Normal cell growth is governed by a complicated biological system, featuring multiple levels of control, often deregulated in cancers. The role of microRNAs (miRNAs) in the control of gene expression is now increasingly appreciated, yet their involvement in controlling cell proliferation is still not well understood. Here we investigated the mammalian cell proliferation control network consisting of transcriptional regulators, E2F and p53, their targets and a family of 15 miRNAs. Indicative of their significance, expression of these miRNAs is downregulated in senescent cells and in breast cancers harboring wild‐type p53. These miRNAs are repressed by p53 in an E2F1‐mediated manner. Furthermore, we show that these miRNAs silence antiproliferative genes, which themselves are E2F1 targets. Thus, miRNAs and transcriptional regulators appear to cooperate in the framework of a multi‐gene transcriptional and post‐transcriptional feed‐forward loop. Finally, we show that, similarly to p53 inactivation, overexpression of representative miRNAs promotes proliferation and delays senescence, manifesting the detrimental phenotypic consequence of perturbations in this circuit. Taken together, these findings position miRNAs as novel key players in the mammalian cellular proliferation network.
Precise regulation of gene expression is crucial for maintaining homeostasis in healthy tissues and for the execution of cellular programs such as proliferation, differentiation and cell death. In the last decade, microRNAs (miRNAs) have been uncovered as an expanding family of gene expression regulators. These short non‐coding RNAs regulate gene expression at the post‐transcriptional level by promoting translational inhibition or mRNA degradation (Bartel, 2004). Similar to protein‐coding genes, the expression of miRNAs is also regulated by transcription factors (TFs), and induction or repression of miRNAs has been demonstrated to play a role in physiological processes such as immune response (Thai et al, 2007) and apoptosis (Chang et al, 2007; Raver‐Shapira et al, 2007). Accordingly, deregulation of miRNAs is associated with diverse types of diseases, including a variety of cancers (Esquela‐Kerscher and Slack, 2006; Volinia et al, 2006).
In an earlier computational study, we predicted the presence of several types of regulatory network motifs that involve TFs and miRNAs (Shalgi et al, 2007), and may provide a mechanism for fine‐tuned coordination between transcriptional and post‐transcriptional regulation of gene expression. Here, we describe and experimentally demonstrate one such regulatory motif, termed feed‐forward loop (FFL), which involves the TF E2F1, a set of miRNAs, and their common targets (Figure 8). In this FFL, E2F1, a key regulator of cell‐cycle progression, transcriptionally activates a family of 15 miRNAs that are organized in three paralogous polycistrons on three different chromosomes. These miRNAs silence a group of antiproliferative regulators including the pocket proteins pRb and p130 and the CDK inhibitors p21 and p57. Importantly, these genes are themselves transcriptional targets of E2F1. Thus, a TF activates a set of genes as well as a set of miRNAs, which in turn post‐transcriptionally regulate that set of genes. Increasing the complexity of this regulatory FFL, many of the shared targets of E2F1 and the miRNAs function as regulators of the cell cycle; some negatively regulate E2F itself. For example, the pocket proteins pRB and p130 are the major components that regulate the activity of E2F family members throughout the phases of the cell cycle through direct protein–protein interaction.
The TF p53 is regarded as one of the key proteins that prevent malignant transformation (Ryan et al, 2001), and deactivating mutations of this tumor suppressor are highly common in a wide variety of tumors (Hussain and Harris, 1999). A hallmark activity of p53 is the inhibition of proliferation and the induction of cellular senescence on diverse types of stress signals with oncogenic potential, including DNA damage, telomere shortening and oncogene activation. There are several known mechanisms by which p53 negatively regulates proliferation, the key one being the transcriptional activation of the CDK inhibitor p21, which indirectly inhibits the activity of E2F family members. Another recently discovered mechanism for inhibiting proliferation by p53 is the induction of miRNAs from the miR‐34 family, which also modulate the E2F pathway (He et al, 2007; Tarasov et al, 2007; Tazawa et al, 2007; Kumamoto et al, 2008). Additionally, direct and indirect transcriptional repression by p53 is considered important for its ability to inhibit proliferation (Ho and Benchimol, 2003).
Using miRNA microarrays, we discovered that p53 activation during cellular senescence in primary human fibroblasts leads to a decrease in the expression of the above‐mentioned family of miRNAs, including members of the miR‐17‐92, miR‐106b/93/25 and miR‐106a‐92 polycistronic miRNA clusters. A similar decrease in miRNA expression was observed in human breast cancer specimens that harbor wild‐type p53 as compared with those that harbor mutant forms of p53. We further investigated the mechanism by which p53 represses the expression of this group of miRNAs, and found that activation of p53 leads to a dramatic reduction of E2F1 mRNA, protein and activity levels, which in turn leads to a decrease in the E2F1‐dependent transcriptional activation of these miRNAs.
To study the consequence of deregulation of this FFL and importance of its inhibition by p53, we ectopically expressed representative members from the set of p53‐repressed miRNAs, namely the miR‐106b/93/25 polycistron, in primary human fibroblasts. Consequently, these cells acquired an enhanced proliferative phenotype manifested by increased growth rate, increased colony formation efficiency and delayed entry into replicative senescence. These results position the repression of this set of miRNAs as a novel mechanism by which p53 inhibits proliferation and controls cell fate.
Here we identified a group of 15 co‐regulated paralogous miRNAs which are transcriptionally activated by E2F1. This group includes the miR‐17‐92, miR‐106a‐92 and miR‐106b/93/25 polycistronic miRNAs.
These miRNAs silence anti‐proliferative genes, which themselves are E2F1 targets and function as negative regulators of proliferation.
Thus, E2F1 and this group of microRNAs cooperate in a feed‐forward loop that involves transcriptional and post‐transcriptional modes of regulation.
The key tumor suppressor p53 disrupts this feed‐forward loop by inactivating E2F1 in senescent cells and in human cancers. This inhibition serves as another arm of p53's tight control of proliferation.
