, ALK2, ALK3 and ALK6 (Figure 1.3).

1.1.3 Receptor-Smad interaction

The intracellular mediators of TGFβ family signaling pathway are the Smad family, which is comprised of eight Smad proteins including receptor-activated Smads (R-Smads: Smad1, Smad2, Smad3, Smad5 and Smad8), the common Smad (Co-Smad: Smad4), and the inhibitory Smads (I-Smads: Smad6 and Smad7) [3, 34]. The MH1 and MH2 domains are highly conserved in R-Smads and Co-Smad and the two domains are connected by a less conserved Pro-rich linker region [35, 36]. I-smads also contain a MH domain, however, the N-terminal and central regions are divergent (Figure 1.3).

Phosphorylation of the GS-domain of type I receptor results in the recruitment and activation of R-Smads [37, 38]. The interaction between the L45 loop of type I receptor kinase domain and the L3 loop of R-Smad MH2 domain are selective, such as Smad2 and Smad3 are specific to ALK4, ALK5, and ALK7 whereas Smad1, Smad5, and Smad8 are specific to ALK1, ALK2, ALK3 and ALK6 (Figure 1.3) [39]. Another interaction between the phosphorylated GS motif and the L3 loop downstream sequence further stabilizes the receptor-Smad complex [40]. Subsequently, R-Smads are phosphorylated by the type I receptor at the Ser residues of C-terminal region, thereby leading to the conformational changes of R-Smads and their dissociation from the type I receptors [41].

TGFβ and BMPs transcriptionally regulate the expression of I-Smads and induce an inhibitory feedback loop to TGFβ family signaling. The mRNA expression of Smad6 and Smad7 are respectively regulated by BMP and TGFβ in a Smad-dependent manner [42, 43] . For example, Smad1 and 5 mediate the expression of Smad6 gene while Smad3 regulates Smad7 gene expression.

Several molecular mechanisms have been proposed for I-Smads inhibition of TGFβ family signaling. First, the inhibitory functions of Smad6 and Smad7 are mediated through their MH2 domains binding to the type I receptors and competitively interfere with the R-Smad recruitment and phosphorylation. STRAP1, a WD repeat protein, was identified for binding to the TGFβ receptors and cooperating with Smad7 to inhibit TGFβ

signaling [44]. Smad6 and Smad7 can also interact directly with Smurf E3 ubiquitin ligases and recruiting these ligases to the type I receptors [45, 46]. Subsequently, the receptors undergo proteasomal degradation leading to inhibition of TGFβ signaling.

1.1.4 Accessory proteins for receptor activation of Smads

SARA and HRS/HGS: The efficient activation of R-Smad requires an accessory protein Smad anchor for receptor activation (SARA) in response to TGFβ or activin [47, 48]. SARA is a multidomain protein that contains an 80 amino acid Smad-binding domain and a FYVE domain which localizes SARA at the plasma membrane by binding to the membrane lipid PtdIns(3)P [48]. Deletion of the FYVE domain in SARA alters the localization of Smad2/3 and inhibits TGFβ-induced transcriptional responses. At steady state, SARA can interact with both type I receptor and Smad2/3 in early endosomes. When Smad2/3 is phosphorylated by the receptors, they can dissociate from both SARA and type I receptor. Hepatocyte growth factor-regulated tyrosine kinase substrate (HRS/HGS) is another FYVE domain protein which cooperates with SARA and interacts with Smad2/3 to activate TGFβ/activin signaling [49].

DAB2: Disabled-2 (DAB2) function as an adaptor protein which contains a N-terminal phosphotyrosine binding domain (PTB) or phosphotyrosine interacting domain and a C-terminal Pro-rich domain [50, 51]. The PTB domain of DAB2 directly interacts with the MH2 domain of Smad2 in a TGFβ-dependent manner. TGFβ also enhances the association of DAB2 with both type II and I receptors. DAB2 reintroduction into a TGFβ signaling-deficient cell line restores TGFβ-mediated responses, suggesting DAB2 plays an important role in stabilizing and activating the TGFβ receptor and Smad complex [52].

Moreover, other accessory proteins have been reported to facilitate TGFβ signaling by connecting the receptor to Smad2/3 complex, such as Axin [53], ELF β-spectrin [54], and Dok-1 [55].

1.1.5 Smads nuclear translocation and heteromeric complex formation

In the absence of TGFβ stimulation, R-Smad proteins predominately exist as a monomer in the cytoplasm, whereas Smad4 is distributed through the cell and shuttle between the cytoplasm and the nucleus [56-59]. Nuclear import and export of individual Smads is regulated by binding to different importins, exportins, and nucleoporins. Nuclear localization signal (NLS) in the Smad MH1 domain is required for nuclear import, for example, NLS of Smad3 directly binds to an importing receptor importin-β1 [60, 61], whereas SMAD4 is imported by binding to importin-α1 [62]. In addition, nucleoporins CAN/Nup214 and Nup153 have been proposed for nuclear import and export of Smad2 and Smad3 through their MH2 domain directly binding to the nucleoporins [63]. The nuclear export signal in Smad4 linker region is required for nuclear export by binding to exportin1 (also known as CRM1) [58].

