Introduction

Cancer cells show characteristic gene expression profiles, and both genetic and epigenetic abnormalities underlie these changes (Hanahan and Weinberg, 2000). Recent studies indicate that cancer cells show characteristic microRNA (miRNA) expression profiles, which, at least in part, reflect the upregulation of oncogenic miRNAs (targeting tumor-suppressor genes) and downregulation of tumor-suppressor miRNAs (targeting oncogenes), which lead to misbehavior of cancer cells, including invasion and metastasis (Calin and Croce, 2006; Ma and Weinberg, 2008; Nicoloso et al., 2009). It is now important to clarify their clinically relevant targets, their functional relationships and the mechanisms of their altered expression.

Epithelial–mesenchymal transition (EMT) is a crucial event during animal development and cancer metastasis (Polyak and Weinberg, 2009). Several transcription factors such as SNAIL, ZEB and TWIST (Gregory et al., 2008; Yang et al., 2008), and several signaling molecules such as transforming growth factor-β, Notch, Shh and receptor tyrosine kinases (Polyak and Weinberg, 2009) have been implicated in the induction of EMT. Hypoxia, a micro-environmental factor known to promote tumor growth and angiogenesis, is also a potent EMT inducer (Sahlgren et al., 2008). The entire network of signaling pathways regulating EMT, however, remains unclear at present.

The membrane-anchored metalloprotease regulator RECK was identified as a gene inducing flat reversion in v-K-ras-transformed NIH3T3 cells (Takahashi et al., 1998). In several common cancers, RECK expression is frequently reduced and the level of residual RECK expression in cancer tissues correlates with better prognosis (Noda and Takahashi, 2007). Forced expression of RECK in cancer cells results in reduced angiogenesis, invasion and metastasis in animal xenograft models (Takahashi et al., 1998; Oh et al., 2001). Mice lacking a functional Reck gene die around embryonic day 10.5 with reduced tissue integrity and defects in various organs, including blood vessels and the central nervous system (Oh et al., 2001; Muraguchi et al., 2007). Studies on RECK gene regulation may therefore yield important insights into the mechanisms of carcinogenesis as well as mammalian development.

In this study, we found that RECK is a target of at least three groups of miRNAs (that is, miR-15b/16, miR-21 and miR-372/373) and that miR-372/373 mediate hypoxia-induced downregulation of RECK through the hypoxia-inducible factor-1α (HIF1α)/TWIST1 pathway. Our data indicate that two (that is, hypoxia-induced and receptor tyrosine kinase-mediated) signaling pathways converge on RECK through miRNAs to cooperatively downregulate this potent suppressor of malignancy and to promote the malignant behavior of cancer cells.

Results

RECK is a target of three groups of miRNAs

Although RECK protein is readily detectable in non-malignant cells, it is undetectable in a variety of cancer-derived cell lines (Figure 1a, top panel). RECK mRNA, however, is detectable in many cancer-derived cell lines (Figure 1a, third panel), suggesting post-transcriptional downregulation of RECK in cancer cells.