The tumor suppressor p53 is considered a central regulator of cell‐fate decisions. Activation of p53 can induce several cellular responses, including cell‐cycle arrest, senescence and apoptosis. Thus, absence of functional p53 predisposes cells to neoplastic transformation. Accordingly, mutations of this gene are highly common in human cancers (Hussain and Harris, 1999). p53 is a sequence‐specific transcription factor (TF) that exerts many of its downstream effects by activating gene transcription (Ryan et al, 2001). Nevertheless, additional transactivation‐independent functions of p53 contribute to its tumor suppressive activity, including protein–protein interactions with additional TFs and other cell‐fate regulators. The importance of transcriptional regulation by p53 is exemplified by the fact that most p53 tumor‐derived mutants are defective in DNA binding and incapable of transactivation (Kern et al, 1991). In addition to its capability to induce gene transcription, p53 activation results in extensive gene repression (Ginsberg et al, 1991). Direct and indirect transcriptional repression by p53 is considered important for its tumor suppressive functions, such as induction of cell‐cycle arrest and apoptosis (Ho and Benchimol, 2003).
microRNAs (miRNAs or miRs in short) are a recently discovered class of small non‐coding RNA species that regulate gene expression at the post‐transcriptional level. Approximately half of the known miRs are encoded in regions of the genome that are distal to known genes, whereas the remaining reside in introns, or in rare cases in exons, of coding genes, usually in the same orientation as the mRNA. Additionally, some miRs are clustered in the genome and are transcribed as polycistrons that may contain up to ∼50 mature miRs (Bartel, 2004). With the recent identification of miRs that regulate cancer‐related processes such as apoptosis, proliferation and differentiation, these RNA species emerge as important regulators of cancer initiation and progression. Accordingly, mutation and transcriptional deregulation of miRs have been linked to cancer (Esquela‐Kerscher and Slack, 2006). Deregulated miRs were suggested to exert their function in cancer through silencing of key cell‐fate regulators, as shown for let‐7 and Ras (Johnson et al, 2005), as well as for miR‐106b and p21 (Ivanovska et al, 2008; Petrocca et al, 2008).
In a previous work we suggested that miRs cooperate with certain TFs in the regulation of mutual sets of target genes, allowing the coordinated modulation of gene expression both transcriptionally and post‐transcriptionally. Specifically, we found a recurring network motif in which a TF regulates the miR with which it cooperates in regulating a common set of targets, creating a feed‐forward loop (FFL). One such case involved E2F and the miR‐106b/93/25 polycistron (Shalgi et al, 2007).
Several studies have implicated p53 in the regulation of miR expression (Xi et al, 2006; Chang et al, 2007; He et al, 2007; Raver‐Shapira et al, 2007; Tarasov et al, 2007; Kumamoto et al, 2008). These studies exploited various high‐throughput methods to identify p53‐regulated miRs in several cellular systems with differential p53 status. Although the resulting candidate lists from each study differed considerably, probably due to differences in the cellular contexts and p53 activation signals, all studies identified members of the miR‐34 family as direct transactivation targets of p53. In line with p53 function, induction of miR‐34 family members was suggested to mediate cell‐cycle arrest, apoptosis and senescence. Importantly, none of these studies focused on miRs, whose expression negatively correlated with p53 activation, and which are likely repressed by this tumor suppressor.
Here, we report the identification of a large set of miRs, the expression of which constitutes a recurring signature in several experiments. The members of this signature are transcriptionally repressed by p53 in primary cells and in human breast cancers. This signature is comprised of known cancer‐associated miRs as well as newly proposed ones, and includes the miR‐106b/93/25 polycistron. We implicate E2F1 in the p53‐dependent repression of these miRs, and demonstrate the oncogenic potential of the miR‐106b/93/25 polycistron. Finally, we delineate a network architecture that includes the transcription factor E2F1 and a family of miRs, which co‐regulate mutual target genes transcriptionally and post‐transcriptionally, thereby enhancing cellular proliferation. This FFL is repressed by p53, possibly to promote senescence and suppress cancer progression.
Identification of p53‐regulated miRNAs in primary human cells and in human breast cancers
To identify novel p53‐regulated miRs, we established two isogenic cell cultures that differ in their p53 status and analyzed their miRNA profiles both under normal conditions and in contexts involving p53 activation. WI‐38 primary human fibroblasts were infected with a retrovirus encoding for the p53‐inactivating peptide, GSE56 (Ossovskaya et al, 1996). These cells (GSE) and their active p53 counterparts (con) were treated with the DNA‐damaging agent doxorubicin or grown until the onset of replicative senescence (the establishment of the system is depicted in Supplementary Figure S1). Analysis of miRNA expression patterns revealed several expression clusters (see Materials and methods). Notable among these was a cluster populated with miRs, the expression of which was negatively regulated by p53 under normal conditions. The cluster showed additional downregulation in senescent cells, which was attenuated upon p53 inactivation. We named this cluster the ‘p53‐repressed miR cluster’ (Figure 1A). Notably, doxorubicin treatment, which resulted in a considerable activation of p53 and its mRNA targets (Supplementary Figure S1), did not significantly affect the levels of these miRs.
Interestingly, a significant number of miRs from this cluster were also clustered together in a similar experiment, in which miR expression was profiled in young and senescent human embryonic fibroblasts (WI‐38 and MRC5). This cluster is termed ‘senescence‐repressed miR cluster’ (Figure 1B). The significant overlap between the clusters (P‐value=3.2 × 10−4) is interesting as the second experiment was not designed to discover p53‐regulated miRs, but rather to identify a general signature of miRs that are altered upon replicative senescence. However, p53 activity was increased in both senescent fibroblast cultures (data not shown). Strikingly, clustering analysis of miR expression data derived from a set of breast cancer tumors with differing p53 status also resulted in a cluster highly overlapping the ‘p53‐repressed miR cluster’ (Figure 1C; P‐value=1.11 × 10−5) (samples, p53 status and histological grading were described by Naume et al (2007) and Sorlie et al (2006), and detailed description of the mutation status is listed in Supplementary Table S1). The ‘breast cancer p53‐repressed miR cluster’ was comprised of miRs, the expression of which was negatively correlated with the presence of a wild‐type p53 in the tumors. Additionally, the miRNA expression and p53 status partially correlated with tumor grade, as almost all cancer samples that contained a mutant p53 and expressed high levels of the miRs were derived from high‐grade tumors.