Once R-Smads are activated and phosphorylated by the type I receptor, they dissociate from the receptor and form a homotrimer or a heterotrimer with one Smad4 [64]. Structural analysis showed that the interaction between pSer-X-pSer (where p is phospho and X is any amino acid) motif of R-Smads with the L3 loop of another Smad MH2 domain is required for their oligomerisation [65]. In addition, Smad2 and Smad4 can form a heterodimeric complex [66]. The existence of Smads as a heterodimer or heterotrimer may depend on the gene promoter context as well as interacting transcription factors [67]. However, the trimers consisting of two R-Smads molecules and one Smad4 are generally considered to be the principal functional complex [3]. Phosphorylation of cytoplasmic R-Smads is essential for Smad complex formation, but also critical for Smad nuclear accumulation [68, 69]. During active signaling, Smad complexes translocate into nucleus, bind to DNA and various transcriptional factors which result in the complex accumulation in the nucleus. In addition, nuclear export motif of monomeric Smads is masked in complexes, thus decreasing monomeric Smad export.

On the other hand, reduced receptor activity and Smad dephosphorylation are required to attenuate TGFβ signaling and balance the proportional transcriptional output [59]. Several phosphatases including metal-dependent monomeric phosphatase of PPM family (PPM1A) have been identified for the dephosphorylation of R-Smads [70, 71], whereas pyruvate dehydrogenase phosphatases (PDP1 and PDP2) are specific for Smad1, Smad5 and Smad8 [72]. Once continuous Smad dephosphorylation accurs, Smad complex formation is disrupted and monomeric R-Smads exit into cytoplasm, in which the signaling is maintained at the basal level [73].

1.1.6 DNA-bound Smads

Smad complexes activate gene transcription by binding to specific DNA sequence of the promoter or enhance regions. DNA sequence 5’-GTCT-3’ and its reverse complement, 5’-CAGA-3’ were defined as the Smad binding element (SBE) [35, 74]. A β-harpin structure of Smad3 or Smad4 MH1 domain is required for binding to SBE [75]. All R-Smads contain this conserved β-harpin structure, thus other R-Smads are capable of binding to SBE as well [35]. In addition, BMP can activate Smad1 and Smad5 to bind GC-rich sequences [76, 77]. However, a unique 30 residue insert near the β-harpin structure of Smad2 results in its poor DNA-binding activity [35].

The SBE CAGAC sequence appears in approximately every 1024 bp of the human genome [78]. However, all Smads have low affinity and lack selectivity to SBE binding for specific sets of genes, thus multiple Smad binding sites and other transcription factor binding to adjacent DNA are required for sufficient Smad-mediated transcriptional activation.

1.1.7 Smad-interacting transcription factors

Smads directly interact with diverse transcription factors to mediate TGFβ signaling in a context-dependent and cell-specific manner. The Smad-interacting transcription factors, for example, the forkhead, homeobox, zinc-finger, bHLH, and AP1 families provide the specificity of TGFβ signaling on target genes and various biological responses in different cell types [79]. Based on the targeted gene and the cellular context, TGFβ can positively or negatively regulate gene transcription.

Forkhead transcription factor (FAST/FoxH1) was identified as the first partner for Smad2/4 complex in response to activin [80]. In the Smad2/Smad4/FoxH1 complex, Smad4 and FoxH1 contacts SBE and neighboring DNA respectively while FoxH1 interacts with the MH2 domain of Smad2 in response to activin [81]. Moreover, other forkhead proteins FoxOs interact with Smad3/4 complex to induce mRNA expressions of cyclin dependent kinase inhibitors p21CIP1 (p21) and p15INK4b (p15) in the growth-inhibitory responses to TGFβ [82]. In this FoxO/Smad3/Smad4 complex, Smad3 MH1 domain binds to FoxO proteins whereas FoxO binds to the distal region of p21 promoter. Smads also regulate the transcription of p21 gene by interacting with a Zn finger transcription factor Sp1 which occupies the p21 proximal promoter sequences [83].

GATA3, a Zinc finger-containing transcription factor, associates with Smad1 and Smad3 in response to BMP and TGFβ respectively [84, 85]. In T cells, the interaction between Smad3 and GATA3 in response to TGFβ results in the recruitment of Smad3 to the promoter sequence of GATA3-targeting interleukin 5 [85]. In response to TGFβ, Smad3 also interacts with transcription factors including C/EBPβ, myogenic bHLH proteins (MyoD and myogenin, E2F4, c-Myc and Max) and several bZIP family members (c-Jun, JunB, ATF-2, and ATF-3) [86]. Some of these Smad partners are also involved in downregulating gene expression. For example, TGFβ-induced complex formation of Smad3/4 and C/EBPβ represses C/EBP transcriptional activation and inhibiting adipocyte differentiation [86]. Similarly, the interactions between Smad3 and bHLH transcription factors result in transcriptional inhibition [87-90]. In some cases, the Smad partner itself is also a targeted gene in response to TGFβ, such as Mixer and ATF3, resulting in a delayed response for transcriptional activation. This self-enabling TGFβ response is mediated by initially inducing ATF3 expression and then forming an ATF3/Smad3/4 complex to repress ID1 gene expression [91].

The context-dependent and cell-type specific responses to TGFβ signaling is further supported by a recent study that identified Smad3-interacting master transcription factors in specific cell type, such as Oct4 in embryonic stem cells, Myod1 in myotubes, and PU.1 in pro-B cells [92]. Interestingly, reintroducing Myod1 to embryomic stem cell, which is deficient for Myod1 expression, directs a fraction of Smad3 to Myod1-binding sites. Collectively, the specific TGFβ response in different types of cells ranging from embryonic stem cell, lineage-committed progenitors, differentiated cells to malignant cancer cells is largely determined by the abundance and activity of TGFβ ligands, receptors and transcription regulators.

