The network of non-coding RNAs and their molecular targets in breast cancer

Background Non-coding RNAs are now recognized as fundamental components of the cellular processes. Non-coding RNAs are composed of different classes, including microRNAs (miRNAs) and long non-coding RNAs (lncRNAs). Their detailed roles in breast cancer are still under scrutiny. Main body We systematically reviewed from recent literature the many functional and physical interactions of non-coding RNAs in breast cancer. We used a data driven approach to establish the network of direct, and indirect, interactions. Human curation was essential to de-convolute and critically assess the experimental approaches in the reviewed articles. To enrol the scientific papers in our article cohort, due to the short time span (shorter than 5 years) we considered the journal impact factor rather than the citation number. The outcome of our work is the formal establishment of different sub-networks composed by non-coding RNAs and coding genes with validated relations in human breast cancer. This review describes in a concise and unbiased fashion the core of our current knowledge on the role of lncRNAs, miRNAs and other non-coding RNAs in breast cancer. Conclusions A number of coding/non-coding gene interactions have been investigated in breast cancer during recent years and their full extent is still being established. Here, we have unveiled some of the most important networks embracing those interactions, and described their involvement in cancer development and in its malignant progression.


Background
The non-coding RNAs are a still growing and heterogeneous set of genes that act upon other non-coding, or coding, RNAs and ultimately regulate most biological processes in the human cell. They have been extensively studied, mainly after year 2000, in human malignancies and particularly in the cancers of the mammary gland.
The studies on non-coding RNAs and breast cancer (BC) prevalently investigate one or few RNAs that have been selected from clinical genomics. Typically, such works analyze the BC transcriptomes from retrospective cohort studies.
We decided to apply a data-driven study selection rather than use only our human and scientific sensitivity. Firstly, we performed two queries to isolate from PubMed all the articles on ncRNAs and miRNAs published in the last 5 years on BC (Table 1). To triage the studies considered for this review we then selected the journals based on their impact factors. A different, and probably fairer, criterion would have been the citation number, but this is impractical for articles with recent publication time, such as those we wanted to consider here. Furthermore, we let the skeleton of our work to self-assemble using the data themselves. We explored this procedure in our earlier organized view of the role of non-coding RNAs in drug resistance. Using an approach where the nodes are the non-coding RNAs, or their target genes and the edges (connections) are the PMIDs of their relative articles, we obtained a network that was used to organize this review. Separate groups of RNAs and genes that were not linked will be discussed as separate entities or 'sub-networks'. A statistical analysis of the network helped to identify nodes (RNAs or genes) with particular properties (i.e. degree, or number of interacting RNA/genes) and ultimately for prioritization. The number of citation of an RNA/gene depends both on its 'real' importance as determined by the experimental method, or on its 'perceived' importance, making it an element of choice by the investigators. The network of non-coding RNAs (ncRNAs) and their targets in BC, defined using this approach is shown in Fig. 1. The graph shows the non-coding RNAs, and their targets, validated in at least two independent sources from literature. The edges are directed (i.e. from the non-coding RNA to its target). In red are depicted the links indicating a repressive action (flat arrowhead), while in black are those showing activation (with traditional arrowhead). Dashed lines correspond to edges indicating indirect effects. The network in Fig. 1 is the essential core showing what remains after filtering the nodes (non-coding RNAs) based on their degrees (i.e. the number of connections to targets). Detailed information about the network composition are reported in Table 2. The filtered out nodes, basically un-replicated findings, are shown in Table 3. They are still worthy of consideration, but were strictly left out of the major network.
We will discuss here the most prominent subnetworks and their single components and interactions, with the goal of understanding the involvement and roles of non-coding RNAs in BC.