Figure 1
figure 1

RECK mRNA is under the control of three groups of miRNAs. (a) Expression of RECK protein (immunoblot assay, top panel) and RECK mRNA (RT–PCR, third panel) in human cell lines/strains. α-tubulin (second panel) and GAPDH (bottom panel) were used as internal controls. Origins: MRC-5, fibroblasts; MCF-10A, mammary epithelium ((S), (R): spontaneous variants with different morphologies); HT1080, fibrosarcoma; MCF-7, MDA-MB-231 and MDA-MB-468, mammary adenocarcinomas; A375, malignant melanoma; A549, lung adenocarcinoma; SW480 and SW620, colon adenocarcinomas; and Panc-1, pancreatic carcinoma. (b) The predicted interaction sites for miR-15b/16 (site A: red, 793-817), miR-21 (site B: blue, 1118–1144) and miR-372/373 (site C: green, 1249–1282) in the RECK 3′-UTR (GenBank D50406; position-1 is the nucleotide next to the termination codon). These sites are located in highly conserved regions: the lower histogram represents the nucleotide sequence identity between mouse and human RECK genes (Vista Genome Browser). (c) Complementarity between RECK 3′UTR and miR-15b, miR-21, or miR-373 predicted by TargetScan. Red/blue dots: Typical Watson–Crick interaction (A–U and G–C, respectively). Green dots: Weak non-typical basepair interactions (Lewis et al., 2005). (d) Effects of overexpressed miRNAs on RECK protein expression. RECK-expressing HT1080 cells (RGC-1R) were transfected with the indicated miRNA precursor(s). RECK protein and RECK mRNA were analyzed 48 h after transfection as outlined in A. The cells in lane-1 were transfected with the precursor of an irrelevant miRNA (cel-miR-67). The numbers below the top photograph indicate the relative intensity of RECK bands normalized against the α-tubulin bands and then divided by the value for the control (lane-1). (e) Effects of RECK 3′UTRs and their mutants on the expression of the linked luciferase gene. Caco-2 cells were co-transfected with pRL-TK and the luciferase gene linked with the 3′RECK-wt or the luciferase gene linked with the RECK 3′UTR containing mutations at the indicated miRNA-interaction site(s); luciferase activity was determined 48 h after transfection (mean±s.e.m., n=3; normalized against the data with pRL-TK). **P<0.01. (f) Effects of inhibiting three groups of miRNAs on the expression of the luciferase gene linked to the RECK 3′UTR. Caco-2 cells were co-transfected with the indicated LNA(s), pRL-TK and a 3′RECK-wt reporter plasmid; luciferase activity was determined 48 h after transfection (n=3). The control cells were co-transfected with an irrelevant LNA (targeting ath-miR-176) and pRL-TK (lane 1). Difference from bar 2: *P<0.05, **P<0.01, ***P<0.005. (g) Effects of miRNA-antagonists (LNAs) on endogenous RECK expression. Caco-2 cells were transfected with the indicated LNA(s), and the expression of **RECK protein and RECK mRNA were analyzed as outlined in A. The control cells (lane 1) were transfected with the irrelevant LNA. GAPDH, glyceraldehyde-3-phosphate dehydrogenase; miRNA, micro RNA; 3′RECK-wt, wild-type RECK 3′UTR; RT–PCR, reverse transcription–PCR; UTR, untranslated region; wt, wild type.

Full size image

RECK mRNA contains multiple islands of highly conserved sequences in its 3′-non-coding region (3′UTR) (Figure 1b, lower panel). Database search allowed us to predict potential target sites for at least three groups of miRNAs in these conserved regions (Sites A, B, C; Figures 1b and c). We tried to test this prediction by three sets of experiments. First, we introduced synthetic miRNA precursors, either individually or in combination, into the cells and examined their effects on the level of RECK protein (Figure 1d). The recipient cell (RGC-1R) was derived from the HT1080 human fibrosarcoma cell line by transfection with a vector expressing the full-length RECK mRNA. Introduction of the miRNAs resulted in downregulation of RECK protein (Figure 1d, Protein) with no appreciable changes in the level of RECK mRNA (Figure 1d, mRNA). Notably, miRNAs targeting separate sites (shown in Figure 1b) seemed to reduce RECK expression in a cooperative manner (Figure 1d, Protein).

In the second set of experiments, a luciferase-reporter gene linked to the wild-type RECK 3′UTR or a mutated version was expressed in Caco-2 cells, and the luciferase activity was determined (Figure 1e). The recipient cell line Caco-2 (derived from colon carcinoma) endogenously expressed all three groups of miRNAs (Figure 2). Luciferase activity was strongly suppressed when the reporter gene was linked to the wild-type RECK 3′UTR (3′RECK-wt) (Figure 1e, bar 2). This effect was relieved when the predicted miRNA-target sites were mutated (Figure 1e and Supplementary Figure 1), suggesting that these sites are involved in suppression of luciferase expression. As with the above experiment, the effect of mutating more than one miRNA-target site was cooperative.