1.1.8 Coactivators and corepressors of Smads

Additional factors which act as coactivators or corepressors are recruited by Smads for transcriptional regulation. The coactivator p300 and CBP directly interacts with the MH2 domain of R-Smads and C-terminal Ser-X-Ser phosphorylation of Smads is required for their interactions. p300 and CBP then acetylate the Lys residues on the MH1 domain of Smads, resulting in the efficient Smad binding to DNA. Two other acetylytransferases, p300/CBP associated p/CAF and GCN5 also associate with Smad2/3 to enhance its DNA binding and activate TGFβ transcriptional regulation. In contrast, poly(ADP-ribose) polymerase 1 (PARP1) induce a poly(ADP)-ribosylation of MH1 domain and inhibits Smad binding to DNA [93].

Smad transcriptional regulation is also inhibited by corepressors including c-Ski [94, 95], SnoN [96], c-Myc and Evi-1 [97]. c-Ski directly interacts with the MH2 domains of Smad2 and Smad3 to regulate TGFβ signaling [98, 99]. The growth inhibitory function of TGFβ is abolished by overexpression of c-Ski through transcriptional regulation on p15 and c-myc expression. This repressive effect of c-Ski is mediated through inhibiting Smad2 phosphorylation and recruiting co-repressor N-CoR or mSin3 and histone deacetylases into the transcription complex [100, 101].

1.1.9 Phosphorylation and dephosphorylation of R-Smads in the linker regions

In addition to C-terminal Ser-X-Ser phosphorylation, the Smad linker region is also a target for intracellular protein kinases to integrate the signaling events [102]. The linker region is rich in Ser and Thr residues, which can be phosphorylated by Pro-primed mitogen-activated protein kinases (MAPKs), cyclin-dependent kinases (CDKs), glycogen synthase kinase 3 (GSK3) [103].

Phosphorylation of the Smad linker by MAPKs can positively and negatively regulate TGFβ/Smads signaling. For instance, mitogen epidermal growth factor (EGF) or oncogenic Ras-activated Erk MAPK phoshphorylates the linker regions of Smad2 at Ser245/250/255 and Thr220, as well as Smad3 at Ser204/208 and Thr179, leading to the inhibition of TGFβ-induced Smads nuclear translocation and transcriptional activity [104]. In contrast, the linker phosphorylations of Smad2, Smad3 and Smad4 by c-Jun N-terminal kinase (JNK), p38 MAPK and Raf/Erk pathways respectively, enhance Smad-dependent transcriptional activiation and TGFβ signaling responses [105-108]. CDK2 and CDK4 phosphorylate Smad3 at the different sites Thr8, Thr 178, and Ser212 of the linker region [109]. Phosphorylation of these sites in Smad3 diminishes its transcriptional activity and TGFβ-induced growth inhibitory effects [109].

The full activation of Smad transcriptional complex requires the linker phosphorylation directed by cyclin-dependent kinase (CDK) 8 and CDK9 [110]. It is noteworthy that RNA polymerase II (Pol II) transcriptional cycles share the same kinases CDK8 and CDK9 which phosphorylate Ser/Pro site of Pol II in the C-terminal domain, resulting in the recruitment of DNA transcription, transcript capping and splicing proteins [111]. Furthermore, CDK8/9-phosphorylated R-Smads is recognized and further phosphorylated by GSK3 [112, 113]. GSK3-driven R-Smad phosphorylation targets transcriptionally active R-Smad for subsequent degradation mediated by E3 ubiquitin ligases Smurf1/2 (Smad1/5-specific) and Nedd4l (Smad2/3-specific) [110, 114].

The phosphorylation of R-Smad linker region is a reversible event that is controlled by small C-terminal domain phosphatases (SCP1, SCP2 and SCP3). SCPs dephosphorylate the specific sites of Smad2 and Smad3 at both linker and N-terminal regions but not at the C-terminus [115, 116]. This dephosphorylation event of Smad2/3 linker prolongs Smad transcriptional activation and enhances TGFβ signaling before the Smurf1 and Nedd4L-mediated degradation of R-Smad occurs [116, 117]. In contrast, C-terminus and linker dephosphorylation of Smad1 by SCP2 results in termination of BMP signaling [116, 118].

1.1.10 Smads in chromatin remodeling

The action on chromatin remodeling is essential for Smad proteins to access to loci and activate/deactivate gene transcription. Several key partners are involved in this process including histone acetyltransferases p300 and CBP, Switch/sucrose nonfermentable SWI/SNF nucleosome positioning complex, and DNA demethylating complex for certain target genes [119]. For instance, Smad-mediated chromatin remodeling requires the recruitment of p300 and CBP which specifically acetylate Lys9, Lys14, Lys18 and Lys23 on histone H3, leading to an open chromatin for gene transcription activation [120], while histone deacetylases (HDACs) are recruited by Smad to repress gene expression [121]. In addition, the SWI/SNF complex mediates nucleosome sliding for transcriptional complex access to DNA and activate or repress genes in response to TGFβ signaling [122]. DNA demethylation is also involved in activating gene expression of a cell cycle inhibitor p15 which mediates TGFβ anti-proliferative effect [123]. At the inactive state, DNA methyltransferase 3A (DNMT3A) together with zinc-finger protein 217 (ZNF217) and co-repressor RE1-silencing transcription factor (CoREST) repress gene expression by binding to the promoter and methylating a CpG island. Upon TGFβ activation, a base excision repair complex is recruited by the Smads proteins to replace the repressive ZFN217-CoREST-DNMT3A complex and prevent DNA methylation.