The miR-200/205 ZEB2 sub-network Figure 2 shows that ZEB2 is a pivotal actor in this subnetwork, interconnecting the cluster composed by miR- Fig. 1 The network of non-coding RNAs and its targets in breast cancer. The graph shows the non-coding RNAs (in the square nodes) cited in at least 2 different sources from literature. Empty circles correspond to the coding genes. Each connecting line (or edge) indicates a publication (PMID) from PubMed. When multiple edges connect the same two RNAs in the network, then multiple publications described this interaction. The edges are directed (i.e. from the non-coding RNA to its target, being either coding or non-coding). In red are depicted the links indicating a repressive action (flat arrowhead), while in black are those showing activation (with traditional arrowhead). Dashed lines correspond to edges indicating indirect effects. The network is the essential core showing what remains after filtering the nodes (non-coding RNAs, in orange, and miRNAs, in light blue) based on their degrees (i.e. the number of connections to targets or other non-coding RNAs). The network's details are reported in the Table 2 200a/b/c and miR-205 with that of miR-30a/e and miR-181. Several research groups independently asserted that miR-200a/b/c are down-regulated in triple negative breast cancer (TNBC) and function as metastasis suppressor reducing epithelial mesenchymal transition (EMT), tumour invasion and drug resistance [142]. MiR-200 family's components target other genes that antagonize malignant processes, among them Rho GTPase-activating protein 18 (ARHGAP18), an important regulator of cell shape, spreading, migration, and angiogenesis [143] and the leptin receptor (OBR), which promotes the formation of cancer stem-like cells (CSCs) and up-regulates the obesityassociated adipokine itself associated to BC [144]. Furthermore, in this subnetwork miR-205 is involved in the modulation of basal-like BC motility mediated by the ΔNp63α pathway, by preserving the epithelial cells characters [145]. Mir-205 also is negatively correlated with DNA damage repair, promoting radio-sensitivity in TNBC, by targeting the ubiquitin conjugating enzyme E2N (UBC13) [146]. In contrast, Le et al. demonstrated that delivery of miR-200 family (miR-200a/b/c) by extracellular vesicles, through the circulatory system from highly metastatic tumour cells to poorly metastatic cells, in which ZEB2 and SEC23A were down-regulated, induced EMT and conferred the ability to colonize distant tissues [147]. Further considerations on opposite effects of ncRNAs could be drawn by the second cluster, where the miR-30's family members suppressed cell invasion in vitro and bone metastasis in vivo by targeting genes implicated in invasiveness (ITGA5, ITGB3) and osteo-mimicry (CDH11) in TNBC [148]. Consistently, miR-30a was involved in EMT regulation, upon TP53 stimulation, by targeting ZEB2 [149], while miR-30e displayed an onco-suppressor role through the modulation of ataxin 1 (SCA1) and EIF5A2, two disruptors of the BC acini morphogenesis promoted by laminin111 (LN1) [100]. MiR-181a could also lead to a reduction in the activation of pro-MMP-2, cell migration and invasion of BC cells through matrix-metalloproteinase MMP-14 [150]. In an apparently opposed fashion, Kuancan et al. demonstrated that miR-181a and miR-30e, once stimulated by SOX2 activation, could promote migration and metastasis dissemination in Basal and Luminal BC via silencing of Tumour Suppressor Candidate 3 (TUSC3) [151]. This subnetwork includes another crucial connection between miR-200c and miR-9, as antagonistic modulators of PDGFR ß-mediated vasculogenesis in TNBC.
