Although ectopic overexpression of miRNAs can influence mammary normal and cancer stem cells (SCs/CSCs), their physiological relevance remains uncertain. Here, we show that miR-146 is relevant for SC/CSC activity. MiR-146a/b expression is high in SCs/CSCs from human/mouse primary mammary tissues, correlates with the basal-like breast cancer subtype, which typically has a high CSC content, and specifically distinguishes cells with SC/CSC identity. Loss of miR-146 reduces SC/CSC self-renewal in vitro and compromises patient-derived xenograft tumor growth in vivo, decreasing the number of tumor-initiating cells, thus supporting its pro-oncogenic function. Transcriptional analysis in mammary SC-like cells revealed that miR-146 has pleiotropic effects, reducing adaptive response mechanisms and activating the exit from quiescent state, through a complex network of finely regulated miRNA targets related to quiescence, transcription, and one-carbon pool metabolism. Consistent with these findings, SCs/CSCs display innate resistance to anti-folate chemotherapies either in vitro or in vivo that can be reversed by miR-146 depletion, unmasking a “hidden vulnerability” exploitable for the development of anti-CSC therapies.

Introduction

Cancer stem cells (CSCs) lie at the apex of the hierarchical cellular organization of different types of solid tumors and are thought to drive tumor initiation, therapy resistance, relapse, and metastasis (Al-Hajj et al., 2003; Dalerba et al., 2007; Visvader and Stingl, 2014). There is evidence that the natural history and clinical outcome of cancers are directly related to CSC content. For instance, poorly differentiated breast cancers (BCs), characterized by unfavorable outcome, display a higher CSC content compared with well-differentiated, good-prognosis BCs (Pece et al., 2010), and a transcriptional signature measuring the degree of “stemness” of BCs was shown to be an independent predictor of prognosis (Pece et al., 2019). Moreover, because of their relative quiescent state, CSCs display resistance to conventional anti-cancer therapies, which typically target highly proliferating cancer cells (Creighton et al., 2009; Diehn et al., 2009; Li et al., 2008; Liu and Wicha, 2010).

The emergence of CSCs has been associated with multiple intrinsic (i.e., genetic) and extrinsic cues, leading to different hypotheses about their origin (Visvader and Stingl, 2014). Stem cell (SC) identity is associated with distinctive features connected to the enactment of vast transcriptional and metabolic programs. For instance, the activation of the epithelial-to-mesenchymal (EMT) transcriptional program has frequently been associated with the acquisition of SC properties, and ectopic expression of EMT transcription factors, such as Snail, Twist, and Zeb1/2, has been shown to induce CSC-like phenotypes in vitro and in vivo (Mani et al., 2008; Scheel et al., 2011).

Metabolic reprogramming is also emerging as a key process supporting both normal and cancer SC biology, with particular catabolic and anabolic pathways associated with, and necessary for, the maintenance of an undifferentiated and pluripotent state (Penkert et al., 2016; Shyh-Chang and Ng, 2017). The switch from oxidative phosphorylation to aerobic glycolysis is a common metabolic trait of CSCs, needed to survive under stressful conditions, to fulfill the demand of essential amino acids, nucleotides, and lipids, and to adapt to changes in the tumor microenvironment (Wong et al., 2017).

miRNAs are a class of small noncoding RNAs (18–22 nt) that function in post-transcriptional regulation of gene expression, acting as “sculptors” of the transcriptome and influencing almost every developmental and disease processes (Bartel, 2018). In BC, a number of miRNAs have been linked to inhibition of the CSC phenotype, namely Let-7a, miR-200c, miR-34a, and miR-93 (reviewed in Tordonato et al., 2015). However, their direct involvement in SC/CSC biology is uncertain, as they are poorly expressed or absent in SCs/CSCs. In addition, these miRNAs inhibit SC phenotypes only upon overexpression, through the induction of differentiation and the inhibition of self-renewal determinants (BMI-1 and Notch), transcription factors (ZEB1/2), or signaling pathways involved in EMT (ZEB1/2, MAPK, and STAT3; Aceto et al., 2012; Iliopoulos et al., 2009; Scheel et al., 2011; Shimono et al., 2009; Wellner et al., 2009).

Here, we report that members of the miR-146 family (miR-146a-5p and -146b-5p) are specifically expressed in the SC compartment of the normal mammary gland and in BC cells displaying CSC features. miR-146 controls SC/CSC identity and highlights a metabolic state, likely coopted from normal SCs, that is associated with an intrinsic resistance to anti-cancer drugs, thus providing evidence of a crosstalk between transcriptional and metabolic programs through miRNA activity.

Results

Identification and characterization of mammary SC-specific miRNAs

To identify miRNAs differentially expressed in mammary SCs versus progenitors, we employed a previously described FACS-based assay that uses the lipophilic dye PKH26 to isolate highly enriched SC versus progenitor populations from mammospheres (Cicalese et al., 2009; Pece et al., 2010). During mammosphere growth, PKH26 is selectively retained by slowly dividing/quiescent SCs (PKHpos), while it is progressively diluted in actively dividing progenitors (PKHneg), permitting the separation of these two populations by FACS.

We analyzed miRNA expression (details in Fig. S1, A–C; and Table S1) in PKHpos (SCs) and PKHneg (non-SCs) cells purified from mammospheres generated from (1) primary mouse mammary epithelial cells (MECs), and (2) the human normal MEC line (MCF10A), which contains a SC-like population that is able to differentiate in vitro (Fig. 1 A; Debnath et al., 2003). In these two cell models, we identified three miRNAs commonly regulated in PKHpos cells, defined here as “SC-specific miRNAs”: miR-146a/b, miR-331, and let-7a (P value of the overlap, <0.01; Fig. 1 B).

In BC, the proportion of cells with tumor-initiating ability (herein operationally equaled to CSCs) correlates with the molecular/biological characteristics of the tumor and its aggressiveness (Clevers, 2011; Pece et al., 2010; Visvader and Lindeman, 2012). We therefore speculated that the SC-specific miRNAs might be differentially expressed in BCs displaying aggressive features. In the cohort of BC patients from The Cancer Genome Atlas (TCGA; Cancer Genome Atlas Network, 2012), the three miRNAs identified a subgroup of cancers with a SC-like expression pattern (Fig. 1 C). These tumors displayed characteristics of aggressive BC associated with poor prognosis, including (1) a predominant basal-like subtype (Fig. 1 D and Table S2); (2) hormone receptor (estrogen receptor [ER] and progesterone receptor [PgR])–negative status (Fig. 1 E and Table S2); and (3) enrichment of p53 mutations/deletions or Myc amplification (Fig. 1 E and Table S2; Deming et al., 2000; Green et al., 2016; Miller et al., 2005).

Similar findings were obtained with an independent BC cohort from the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC; Fig. S1, D–F; and Table S3; Curtis et al., 2012), thus confirming that the SC-miRNAs signature can stratify breast tumors according to their biological and molecular features.

miR-146a/b are enriched in mammary SCs/CSCs versus their progenitors

Of the three identified miRNAs, the relevance of let-7a and miR331-3p to SCs and BC homeostasis has been previously reported (Copley et al., 2013; Leivonen et al., 2014; Yu et al., 2007). We therefore concentrated on miR-146a/b.

Previous studies have shown that mammary SCs and CSCs share common transcriptional traits (Lim et al., 2010; Shipitsin et al., 2007; Visvader and Stingl, 2014). Consistently, our results (Fig. 1, B–E) suggest that the higher levels of expression of miR-146a/b in SCs versus progenitors might be a hallmark of the mammary SC compartment, both in the normal and cancer settings. To investigate this possibility, we initially examined a published list of miRNAs expressed in mammary CSCs, purified using the CD44high/CD24low configuration (Shimono et al., 2009). We found that both miR-146a/b were expressed at higher levels in CSCs versus non-CSCs (Fig. 1 F). We purified CSCs from six primary human BC mammosphere cultures using the PKH26 method and observed higher levels of miR-146a/b in PKHpos versus PKHneg cells (Fig. 1 F and Fig. S1 G). Accordingly, we found miR-146a/b up-regulated in a CD44high/CD24low subpopulation from the human normal mammary cell line, HMLE, which is enriched in SC-like cells (Fig. 1 F; Al-Hajj et al., 2003; Mani et al., 2008). Thus, miR-146a/b are up-regulated in normal and cancer mammary SCs versus non-SCs, regardless of the methodology used for their purification. miR-146 levels were consistently higher in tumor cell lines (Fig. 1 G) and primary tumors (Fig. S1, H–J) displaying basal-like and mesenchymal features. These tumors exhibit the most aggressive disease course, among the various molecular subtypes of BC, and the highest enrichment in CSCs (Pece et al., 2019). Finally, in the METABRIC dataset, we demonstrated that a high level of miR-146a correlated with reduced overall survival at 20 yr (HR 1.22; P = 0.04; Fig. S1 K).