We thus revealed a recurring signature of miRs that are coordinately regulated both in primary human cells in vitro and in human breast tumors in vivo. We suggest that these miRs are repressed by wild‐type p53 during both normal growth and cancer progression.
The presented clusters contain families of paralogous cancer‐related miRNAs
Interestingly, 15 miRs represented in the three clusters (Figure 1) are transcribed from three homologous genomic loci, reported earlier to be paralogs that evolved from a common evolutionary origin (Tanzer and Stadler, 2004). These include miRs‐106b/93/25, which reside within an intron of the cell‐cycle gene ‘minichromosome maintenance protein 7’ (MCM7); miRs‐17/18a/19a/20a/19b‐1/92a‐1 (miR‐17‐92 polycistron), which are transcribed as the non‐coding RNA c13orf25, and miRs‐106a/18b/20b/19b‐2/92‐2 (miR‐106a‐92 polycistron), which are clustered on chromosome X. Our data indicate that not only were these miRNA sequences and genomic organization conserved during evolution but also was their transcriptional regulation. Additional well‐represented miRs in the clusters include the miR‐15b/16 polycistron and miR‐155.
Many members of the clusters are overexpressed in various tumors, consistent with the frequent p53 loss of function in cancer, and some were shown to possess oncogenic functions. For example, miR‐92, miR‐106a, miR‐17‐5p, miR‐20a and miR‐155, which appear in at least two expression clusters, were reported to be overexpressed in solid tumors (Volinia et al, 2006). Members of the miR‐17‐92 polycistron are overexpressed in lymphomas and in lung and colorectal carcinomas (He et al, 2005; Schetter et al, 2008), and were shown to accelerate tumor growth (O'Donnell et al, 2005). Interestingly, the MCM7 gene that contains three of the clusters' miRs in its intron (miRs‐106b/93/25) is amplified or overexpressed in diverse types of cancers (Ren et al, 2006), as are its resident miRs (Petrocca et al, 2008).
Representative miRNAs show p53‐dependent repression during senescence in many cell types
To further validate our data, we generated two additional human isogenic cell culture pairs from the IMR90 lung primary fibroblasts and from prostate‐cancer‐associated fibroblasts. Each culture was infected with a retrovirus encoding for either a small hairpin RNA (shRNA) targeting p53 (p53i) or a control shRNA (con), and grown until the onset of replicative senescence. p53 knockdown, which significantly reduced the mRNA and protein levels of both p53 and its target p21, delayed the onset of senescence by approximately 10 population doublings (PDLs) (Figure 2A; Supplementary Figure S2C). For these cell types, as well as for the WI‐38 cells, we compared the levels of representative miRs that appear in the three expression clusters using TaqMan miRNA assays. Analyses of miR‐106b and miR‐17‐5p, as well as their host transcripts MCM7 and c13orf25, respectively, revealed transcriptional repression upon replicative senescence in all three tested cell cultures in a manner that was partially or completely p53 dependent (Figure 2B). Additionally, the non‐coding RNA BIC and its resident miR‐155 were also transcriptionally repressed in a p53‐dependent manner upon replicative senescence (Supplementary Figure S2D).
Co‐clustering of miRNA and mRNA expression data from human breast cancers reveals two distinct groups of p53‐repressed miRNAs
To gain further insights into the regulation and function of the identified miRNA signature, we exploited previously published mRNA profiling (Sorlie et al, 2006; Naume et al, 2007), performed on the same set of breast cancer specimens from which the ‘breast cancer p53‐repressed miR cluster’ was derived. The mRNA and miRNA array data were combined into one set of expression profiles, and were clustered into 40 co‐clusters; each may consist of both miRs and mRNAs (Figure 3). Interestingly, the members of the ‘breast cancer p53‐repessed miR cluster’ were separated into two distinct co‐clusters with dissimilar characteristics, as revealed by functional annotation using DAVID (Dennis et al, 2003) and by promoter motif analyses using AMADEUS (Linhart et al, 2008) of the mRNAs from each co‐cluster. Of 37 miRs, 31 were co‐clustered (Figure 3A, red cluster; Supplementary Table S2) with genes highly enriched for cell‐cycle‐related annotations (Figure 3D; enrichment P‐value for ‘cell‐cycle’ annotation=8.5 × 10−20, see Supplementary Table S3A for other enriched annotations) and for regulatory binding motifs of known cell‐cycle‐related TFs such as E2F (Figure 3F; P‐value=2.5 × 10−16). All members of the three paralogous polycistrons described above were included in this ‘cell‐cycle‐associated co‐cluster,’ as were their hosts MCM7 and c13orf25. Supporting the notion that this cluster consists mainly of cell‐cycle‐related genes, we compared the mRNA dataset to a previously published list of genes expressed in a cell‐cycle periodic manner (Whitfield et al, 2002), and found a significant enrichment of periodically expressed genes in this cluster (Figure 3H; P‐value=4.9 × 10−17). Another co‐cluster, which contained six miRs (Figure 3A, purple cluster; Supplementary Table S2), was comprised of genes highly enriched for immune response‐related functions (Figure 3E; P‐value for ‘immune response’ annotation=1.2 × 10−48, see Supplementary Table S3B for other enriched annotations). The promoters of these genes were enriched for immune response‐related motifs such as interferon‐stimulated‐responsive element (Figure 3G; P‐value=1.62 × 10−12) as well as for the IRF and NF‐κB motifs (data not shown). Importantly, both mRNAs and miRNAs of both co‐clusters were downregulated in tumors that harbor wild‐type p53, suggesting that our miR signature belongs to a larger transcriptional program that mediates p53‐dependent gene repression of both RNA types.