1.1.11 Non-canonical pathways

In addition to Smad-dependent signal transduction, TGFβ can also activate multiple signaling pathways independently of Smads which are collectively referred as non-canonical TGFβ signaling. These non-Smad cascades, including MAP kinases (Erk/JNK/p38), Rho-like GTPases (RhoA/Cdc42) and phosphatidylinositol-3-kinase (PI3K), are activated by TGFβ through its type II/I receptors [124] (Figure 1.4).

Erk pathway: In addition to Ser and Thr site phosphorylation, TGFβ can induce Tyr phosphorylation of TGFβ type II/I receptors [125, 126]. Tyr-phosphorylated TGFβ receptors then recruit and phosphorylate an adaptor protein Src homology domain 2 containing (ShcA). ShcA further forms a complex with growth factor

receptor binding protein 2 (Grb2) and Sos to sequentially activate receptor tyrosine kinase Ras, its downstream Raf, MEK1/2 and Erk1/2. Importantly, the activation of Shc-Grb2-Erk pathway is required for TGFβ-mediated epithelial-to-mesenchymal transition (EMT), cell migration and invasion which play an important role in breast cancer progression [127-130].

JNK/p38 pathway: Two other MAPK cascades JNK and p38 are activated by TGFβ through MKK4 [105, 131, 132] and MKK3/6 [133-135] respectively. Upon TGFβ stimulation, TGFβ type II/I receptors interact with TRAF6 and induce its poly-ubiquitination, leading to recruitment of TGFβ-activated kinase 1 (TAK1), which is an upstream MAP3K for activating MKKs. TGFβ/BMP-activated JNK/p38 not only induces cell apoptosis but also is another route for TGFβ-mediated EMT and cell migration [136-140].

Rho-like GTPase pathway: RhoA and Cdc42 are small Rho-like GTPases which are essential for cytoskeleton reorganization and cell motility [141]. TGFβ and its receptor regulate these two modules of RhoA-Rock1 and Cdc42/Rac-Pak2 signaling pathway, however, the intermediator between TGFβ receptor and activated RhoA/Cdc42 has not been identified yet [142, 143]. In addition, RhoA protein can also be degraded by Smurf1 in response to TGFβ, resulting in the dissolution of epithelial cell tight junctions and EMT [144]. This signal is mediated through TGFβ type II receptor by direct phosphorylation of partitioning defective 6 (PAR6), a cell polarity regulator. Par6 then recruits Smurf to degrade RhoA at the tight junction.

PI3K/Akt pathway: The PI3K/Akt pathway is also required for EMT induced by TGFβ [145]. The activation of PI3K/Akt is mediated by TGFβ type I receptor, as using a chemical inhibitor for type I receptor blocks the phosphorylation of Akt in response to TGFβ. The activation of mammalian target of rapamycin (mTOR), a downstream effector of Akt, is also involved for TGFβ-mediated cell migration and invasion in murine NMUMG mammary gland epithelial cells and human HaCat keratinocytes [146].

TGFβ Signaling in Breast Cancer

1.2.1 Overview on breast cancer

Breast cancer is the most common type of cancer and is the leading cause of cancer-related death in women [147]. Breast cancer is a complex and heterogeneous disease comprised of largely various entities which are different in molecular profiles, pathologies, and clinical behaviours. Histologically, breast tumors may undergo progressive development, starting from hyperplasia to in situ, invasive or metastatic carcinomas [148]. Significant advances in breast cancer diagnosis and treatment have increased five-year survival rate up to 86 percent [149]. However, once local advanced tumor or distant metastasis occurs, breast cancer patients are largely incurable. Moreover, breast tumor local or distant recurrences are often detected after years of remission, and patients with relapse disease are highly resistant to existing therapeutic agents [150]. The heterogeneity of breast cancer is highly relevant to these unresolved clinical problems [151, 152].

1.2.1.1 Histological types of breast cancer

Ductal carcinoma in situ (DCIS) is the most common non-invasive or pre-invasive breast cancer which is defined by the constraint of proliferative lesions in the mammary duct. Lobular carcinoma in situ is referred as abnormal cells existing in the lobules of the breast which is not truly cancer but with the increased risk of developing cancer. Invasive ductal carcinoma not otherwise specified (IDC NOS) accounts for approximately 75% of breast carcinomas and displays the histological features with loss of myoepithelial cell layer and basement membrane [153]. The reminder 25% histologic specific types of breast tumor include the second frequent invasive lobular carcinoma (about 10%) and the rare apocrine, adenoid cystic, inflammatory, tubular, medullary, mucinous, micropapillary and metaplastic carcinomas [153, 154]. Compared with IDC NOS type, individual histological specific type tends to be more homogenous based on gene expression profiling [155].