Loss of the miR-146 reduces SC/CSC self-renewal in vitro and in vivo

miR-146a/b might simply represent “markers” of the SC state or be directly involved in the specification/maintenance of stemness traits. We investigated these possibilities in primary mouse MECs and SUM159 cells, a BC cell line containing a subpopulation of cells that behaves as CSCs in vitro and in vivo (Fillmore and Kuperwasser, 2008; Gupta et al., 2011; see also Fig. 1 G). In these cells, we silenced miR-146 family expression with a lentiviral sponge (146 kD), which reduced total miR-146 levels by >50% (Fig. 2, A and B) and the levels of miR-146 loaded on RNA-induced silencing complex (where miRNAs function) by >80%, as assessed by Ago2-RNA immunoprecipitation (RIP; Fig. 2 C). While miR-146 knockdown (KD) had no effect on 2D proliferation (Fig. S2, A–D), it significantly impaired sphere-forming efficiency (SFE; Fig. 2, D and E), suggesting a role of miR-146 in the regulation of self-renewal of normal and cancer mammary SCs. This was assessed directly by limiting dilution transplantation and measuring the frequency of tumor-initiating cells (TICs). We found that the frequency of TICs was significantly reduced in miR-146 KD SUM159 cells as compared with controls (Fig. 2 F). We next used three BC patient-derived xenografts (PDXs) from a triple-negative subtype, which all maintained the histopathological characteristics of their matched primary tumor (Fig. 3 A and Fig. S2 E) and expressed high levels of miR-146 (Fig. 3 B). The silencing of miR-146 in these PDXs reduced the frequency of TICs by fivefold (Fig. 3, C–F; and Fig. S2 F). When we measured proliferation effects by Ki67 staining (which was possible in two out of three PDXs because of availability of material), we did not score differences in KD versus SCRAMBLED (SCR; Fig. S2 G). Together, the in vitro and in vivo data support a role for miR-146 in the maintenance of the homeostasis of the SC compartment in the breast gland.

miR-146 levels stratify cells with SC-like properties

To understand the physiological impact of miR-146 under unperturbed conditions, we generated an miR-146 sensor (Fig. 4, A and B), in which the GFP transgene contained four repeats of a sequence complementary to miR-146a/b in its 3′ UTRs, so that the levels of GFP inversely correlated with miR-146 levels. A second transgene, a truncated form of NGFR (nerve growth factor receptor; ΔNGFR), was used to normalize for lentiviral integrations (Brown et al., 2007).

SUM159 cells, which express high levels of miR-146, showed heterogeneous single-cell expression of miR-146 (Fig. 4, C and D). Thus, we sorted subpopulations with high versus low GFP levels into three different mammary cell types (murine MECs, HMLE, and SUM159) and characterized their biological properties. In all cell types, miR-146high cells (GFPlow) displayed significantly increased SFE versus miR-146low cells (GFPhigh; Fig. 4 E–G), suggesting that high miR-146a/b levels distinguish a subpopulation endowed with SC traits. Consistently, purified CD44high/CD24low SC-like from HMLE cells, which express higher levels of miR-146a/b compared with their CD44low/CD24high counterparts (Fig. 1 F), displayed characteristics similar to miR-146high cells, including a fibroblast-like appearance (Fig. 4 H), high expression of mesenchymal/SC markers (CD44, CDH2, Snai1, and Serpine1), and low levels of the epithelial/differentiation marker CDH1 (Figs. 4, I and J). Ablation of miR-146 in HMLE cells decreased SC-like properties, including the efficiency of mammosphere formation (Fig. 4 K), whereas the KD in CD44high/CD24low cells increased the expression of epithelial/differentiation markers (CDH1, MUC1, CD24, and KRT5/18) with concomitant decrease in the mesenchymal marker CDH2 (Fig. 4, L and M). Finally, we isolated CD44highCD24low cells and followed their reconversion, over time, toward the initial cell heterogeneity in the presence or not of miR-146. As shown in Fig. 4 N, miR-146 KD accelerated the conversion (Mani et al., 2008), confirming that miR-146 is necessary for the maintenance of the mammary SC-like pool.

miR-146a/b modulates multiple pathways in the mammary SC-like compartment and targets quiescence, DNA, and RNA metabolism genes

To gain mechanistic insights into miR-146 functions in the mammary SC compartment, we performed a comparative analysis of the transcriptomes of: (1) miR-146low versus miR-146high cells, and (2) miR-146high cells in which the expression of miR-146 was ablated (miR-146high KD versus SCR; Fig. 5 A). This strategy could enrich for direct miR-146 targets, as these transcripts should (1) inversely correlate with miR-146 levels and (2) be induced upon miR-146 KD. We looked for predicted miR-146 targets using TargetScan7.1 (Agarwal et al., 2015) with stringent criteria (context score below −0.15, n = 945). Overall, we found that the predicted miR-146 target genes were slightly (+0.10 median log2 fold-change) but significantly up-regulated upon either miRNA 146 inhibition (KD) or by comparing cells with different miR-146 levels (146-low versus 146-high), with respect to not-targets (P < 0.001; Fig. 5, B and C.) This magnitude of change is comparable to the change observed in other published work investigating loss of miRNA function (Baek et al., 2008; Wen et al., 2015).

We next investigated whether gene signatures related to mammary SCs correlate with miR-146 levels using gene set enrichment analysis (GSEA). Initially, we used a gene expression dataset generated in our laboratory by comparing the HMLE SC-like (CD44high/CD24low) versus differentiated (CD24high/CD44low) subpopulations and selecting pathways up-regulated or repressed in SC-like cells. Genes up-regulated in SC-like cells (herein referred to as the CD44_UP gene set) were significantly associated with miR-146high cells and down-regulated by miR-146 KD (Fig. 5 D). Conversely, genes down-regulated in SC-like cells (the CD44_DOWN gene set) did not show a coherent reciprocal correlation with miR-146 levels (Fig. S3 A). Using independent mammary SC signatures from normal mouse (Lim et al., 2010; Stingl et al., 2006) or human normal/cancer mammary tissues (Shipitsin et al., 2007), we confirmed that SC-specific gene sets were down-regulated by miR-146 KD and associated with miR-146high cells (Fig. 5 E). In contrast, no coherent correlation was observed for signatures down-regulated in SCs (Fig. 5 E), suggesting that miR-146 expression might be required to sustain genes specifying SC functions, but not directly connected to differentiation. Finally, using the molecular signature database (MSigDB), we noticed that the transition from the miR-146high to the miR-146low state is accompanied by (1) the reduction of several adaptive response pathways involved in mammary SC maintenance as inflammatory (i.e., TNF-α and IFN), p53, hypoxia, and EMT pathways; and (2) the activation of oxidative phosphorylation metabolism, basal transcriptional activity (e.g., Myc targets), and exit from the quiescent state (with activation of the G2–M transition and E2F targets; Fig. S3 B).

To search for direct miR-146–regulated targets in SCs, we adopted a ranking strategy instead of a fold-change cutoff, since transcriptional effects following loss-of-function of miRNAs are typically mild (Baek et al., 2008; Selbach et al., 2008) and may fall out conventional thresholds (see also Fig. 5 B). The four datasets containing miR-146–related gene expression profiles (two datasets each for the 146low/146high and KD/SCR comparisons) were ordered from the most up-regulated to the most down-regulated transcript and divided into 10 bins, to select for transcripts with consistent regulation. The first four bins, of all datasets, were significantly enriched in miR-146–predicted targets (Fig. 5 F, highlighted in red); they were therefore merged to select for commonly induced transcripts (1765_UP genes; Fig. 5 G), including 221 predicted direct targets (miR-146 SC targets; Fig. 5 H). Similarly, the last four bins were depleted of predicted direct targets (Fig. 5 F, highlighted in blue) and were merged to obtain commonly repressed genes (1875_DOWN; Fig. 5 G). In the subset of the 1875_DOWN genes, we detected enrichment of pathways connected to BC aggressiveness, stemness, and SC-related properties, such as EMT, inflammatory pathways, or hypoxia (Fig. S3 C and Table S4), suggesting that miR-146 maintains the SC identity (as observed by GSEA analysis) by indirect transcriptional effects on pathways mostly related to adaptive response mechanisms. Conversely, pathways related to metabolism, RNA transcription, DNA synthesis/repair, and cell cycle/mitosis were enriched among the 1765_UP genes, suggesting a direct role of miR-146 in repressing pathways connected to “exit from quiescence” (Fig. S3 D and Table S4).

When the 221 direct miR-146 targets in SCs were considered alone, they showed a high degree of interconnection, with the most significantly enriched category represented by metabolic processes (one-carbon metabolism, purine synthesis and folate biosynthesis), cell cycle/mitosis, and RNA processing/transcription (Fig. 5 I). Within the one-carbon metabolism category, the direct targets of miR-146 (MTDHF1, MTDHF2, phosphoribosylglycinamide formyltransferase, and dihydrofolate reductase) were also confirmed at the biochemical level by immunoblot analysis on SC-like cells (Fig. S3 E).

miR-146 role in the determination of resistance to therapy

The sum of the previous data suggests that the loss of miR-146 in SC-like cells has pleiotropic effects, reducing adaptive response mechanisms and activating the exit from quiescent state. These pathways might concur with the determination of the SC state imposed by the expression of miR-146.