The miRNAs from the ‘cell‐cycle‐associated co‐cluster’ are associated with p53 and E2F in a proliferation‐related regulatory network
We have reported earlier the identification and characterization of an mRNA cluster termed the ‘proliferation cluster’ that consists mainly of cell‐cycle‐related genes (Milyavsky et al, 2005). This cluster emerged from mRNA profiling of an in vitro transformation process, in which primary WI‐38 cells were gradually transformed into tumorigenic cells. Importantly, the ‘proliferation cluster’ is one of the most prominent expression signatures revealed when tumors are compared to normal tissues or when highly proliferating cells are compared to slow growing cells, and contains many cell‐cycle periodic genes (Whitfield et al, 2006). The expression pattern of the ‘proliferation cluster’ is highly similar to that of the ‘p53‐repressed miR cluster’; i.e. the ‘proliferation cluster’ mRNAs display p53‐dependent downregulation. The similarity in expression patterns prompted us to hypothesize that both clusters share a common transcriptional program. It was shown earlier that the p53‐mediated repression of the ‘proliferation cluster’ was mediated through E2F (Tabach et al, 2005). Providing further support, the promoters of the mRNAs from the ‘cell‐cycle‐associated co‐cluster’ are highly enriched with E2F‐binding motifs (Figure 3F), and in particular, a conserved E2F‐binding site is found upstream of the three miRNA polycistrons (Supplementary Figure S3A and Supplementary Table S4). Confirming these predictions, chromatin immunoprecipitation analysis revealed that E2F1 binds to its conserved motifs upstream of each of the three polycistrons (Figure 4A). Similarly, it was shown recently that the miR‐17‐92 and the miR‐106b/93/25 polycistrons are transcriptionally activated by E2F family members (O'Donnell et al, 2005; Sylvestre et al, 2007; Woods et al, 2007; Petrocca et al, 2008).
In view of the above, it appears conceivable that the miRs from the ‘cell‐cycle‐associated co‐cluster’ are transcriptionally activated by E2F, and that p53 exerts its repression through E2F inhibition. In agreement with the observed downregulation of the ‘p53‐repressed miR cluster’ in senescence, it was shown that E2F activity is significantly downregulated in senescent cells (Campisi and d'Adda di Fagagna, 2007). In addition, the miRs presented here are proposed to be novel members of the well‐established ‘proliferation cluster.’
The p53‐dependent repression of miRNAs from the ‘cell‐cycle‐associated co‐cluster’ is mediated through E2F1
To experimentally test whether the cell‐cycle‐associated miRs and their host mRNAs are transcriptionally activated by E2F1, we infected primary WI‐38 cells with E1A, a viral oncoprotein that disrupts pRb‐E2F complexes and leads to an upregulation of the endogenous E2F activity (Fattaey et al, 1993). As expected, stable overexpression of E1A resulted in elevated levels of candidate miRNAs and host mRNAs, which were part of the ‘cell‐cycle‐associated co‐cluster,’ that together represent all three paralogous polycistrons (Figure 4B). Specifically, the MCM7 gene and its resident miRNAs miRs‐106b/93/25; the non‐coding RNA c13orf25 and its resident miR‐17‐5p; and miR‐106a, which represents the miR‐106a‐92 polycistron, were all upregulated following E2F activation. We note that the level of miR‐155, which belongs to the ‘immune response co‐cluster’, was not upregulated by E2F (Supplementary Figure S3E).
To investigate the kinetics of the miRNA's transcriptional activation by E2F1, we infected WI‐38 cells with a retrovirus encoding for an E2F1 protein fused to a modified estrogen receptor (ER) ligand‐binding domain. Treatment of ER‐E2F1‐expressing cells with 4‐hydroxytamoxifen (4‐OHT) permits ER‐E2F1 translocation to the nucleus, thereby inducing its transactivation activity. As depicted in Figure 4C, as early as 4 h after ER‐E2F1 activation by 4‐OHT, significant upregulation of the candidate miRNAs and host transcripts was already evident. The miRNAs and their hosts peaked after 8–10 h of 4‐OHT treatment, similarly to Cyclin E, a known E2F1 target. It is noted that upregulation of MCM7 and its resident miRNAs following 4‐OHT treatment was also observed in ER‐E2F1 expressing lung carcinoma cells (H1299) and osteosarcoma cells (U2OS) (Supplementary Figure S3B and C). Finally, to strengthen the notion that E2F1 directly transactivates the miRNAs, we treated ER‐E2F1 expressing WI‐38 cells with 4‐OHT in the presence or absence of cycloheximide, which inhibits protein biosynthesis and should attenuate the induction of the miRNAs if translation of a secondary mediator is required. As depicted in Supplementary Figure S3D, the induction of the miRNAs was not inhibited by cycloheximide. Altogether, these results indicate that E2F1 can directly bind its cognate sites upstream of the polycistronic miRNAs and activate their transcription.
Having shown that representative miRs are activated by E2F1 in our system, we set to test whether their p53‐dependent repression is mediated through modulation of E2F1 activity. To that end, we infected WI‐38 cells with a retrovirus encoding for either an shRNA targeting p53 (p53i) or a control shRNA (con) and treated them with Nutlin‐3, a small molecule that stabilizes the p53 protein by inhibiting its Mdm2‐dependent ubiquitylation and degradation (Vassilev et al, 2004), thereby inducing p53 activation in a non‐genotoxic manner. Nutlin treatment resulted in a robust p53 protein accumulation, accompanied by p21 mRNA and protein induction (Figure 5A and B), which was completely abrogated in the p53i cells. Remarkably, E2F1 mRNA and protein levels were downregulated upon Nutlin treatment in a p53‐dependent manner. Cyclin E showed a similar pattern, supporting the notion that E2F1 downregulation was accompanied by a reduction in E2F activity. Accordingly, both MCM7 and its resident miR‐106b were significantly downregulated in a p53‐dependent manner (Figure 5A) along with other miRs from the ‘cell‐cycle‐associated co‐cluster’ (data not shown). Thus, treatment with Nutlin, a non‐genotoxic p53 activator, resulted in a p53‐dependent transcriptional repression of mRNAs and miRNAs with associated cell‐cycle functions. To substantiate the causal relationship between the p53‐dependent reduction of E2F1 activity and the repression of the miRs and their hosts, we treated control and E1A‐expressing WI‐38 cells with Nutlin. As depicted in Figure 5C and D, E1A induced the expression and prevented the Nutlin‐dependent repression of E2F1 as well as of its target Cyclin E. Most significantly, E1A abolished the downregulation of MCM7 and miR‐106b upon Nutlin treatment. A similar pattern was observed for miR‐17‐5p and its host c13orf25 (data not shown). Finally, we stably knocked down E2F1 using retroviral‐encoded shRNA in WI‐38 cells in combination with Nutlin treatment, and measured the levels of miRs‐106b/25/93 (Figure 5E), as well as the protein levels of p53, p21 and E2F1 (Figure 5F). Indeed, the knock down of E2F1 resulted in reduced levels of the miRNAs. Supporting the notion that repression of the miR‐106b/93/25 polycistron by p53 is mediated through E2F1 inhibition, Nutlin treatment of the E2F1‐knockdown cells had very little effect as compared with the control cells.