1.2.1.2 Molecular classification of breast cancer subtypes

Based on differential expression of genes by microarray analysis, Perou et al. [151] and Sorlie et al. [152] identified five major breast cancer molecular subtypes on the basis of estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) expression: luminal A, luminal B, normal breast-like, HER2+, and basal-like subtype. Luminal A and B subtypes were characterized by positive ER and PR expression [152, 156]. In contrast, three groups including normal-like, HER2+ and basal-like subtypes were identified as low ER and PR expression. Normal-like breast cancer has similar molecular features of normal breast tissue, however this category remains uncertain due to the potential contamination of normal tissue samples [157]. HER2+ tumors overexpress the HER2 oncogene, a member of a family of four receptor Tyr kinases (HER1/ERBB1 through HER4/ERBB4). Basal-like subtype was defined by the absence of or low levels of expressions of ER and PR, as well as an absence of HER2 overexpression [158]. This subtype of tumor expresses genes normally expressed in the basal or myoepithelial cells of the human breast including cytokeratins 5, 14 and 17, vimentin, p-cadherin, αB crystalline, caveolins 1 and 2, and epidermal growth factor receptor (EGFR/HER1) [159-164]. However, about 5% of ER positive tumors and between 6 and 35% of HER2+ tumors also show a basal-like gene expression profile.

1.2.1.3 Clinical outcomes of breast cancer subtypes

Each of these subtypes has different clinical outcomes and responses to treatment; therefore, individualized treatment has been used for the efficacious therapy [165]. The majority of luminal A subtype has good prognosis and respond well to hormonal interventions. Although HER2 is amplified in 15 to 20% of breast cancers and associated with poor prognostic features [166], this type of tumor can be effectively treated with anti-HER2 therapies such as humanized monoclonal antibody trastuzumab. In contrast, basal-like tumors have worst clinical outcomes without effective therapeutic treatment.

The majority of basal-like tumors (about 70%) are also called triple-negative breast cancer (TNBC) due to lack of ER, PR, and HER2 expression. TNBC is a highly heterogeneous group of cancers including two basal-like, an immunomodulatory, a mesenchymal a mesenchymal stem–like, and a luminal androgen receptor subtypes [167]. High histologic grade (Grade 3) is often detected in TNBC patient cohorts ranging from 66-91% [168]. Lymph node spread and distant metastasis also frequently occur in TNBC patients [169, 170]. Despite adjuvant chemotherapy, less than 30% of TNBC patients survive within 5 years of diagnosis due to local and distant tumor recurrence [170]. Unfortunately, TNBC or basal-like subtype patients cannot be treated with conventional therapies such as endocrine or targeted HER2 antibody treatment due to lack of expressions of ER, PR, and HER2. Thus, identification of molecular targets is still required for the successful therapy of this subtype of breast cancer.

1.2.2 TGFβ tumor-suppressive functions in breast cancer

In breast cancer, TGFβ signaling pathway exerts its tumor-suppressive effects in specific context, primarily under conditions of oncogenic stresses. For instance, MMTV-TGFβ1 transgenic mice overexpressing constitutively active TGFβ1 in mammary epithelium exhibited a tumor-suppressive function in mammary tumor growth induced by a chemical carcinogen 7,12-dimethylbenz[α]-anthracene or when crossed with MMTV-TGFα transgenic mice [171]. Similarly, overexpression of constitutively active TGFβ1 ligand or its type I receptor in mammary epithelium suppressed mammary tumor formation initiated by the mouse mammary tumor virus infection or oncogenic ErbB2/HER2 expression respectively [172-174]. Conversely, deletion of TGFβ type II receptor in mammary epithelium accelerated mammary tumor formation induced by polyoma middle T expression [175]. Furthermore, attenuation of TGFβ signaling by expressing a dominant negative type II receptor also prevented mammary tumor cell metastases to the lung in the MMTV-c-Neu mouse model [174].

1.2.3 Cytostatic program responsive for tumor-suppressive function

The mechanisms of TGFβ tumor-suppressive effects are well characterized. The effects of TGFβ on tumor proliferation are largely regulated by a cytostatic response to TGFβ canonical Smad signaling. The cytostatic program is mediated through transcriptional induction of the cell cycle inhibitors p15 and p21 [176, 177], leading to G1 cell cycle arrest and growth inhibition. Induction of p15 and p21 gene expression requires Smad3/4 complex with FoxO and SP1 to activate gene transcription regulation [82, 178, 179]. Conversely, TGFβ represses c-myc gene expression, which is a transcriptional activator of cell proliferation and growth [180]. The repression of c-myc expression is mediated by TGFβ-induced transcriptional complex of Smad3/4, corepressor p107, E2F4/5 and C/EBPβ [88]. In addition, TGFβ represses Id proteins (Inhibitor of Differentiation/DNA binding) to promote cell differentiation and senescence. Cell senescence often functions as a tumor-suppressive barrier, as Id1 bypasses senescence to potentiate mammary tumorigenesis initiated by oncogenic Ras [181]. The downregulation of Id1 expression is mediated by Smad3 and repressor ATF3 in response to TGFβ [91]. Importantly, this inhibitory effect of TGFβ on Id1 prevented mammary tumor formation, resulting in a less proliferative phenotype in a breast cancer xenograft model [182].