Among the many properties of CSC, one that is of particular interest for patients’ management is resistance to therapy, a widely reported attribute of CSC (Shibue and Weinberg, 2017) that might be at the basis of therapy failure, especially in the metastatic setting (Oskarsson et al., 2014). We therefore reasoned that drug resistance might represent an exploitable tool to probe into one of the molecular mechanisms through which miR-146 operates, with potential clinical relevance. We employed the SUM159 cell line, which has high CSCs content (SFE ∼15–20%; Fig. 2 D) and high miR-146 levels (Fig. 1 G), and exposed it to several chemotherapeutic drugs, under conditions of miR-146 KD. The miR-146 KD induced a modest effect on the drug sensitivity, as measured by IC50 (half-maximal inhibitory concentration), of almost all tested drugs (Fig. 6, A–C; and Fig. S4 A). In sharp contrast, the effect of methotrexate (MTX), which selectively interferes with the folate pathway (Friedman and Cronstein, 2019), was increased by more than 20-fold (Fig. 6 D). Of note, one-carbon pool and folate metabolism emerged as one the main metabolic pathways targeted by miR-146 in SC-like cells in previous analysis.

To extend the validity of the findings, we employed the mammary cell line HMLE, probing the drug sensitivity in the SC-like fraction as compared with the non-SC one. As shown in Fig. 6 E and Fig. S4 B, the SC-like (CD44highCD24low) subpopulation showed an ∼50-fold reduced sensitivity to MTX versus non-SC cells (CD44lowCD24high) or the bulk population (IC50: 3,560, 71, and 65 nM, respectively). This behavior was dependent on miR-146 levels, since SC-like cells increased sensitivity to MTX (511 nM, 6.9-fold) upon miR-146 KD (Fig. 6 E and Fig. S4 C). Importantly, the non-SC population did not show any change in MTX sensitivity upon miR-146 manipulation (KD or overexpression; Fig. S4 D).

To obtain formal proof for this concept in a tumoral context, we explored the effects on MTX sensitivity upon miR-146 manipulation using in vivo transplantation experiments. SUM159 cells (with or without miR-146 KD) were transplanted orthotopically in the mammary gland of mice by intra-nipple injection and grown either in untreated (saline) or with MTX (60 mg/kg, four cycles) conditions (Fig. 6 F). The KD of miR-146 resulted in slight longer latency in the appearance of palpable tumors versus controls, followed by tumor development with comparable kinetics (Fig. 6 G, left). Though, the number of CSCs, as measured in retransplantation experiments by TIC frequency, was significantly diminished (Fig. 6 H, upper), in agreement with results obtained previously (Fig. 2 D) and in PDXs (Fig. 3, D–F).

The treatment with MTX had a modest effect on tumor growth in control cells, while the combination of miR-146 KD and MTX displayed a potent synergistic effect (Fig. 6 G, right). This interaction was also evident on CSC number, measured in limiting dilution experiments without any additional further treatment. As summarized in Fig. 6 H, MTX had no effects on TIC frequency of control cells, but enhanced significantly the effects of miR-146 KD, with TIC frequency reduced from fourfold to 15-fold as compared with control.

The sum of the above data strongly argues that loss of miR-146 sensitizes tumors to chemotherapy and antifolate treatment in particular, directly targeting the CSC pool.

Discussion

In this study, we sought to identify miRNA(s) required for the maintenance of the mammary SC phenotype and also “inherited” by the CSC compartment, which could represent potential therapeutic targets in BC. The miR-146 family fulfills these characteristics as (1) they are expressed at high levels in mammary SCs and CSCs versus more differentiated progeny; (2) their depletion leads to loss of SC features in vitro and in vivo and accelerates the conversion of SCs to non-SCs; and (3) their depletion in SC and CSCs causes increased sensitivity to the chemotherapeutic agent MTX. Thus, the miR-146 family appears to be specifically required to maintain the SC identity in the mammary tissues. In this regard, miR-146s are functionally similar to other SC-specifying miRNAs, such as the miR-290/302 family in embryonic SCs (Wang et al., 2008) or miR-125b in the skin (Zhang et al., 2011).

The role of miR-146 in cancer is, perhaps not surprisingly, rather complex. In BC, previous reports have linked miR-146 expression to the basal subtype (Forloni et al., 2014; Garcia et al., 2011). While we confirmed this association using large BC datasets (TCGA or METABRIC), we have reasons to believe that this is not an intrinsic property of basal BCs per se, but rather a reflection of the high CSC content of these tumors (Pece et al., 2010). Indeed, in every condition that we analyzed, the expression of miR-146 was heterogeneous at the single-cell level and segregated with SC-like phenotypes. In other cancers, miR-146 expression has been reported to be either down- or up-regulated, depending on the context (Testa et al., 2017). However, in most cases, bulk cell populations were analyzed, and some ambiguity in miR-146 expression levels might derive from the presence of nonepithelial contaminants, such as macrophages and regulatory T cells, which are known to express miR-146 at high levels, in particular during inflammatory response (Lu et al., 2010). Thus, it is yet to be established how exactly miR-146 expression is in other epithelial cancers, if it is heterogeneous at the intratumoral level, and, most importantly, whether it demarcates the CSC population. A role for miR-146 in CSCs has been described in colorectal cancer, where it promotes a symmetric mode of division through the Snail/miR-146a/β-catenin/Numb axis (Hwang et al., 2014), and in glioma, where it was shown to inhibit neurosphere formation and tumor development by targeting NOTCH1 (Mei et al., 2011). Therefore, while miR-146 has been linked to CSC behavior in various contexts, the molecular mechanism through which it operates could be rather context-specific.

Different functions, sometimes underlying opposing effects on cancer phenotypes (oncogenic versus tumor-suppressive), have been reported for the miR-146 family, which could be explained by the wide spectrum of miR-146 target genes. With the exception of a few common genes belonging to the inflammatory signaling cascade (e.g., TRAF6 and IRAK1/2), miR-146 targets appear to be context specific (Taganov et al., 2006). We therefore investigated the genes and the pathways under the control of miR-146 within the mammary SC compartment. We combined the isolation of a miR-146high SC-like population by miR sensor with a loss-of-function approach, to identify broad transcriptional effects of miRNAs under physiological conditions of expression, which might be very different from those observed upon ectopic (and nonphysiological) overexpression. By this approach, we identified a plethora of transcripts that are regulated, directly or indirectly, by miR-146 and might participate in its control over the maintenance of the SC/CSC-like state. Indeed, regulated genes belong to pathways and cellular functions that have been widely linked either to the exit from quiescence (i.e., activation of oxidative phosphorylation metabolism and of the G2–M transition, E2F targets, and cell cycle genes) or to transcriptional programs that promote SC phenotype (inflammatory pathways, hypoxia, and EMT).

The significance and the impact of miR-146 within each of these pathways remain hard to be established, as they are frequently interconnected; however, by narrowing down the candidate list to the set of putative direct miR-146 targets (221 genes), there was significant enrichment of genes acting in a few specific pathways: metabolism, RNA transcription, DNA synthesis/repair, and cell cycle/mitosis.

This allowed us to establish a mechanistic proof of principle linking miR-146 to at least one of the phenotypic properties of CSCs, i.e., drug resistance, in particular resistance to MTX, a chemotherapy agent and immune system suppressant, widely used for the treatment of a variety of cancers, including advanced-stage BC (Yang et al., 2020b). The genetic interaction between miR-146 and MTX has been revealed as highly effective both in vitro and in vivo and further supported by the regulation of the enzymes of the folate biosynthetic pathway by miR-146 observed in the transcriptomic analyses. In addition, drug sensitivity was affected only and specifically in the SC-like population, while no major effects were observed in the non-SC population (Fig. S4 D), which further suggests a context-specific role for miR-146 in the breast SC, rather than a more general effect on bulk epithelial cells.

Our results suggest that miR-146 KD did not induce a general increase in sensitivity to anti-cancer drugs. This would have indicated, in all probability, that drug sensitivity/resistance followed cell identity rather than the specific miR-146–dependent metabolic profile. Rather, loss of expression of miR-146 seemed to confer sensitivity to a specific drug, MTX, as hypothesized based on the transcriptional metabolic pattern; this, in turn, asks questions about the exact mechanism of miR-146–induced MTX resistance.

In this regard, we envision a possible explanation. Folate is the critical cofactor of one-carbon pool metabolism, a process that directly controls nucleotide biosynthesis (purines and pyrimidines), amino acid homeostasis (glycine, serine, and methionine), availability of methyl-groups (methionine/homocysteine), and redox defense (glutathione; Ashkavand et al., 2017; Ducker and Rabinowitz, 2017; Locasale, 2013). In SCs/CSCs, the limited availability of the pathway (determined by the high miR-146 levels) might impose (or contribute to) a quiescent state, which is a hallmark of the SC-like state. Under these conditions, the cell might be refractory to the inhibition of the pathway, simply because it does not depend on it. Upon exit from the SC-like state and entrance in the transit-amplifying compartment, anabolic cellular demands might require the switch of miR-146 (or the switch of miR-146 might license the cell to fulfill these demands). This situation is mimicked by miR-146 ablation, which we note is indeed accompanied by increased basal transcriptional activity (e.g., Myc targets) and exit from the quiescent state (with activation of the G2–M transition and E2F targets; Fig. S4 B). Under these conditions, the cellular demand and dependency on folate metabolism would represent a “fragility” point that can be unmasked by anti-folate treatment. Alternatively, MTX, which is also known as an anti-inflammatory drug, could cooperate with miR-146 loss in the suppression of the inflammatory (IL1, IL6, and TNF-α) signaling pathway, which is required for sustaining the identity of SCs/CSCs (Yang et al., 2020a). In the context of SCs/CSCs, the optimal output of the inflammatory signaling pathway could be provided through endogenous miR-146, which modulates (and is modulated by) NF-κB activity through a negative feedback loop (Taganov et al., 2006). In the absence of miR-146, NF-κB signaling pathway loses robustness, and thus, SCs/CSCs might become susceptible to the anti-inflammatory action of MTX.