High levels of miR-9 exerted pro-metastatic function and mediated the acquisition of a mesenchymal and aggressive phenotype. In addition, miR-9 enhanced the generation of vascular lacunae both in vitro and in vivo, in part by direct repression of STARD13, and was also required for PDGFRßmediated activity. On the other hand, miR-200c in TNBC models strongly inhibited tumour growth and impaired tumour cell-mediated vascularization, by inhibiting PDGFRß activity in vascular lacunae and acting on ZEB1, one of the main transcriptional factors in EMT induction [152]. Furthermore, miR-9 in collaboration with miR-203a could lead to a CSC phenotype and to drug resistance after their release from exosomal vesicles (EV), upon treatment with chemotherapeutic agents. These miRNAs target the transcription factor One Cut Homeobox 2 (ONECUT2), whose reduction induces the expression of a variety of stemness-associated genes, including NOTCH1, SOX9, NANOG, OCT4, and SOX2 [83]. Blocking the EV miRNA-ONECUT2 axis could constitute a potential strategy to maximize the anticancer effects of chemotherapy, as well as to reduce chemoresistance. MiR-203a can collaborate with miR-135 (not showed in this subnetwork) to inhibit cell growth, migration and invasion, by the down-regulation of Runx2 and IL11, MMP-13 and PTHrP targets. Indeed, an aberrant expression of Runx2, which promotes tumour growth and bone metastasis formation, was detected in BC [67]. This subnetwork highlights another connection of miR-203a, occurring with the long non coding UCA1 which affects directly and indirectly the snail family transcriptional repressor 2 (SLUG). MiR-203 prevents the induction of motility in luminal BC cells, through downregulation of ΔNp63α activity, and the inhibition of its SLUG and AXL targets [145]. Of interest, UCA1 expression in BC cells correlated with TGF-β-induced EMT and tumour metastasis. Mechanistically UCA1 is up-regulated by TGF-β and cooperates with the LINC02599 (AC026904.1) in order to promote SLUG activation and maintenance [1]. Furthermore, UCA1 was proposed to act as a competing endogenous RNA (ceRNA) to sequester miR-122, thus promoting BC invasion. Interestingly, a mechanism mediated by insulin-like growth factor 2 messenger RNA binding protein (IMP1) and repressing invasion has also been hypothesized, via UCA1 decay through the recruitment of the CCR4-NOT1 deadenylase complex. According to this model, IMP1 could compete with UCA1 for binding to miR-122 and restore miRNA targets to inhibit cell invasion [153]. "Search ((((microRNA OR miRNA OR ncRNA OR ""non coding RNA"" OR lncRNA) AND ""last 5 years""[PDat]) AND (""breast neoplasms""[MeSH Terms] OR ""breast carcinoma"" OR ""breast cancer"")) AND ""last 5 years""[PDat]) Sort by: [pubsolr12]" 5219 "Search (ncRNA OR ""noncodingRNA"" OR ""non coding RNA"" OR lncRNA) AND ""last 5 years""[PDat]) AND (""breast neoplasms""[MeSH Terms] OR ""breast carcinoma"" OR ""breast cancer"")) AND ""last 5 years""[PDat]) Sort by: [pubsolr12]"  The LINC0511-HOTAIR subnetwork The intergenic non-protein coding RNA 00511 (LINC00511) participates in a subnetwork with HOTAIR (HOX transcript antisense RNA), which is linked to the methyltransferase EZH2 and causes impaired cell proliferation and inhibition of apoptosis in estrogen receptor (ER) negative BC cells [154]; indeed, LINC00511 promotes metastasis dissemination by silencing NLK [155]. In this subnetwork LINC00511 was proposed to function as a competitive endogenous RNA, sequestering miR-185, with the effect of inducing E2F1 expression, ultimately leading to stemness and tumorigenesis in all BC subtypes [80]. The other subnetwork member HOTAIR, can act as a scaffold for the late endosomal/lysosomal adaptor, MAPK and MTOR activator 5 (HBXIP), which promotes the expression of three MYC targets, i.e. CCNA1, EIF4E and LDHA, as well as of the lysine demethylase 1A (LSD1), recruited by HBXIP itself [156]. A novel isoform of HOTAIR, named HOTAIR-N, was observed in association with an increase of invasion and metastasis in laminin-rich extracellular matrix-based threedimensional organotypic cultures (lrECM 3D), compared with traditional "Claudin-low" culture. HOTAIR-N, once   Table 3 List of non-codingRNAs, their targets and the type of interactions, cited by only one scientific article, and therefore excluded from the network illustrated in Fig. 1 and further described and discussed in the main text of the review  Table 3 List of non-codingRNAs, their targets and the type of interactions, cited by only one scientific article, and therefore excluded from the network illustrated in Fig. 1 and further described and discussed in the main text of the review (Continued)  Table 3 List of non-codingRNAs, their targets and the type of interactions, cited by only one scientific article, and therefore excluded from the network illustrated in Fig. 1 and further described and discussed in the main text of the review (Continued)  Table 3 List of non-codingRNAs, their targets and the type of interactions, cited by only one scientific article, and therefore excluded from the network illustrated in Fig. 1 and further described and discussed in the main text of the review (Continued) *pos positive interaction, activation, neg negative interaction, repression cells are attached to extracellular matrix, binds BRD4, a reader of histone markers that recognizes trimethylation on histone H3 lysine 4 [157].