1.2.4 Escape of TGFβ tumor-suppressive response

Central components of TGFβ signaling pathway including TGFBRII, TGFBRI, Smad2 and Smad4 are frequently mutated or deleted, resulting in inactivated signaling in large subsets of colorectal, pancreatic, ovarian, gastric, and head and neck carcinomas [183, 184]. In addition, epigenetic alterations also repress the transcriptional expression of the key TGFβ signaling components [185]. For instance, hypermethylation of TGFBRI gene results in TGFBRI transcriptional silencing and its reduced expression in gastric cancer [186]. Promoter methylation of Smad4 gene correlates with low Smad4 expression in advanced prostate cancer [187]. These defects in TGFβ signaling allow tumor cells to evade the TGFβ growth inhibitory effects. However, complete inactivation by genetic mutation or deletion of TGFβ signaling cascades rarely occurs in human breast cancer. Instead, these cancers are resistant to growth-inhibitory and tumor-suppressive effects of TGFβ preferentially through a defective cytostatic response. In primary breast cancer cells isolated from pleural effusion of patients with metastatic disease, these tumor cells retained operational TGFβ receptors and downstream mediator Smad as well as the regulation of certain target genes. However, TGFβ failed to induce cell cycle inhibitor p15 or to repress c-myc expression, causing a defective growth inhibition response [178]. As mentioned earlier, the transcriptional complex of Smad and transcription regulator C/EBPβ is required for the regulation of p15 and c-myc gene expression. There are two C/EBPβ isoforms which are transcriptional activator LAP and transcriptional inhibitor LIP. LIP is able to bind the active isoform LAP and inhibit its transcriptional activity. Thus, the inhibitory isoform LIP is overexpressed in these breast cancer samples to prevent growth inhibition induced by activator LAP isoform. Moreover, oncogenic driver HER2 can also weaken tumor suppressor Smad co-factor FoxO and C/EBPβ by switching expression of C/EBPβ isoform [188]. Another study reported a correlation between higher LIP expression and enhanced breast cancer aggressiveness [189].

In addition to deficient regulation on p15 and c-myc gene expression, the repression of cell differentiation and senescence inhibitor Id1 is even replaced by an induction of Id1 expression in response to TGFβ in patient-derived metastatic breast cancer cells [190]. Id1 and Id3 mediate tumor reinitiation during breast cancer lung metastasis in Xenograft assays in mice using human breast cancer cell lines [191]. Id1 has been identified as one of lung metastasis signature genes which correlate with tumor relapse in ER negative breast cancer patients [192]. Together, as breast cancer progression, tumors preferentially turn off the TGFβ growth-inhibitory arm, in addition, they use the remaining signaling capacity of operational TGFβ receptor and Smads to activate genes that are benefit for metastasis.

1.2.5 TGFβ pro-metastatic functions in advanced breast cancer

Tumor metastasis is a complex and multifactorial process initiated with the detachment of cancer cells from the primary tumor followed by local invasion to the stroma, migration and intravasation of these cells through the blood or lymphatic vessels, survival and re-implantation at a site secondary to the initial tumor bed into the parenchyma of distant tissues, leading to the establishment of micrometastases and ultimately the development of clinically relevant metastatic tumors [193]. Prevention of breast cancer metastasis is critical to cancer patient survival as it represents the last stage in tumorigenesis and the leading cause of patient mortality. Despite being responsible for 90% of human cancer deaths [194], the molecular mechanisms that govern the metastatic process remain largely uncharacterized.

TGFβ is thought to play a dual role in breast cancer progression, acting as a tumor suppressor in normal and early carcinoma and as a pro-metastatic factor in aggressive carcinoma [195]. The tumor-promoting functions of TGFβ in late stages of breast cancer are complex. TGFβ can be released from extracellular membrane and bone matrix, as well as produced in paracrine and autocrine manners by platelets, myeloid, mesenchymal and cancer cells [196-198]. The increasing amount of TGFβ1 is of high incidence to the formation of distant metastasis, as TGFβ acts on the tumor cells for EMT and the surrounding stroma to promote degradation of the ECM, cell migration, cell invasion, angiogenesis, immunosuppression and modification of the microenvironment [184, 199-201]. TGFβ pro-metastatic effects are also thought to be correlated with specific breast cancer subtype. For example, a TGFβ1-response gene signature predicts the incidence of lung metastases and poor outcomes in ER-negative, but not ER-positive breast cancer [190].

1.2.5.1 Epithelial-to-mesenchymal transition

Cells undergoing EMT process are not restricted to embryonic development, but also occur in cancer progression [202]. The transition of EMT can endow cancer cells with the stem-like characteristics and the capacity of migration and invasion which are critical steps for breast cancer metastasis [203]. This process starts with the loss of epithelial structures including apical-basolateral cell polarity, tight junctions and adherens junctions by downregulation of ZO1 and E-cadherin. The full EMT process is completed by gain of a migratory mesenchymal phenotype which is generated by actin reorganization, induction of N-Cadherin, intermediate filament vimentin and ECM protein collagens and fibronectin [204].

The roles of EMT in human cancer are likely contributed by the migratory mesenchymal traits which drive cancer cells to invade locally and disseminate from the primary tumor. Except for diffuse lobular carcinoma [205] and spindle-cell carcinoma tumors [206], EMT rarely appears homogenous in the entire tumor. In fact, tumor cells with mesenchymal phenotypes often occur in the invasive front of carcinomas [207-209]. TGFβ is a strong inducer of EMT during mouse heart formation and palate fusion [210]. As mentioned earlier, high expression of TGFβ is also coincidently apparent at the invasive front location of advanced breast tumor, and correlates with lymph node metastasis [211]. TGFβ-induced EMT has been identified in normal mammary epithelial cell and various types of cancer cells including breast carcinoma [212-215], squamous carcinoma [216, 217], ovarian adenosarcoma [218] and melanoma [219].