While the molecular validation of these scenarios would require further analysis and metabolic profiling of SC-like versus non-SCs in the presence/absence of miR-146, at the biological level, our results clearly show that interference of miR-146 expression represents an attractive approach to overcome some forms of drug resistance in the clinical settings.

Materials and methods

Cell biology procedures and flow cytometry

The SUM159PT cell line (Asterand) was cultured in Ham’s F12 medium with 5% fetal bovine North American serum, human insulin (5 μg/ml), hydrocortisone (1 μg/ml), and Hepes (10 mM). The MCF10A cell line (American Type Culture Collection) was cultured in DMEM/F-12 (1:1) with 5% horse serum, hydrocortisone (500 ng/ml), human insulin (10 μg/ml), cholera toxin (100 ng/ml), and human EGF (20 ng/ml). HMLE cells were kindly provided by Robert Weinberg’s laboratory (Whitehead Institute for Biomedical Research and Department of Biology, Massachusetts Institute of Technology, Cambridge, MA) and were grown in mamamry epithelial cell growth medium (Lonza) according to the manufacturer’s protocol. Mammary glands from 5-wk-old FVB/Hsd females (Harlan Laboratories) were established as described previously (Cicalese et al., 2009). Briefly, glands were mechanically and enzymatically digested in EDM medium: DMEM plus Ham’s F12 (1:1) medium supplemented with human insulin (1 µg/ml,) hydrocortisone (0.5 µg/ml), human EGF (20 ng/ml), 200 U/ml collagenase type 1A (Sigma-Aldrich), and 100 U/ml hyaluronidase (Sigma-Aldrich) for 3 h at 37°C. After digestion, cell suspension was filtered through 100-, 70-, 40-, and 20-µm filters, and red blood cells were lysed using ammonium-chloride-potassium lysis buffer. MECs from primary tumors were collected at the European Institute of Oncology (Milan, Italy) from patients who had given the informed consent to use of biological materials for scientific purposes. Primary tissues were digested as described in Dontu et al. (2003). All the cells were grown in a humidified atmosphere at 5% CO2 at 37°C, except for SUM159PT, which were grown at 10% CO2.

For mammosphere culture, SUM159, MCF10A, and HMLE cells were plated in ultra-low attachment dishes (Falcon) coated with Poly(2-hydroxyethyl methacrylate) (Sigma-Aldrich) at a density of 1,000 cells/ml in serum-free mammary epithelial medium (Lonza) supplemented with 5 µg/ml insulin, 0.5 µg/ml hydrocortisone, 2% B27 (Invitrogen), 20 ng/ml EGF, 20 ng/ml human b-FGF, and 4 µg/ml heparin. Mammosphere cultures of human and mouse primary samples were plated at a density of 5,000 cells/ml. For serial propagation, mammospheres were collected after 7 d of culture, enzymatically dissociated with trypsin-EDTA (0.025%), and plated at the same density for successive generations. PKH26 (Sigma-Aldrich) staining was performed on MCF10A and primary tissues as described in Cicalese et al. (2009) and Pece et al. (2010; Fig. S1 G). PKH-labeled mammospheres were collected after 7–8 d and enzymatically dissociated with trypsin-EDTA to a single-cell suspension. Each human PKH-labeled mammosphere preparation was depleted of contaminants with CD31 and CD45 microbeads (MACS technology) and subsequently stained with DAPI (1 mg/ml, diluted 1:200 in PBS) for 1 min at RT to select for living cells. Finally, cells were FACS sorted to collect PKHpos and PKHneg cells in 96-well plates.

Cells infected with miR-146 sensors were blocked with PBS 10% BSA for 10 min at 4°C and stained with anti-ΔNGFR/PE-cy7 (CD271-PeCy7; clone C40-1457; BD Pharmigen) for 15 min at 4°C. Two-color flow cytometry (GFP and PE-cy7) was used to collect ΔNGFR+-GFPhigh and/or ΔNGFR+-GFPlow populations. To FACS sort CD44highCD24low and CD44lowCD24high populations from HMLE, cells were blocked with PBS-BSA 10% for 1 h at 4°C and then stained with CD44-APC (clone C26; BD Pharmigen) and CD24-PE (clone ML5; BD Pharmigen) antibodies for 45 min at 4°C.

Lentiviral constructs and viral infection

Lentiviral backbone (Bd.LV.miRT vector) for miR-146 sensor was courtesy of L. Naldini (San Raffaele Telethon Institute for Gene Therapy, San Raffaele Scientific Institute, Milan, Italy) and modified as follows: two DNA sequences containing four miRNA response elements with perfect complementarity to miR-146a/b at the 3′ UTR of GFP (miRT) were synthesized (Primm) as follows: sensor 146 sense: 5′ → 3′, CTA​GAA​AGC​CTA​TGG​AAT​TCA​GTT​CTC​ACG​ATA​AGC​CTA​TGG​AAT​TCA​GTT​CTC​AAC​CGG​TAA​GCC​TAT​GGA​ATT​CAG​TTC​TCA​TCA​CAA​GCC​TAT​GGA​ATT​CAG​TTC​TCA​C; sensor 146b antisense: 5′ → 3′, CCG​GGT​GAG​AAC​TGA​ATT​CCA​TAG​GCT​TGT​GAT​GAG​AAC​TGA​ATT​CCA​TAG​GCT​TAC​CGG​TTG​AGA​ACT​GAA​TTC​CAT​AGG​CTT​ATC​GTG​AGA​ACT​GAA​TTC​CAT​AGG​CTT.

1 µl of each oligo (100 µM) was annealed in a final volume of 50 µl Annealing Buffer (Promega) for 4 min at 95°C, then 10 min at 70°C. Diluted annealed oligos (1:10) were ligated with 100 ng of lentiviral backbone (Bd.LV.miRT vector) doubled-digested with XhoI and XbaI. Ligation protocol was performed with Quick T4 DNA Ligase according to the manufacturer’s indications (New England Biolabs). After cloning, each positive clone sequence was verified by DNA sequencing. miR-146 KD and nontargeting scramble control (CTRL) were commercially available vectors (pmiRZIP lentivector) from System Bioscience (clone MZIP000-PA-1 for CTRL and MZIP146b5p-PA-1 for miR-146 KD).

For virus packaging, pRSV-Rev, pMDLg/pRRE (gag&pol), pMD2.G (VSV-G), and lentiviral vectors (pmiRZIP lentivector or miR-sensor) were cotransfected in HEK293T cells via the calcium phosphate method. The viral supernatant was collected at 36 h after transfection, filtered with a 0.22-µm syringe filter, and ultra-centrifuged for 2 h at 19,800 rpm at 4°C. The viral pellet was then resuspended in mammary epithelial medium at 100× concentration. Viral stock was frozen (−80°C) or directly used to infect target cells in the presence of 1 µg/ml of polybrene. Cells infected were then selected with puromycin for 2–3 d to select stable clones.

miR-146 overexpression and KD

For miR-146 overexpression, cells were transfected with HiPerFect (Qiagen) according to the fast-forward protocol with miRNA Mimic (we used for miR-146 overexpression mimic MSY0002809; and for control, the all-star negative control siRNA SI03650318; Qiagen) at a final concentration of 50 nM. For miR-146 KD, cells were transfected with HiPerFect (Qiagen) according to the fast-forward protocol with the miRNA power family inhibitor at a final concentration of 100 nM (hsa-miR-146 miRCURY LNA microRNA Power family inhibitor; and as control, Negative Control A; Exiqon).

Cell viability analysis

SUM159 cells infected with miR-146 KD lentivirus were plated in 96-well plates (5,000 cells/well) and treated with drugs at different concentrations for 72 h. HMLE CD44high24low or CD44low24high were plated in 12-well plates, then transfected with anti-miR146 or CTRL oligos. 24 h after transfection, cells were treated with MTX for 72 h. Viability was assessed using the Cell-Counting Kit-8 viability kit (CCK-8; Dojindo) according to the manufacturer’s protocol.