The H19/LINK-A/MIR2052HG/miR-25/miR-10b/ Eleanor sub-network
This relatively large sub-network is depicted in Fig. 3. Lnc-H19 and Long intergenic non-coding RNA for kinase activation 01139 (LINK-A) are both indirectly involved in the regulation of the expression of HIF1A. In particular, H19 could induce CSC properties and tumorigenesis possibly via LIN28 by acting as a competitive endogenous RNA towards let-7 miRNA. Furthermore, H19 can indirectly stimulate the expression of HIF1A and PDK1, thus promoting the glycolysis pathway, a crucial step in CSC reprogramming. H19 and PDK1 therefore may represent possible therapeutic targets, to contrast glycolysis and cancer stemlike properties [158,159]. Consistently, LINK-A is involved in the normoxic HIF1A stabilization pathway, through the recruitment of the protein tyrosine kinase 6 (BRK) and of LRRK2, that phosphorylate and activate HIF1A itself. From a functional point of view, LINK-A is associated with glycolysis reprogramming in TNBC and promotes tumorigenesis [160]. H19 promotes tamoxifen resistance and autophagy in MCF7 cells, by downregulating Beclin-1 methylation via epigenetic mechanisms. In details, H19 inhibits adenosylhomocysteinase (SAHH), with subsequent acyl-CoA synthetase medium chain family member 3 (SAH) accumulation, which in turn inhibits Beclin-1 promoter methylation by DNMT3B. Therefore the H19/SAHH/DNMT3B axis was proposed as a therapeutic target against tamoxifen resistance [161]. LINK-A is further connected with MIR2052HG, miR-25 and miR-10b, all known activators of AKT1. In this subnetwork a single nucleotide polymorphism (SNP), rs12095274: A > G, in LINK-A affects the phosphorylation status of AKT1 and is associated with AKT inhibitorresistance by AKT-PREX1 interactions, which results in a worse prognosis for patients [162]. Also MIR2052HG presents a SNP (rs13260300), which have been associated with a higher recurrence of BC and resistance to aromatase inhibitors. MIR2052HG positively regulates estrogen receptor alpha (ERα) via the AKT/FOXO3 pathway, and limiting ERα ubiquitination [163]. MIR2052HG has shown to regulate ERα expression by: i) promoting the recruitment of EGR1 on LMTK3 promoter with reduction of PKC activity, indirectly enhancing ERα protein levels; ii) limiting ERα ubiquitination via PKC/MEK/ERK/RSK1 pathway. Both mechanisms have been identified as active in the presence of the MIR2052HG SNP rs13260300 and of aromatase inhibitors in ERα-positive BC [164]. MiR-25 can promote cell proliferation in TNBC by silencing B-cell translocation gene 2 (BTG2) and, indirectly, by the activation of AKT and ERK-MAPK pathways [165]. Additionally it has been reported that miR-25 interacts with miR-93 (not present in this network), to down-regulate CGAS, by targeting NCOA3 at its promoter. Hence, it could determine immune evasion and accelerated cell cycle progression under hypoxia in Luminal A cells [114].