TGFβ-induced EMT is mediated by Smad transcriptional complex with a set of transcriptional factors including the zinc-finger proteins Snail and Slug, the bHLH factor Twist, high-mobility group A2 (HMGA2) and the zinc-finger/homeodomain proteins ZEB1 and ZEB2 [220]. A Snail1-Smad3/4 complex was identified to transcriptionally inhibit the expression of E-Cadherin, occludin and CAR (coxsackie- and adenovirus receptor). Importantly, the nuclear location of Snail-Smad3/4 is coincident with the loss of CAR and E-Cadherin expression at the invasive front of breast tumor [221]. Oncogenic Ras can further potentiate the TGFβ-induced EMT in mouse tumors and cell lines [210]. In addition to Smad-dependent EMT, the activation of non-canonical pathways is required for TGFβ-induced EMT. For example, blocking activated p38 and Rho Kinase prevented stress fiber formation and relocalization of E-cadherin in response to TGFβ [135]. TGFβ-activated PI3K and Akt resulted in relocalization of tight junction ZO1 and mesenchymal morphology [145]. TGFBRII-phosphorylated Par6 also promoted the dissolution of cell junction complexes [144].

1.2.5.2 Breast cancer cell migration and invasion

Acquisitions of cellular motility and invasiveness are importantly initial steps for cancer cell spreading. To establish a secondary tumor in distant organ, breast cancer cells must disseminate from primary tumor and invade through basement membrane into lymphatic or blood vessels. The molecules are involved in this process are becoming attractive therapeutic targets [222]. Two different types of cell motility have been identified in human cancer including fast moving single cells and slower moving collective chains [223]. In breast cancer patients, single cell or strand migration is frequently observed in lobular carcinoma [224]. Conversely, collective clusters are often displayed in invasive ductal carcinoma. Singly moving cells lose cell-cell contracts and display an amoeboid or mesenchymal phenotype which enables these cells to invade surrounding stroma and extracellular matrix [225, 226]. Amoeboid cells squeeze through matrix pores, whereas mesenchymal cells carry the proteolytic proteins at the leading front to digest ECM. Collectively moving cells proceed slowly in a cluster or strand consisting of leading cells and following groups. The leading cell has a mesenchymal phenotype, capacity of proteolytic ECM degradation, and serves as a guide for the following cells, while the following cluster or strand cells maintain cell-cell contacts and epithelial morphology [227-229].

Intravital imaging of live, tumor-bearing nude mice demonstrated that migratory cells are not evenly distributed in primary tumor, and only a small proportion of cancer cells between 1 to 5% exhibited motility [230]. Furthermore, Sahai and collegues also using intravital imaging showed that active TGFβ signaling is heterogeneously distributed in a minority of cancer cells within the primary mammary tumor [231]. They found that activation of TGFβ signaling regulates the mode of tumor cell migration and the route of metastatic spread, as singly migratory cells have highly active TGFβ signaling measured by a fluorescent labelled Smad3-dependent reporter while collectively moving cells proceed with inactive TGFβ signaling. Eventually, these single cells spread into blood vessel and lymph node, resulting in lung metastasis. In contrast, cells lacking TGFβ signaling by expressing a dominant negative TGFβ type II receptor collectively and exclusively invade lymphatics. However, not all cells with active TGFβ signaling are migratory, and transient active TGFβ signaling is required for sufficient lung metastasis as TGFβ downregulation occurred in tumor cells during lung colonization.

TGFβ drives the switch of breast cancer cell motility mode from collective to single cell migration through Smad4-dependent transcriptional genes including EGFR, Rho GTPases RhoC, Rac activator Nedd9, Rho-interacting regulator myosin phosphatase activity M-RIP, and FERM domain-containing Rho exchange factor FARP-1. Several of these TGFβ target genes are also overexpressed in highly motile breast cancer cells [232], and implicated in tumor progression and metastasis [233-236].

Several TGFβ inducible genes that mediate its migratory and invasive effect have been identified in breast cancer cell models. In a spheroid invasion assay, TGFβ-induced invasion is regulated by Smad3/4-targeted matrix metalloproteinases (MMP) 2 and 9 in Ras-transformed human MCF10A cells models [237]. TMEPA1 is a TGFβ inducible gene encoding a NEDD4 E3 ubiquitin ligase that is often overexpressed in many types of cancer including prostate, breast, lung, and liver [238-240]. TMEPA1 functions as a negative regulator of TGFβ signaling by sequestering R-Smads [241]. More recently, TMEPA1 has been implicated in TGFβ-induced cell motility in breast cancer cells and was proposed to act as a molecular switch that converts TGFβ from tumor suppressor to tumor promoter [242]. Fascin is the primary actin cross-linker in invadopodia and provides cells with powerful invasive capacity by stabilizing actin into straight, compact, rigid bundles [243]. Overexpression of fascin has been correlated with aggressiveness and poor prognosis in cancer [244] .