Serial transplantation of human PDXs and in vivo experiments

Immunodeficient NOD.Cg-PrkdcscidIL2rgtm1Wjl/SzJ mice were anesthetized by intraperitoneal injections of 150 mg/kg of tribromoethanol (Avertin), and fresh specimens from human primary tumors were implanted in the fourth inguinal mammary gland of 4–5-wk-old animals. Mice were monitored twice weekly by an investigator and were euthanized after 3–5 mo when the tumors were ∼0.5–1 cm in the largest diameter (depending on the intrinsic variability of human specimens). Human PDXs were collected and mechanically/enzymatically digested in EDM medium for 4 h at 37°C. Cell suspensions were filtered through 100-, 70-, 40-, and 20-µm filters, and red blood cells were lysed with ammonium-chloride-potassium lysis buffer. After 24 h in mammosphere culture, cells were cleaned of murine contaminants with the mouse epithelial cell enrichment kit (StemCell Technologies) and the dead cells removal kit (Miltenyi Biotec). Pure human epithelial populations were then infected with CTRL or miR-146 KD lentivirus and puromycin-selected. SUM159 were infected with CTRL or miR-146 KD lentivirus and puromycin-selected before injection. For in vivo limiting dilution transplantation experiments, decreasing concentrations of infected cells (SUM159 or human PDXs) were resuspended in a mix of 14 µl PBS and 6 µl Matrigel and transplanted via intra-nipple injection in the fourth inguinal mammary gland of 6–8-wk-old animals. Animals were euthanized after 1–5 mo (depending on tumor latency) when the tumors were ∼0.5–1.2 cm in the largest diameter. Transplantation frequency was calculated with the Extreme Limiting Dilution Analysis (ELDA) web tool, and single-hit assumption was tested for each experimental setting. For therapeutic treatment with MTX, SUM159 tumors with miR-146 KD or CTRL were monitored until they reached a mean volume of 4–6 mm3. Tumor volume was calculated according to the formula (L * W2/2), where L is the length of the longest diameter and W is the length of the shorter diameter. Mice were randomly assigned to different groups (treated or saline) with a minimum of three to five animals/group. Animals received intraperitoneal injection of vehicle drug (saline) or MTX at 60 mg/kg dosed every 5 d, for a total of four injections. Changes in tumor burden were assessed every 3 d with calipers. Animals were euthanized after 40 d when the tumors reached ∼1.2 cm in the largest diameter. Tumors, treated or not with MTX, were digested as previously described. After 24 h in mammosphere culture, we checked for GFP expression, we removed murine contaminants and dead cells, and then we reinjected the cells (SCR ± MTX or KD ± MTX) at limiting dilutions. All animal studies were conducted with the approval of the Italian Minister of Health (762/2015-PR) and were performed in accordance with Italian law (D.lgs. 26/2014), which enforces Directive 2010/63/EU of the European Parliament and of the Council of September 22, 2010, on the protection of animals used for scientific purposes.

Immunoblotting and immunohistochemistry

Cell lysates were extracted with RIPA lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 1% sodium deoxycholate, and 0.1% SDS), supplemented with a protease inhibitor cocktail (Calbiochem) and phosphatase inhibitors. Lysates were clarified by centrifugation at 16,000 g for 10 min at 4°C, and protein concentration was measured by the Bradford assay (Bio-Rad) according to the manufacturer’s instructions. Proteins were resolved in 4–20% Protean TGX Precast gel (Bio-Rad), then transferred to nitrocellulose filters. Filters were blocked in 5% milk in TBS 0.1% Tween. After blocking, filters were incubated with the following primary antibodies: phosphoribosylglycinamide formyltransferase (4D6-1D5; NovusBio), dihydrofolate reductase (EPR5284; Abcam), MTHFD1 (C-3; Santa Cruz), and MTHFD2 (D8W9U; Cell Signaling). As a normalizer, we used γ-tubulin (homemade clone).

Filters were finally incubated with the appropriate HRPconjugated secondary antibody (anti-mouse IgG HRP-linked 7076 or anti-rabbit IgG HRP-linked 7074; Cell Signaling) diluted 1:2,000 in TBS 0.1% Tween for 30 min. The signal was revealed using the ECL method (Amersham) with Image Lab software (v3.0; Bio-Rad).

Paraffin sections were twice deparaffinized with Bio Clear (Bio-Optica) for 15 min and hydrated through graded alcohol series (100%, 95%, and 70% ethanol and water) for 5 min. Antigen unmasking was performed with 0.1 mM citrate buffer (pH 6) or EDTA (pH 8) for 50 min at 95°C. Slides were cooled for 20 min at RT then washed in water and treated with 3% H2O2 for 5 min at RT. Then, slides were preincubated with an antibody mixture (2% BSA, 2% normal goat serum, 0.02% Tween 20 in TBS) for 20 min at RT and stained with primary antibody for 1 h at 37°C. As primary antibodies, we used rabbit anti-human estrogen receptor (1:40; Dko), mouse anti-hKi67 (1:200; Dko), mouse anti-cytokeratin 5 (1:200; Abcam), mouse anti-cytokeratin 8 (1:10; Abcam), and mouse anti-vimentin (1:50; Dko). Slides were then incubated with a secondary antibody (DAKO Envision system HRP rabbit or mouse) for 30 min at RT and finally incubated in peroxidase substrate solution (DAB DAKO) for 2–10 min. Stained slides were digitalized at 20× magnification using the Aperio Scanscope XT (Leica Biosystems) and acquired with the Aperio ImageScope software (Leica Biosystems).

Total RNA extraction and RT quantitative PCR (qPCR)

Cells were lysed in TRIzol reagent (Invitrogen), and total RNA was extracted with miRNeasy Mini columns or miRNeasy micro-columns according to the manufacturer’s protocol. miRNAs were reverse-transcribed with an miScript II reverse transcription kit (Qiagen), and mature miRNAs were detected with miR-146a (MS00003535) and miR-146b-5p (MS00003542) assays from Qiagen. As controls, we used SNORD61 (MS00033705) or SNORD72 (MS00033719). For gene detection, total RNA was reverse-transcribed with the SuperScript VILO cDNA Synthesis Kit (Life Technologies), and genes were analyzed with Quantifast SYBR green master mix (Qiagen) or SsoFast supermix (Bio-Rad). The complete list of primers used in this study is as follows: CD44: (forward) 5′-ATA​GCA​CCT​TGC​CCA​CAA​TG-3′, (reverse) 5′-TTG​CTG​CAC​AGA​TGG​AGT​TG-3′; CD24: (forward) 5′-TCA​GGC​CAA​GAA​ACG​TCT​TC,-3′, (reverse) 5′-TCC​TTG​CCA​CAT​TGG​ACT​TC-3′; CDH1: (forward) 5′-TGC​CCA​GAA​AAT​GAA​AAA​GG-3′, (reverse) 5′-GTG​TAT​GTG​GCA​ATG​CGT​TC-3′; CDH2: (forward) 5′-ACA​GTG​GCC​ACC​TAC​AAA​GG-3′, (reverse) 5′-CCG​AGA​TGG​GGT​TGA​TAA​TG-3′; RPLP0: (forward) 5′-TTC​ATT​GTG​GGA​GCA​GAC-3′, (reverse) 5′-CAG​CAG​TTT​CTC​CAG​AGC-3′; ACTB: (forward) 5′-TCT​ACA​ATG​AGC​TGC​GTG​TG-3′, (reverse) 5′-TGG​ATA​GCA​ACG​TAC​ATG​GC-3′; SNAI1: (forward) 5′-GGT​TCT​TCT​GCG​CTA​CTG​CT-3′, (reverse) 5′-TAG​GGC​TGC​TGG​AAG​GTA​AA-3′; SNAI2: (forward) 5′-ACG​CCT​CCA​AAA​AGC​CAA​AC-3′, (reverse) 5′-ACA​CAG​TGA​TGG​GGC​TGT​ATG-3′; SERPINE1: (forward) 5′-AAG​ACT​CCC​TTC​CCC​GAC​TC-3′, (reverse) 5′-CAG​TGC​TGC​CGT​CTG​ATT​TGT-3′; MUC1: (forward) 5′-TGC​CGC​CGA​AAG​AAC​TAC​G-3′, (reverse) 5′-TGG​GGT​ACT​CGC​TCA​TAG​GAT-3′; KRT5: (forward) 5′-AGG​AGT​TGG​ACC​AGT​CAA​CAT-3′, (reverse) 5′-TGG​AGT​AGT​AGC​TTC​CAC​TGC-3′; KRT14: (forward) 5′-TGA​GCC​GCA​TTC​TGA​ACG​AG-3′, (reverse) 5′-GAT​GAC​TGC​GAT​CCA​GAG​GA-3′; KRT8: (forward) 5′-CAG​AAG​TCC​TAC​AAG​GTG​TCC​A-3′, (reverse) 5′-CTC​TGG​TTG​ACC​GTA​ACT​GCG-3′; KRT18: (forward) 5′-TCG​CAA​ATA​CTG​TGG​ACA​ATG​C-3′, (reverse) 5′-GCA​GTC​GTG​TGA​TAT​TGG​TGT-3′; CD49f: (forward) 5′-ATG​CAC​GCG​GAT​CGA​GTT​T-3′, (reverse) 5′-TTC​CTG​CTT​CGT​ATT​AAC​ATG​CT-3′; EPCAM: QT00000371 (Qiagen); mKI67: HS01032443_m1 (TaqMan); and RPLP0_TaqMan: (forward) 5′-CCA​TTG​AAA​TCC​TGA​GTG​ATG​TG-3′, (reverse) 5′-TCG​CTG​GCT​CCC​ACT​TTG-3′.

miRNAs high-throughput (HT) profile and low sample input (LSI) protocol

For the analysis of PKHpos and PKHneg cells isolated from murine primary MECs, we reverse-transcribed total RNAs with Megaplex RT Primers mix and amplified with Megaplex PreAmp Primers (rodent pool A). For miRNA HT profiling, we used the TaqMan Low Density Array Rodent V2.0 (Applied Biosystems), following the manufacturer’s instructions. For the analysis of PKHpos and PKHneg cells from MCF10A mammospheres, we isolated 40 PKHpos and PKHneg cells, lysed directly in single-cell Lysis Buffer (Ambion). Total RNA was reverse-transcribed with Human Megaplex RT Primers mix and amplified with Human Megaplex PreAmp Primers (pool A). For miRNA HT profiling, we used the TaqMan Low Density Array Human V2.1 (Applied Biosystems; Table S4).