The other microRNA engaged in this network is miR-10b which targets HOXD10 and KLF4 to play a prooncogenic role. It can promote cell invasion and metastasis formation in the TNBC subtype through its secretion via exosomal vesicles, mediated by neutral sphingomyelin phosphodiesterase 2 (nSMase) indeed and it is capable of transforming non malignant HMLE Fig. 3 The H19/LINK-A/MIR2052HG/miR-25/miR-10b/Eleanor sub-network cells into cells with invasion-ability [166]. Metastasis generation and self-renewal of CSCs driven by miR-10b are the results of the directly inhibition of miRNA target, PTEN, and the indirectly increase of the expression of AKT [167], as well as that of HOXD10 and BCL2 like 11(BIM) [168].
For this reason, miR-10b has been proposed as a "metas-tamiR", re-asserted by Kim and co-workers who focused on its targets onco-suppressors Tbx, PTEN, DYRK1A and the anti-metastatic gene HOXD10 [169]. Finally, Eleanor also plays a role in the cluster of non-coding RNAs, cisactivating both ESR1 and FOXO3 [170]. The inhibition of Eleanor could represent a key to switch off topologically associating domain (TAD) containing proteins and to target cells resistant to endocrine therapy [171].

The MALAT1/miR-100 partnership
The sub-network shown in Fig. 4 evidences long noncoding MALAT1 and miR-100. These non-coding RNAs are indirectly interconnected by VEGFA. MALAT1 modulates VEGFA isoforms expression enhancing TP53 mutations in basal-like BC subtype (BLBC). The interaction between MALAT1 and mutant TP53/ID4 is mediated by SRSF1 splicing factor and promotes MALAT1 delocalization from nuclear speckles and its recruitment on VEGFA pre-mRNA [172].. In addition, MALAT1 acts as competitive endogenous RNA to sponge miR-216b, thus restoring the expression of PNPO, which is associated with promoted cell proliferation, migration and invasion in invasive ductal carcinoma (IDC). MALAT1/ miR-216/PNPO pro-metastatic axis represents a target for molecular therapy, as validated in Luminal A and TNBC subtypes [91]. However the role of MALAT1 is still debated. Other studies reported that MALAT1 inhibits the transcription of the pro-metastatic factor TEAD, hindering the interaction between the YAP1 at the TEAD promoters; suggesting MALAT1 as a metastasis-suppressing factor in BLBC [173]. The transfer of miR-100 via MSC-derived exosomes in cancer cells determines the down-regulation of VEGFA secretion by directly targeting mammalian target of rapamycin (mTOR) and modulating mTOR/HIF-1α axis, in fact the miR-100 up-regulation could inhibit angiogenesis and endothelial cell proliferation in the BC microenvironment [174].
Furthermore, mir-100 is negatively correlated with CSC-like self-renewal by inhibiting the SMARCA5, SMARCD1 and BMPR2 regulatory genes in TNBC and Luminal A subtypes. The miR-100 involvement in the inhibition of metastasis has also been validated in vivo [175].
The miR-125a/b-miR196 sub-network Figure 5 shows the miR-125/HER2 subnetwork. MiR-125a/b target the 3'UTR region of both HER2 which elevates HER3 expression levels, thus reducing HER2 mRNA levels and consequently their oncogenic effects in cellular models, including increase of tumour growth rates and trastuzumab resistance [176]. Consistently, the loss of miR-125b promotes HER2 signalling, and is Fig. 4 The MALAT1/miR-100 sub-network Fig. 5 The miR125/HER2 subnetwork associated with poor prognosis in patients with Luminal A tumours [116]. MiR-125a exerts also a crucial role in the regulation of apoptosis by silencing of HDAC5, upon stimulation of the RUNX3/p300 pathway, representing a novel anticancer strategy able to activate caspase 3/9 [177]. Indirectly, also miR-196 contributes to inhibit HER2 expression, by altering HOXB7 and HOXB7-ERα interaction. Nevertheless, miR-196 is down-regulated by MYC, which restores HOXB7 and promotes Luminal A breast cancer tumorigenesis and tamoxifen resistance [178]. On the contrary, Jiang et al. demonstrated that miR-196a, upon stimulation by ER-α interaction, promotes growth of Luminal A breast cancer inhibiting SPRED1, a negative regulator of the RAS/RAF/MAPK signalling, indirectly activated by miR-196 [179].