We therefore conclude that E2F1 inhibition by p53 is necessary for the downregulation of MCM7 and its resident miRNAs. The same mechanism may underlie the p53‐dependent downregulation of additional miRs from the ‘cell‐cycle co‐cluster’ and, more specifically, the three paralogous polycistrons.
Cell‐cycle‐associated miRNAs target key cell‐cycle regulators and affect pivotal characteristics of proliferation
Next, we set out to identify the functions of the p53‐repressed miRs. We focused on the miR‐106b/93/25 polycistron as a representative member of the large family of miRs that includes also the miR‐17‐92 and miR‐106a‐92 polycistrons. We overexpressed the genomic region encoding miRs‐106b/93/25, which corresponds to an intron of the MCM7 gene in young WI‐38 cells and in MCF10A mammary epithelial cells, both characterized by low basal expression of these miRs. Following our previous computational prediction of E2F and miR‐106b/93/25 involvement in a FFL, in which they both target a mutual set of genes (Shalgi et al, 2007), we compiled a list of their mutual predicted targets (Supplementary Table S5). Interestingly, many of these predicted targets participate in cell‐cycle regulation (P‐value for ‘cell‐cycle’ annotation enrichment=1.4 × 10−10). Another key cell‐cycle regulator, p21, was recently reported as a target for miR‐106b (Ivanovska et al, 2008), and is a known target of both E2F1 (Gartel et al, 1998) and p53. We then measured the protein levels of selected predicted targets in the miR‐106b/93/25‐overexpressing cells (Figure 6). We observed downregulation of p21, as well as of pRB and p130, which were suggested earlier, based on reporter assays, as potential targets of miR‐106a and the miR‐17‐92 cluster, respectively (Volinia et al, 2006; Wang et al, 2008). Interestingly, E2F1, which was shown to be a target of miR‐17‐5p and miR‐20a (O'Donnell et al, 2005), and is predicted by PicTar (Krek et al, 2005) to be a target for both miR‐106b and miR‐93, was significantly downregulated as well. We also observed downregulation of p57 in WI‐38 cells, in agreement with PicTar predictions (Figure 6A; Supplementary Figure S4). Notably, these proteins have defined functions in the regulation of the cell cycle, most of them being negative regulators of proliferation. As the mRNA levels of p57, p21, pRb and p130 did not decrease in WI‐38 cells and only marginally in MCF10A cells (Figure 6B), the reduction in their protein levels most likely stems from translational inhibition and not mRNA degradation. Considering the above, it is unlikely that the reduction in the targets' protein levels stems from reduced E2F transcriptional activity. In contrast, E2F1 mRNA levels were reduced in both cell lines that express the miR‐106b/93/25 polycistron, in agreement with Petrocca et al (2008).
To gain independent support for the role of the ‘p53‐repressed miR cluster,’ we investigated the expression pattern of genes that are targeted by these miRs in the above‐mentioned transformation system, where primary WI‐38 cells were gradually transformed into tumorigenic cells (Milyavsky et al, 2005). Interestingly, targets harboring predicted sites for multiple miRs from the ‘p53‐repressed miR cluster’ within their 3′‐UTR had significantly coherent expression patterns (Pilpel et al, 2001) during the transformation process (Figure 6C). This observation means that genes that are targeted by multiple miRs from the ‘p53‐repressed miR cluster’ are significantly co‐expressed during the transformation process as compared with random sets of genes. Furthermore, not only are these targets expressed similarly to one another but also the actual expression pattern of many of them is consistent with the pro‐proliferative role of the miRs that regulate them, i.e. the expression of the majority of the genes in this target set was decreased when cells gained the accelerated proliferation phenotype (designated as ‘fast growing’). Furthermore, promoter analysis, using the AMADEUS algorithm (Linhart et al, 2008), of these predicted target genes, revealed the E2F‐binding site as one of the most highly enriched motifs (P‐value=2.2 × 10−13). This supports our general notion that E2F cooperates with the ‘p53‐repressed miR cluster’ in regulating shared targets at the transcriptional and post‐transcriptional levels. This suggests that the FFL consisting of E2F and miRs‐106b/93/25 may be deregulated during cancer progression. We note that a fifth of the predicted targets of the miR cluster show a very different expression pattern (Figure 6C, bottom part), which may indicate more complex regulatory interactions.
Having shown the molecular effects of the overexpression of the miR‐106b/93/25 polycistron, we tested whether proliferation‐related parameters such as growth rate, colony formation efficiency (CFE) and replicative senescence are affected by these miRs. As these miRs are significantly repressed by p53 during senescence, and considering the fact that they target several antiproliferation regulators, we predicted that their overexpression, similarly to p53 inactivation, would accelerate cellular growth rate and delay senescence. Indeed, as depicted in Figure 7, the miR‐106b/93/25‐overexpressing WI‐38 cells demonstrated a moderate acceleration in proliferation rate and an increased fraction of S‐phase cells (24% compared to 18%). Strikingly, these cells displayed a pronounced increase in the efficiency of young cells to form colonies when seeded at low density and reduced senescence‐associated beta‐galactosidase (SA‐β‐Gal) staining at late passages, indicating a delay of replicative senescence. Additionally, we evaluated the effect of miR‐106b/93/25 overexpression in WI‐38 cells on their CFE in combination with p53 inactivation. As depicted in Supplementary Figure S5, miR‐106b/93/25 enhanced the CFE of both active and inactive p53‐expressing cells. However, the effect of the overexpressed miRs was much more pronounced in the active p53 cells, augmenting their CFE by 20‐fold as compared with only 2.6‐fold increase in CFE in the inactive p53 cells. In fact, the effect of overexpression of the miRs on the CFE of the control cells was comparable to that of p53 inactivation. These observed phenotypes suggest that the transcriptional repression of miR‐106b/93/25 and their paralogs mediates part of the antiproliferative effects of p53.