For the LSI setup, we collected by FACS sorting no more than 200 PKHpos and 200 PKHneg cells in 96-well plates in 10 µl of Single Cell Lysis Kit plus DNase (Ambion). Total RNAs were reverse-transcribed using Human Megaplex RT Primers mix, followed by preamplification with Human Megaplex PreAmp Primers (pool A). Then HT qPCR profiling was performed on TaqMan human platform A V2.1 (Applied Biosystems). We improved the original protocol from Applied Biosystems (Table 1).Raw data (i.e., cycle threshold [Ct] values) were exported to Excel (Microsoft). miRNAs with raw Ct >28 or not expressed (e.g., not amplified) were excluded from the analysis. Expressed miRNAs (Ct <28) were then normalized over the median of housekeeping controls (RNU44, RNU48, and RNU6B) for human array and over the median of U6b, SnoRNA135, and SnoRNA202 for rodent array. Regulated miRNAs were selected based on the following criteria: P value < 0.05, |log2 fold| > 0.5.

Ago2 RIP

The Ago2 RIP experiment was performed using the Imprint RNA Immunoprecipitation kit (Sigma-Aldrich). Briefly, 107 cells were lysed in mild lysis buffer (plus Protease Inhibitor Cocktail and RNase inhibitor) for 15 min on ice. Then the lysate was pelleted at 16,000 g, 4°C, for 10 min. A fraction (5%) of supernatant was collected as input for RNA control. For each RIP, protein A magnetic beads were preloaded with 2.5 µg Ago2 antibody (rat monoclonal; clone 11A9; Sigma-Aldrich) or 2.5 µg of IgGs from rat serum, at RT for 30 min with rotation. RNAs were immunoprecipitated with Ago2 antibodies or rat IgGs overnight at 4°C with rotation. RIPs were then washed, and RNA was purified with TRIzol LS reagent (Life Technologies) plus miRneasy Micro kit and analyzed with RT-qPCR.

RNA sequencing (RNA-seq) and GSEA analysis

Total RNA was extracted with the miRNeasy Micro kit (Qiagen) and treated on-column with DNase (Qiagen). Then 500 ng was purified with the Ribozero rRNA removal kit (Illumina). Libraries were generated with the TruSeq RNA Library Prep Kit v2 (Illumina). Next, sequencing was performed on Illumina HiSeq 2000 at 50-bp single-read mode and 50 million reads depth. RNA-seq Next Generation Sequencing reads were aligned to the human hg38 gencode v25 reference genome using the TopHat aligner (version 2.0.6) with default parameters. Differentially expressed genes were identified using the Bioconductor package DESeq2 based on read counts, considering genes whose q value relative to the control is lower than 0.05 and whose maximum expression is higher than reads per kilobase of exon per million mapped reads of 1.

GSEA (http://www.broadinstitute.org/gsea/index.jsp) was performed using the 11,000 genes expressed in HMLE cells obtained from RNA-sequencing in Fig. 5. As gene sets to calculate the normalized enrichment scores, we used four SC signatures (CD44high, Polyak, Stingl, and Visvader) subdivided in STEM_UP and STEM_DOWN genes. P values were calculated by performing 1,000 random permutations of gene labels to create ES-null distribution.

Data availability

The RNA-seq dataset for this study has been deposited in GEO under accession no. GSE131876.

Statistics

All the analyses (Oneway, Contingency, Principal Component Analysis, IC50 calculation) and statistics related were produced using JMP 12 (SAS) software. Microsoft Excel was used to generate bar graphs with average and SD of repeated experiments, with number of replicates and the statistical test indicated in figure legends. Hierarchical clustering was generated by Cluster 3.0 software (C Clustering Library 1.53) and heatmaps by Java TreeView software (http://jtreeview.sourceforge.net) for Mac OSX.

Clinical samples

Fresh or archival formalin-fixed paraffin-embedded mammary primary specimens were collected at the European Institute of Oncology, via standard operating procedures approved by the Institutional Ethical Board. Only samples for which patients gave informed consent were used in the present study.

Online supplemental material

Fig. S1 shows generation of the LSI protocol and analysis of miR-146 levels in BC datasets. Fig. S2 shows characterization of human BC PDXs and effects of miR-146 KD on proliferation. Fig. S3 shows analysis of miR-146 transcriptional effects. Fig. S4 shows sensitivity to chemotherapy treatment upon miR-146 manipulation. Table S1 shows raw data from the LSI protocol.Table S2 shows analysis of SC-specific miRNAs’ signature in the TCGA dataset. Table S3 shows analysis of SC-specific miRNAs’ signature in the METABRIC dataset. Table S4 shows a list of pathways regulated by miR-146 in SC-like cells.

Acknowledgments

We thank first all the patients who donated their biopsy specimens for research purposes; the Genomic Unit at the Italian Institute of technology for sequencing runs; the European Institute of Oncology (IEO) Imaging Unit for FACS-sorting experiments; the Veterinary Facility at FIRC Institute of Molecular Oncology; C. Luise, G. Jodice, and G. Bertalot at the IEO Molecular Pathology Unit for the immunohistochemistry analyses; M. Coazzoli for technical assistance with in vivo experiments; S. Confalonieri for the survival analysis on the METABRIC dataset; the IEO Pharmacy for providing drugs; and R. Gunby for critically editing the manuscript.

This work was supported by grants from the Associazione Italiana per la Ricerca sul Cancro (MCO10000, IG18988, and IG23060 to P.P. Di Fiore; IG14085, IG18774, and IG22851 to F. Nicassio), the Fondazione Italiana per la Ricerca sul Cancro triennial fellowship “Livia Perotti” (project code 18224 to C. Tordonato), the Associazione Italiana per la Ricerca sul Cancro fellowship “Isabella Gallo” (project code 22386 to G. Giangreco), the Fondazione Cariplo (2015-0590 to F. Nicassio), the Italian Ministry of University and Scientific Research (to P.P. Di Fiore), and the Italian Ministry of Health (RF-2016-02361540 to P.P. Di Fiore). This work was also partially supported by the Italian Ministry of Health with Ricerca Corrente and 5x1000 funds.

The authors declare no competing financial interests.

Author contributions: Experiments were performed by C. Tordonato; M.J. Marzi performed RNA-seq analyses; S. Freddi and C. Tordonato performed sorting and imaging; D. Tosoni initially established human PDXs; and P. Bonetti and G. Giangreco helped with mouse experiments. C. Tordonato, F. Nicassio, and P.P. Di Fiore planned the experiments, analyzed the results, and wrote the manuscript.