The miR-182 and miR-96 microRNAs
A study by Yu et al. focuses on the pro-metastatic miR-182, which is associated with EMT, invasion, as well as distant metastasis formation. MiR-182 inhibits the expression of SMAD7, which is both a transcriptional target of TGFβ and a negative regulator of TGFβ signalling [180]. Also, miR-96 modulates the pro-apoptotic FOXO1, a relevant target for precision therapies, and inspired the rational design of TargaprimiR-96 [181]. As a proof of concept, the development of a conjugate small molecule that selectively binds the oncogenic miR-96 hairpin precursor (RIBOTACs), is able to recruit a latent endogenous ribonuclease (RNase L) to FOXO1 transcript, inducing its cleavage. Functionally the silencing of miR-96 de-repressed FOXO1 and induced apoptosis exclusively in TNBC [182]. Other articles highlight an opposite role for these two miRNAs. MiR-96 and miR-182 both target the 3′-UTR region of the PALLD gene. Down-modulation of Palladin transcript expression leads both to decreased migration and invasion of Luminal A breast tumour cells. However, when it is present rs1071738 SNP, a common functional variant of PALLD gene, at the miR-96/miR-182-binding site, the 3'UTR fails to bind the target microRNAs, compromising cell invasion, as verified in in vitro experiments [183].

miR29b and miR-29c
MiR-29b and miR-29c both target chaperone Hsp47, a modulator of the extracellular matrix (ECM) and promoter of BC development; their indirect regulation of ECM genes reduces collagen and fibronectin deposition [184].
Interestingly, miR-29b can act as both inhibitor and promoter of cell proliferation, in Luminal A and TNBC subtypes respectively, based on differential regulation of activation of NFkB and TP53 pathway, mediated by S100A7. In MCF7 cells, S100A7 inhibits NFKB signalling with a consequent upregulation of miR-29b that in turn targets CDC42 and PIK3R1 and indirectly activates TP53 leading to the activation of anti-proliferative pathways. In contrast, in MDA-MB-231 cells, miR-29b which has a lower expression than in MCF7 cells, is suppressed by NFkB with consequent repression of TP53 and promotion of metastasis dissemination [185].

Other non-coding RNAs relevant in breast cancer
In Fig. 1 we showed all sub-networks, whose ncRNAs have been described in at least two different sources from literature.
One of these ncRNAs is the estrogen-inducible long non-coding NEAT1, which has been proposed to act as ceRNA and 'sponge' miR-204. MiR-204 inhibition in turn induced impaired cell proliferation and inhibition of apoptosis. These two processes were supported by the H19 lncRNA [186], to promote para-speckle formation under hypoxia condition, mediated by sequestration of HIF2A and F11 receptor (JAM1) [187]. NEAT1 was also involved in the promotion of invasion, EMT and metastasis dissemination in Luminal A cells by interfering with FOXN3/SIN3A interactions and leading to the repression of GATA3, a crucial regulator of EMT [188].
Another miRNA, miR-27b negatively regulates the acquisition of drug resistance, and is able to induce tumour seeding, two critical properties of CSCs. These effects are mediated by the targeting of ENPP1 and by indirect prevention of the over-expression of ABCG2 transporter. This function was supported by anti-type II diabetes (T2D) drug metformin, that counteracted the generation of CSCs [189]. MiR-27b was also shown to promote the Warburg effect, by inhibiting the PDHX with subsequent dysregulation of the levels of pyruvate, lactate and citrate that increase cell proliferation in the Luminal A and TNBC subtypes [190].