In the present study, we elucidate a complex regulatory network involving a group of cancer‐related miRs. In this network, E2F1 transcriptionally controls the miR‐106b/93/25 polycistron and its paralogs, and together they regulate a mutual set of target genes. In concordance with the growth acceleration that resulted from the overexpression of these miRs, many of their targets are considered antiproliferative cell‐cycle regulators. Importantly, this intricate FFL is repressed by p53 through inhibition of E2F1. A schematic model for the proposed network is presented in Figure 8.
Employing three independent experiments, we identified a novel miR signature that is transcriptionally repressed by p53 in human primary cells and in breast cancers. Consistent with p53 function, many signature members, including the three paralogous polycistrons and miR‐155, are considered oncogenic miRs and are overexpressed in diverse types of tumors (Eis et al, 2005; He et al, 2005; Volinia et al, 2006; Yanaihara et al, 2006). miR‐15b and miR‐16, which are considered to be tumor suppressor miRs (Cimmino et al, 2005), are exceptions in this regard.
Upon diverse stress stimuli, p53 is known to regulate different subsets of genes, resulting in alternative cellular outcomes (Oren, 2003). Consistently, the repression of the miRs was restricted to non‐genotoxic contexts, namely, replicative senescence and Nutlin‐induced Mdm2 inhibition, as doxorubicin treatment did not result in transcriptional repression of the miR cluster despite p53 activation. Our earlier studies have shown that Nutlin treatment induces p53‐dependent senescence accompanied by upregulation of miR‐34 (Kumamoto et al, 2008). Therefore, the repression of the miRs might be specific to p53‐induced senescence triggered in this study by cellular aging or Nutlin treatment. As p53 is capable of inducing senescence in vivo (Xue et al, 2007), the observed miRNA repression in the wild‐type p53‐harboring breast tumors may be associated with increased senescence in these samples.
Co‐clustering of coding mRNAs and microRNAs from the breast cancer study separated the p53‐repressed miRs into two functional and regulatory categories, namely ‘cell cycle’ and ‘immune response’ (Figure 3). The expression‐based division perfectly mirrored earlier reported functions of these miRs, including members of the three paralogous polycistrons that were co‐clustered with cell‐cycle‐associated mRNAs, and were shown here and in additional reports (Hossain et al, 2006; Lu et al, 2007; Ivanovska et al, 2008) to promote cell proliferation. Another p53‐repressed polycistron, miR‐15b/16, was clustered with the cell‐cycle genes. Indeed, members of this polycistron have been implicated in the regulation of cell‐cycle progression (Linsley et al, 2007). The ‘immune response‐associated co‐cluster’ included miR‐155, which is commonly overexpressed in lymphomas (Eis et al, 2005), participates in the germinal center response (Thai et al, 2007), and is upregulated in chronic gastritis (Petrocca et al, 2008). The remaining members of this co‐cluster were also implicated in immune responses, including miR‐150 and miR‐146 (Lodish et al, 2008) as well as miR‐142 (Wu et al, 2007). Overexpression of miR‐155 in primary cells did not affect the rate of proliferation (Supplementary Figure S3F), suggesting distinct functions for the members of the ‘immune response‐associated co‐cluster.’ Interestingly, in addition to many cell‐cycle‐related coding genes reported to be repressed by p53, recent evidence indicates p53‐mediated repression of immune response‐related genes (e.g. interleukin‐1β, interleukin‐6 and Cxcl1 (Buganim et al, submitted) and SDF‐1 (Moskovits et al, 2006)).
For the first time, we demonstrate that the three paralogous polycistronic miRNAs are coordinately activated by E2F1. Importantly, we establish E2F1 as the mediator of the p53‐dependent repression of miRs‐106b/93/25 and suggest that this mechanism underlies the repression of the two additional paralogous polycistrons. Upon Nutlin treatment, E2F1 protein levels were dramatically downregulated in a p53‐dependent manner. A similar phenomenon was described earlier, and was attributed to enhanced ubiquitylation of E2F1 by an unknown ligase, resulting in proteasome‐mediated degradation (Ambrosini et al, 2007). However, we demonstrate that Nutlin treatment also results in a robust p53‐dependent E2F1 mRNA repression, in agreement with an earlier observation that overexpression of p53, as well as p21, results in downregulation of E2F1 mRNA (Ookawa et al, 2001). It is plausible that p53‐mediated reduction in E2F1 protein inhibits E2F1 transcription as this gene contains an E2F motif in its promoter and is itself an E2F target gene (Johnson et al, 1994). Thus, even a slight reduction in E2F1 protein level might trigger a feedback loop that will result in significant reduction of both protein and mRNA levels. Considering this feedback loop, inhibition of E2F1 activity could also explain the observed repression of its mRNA and protein levels. Such inactivation may be indirectly mediated by the p53 target gene p21 through the inhibition of CDKs that inactivate the pocket proteins, which in turn inhibit E2F activity. Another mechanism for E2F1 inactivation by p53 was recently suggested by identifying BTG3 as a p53 target gene that directly binds E2F1 and inhibits its activity (Ou et al, 2007). Yet another possible mediator for p53‐dependent E2F1 suppression is miR‐34a, a direct transcriptional target of p53, which was recently suggested to induce senescence and to repress the E2F pathway (Tazawa et al, 2007).