References

Aceto
,
N.
,
N.
Sausgruber
,
H.
Brinkhaus
,
D.
Gaidatzis
,
G.
Martiny-Baron
,
G.
Mazzarol
,
S.
Confalonieri
,
M.
Quarto
,
G.
Hu
,
P.J.
Balwierz
, et al
.
2012
.
Tyrosine phosphatase SHP2 promotes breast cancer progression and maintains tumor-initiating cells via activation of key transcription factors and a positive feedback signaling loop
.
Nat. Med.
18
:
529
537
.
Agarwal
,
V.
,
G.W.
Bell
,
J.W.
Nam
, and
D.P.
Bartel
.
2015
.
Predicting effective microRNA target sites in mammalian mRNAs
.
eLife.
4
:e05005.
Al-Hajj
,
M.
,
M.S.
Wicha
,
A.
Benito-Hernandez
,
S.J.
Morrison
, and
M.F.
Clarke
.
2003
.
Prospective identification of tumorigenic breast cancer cells
.
Proc. Natl. Acad. Sci. USA.
100
:
3983
3988
.
Ashkavand
,
Z.
,
C.
O’Flanagan
,
M.
Hennig
,
X.
Du
,
S.D.
Hursting
, and
S.A.
Krupenko
.
2017
.
Metabolic Reprogramming by Folate Restriction Leads to a Less Aggressive Cancer Phenotype
.
Mol. Cancer Res.
15
:
189
200
.
Baek
,
D.
,
J.
Villén
,
C.
Shin
,
F.D.
Camargo
,
S.P.
Gygi
, and
D.P.
Bartel
.
2008
.
The impact of microRNAs on protein output
.
Nature.
455
:
64
71
.
Bartel
,
D.P.
2018
.
Metazoan MicroRNAs
.
Cell.
173
:
20
51
.
Brown
,
B.D.
,
B.
Gentner
,
A.
Cantore
,
S.
Colleoni
,
M.
Amendola
,
A.
Zingale
,
A.
Baccarini
,
G.
Lazzari
,
C.
Galli
, and
L.
Naldini
.
2007
.
Endogenous microRNA can be broadly exploited to regulate transgene expression according to tissue, lineage and differentiation state
.
Nat. Biotechnol.
25
:
1457
1467
.
Cancer Genome Atlas Network
.
2012
.
Comprehensive molecular portraits of human breast tumours
.
Nature.
490
:
61
70
.
Cicalese
,
A.
,
G.
Bonizzi
,
C.E.
Pasi
,
M.
Faretta
,
S.
Ronzoni
,
B.
Giulini
,
C.
Brisken
,
S.
Minucci
,
P.P.
Di Fiore
, and
P.G.
Pelicci
.
2009
.
The tumor suppressor p53 regulates polarity of self-renewing divisions in mammary stem cells
.
Cell.
138
:
1083
1095
.
Clevers
,
H.
2011
.
The cancer stem cell: premises, promises and challenges
.
Nat. Med.
17
:
313
319
.
Copley
,
M.R.
,
S.
Babovic
,
C.
Benz
,
D.J.
Knapp
,
P.A.
Beer
,
D.G.
Kent
,
S.
Wohrer
,
D.Q.
Treloar
,
C.
Day
,
K.
Rowe
, et al
.
2013
.
The Lin28b-let-7-Hmga2 axis determines the higher self-renewal potential of fetal haematopoietic stem cells
.
Nat. Cell Biol.
15
:
916
925
.
Creighton
,
C.J.
,
X.
Li
,
M.
Landis
,
J.M.
Dixon
,
V.M.
Neumeister
,
A.
Sjolund
,
D.L.
Rimm
,
H.
Wong
,
A.
Rodriguez
,
J.I.
Herschkowitz
, et al
.
2009
.
Residual breast cancers after conventional therapy display mesenchymal as well as tumor-initiating features
.
Proc. Natl. Acad. Sci. USA.
106
:
13820
13825
.
Curtis
,
C.
,
S.P.
Shah
,
S.-F.
Chin
,
G.
Turashvili
,
O.M.
Rueda
,
M.J.
Dunning
,
D.
Speed
,
A.G.
Lynch
,
S.
Samarajiwa
,
Y.
Yuan
, et al.
METABRIC Group
.
2012
.
The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups
.
Nature.
486
:
346
352
.
Dalerba
,
P.
,
R.W.
Cho
, and
M.F.
Clarke
.
2007
.
Cancer stem cells: models and concepts
.
Annu. Rev. Med.
58
:
267
284
.
Debnath
,
J.
,
S.K.
Muthuswamy
, and
J.S.
Brugge
.
2003
.
Morphogenesis and oncogenesis of MCF-10A mammary epithelial acini grown in three-dimensional basement membrane cultures
.
Methods.
30
:
256
268
.
Deming
,
S.L.
,
S.J.
Nass
,
R.B.
Dickson
, and
B.J.
Trock
.
2000
.
C-myc amplification in breast cancer: a meta-analysis of its occurrence and prognostic relevance
.
Br. J. Cancer.
83
:
1688
1695
.
Diehn
,
M.
,
R.W.
Cho
, and
M.F.
Clarke
.
2009
.
Therapeutic implications of the cancer stem cell hypothesis
.
Semin. Radiat. Oncol.
19
:
78
86
.
Dontu
,
G.
,
W.M.
Abdallah
,
J.M.
Foley
,
K.W.
Jackson
,
M.F.
Clarke
,
M.J.
Kawamura
, and
M.S.
Wicha
.
2003
.
In vitro propagation and transcriptional profiling of human mammary stem/progenitor cells
.
Genes Dev.
17
:
1253
1270
.
Ducker
,
G.S.
, and
J.D.
Rabinowitz
.
2017
.
One-Carbon Metabolism in Health and Disease
.
Cell Metab.
25
:
27
42
.
Fillmore
,
C.M.
, and
C.
Kuperwasser
.
2008
.
Human breast cancer cell lines contain stem-like cells that self-renew, give rise to phenotypically diverse progeny and survive chemotherapy
.
Breast Cancer Res.
10
:
R25
.
Forloni
,
M.
,
S.K.
Dogra
,
Y.
Dong
,
D.
Conte
Jr
.,
J.
Ou
,
L.J.
Zhu
,
A.
Deng
,
M.
Mahalingam
,
M.R.
Green
, and
N.
Wajapeyee
.
2014
.
miR-146a promotes the initiation and progression of melanoma by activating Notch signaling
.
eLife.
3
:e01460.
Friedman
,
B.
, and
B.
Cronstein
.
2019
.
Methotrexate mechanism in treatment of rheumatoid arthritis
.
Joint Bone Spine.
86
:
301
307
.
Garcia
,
A.I.
,
M.
Buisson
,
P.
Bertrand
,
R.
Rimokh
,
E.
Rouleau
,
B.S.
Lopez
,
R.
Lidereau
,
I.
Mikaélian
, and
S.
Mazoyer
.
2011
.
Down-regulation of BRCA1 expression by miR-146a and miR-146b-5p in triple negative sporadic breast cancers
.
EMBO Mol. Med.
3
:
279
290
.
Green
,
A.R.
,
M.A.
Aleskandarany
,
D.
Agarwal
,
S.
Elsheikh
,
C.C.
Nolan
,
M.
Diez-Rodriguez
,
R.D.
Macmillan
,
G.R.
Ball
,
C.
Caldas
,
S.
Madhusudan
, et al
.
2016
.
MYC functions are specific in biological subtypes of breast cancer and confers resistance to endocrine therapy in luminal tumours
.
Br. J. Cancer.
114
:
917
928
.
Gupta
,
P.B.
,
C.M.
Fillmore
,
G.
Jiang
,
S.D.
Shapira
,
K.
Tao
,
C.
Kuperwasser
, and
E.S.
Lander
.
2011
.
Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells
.
Cell.
146
:
633
644
.
Hu
,
Y.
, and
G.K.
Smyth
.
2009
.
ELDA: extreme limiting dilution analysis for comparing depleted and enriched populations in stem cell and other assays
.
J. Immunol. Methods.
347
:
70
78
.
Hwang
,
W.-L.
,
J.-K.
Jiang
,
S.-H.
Yang
,
T.-S.
Huang
,
H.-Y.
Lan
,
H.-W.
Teng
,
C.-Y.
Yang
,
Y.-P.
Tsai
,
C.-H.
Lin
,
H.-W.
Wang
, and
M.-H.
Yang
.
2014
.
MicroRNA-146a directs the symmetric division of Snail-dominant colorectal cancer stem cells
.
Nat. Cell Biol.
16
:
268
280
.
Iliopoulos
,
D.
,
H.A.
Hirsch
, and
K.
Struhl
.
2009
.
An epigenetic switch involving NF-kappaB, Lin28, Let-7 MicroRNA, and IL6 links inflammation to cell transformation
.
Cell.
139
:
693
706
.
Klijn
,
C.
,
S.
Durinck
,
E.W.
Stawiski
,
P.M.
Haverty
,
Z.
Jiang
,
H.
Liu
,
J.
Degenhardt
,
O.
Mayba
,
F.
Gnad
,
J.
Liu
, et al
.
2015
.
A comprehensive transcriptional portrait of human cancer cell lines
.
Nat. Biotechnol.
33
:
306
312
.
Leivonen
,
S.K.
,
K.K.
Sahlberg
,
R.
Mäkelä
,
E.U.
Due
,
O.
Kallioniemi
,
A.L.
Børresen-Dale
, and
M.
Perälä
.
2014
.
High-throughput screens identify microRNAs essential for HER2 positive breast cancer cell growth
.
Mol. Oncol.
8
:
93
104
.
Li
,
X.
,
M.T.
Lewis
,
J.
Huang
,
C.
Gutierrez
,
C.K.
Osborne
,
M.F.
Wu
,
S.G.
Hilsenbeck
,
A.
Pavlick
,
X.
Zhang
,
G.C.
Chamness
, et al
.
2008
.
Intrinsic resistance of tumorigenic breast cancer cells to chemotherapy
.
J. Natl. Cancer Inst.
100
:
672
679
.
Lim
,
E.
,
D.
Wu
,
B.
Pal
,
T.
Bouras
,
M.-L.
Asselin-Labat