1.2.5.3 Cancer stem cells or tumor-initiating cells

Cancer stem cells (CSCs) or tumor-initiating cells (TICs) are proposed as a distinct subpopulation of cancer cells capable of tumorigenicity, self-renewal, and multilineage differentiation [245]. These stem-like properties are implicated in tumor growth, progression and recurrence in various human cancer types. The term of CSC does not necessarily imply that they derived from normal stem cells, but rather that it possesses stem-like properties. Although TICs are also referred as CSCs, they are generally accepted as a small population of cells with the ability of seeding tumors by experimental transplantation in mice and not always necessarily possess stem-like properties [246]. Current treatments are often effective during initial response leading to cancer remission. However, tumor recurrence, metastasis and drug resistance cause cancer patients to ultimately fail to respond to treatment. CSCs are thought to play critical roles in these processes and have emerged as potential cellular targets for clinical therapeutic strategies.

Cancer stem cells have been identified in the breast [247-249]. Several cell surface markers have been used to identify and isolate breast CSCs. The initial study of breast CSC identified a small subpopulation of breast cancer cells expressing CD44+/CD24-/low/Lineage- markers as capable of seeding new tumors in immunodeficient mice from eight out of nine patients. Further enrichment of tumorigenic activity using an additional marker, epithelial specific antigen (ESA; also known as EpCAM), demonstrated that as few as 200 CD44+/CD24-/low/EpCAM+/Lineage- CSCs were able to form tumors in mice compared with tens of thousands of cells without these markers failed to initiate tumors. Importantly, these breast CSCs generated tumors that resembled the heterogeneity of the primary tumor in patients.

CD44+/CD24-/low breast CSCs are not the only population with stem-like and tumorigenic properties. Recently, another subset of breast CSCs were enriched in positive aldehyde dehydrogenase (ALDH) cells which are not identical with CD44+/CD24-/low/Linage-/EpCAM+ CSCs and only a small proportion overlapped between the two types of CSCs [248]. ALDH+ cells derived from human primary breast cancer have a particular stem/progenitor phenotype. Furthermore, only few ALDH+ cells generated mammary tumor in immunodeficient mice but not larger numbers of cells depleted of ALDH.

In addition to using cell-surface markers for CSC selection, serum-free and non-adherent mammosphere culture has been established for the enrichment of normal mammary stem cells or breast CSCs [250, 251]. Mammospheres generated from normal mammary epithelial cells have the stem cell potential of multilineage differentiation and self-renewal under serial passages [250]. Both CD44+/CD24-/low and ALDH+ cells were able to form tumor mammospheres whereas nonstem-like cells without these markers did not generate mammospheres [251, 252].

Tumorigenic CD44+/CD24-/low CSCs are frequently detectable in metastatic pleural effusions of breast cancer patients or early-disseminated cancer cells in the bone marrow [247, 253]. Moreover, a higher proportion of CD44+/CD24-/low cells correlated with an increased incidence of distant metastasis in breast cancer patients [254, 255]. The comparison between gene expression profiles from CD44+ and CD24+ cells demonstrated that CD44+ cells specifically expressed many known stem cell markers which correlate with poor survival outcomes in breast cancer patients [256]. Furthermore, the proportion of CD44+/CD24-/low tumor cells in breast cancer patients increased after chemotherapy treatment [257], suggesting these CSCs are resistant to chemotherapy.

The TGFβ signaling pathway plays an important role in regulating both normal and cancer stem cells from different tissues [197, 258, 259]. CD44+ cells that specifically express stem cell markers also have active TGFβ signaling and high expression of TGFβ target genes [256]. These CD44+/CD24-/low cells displayed a mesenchymal, migratory phenotype, while a more epithelial phenotype was induced by a TGFβ receptor kinase inhibitor. Furthermore, another study showed that normal and transformed mammary epithelium cells passing through EMT endowed them with stem-like properties [260]. TGFβ and Wnt cooperated to generate CD44+/CD24-/low mesenchymal phenotype and increase mammosphere formation. In addition, the regulation of TGFβ signaling on TICs in different subtypes of breast cancer has been clarified [261, 262]. This study demonstrated that TGFβ promotes mammosphere formation and induction of TICs exclusively in claudinlow breast cancer (also known as basal-b subtype).

1.2.5.4 TGFβ and distant metastasis

In addition to the role of TGFβ in promoting EMT, CSC activity, cell migration and invasion, TGFβ can also contribute to the late stages of metastasis including extravasation, survival in a new host microenvironment and colonization in distant tissues [263]. Inhibition of TGFβ signaling by expressing a dominant negative TGFβ type I receptor and deleting downstream mediator Smad4 prevents lung metastasis of ER negative breast cancer cells. This process of lung metastasis is mediated through TGFβ-induced angiopoitein-like 4 (ANGPTL4) expression in primary tumor cells. ANGPTL4 derived from tumor cells can further disrupt vascular endothelial junctions and facilitate tumor cell entering and growing in lung [192]. Expression of a dominant negative TGFBR2 mutant in metastatic ER negative MDA-MB231 cell line can also prevent the bone osteolytic lesion and metastasis. Introduction of a constitutively active type I receptor in the mutant type 2 receptor expressed cell restored the metastatic ability of MDA-MB231 to bone site [264]. A different set of genes regulated by TGFβ is required for bone osteolytic lesion and metastasis. Parathyroid hormone-related protein (PTHrP) and interleukin (IL) 11 are the primary targets of TGFβ in the bone site [265, 266], and the enhanced production of PTHrP and IL11 further promotes the secretion of osteoclastogenic RANK ligand in osteoblast, resulting in the differentiation of osteoclast precursors and osteolytic lesions.