Earlier studies have reported the effect of a single miR on a single target gene, such as miR‐106b effect on p21 (Ivanovska et al, 2008). Others have described E2F‐dependent activation of a single polycistron, miR‐17‐92 (Sylvestre et al, 2007; Woods et al, 2007), and miRs‐106b/93/25 (Petrocca et al, 2008). We suggest that the three polycistrons, consisting of a family of 15 different miRs, are in fact transcriptionally co‐regulated directly by E2F and indirectly by p53. In addition, many miRs in this family share highly similar seed sequences (Tanzer and Stadler, 2004). Thus, when an entire miR family is coordinately activated, its combinatorial and cumulative effects on mRNA targets may be profound. To recapitulate natural conditions, we combinatorially expressed three of the family miRs (miRs‐106b/93/25), which are naturally co‐transcribed, and demonstrated their effect on a set of target proteins. Future technologies allowing combinatorial knock down of an entire miR family may further establish their effects on other target genes. We show that miRs‐106b/93/25 silence key members of the E2F pathway, including negative regulators of proliferation such as the pocket proteins pRb and p130 and the CDK inhibitors p21 and p57 (Figure 6). These and many other cell‐cycle regulators are known E2F targets and are predicted to be silenced by miRs‐106b/93/25 (Supplementary Table S5). Thus, we provide experimental evidence for our recent in silico predicted FFL motif (Shalgi et al, 2007).
We demonstrate here (Figure 6) a cancer‐related manifestation of the concept of miR target avoidance (Farh et al, 2005; Stark et al, 2005). These studies introduced the concept of ‘spatio‐temporal avoidance,’ showing that miRs and their targets tend to avoid being expressed in the same tissue or at the same developmental time, thereby assisting to determine differentiation boundaries and transitions. This avoidance could reflect a direct negative regulatory effect of the miR on its target. Alternatively, miR target avoidance may be mediated through a common transcriptional program controlling both the miRs and their targets.
The overexpression of miRs‐106b/93/25 phenotypically mimicked p53 inactivation in WI‐38 cells, as evident from an elevated rate of proliferation, increased CFE and delay of senescence. Importantly, induction of senescence, which we suggest to be partially mediated by the repression of the polycistronic miRs described above, is considered one of the main mechanisms by which p53 suppresses tumor formation (Xue et al, 2007).
In summary, we present here another arm of p53's tight control of cell proliferation, senescence and tumor suppression. This involves an elaborate network encompassing miRs and their targets, which modulate cell fate both during normal growth and in cellular senescence.
Materials and methods
WI‐38, MRC5, IMR90 (obtained from the ATCC) and PFCA179 (provided by Dr H Klocker) cells were cultured in MEM with 10% FCS (fetal calf serum), 1 mM sodium pyruvate, 2 mM l‐glutamine and antibiotics. U2OS and H1299 cell lines were cultured in DMEM and RPMI, respectively, with 10% FCS and antibiotics. MCF10A cells were maintained in DMEM F12 supplemented with 5% horse serum, 0.5 μg/mlhydrocortisone, 0.1 mg/ml insulin, 0.1 μg/ml cholera toxin and 10 ng/ml EGF. All cells were maintained in a humidified incubator at 37°C and 5% CO2. Primary fibroblasts were passaged every 5–6 days. PDLs were calculated using the formula: PDLs=log(cell output/cell input)/log2. For colony formation assays, cells were plated at low density (0.1–0.2 cells/mm2), grown for 10–14 days and stained with crystal violet.
Plasmids and retroviral infections
The retrovirus encoding for GSE56 was described by Milyavsky et al (2005). shRNAs targeting p53 (p53i), or mouse NOXA (control shRNA) were stably expressed using pRetroSuper and were described by Berkovich and Ginsberg (2003). pRetroSuper‐E2F1 was described by Korotayev et al (2008). ER‐E2F1 was described by Vigo et al (1999). E1A was expressed from pBabe‐puro‐E1A12S (a gift from K Helin). For expression of miRs‐106b/93/25, a 1‐kb human genomic fragment was cloned with the primers 5′‐ggatcctatcctgcgcctttcc‐3′ and 5′‐cacatggccacagaagac‐3′ into miR‐Vec (Voorhoeve et al, 2006). For expression of miR‐155, a 238‐bp human genomic fragment was cloned with the primers 5′‐gtggcacaaaccaggaag‐3′ and 5′‐tatccagcagggtgactc‐3′ into miR‐Vec. Retrovirus infection procedures were described by Milyavsky et al (2003).
RNA preparation and quantitative real‐time PCR
RNA was extracted with TRI reagent (Molecular Research Center Inc.). For mRNA quantification, a 2 μg aliquot of total RNA was reverse transcribed using Bio‐RT (Biolab) and random hexamers. Quantitative real‐time PCR (QRT–PCR) was performed using Platinum SYBR Green qPCR SuperMix (Invitrogen). mRNA levels were normalized to the level of GAPDH in the same sample. Primer sequences are listed in Supplementary Table S6A. For miRNA quantification, TaqMan miRNA assays (Applied Biosystems) were used according to the manufacturer's protocol. Levels were normalized to the U6 control gene. All QRT–PCR reactions were performed on ABI7300 machine. Results are presented as mean and standard deviation for two or three duplicate runs.
miRNA microarrays, data analysis and clustering
The miRNA profiling presented in Figure 1A was performed as follows: RNA was extracted from WI‐38 cells using TRI reagent as described above, labeled with Hy5 and hybridized on Exiqon's miRCURY™ LNA Array (v.8.1) with a common reference Hy3‐labeled RNA pool. Data are provided as Supplementary dataset S1. Two biological replicates were performed for each sample type. Hy5/Hy3 ratios were log2 transformed and filtered such that miRs that were undetected in 11 or 12 samples were discarded. Duplicates were averaged, such that each miR was represented by six values, corresponding to the six different samples. For each miR, a credibility value was calculated as one minus the average of the six standard deviations (s.d.) between the duplicates. A duplicate that had one missing value was set as the detected value and was assigned with high s.d. The 5% most non‐credible miRs were discarded. Data were clustered using hierarchical clustering (average linkage), with 20 clusters. The miRNA profiling presented in Figure 1B was performed as follows: RNA was extracted from young and old WI‐38 and MRC5 cells using TRIZOL (Invitrogen) according to the manufacturer's protocol and was used for biotin labeling and hybridization on the version 4.0 miR arrays (Ohio State University) as described by Liu et al (2004). Data were normalized and log2‐transformed. Data are provided as Supplementary dataset S2. Data were clustered using hierarchical clustering (average linkage), with 10 clusters. The miRNA profiling presented in Figure 1C was performed as follows: RNA was extracted as described by Sorlie et al (2006). Samples were hybridized on Agilent's miRNA arrays (beta version of V1) at the Agilent's facilities in St Clara, US by HJ. Data are provided as Supplementary dataset S3. Two outlier samples (both belonging to the mutant p53 set) were discarded (see Supplementary Figure S6). Data were clustered using hierarchical clustering (average linkage), with 25 clusters.