,
F.
Vaillant
,
H.
Yagita
,
G.J.
Lindeman
,
G.K.
Smyth
, and
J.E.
Visvader
.
2010
.
Transcriptome analyses of mouse and human mammary cell subpopulations reveal multiple conserved genes and pathways
.
Breast Cancer Res.
12
:
R21
.
Liu
,
S.
, and
M.S.
Wicha
.
2010
.
Targeting breast cancer stem cells
.
J. Clin. Oncol.
28
:
4006
4012
.
Locasale
,
J.W.
2013
.
Serine, glycine and one-carbon units: cancer metabolism in full circle
.
Nat. Rev. Cancer.
13
:
572
583
.
Lu
,
L.F.
,
M.P.
Boldin
,
A.
Chaudhry
,
L.L.
Lin
,
K.D.
Taganov
,
T.
Hanada
,
A.
Yoshimura
,
D.
Baltimore
, and
A.Y.
Rudensky
.
2010
.
Function of miR-146a in controlling Treg cell-mediated regulation of Th1 responses
.
Cell.
142
:
914
929
.
Mani
,
S.A.
,
W.
Guo
,
M.J.
Liao
,
E.N.
Eaton
,
A.
Ayyanan
,
A.Y.
Zhou
,
M.
Brooks
,
F.
Reinhard
,
C.C.
Zhang
,
M.
Shipitsin
, et al
.
2008
.
The epithelial-mesenchymal transition generates cells with properties of stem cells
.
Cell.
133
:
704
715
.
Mei
,
J.
,
R.
Bachoo
, and
C.L.
Zhang
.
2011
.
MicroRNA-146a inhibits glioma development by targeting Notch1
.
Mol. Cell. Biol.
31
:
3584
3592
.
Miller
,
L.D.
,
J.
Smeds
,
J.
George
,
V.B.
Vega
,
L.
Vergara
,
A.
Ploner
,
Y.
Pawitan
,
P.
Hall
,
S.
Klaar
,
E.T.
Liu
, and
J.
Bergh
.
2005
.
An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival
.
Proc. Natl. Acad. Sci. USA.
102
:
13550
13555
.
Oskarsson
,
T.
,
E.
Batlle
, and
J.
Massagué
.
2014
.
Metastatic stem cells: sources, niches, and vital pathways
.
Cell Stem Cell.
14
:
306
321
.
Pece
,
S.
,
D.
Tosoni
,
S.
Confalonieri
,
G.
Mazzarol
,
M.
Vecchi
,
S.
Ronzoni
,
L.
Bernard
,
G.
Viale
,
P.G.
Pelicci
, and
P.P.
Di Fiore
.
2010
.
Biological and molecular heterogeneity of breast cancers correlates with their cancer stem cell content
.
Cell.
140
:
62
73
.
Pece
,
S.
,
D.
Disalvatore
,
D.
Tosoni
,
M.
Vecchi
,
S.
Confalonieri
,
G.
Bertalot
,
G.
Viale
,
M.
Colleoni
,
P.
Veronesi
,
V.
Galimberti
, and
P.P.
Di Fiore
.
2019
.
Identification and clinical validation of a multigene assay that interrogates the biology of cancer stem cells and predicts metastasis in breast cancer: A retrospective consecutive study
.
EBioMedicine.
42
:
352
362
.
Penkert
,
J.
,
T.
Ripperger
,
M.
Schieck
,
B.
Schlegelberger
,
D.
Steinemann
, and
T.
Illig
.
2016
.
On metabolic reprogramming and tumor biology: A comprehensive survey of metabolism in breast cancer
.
Oncotarget.
7
:
67626
67649
.
Scheel
,
C.
,
E.N.
Eaton
,
S.H.
Li
,
C.L.
Chaffer
,
F.
Reinhardt
,
K.J.
Kah
,
G.
Bell
,
W.
Guo
,
J.
Rubin
,
A.L.
Richardson
, and
R.A.
Weinberg
.
2011
.
Paracrine and autocrine signals induce and maintain mesenchymal and stem cell states in the breast
.
Cell.
145
:
926
940
.
Selbach
,
M.
,
B.
Schwanhäusser
,
N.
Thierfelder
,
Z.
Fang
,
R.
Khanin
, and
N.
Rajewsky
.
2008
.
Widespread changes in protein synthesis induced by microRNAs
.
Nature.
455
:
58
63
.
Shibue
,
T.
, and
R.A.
Weinberg
.
2017
.
EMT, CSCs, and drug resistance: the mechanistic link and clinical implications
.
Nat. Rev. Clin. Oncol.
14
:
611
629
.
Shimono
,
Y.
,
M.
Zabala
,
R.W.
Cho
,
N.
Lobo
,
P.
Dalerba
,
D.
Qian
,
M.
Diehn
,
H.
Liu
,
S.P.
Panula
,
E.
Chiao
, et al
.
2009
.
Downregulation of miRNA-200c links breast cancer stem cells with normal stem cells
.
Cell.
138
:
592
603
.
Shipitsin
,
M.
,
L.L.
Campbell
,
P.
Argani
,
S.
Weremowicz
,
N.
Bloushtain-Qimron
,
J.
Yao
,
T.
Nikolskaya
,
T.
Serebryiskaya
,
R.
Beroukhim
,
M.
Hu
, et al
.
2007
.
Molecular definition of breast tumor heterogeneity
.
Cancer Cell.
11
:
259
273
.
Shyh-Chang
,
N.
, and
H.H.
Ng
.
2017
.
The metabolic programming of stem cells
.
Genes Dev.
31
:
336
346
.
Stingl
,
J.
,
P.
Eirew
,
I.
Ricketson
,
M.
Shackleton
,
F.
Vaillant
,
D.
Choi
,
H.I.
Li
, and
C.J.
Eaves
.
2006
.
Purification and unique properties of mammary epithelial stem cells
.
Nature.
439
:
993
997
.
Taganov
,
K.D.
,
M.P.
Boldin
,
K.J.
Chang
, and
D.
Baltimore
.
2006
.
NF-kappaB-dependent induction of microRNA miR-146, an inhibitor targeted to signaling proteins of innate immune responses
.
Proc. Natl. Acad. Sci. USA.
103
:
12481
12486
.
Testa
,
U.
,
E.
Pelosi
,
G.
Castelli
, and
C.
Labbaye
.
2017
.
miR-146 and miR-155: Two Key Modulators of Immune Response and Tumor Development
.
Noncoding RNA.
3
:
22
.
Tordonato
,
C.
,
P.P.
Di Fiore
, and
F.
Nicassio
.
2015
.
The role of non-coding RNAs in the regulation of stem cells and progenitors in the normal mammary gland and in breast tumors
.
Front. Genet.
6
:
72
.
Visvader
,
J.E.
, and
G.J.
Lindeman
.
2012
.
Cancer stem cells: current status and evolving complexities
.
Cell Stem Cell.
10
:
717
728
.
Visvader
,
J.E.
, and
J.
Stingl
.
2014
.
Mammary stem cells and the differentiation hierarchy: current status and perspectives
.
Genes Dev.
28
:
1143
1158
.
Wang
,
Y.
,
S.
Baskerville
,
A.
Shenoy
,
J.E.
Babiarz
,
L.
Baehner
, and
R.
Blelloch
.
2008
.
Embryonic stem cell-specific microRNAs regulate the G1-S transition and promote rapid proliferation
.
Nat. Genet.
40
:
1478
1483
.
Wellner
,
U.
,
J.
Schubert
,
U.C.
Burk
,
O.
Schmalhofer
,
F.
Zhu
,
A.
Sonntag
,
B.
Waldvogel
,
C.
Vannier
,
D.
Darling
,
A.
zur Hausen
, et al
.
2009
.
The EMT-activator ZEB1 promotes tumorigenicity by repressing stemness-inhibiting microRNAs
.
Nat. Cell Biol.
11
:
1487
1495
.
Wen
,
J.
,
E.
Leucci
,
R.
Vendramin
,
S.
Kauppinen
,
A.H.
Lund
,
A.
Krogh
, and
B.J.
Parker
.
2015
.
Transcriptome dynamics of the microRNA inhibition response
.
Nucleic Acids Res.
43
:
6207
6221
.
Wong
,
T.L.
,
N.
Che
, and
S.
Ma
.
2017
.
Reprogramming of central carbon metabolism in cancer stem cells
.
Biochim. Biophys. Acta Mol. Basis Dis.
1863
:
1728
1738
.
Yang
,
L.
,
P.
Shi
,
G.
Zhao
,
J.
Xu
,
W.
Peng
,
J.
Zhang
,
G.
Zhang
,
X.
Wang
,
Z.
Dong
,
F.
Chen
, and
H.
Cui
.
2020
a
.
Targeting cancer stem cell pathways for cancer therapy
.
Signal Transduct. Target. Ther.
5
:
8
.
Yang
,
V.
,
M.J.
Gouveia
,
J.
Santos
,
B.
Koksch
,
I.
Amorim
,
F.
Gärtner
, and
N.
Vale
.
2020
b
.
Breast cancer: insights in disease and influence of drug methotrexate
.
RSC Med Chem.
11
:
646
664
.
Yu
,
F.
,
H.
Yao
,
P.
Zhu
,
X.
Zhang
,
Q.
Pan
,
C.
Gong
,
Y.
Huang
,
X.
Hu
,
F.
Su
,
J.
Lieberman
, and
E.
Song
.
2007
.
let-7 regulates self renewal and tumorigenicity of breast cancer cells
.
Cell.
131
:
1109
1123
.
Zhang
,
L.
,
N.
Stokes
,
L.
Polak
, and
E.
Fuchs
.
2011
.
Specific microRNAs are preferentially expressed by skin stem cells to balance self-renewal and early lineage commitment
.
Cell Stem Cell.
8
:
294
308
.

Author notes

*

P.P. Di Fiore and F. Nicassio contributed equally to this paper.

This article is available under a Creative Commons License (Attribution 4.0 International, as described at https://creativecommons.org/licenses/by/4.0/).