MiR23b has also been subject of recent researches, and itself a notable ncRNA in BC. Its exosome-mediated delivery promoted by Docosahexaenoic acid, an anti-angiogenesis compound, was able to suppress the pro-angiogenic targets PLAU and AMOTL1 in Luminal A and TNBC [96]. Furthermore, in ER-positive endocrine therapy resistant cells, miR-23b was involved in the reprogramming of aminoacid metabolism occurring in association with the downregulation of SLC6A14 aminoacid transporter, the stimulation of autophagy and the import of aspartate and glutamate by SLC1A2 transporter [191].
The lncRNA breast cancer anti-estrogen resistance 4 (BCAR4) is associated with advanced BC and metastasis.
In response to CCL21 chemokine, BCAR4 binds SNIP1 and protein phosphatase 1 regulatory subunit 10 (PNUTS) activating the non-canonical Hedgehog/GLI2 transcriptional program and promoting cell migration [192]. It has been demonstrated that BCAR4 is also involved in the reprogramming of glucose metabolism mediated by YAP1 and favours the transcription of glycolysis promoters HK2 and PFKFB3 via Hedgehogsignalling. The activation of YAP1-BCAR4-glycolisis axis is linked with poor prognosis, and represents an interesting therapeutic target for locked nucleic acids (LNA) delivery, as shown by Zheng et al. [193] In our review, miR-34a appears as the most discussed non-coding RNA, and several independent research groups all pointed it out as an oncosuppressor. MiR-34a, poorly-expressed in TNBC, revealed its antitumorigenic nature by direct targeting of c-SRC [194], GFRA3 [195], and the MCTS1 re-initiation and release factor (MCT-1). Mir-34a also indirectly modulates IL-6, an interleukine associated with breast epithelial acini morphogenesis, and with EMT stimulation in TNBC [196]. Consistently, miR-34a inhibits cancer stem cell properties and promotes doxorubicin sensitivity in MCF7 cells, by targeting NOTCH1. In MCF7 doxorubicin resistant (MCF7/ADR) cells, miR-34a is expressed at low level, possibly due to TP53 mutations [197]. Other effects promoted by miR-34a are the cell-cycle arrest and the apoptosis of TNBC by targeting tRNA i Met and AGO2 [198]. Furthermore, miR-34a negatively regulates the EEF2K and FOXM1 proto-oncogenes, both associated with short-term patient survival [199].
The tumour suppressor miR146a, (and its relative miR-146b) is up-regulated by FOXP3 and targets IRAK1 and TRAF6 causing NF-kB inactivation in the Luminal A subtype. The FOXP3/miR-146/NF-kB axis limits tumour growth and could be a valuable target for therapy [200]. The role of miR-146a includes the reduction of fibronectin and opposing to the epithelial phenotype in TNBC subtype with a pro-metastatic activity supported via the oncosuppressor WWOX, that antagonizes MYC functions [201].

Conclusions
The roles of non-coding RNAs in the establishment and evolution of breast cancer are still under scrutiny by many investigators currently active in the field. In this review we performed an unsupervised and large study of the recent literature in the last quinquennium (2014-2019). We used a data-driven approach in order to produce the most unbiased outcome. Orthogonally, we enforced a strict human based curation of each article selection by the PubMed queries. Only papers that clearly applied mechanistic approaches by using in vitro or in vivo methods were Fig. 6 Synthesis of data-approach used to build the network ncRNAs-target. The PubMed queries used for the articles selection are shown in Table 3 included in this review. Thus, we excluded, and did not report, papers with pure correlative analyses, which albeit revealing would not distinguish a causative action of the non-coding RNAs under scrutiny. All steps of our approach are synthesized in Fig. 6. Funding thank AIRC for their support throughout the years.

Availability of data and materials not applicable
Ethics approval and consent to participate not applicable Consent for publication not applicable

Competing interests
The authors declare that they have no competing interests.