Microarray data can be downloaded from the GEO database (http://www.ncbi.nlm.nih.gov/geo/). Accession numbers are as follows: GSE12450 (the WI‐38 p53‐dependent senescence study), GSE12821 (the WI‐38 and MRC5 senescence study) and GSE12848 (The Breast Cancer Study).
Co‐clustering of miRNA and mRNA expression data from human breast cancer samples
mRNA array data were filtered and normalized as in Sorlie et al (2006) (GEO accession number GSE3155) and only samples that were included in the miRNA analysis were used. Data were log2‐transformed, and replicate mRNA probes were averaged. Only variable mRNAs that differ by at least 1.5‐fold from their median expression in at least 40% of the samples were considered. miRNA data were centered such that the intensity values for each miR were divided by their mean, and log2 transformed. The combined mRNA and miRNA data were then clustered using hierarchical clustering (average linkage) with 40 clusters. miRs from the ‘breast cancer p53‐repressed miR cluster’ were mapped to the resulting co‐clusters, and were found to reside in two clusters, that were then analyzed for functional enrichment using DAVID (Dennis et al, 2003), and for enrichment of sequence motifs in their corresponding ENSMBL gene promoters using AMADEUS (Linhart et al, 2008).
Analysis of miRNA targets expression coherence
The entire set of miRNA expression profiles was clustered into 20 miR clusters based on the above expression data (WI‐38 young versus senescent, along with p53 inactivation). Then, we compiled a set of predicted targets for the miRs from each cluster using PicTar (Krek et al, 2005). Specifically, for each of the 20 miR clusters, a series of potential sets of targets were created. The first set consisted of mRNAs predicted to be targeted by at least one miR from the miR cluster. The second set consisted of mRNAs predicted to be targeted by at least two miRs from the miR cluster, and so on. The expression coherence (EC) score, a measure of expression similarity (Pilpel et al, 2001), was then computed for each set of targets according to their expression described by Milyavsky et al (2005) (expression data are also available as Supplementary dataset S5). The most significantly coherent expression pattern belonged to the set of genes that had target sites for at least five miRs from the ‘p53‐repressed miR cluster’ (EC P‐value=5 × 10−3). The expression values for this gene set in the data from Milyavsky et al (2005) and prediction of E2F sites are found in Supplementary Figure S4.
Western blots were performed as described by Milyavsky et al (2005). The following antibodies were used: α‐p53 pAb H‐47 (produced in our laboratory), α‐p21 sc‐377 (Santa Cruz), α‐E2F1 sc‐193 (Santa Cruz), α‐GAPDH MAB374 (Chemicon), α‐p130 sc‐317 (Santa Cruz), α‐p57 sc‐8298 (Santa Cruz), α‐pRb 554136 (Pharmingen), and α–β‐tubulin T7816 (Sigma).
Cells were labeled for 30 min with 10 μM BrdU (Sigma), fixed with 70% EtOH/HBSS (2 h, −20°C), treated with 2 M HCl/0.5% Triton, washed and treated with 0.1 M Na2B4O7 pH 8.5, and stained with FITC‐conjugated anti‐BrdU (Becton Dickinson) and 10 μg/ml propidium iodide. Samples were analyzed using a FACSort machine (Becton Dickinson). At least 1 × 104 events were recorded per sample.
SA‐β‐Gal activity assay
Cells were fixed with 3% formaldehyde/PBS for 5 min, washed with PBS and incubated for 16 h at 37°C with a solution containing 1 mg/ml X‐gal, 40 mM citric acid, sodium phosphate, pH 6.0, 5 mM potassium ferrocyanide, 5 mM potassium ferricyanide, 150 mM NaCl and 2 mM MgCl2.
DNA–protein complexes were immunoprecipitated from U2OS cells using the ChIP assay kit (Upstate Biotechnology), according to the manufacturer's protocol with the following polyclonal antibodies: α‐E2F1 sc‐193 (Santa Cruz) and α‐HA sc‐805 (Santa Cruz); the latter served as a control for nonspecific DNA binding. The precipitated DNA was subjected to QRT–PCR analysis using specific primers corresponding to each predicted E2F site, as well as primers for normalization (β‐tubulin) and negative control for E2F1 binding (β‐actin coding region). Primer sequences are listed in Supplementary Table S6B.
This study was supported by a Center of Excellence grant from the Flight Attendant Medical Research Institute; EC FP6 funding (contract no. 502983); FP7 funding (ONCOMIRS, agreement 201102); the Ben May Charitable Trust; the Israel Science Foundation and the Yad Abraham Center for Cancer Diagnosis and Therapy. This research was supported in part by the Intramural Research Program, NIH, NCI and CCR. VR is the incumbent of the Norman and Helen Asher Professorial Chair Cancer Research at the Weizmann institute. YP is an incumbent of the Rothstein Career Development Chair in Genetic Diseases. RS is a fellow of the Horowitz Foundation for Complexity Sciences. GHN was supported by the Howard Hughes Medical Institute Grant for Graduate Medical Education. This publication reflects only authors' views. The European Commission is not liable for any use that may be made of the information herein. We thank Agilent Technologies for access to the early version of their miRNA arrays and for their support.
Conflict of Interest
The authors declare that they have no conflict of interest.
Supplementary Information [msb200865-sup-0001.pdf]
Supplementary Dataset 1 [msb200865-sup-0002.xls]
Supplementary Dataset 2 [msb200865-sup-0003.xls]
Supplementary Dataset 3 [msb200865-sup-0004.xls]
Supplementary Dataset 4 [msb200865-sup-0005.xls]
Supplementary Dataset 5 [msb200865-sup-0006.xls]
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.
- Copyright © 2008 EMBO and Nature Publishing Group