Figure 2
figure 2

Relative abundance of three groups of miRNA in MRC-5 human fibroblast strain and six human tumor-derived cell lines. (a) Schematic representation (left panel; not in scale) of the primary transcripts of the three groups of miRNAs examined in this study. Blue regions represent precursors and red regions mature forms. Red arrows indicate the positions of primers used for semi-quantitative RT–PCR. The products were examined by agarose gel electrophoresis. Typical data are shown in the right panels. GAPDH mRNA was detected as an internal control. (b) Mature miRNA detected by miRNA qRT–PCR (top bar graph) and RNA blot hybridization (second panel). U6 was detected as an internal control (bottom panel).

Full size image

In the third set of experiments, the activity of a luciferase gene linked to the wild-type RECK 3′UTR was measured in the presence of antisense oligonucleotides (LNAs) against miRNAs, individually or in combination, to knock down endogenous miRNAs in Caco-2 cells (Figure 1f). Luciferase activity was strongly suppressed in the absence of LNA treatment (Figure 1e, bar 2), and this suppression was relieved by introduction of the LNAs (Figure 1f), suggesting that the LNA-targeted miRNAs were responsible for suppression of luciferase expression in this cell line. Again, recovery of luciferase activity occurred in a cooperative manner. Importantly, the level of endogenous RECK protein was also elevated when Caco-2 cells were treated with these LNAs (Figure 1g).

In addition to these experimental data, multiple linear regression analysis of our data (Figures 1a and 2b) indicated significant negative correlation between the level of RECK protein and the levels of all miRNAs detected among the cell lines and strain examined (R=−0.983, P<0.05).

Taken together, these findings implicate three groups of miRNAs in the post-transcriptional downregulation of RECK protein in cancer cells.

RECK-mediated suppression of invasion and MMP-9 production are inhibited by miRNAs

We next asked whether the miRNAs could modulate the malignant behavior of tumor cells through inhibition of RECK expression. We obtained three lines of evidence indicating that this is indeed the case in cultured cells. First, the miRNA precursors cooperatively relieved the effects of wild-type RECK cDNA to suppress Matrigel invasion and pro-matrix metalloproteinase-9 (pro-MMP-9) secretion in HT1080 cells (Figure 3a). Second, two mutant RECK cDNAs lacking miRNA-target sites (one carrying mutations at the three miRNA-target sites (RECK-mutABC) and the other lacking the entire 3′UTR (RECK-Δ3′UTR)) could express higher levels of RECK protein in HT1080 cells (Figure 3b) and suppress the Matrigel invasion of HT1080 cells more strongly than wild-type cDNA (RECK-wt) (Figure 3c). Third, these mutant cDNAs could suppress the Matrigel invasion of Caco-2 cells to a full extent without the aid of miRNA antagonists (Figure 3d).

Figure 3
figure 3

Three groups of miRNAs cooperatively abrogate the RECK-mediated suppression of gelatinolysis and Matrigel invasion. (a) Effects of miRNA overexpression on Matrigel-invasion activity of HT1080 expressing a wild type RECK cDNA (RGC-1R). RGC-1R cells were transfected with the indicated miRNA precursors (below the bar graph) and, after 48 h, their invasive activity (8 h incubation) was determined in Matrigel invasion chambers using 10% serum as a chemo-attractant (top bar graph; mean±s.e.m., n=3). Difference from bar 1: **P<0.01, ***P<0.005. Gelatinolytic activities released into the culture supernatant were also analyzed by gelatin zymography (bottom panel). Note that the miRNA precursors cooperativelyenhanced both Matrigel invasion and pro-MMP-9 secretion in RGC-1R cells, and these effects were roughly parallel to the reduction in the level of RECK protein (see Figure 1d, Protein). These data support the idea that three groups of miRNAs up-regulate MMP-9 and promote invasion by inhibiting RECK protein expression. (b) Effects of mutations in the 3′UTR of RECK cDNA on the level of RECK protein and mRNA when expressed in HT1080 cells. The vacant expression vector (V) or the vector containing wild type RECK cDNA (wt), RECK cDNA carrying mutations at sites A, B, and C (ABC), or RECK cDNA lacking the entire 3′UTR (Δ) was transfected into HT1080 cells. The levels of RECK protein and RECK mRNA were analyzed as outlined in Figure 1a. Note that the 3′UTR-mutations and deletion-mutation significantly enhanced RECK protein expression (lanes 3, 4), which indicates that the mutant genes escaped the post-transcriptional regulation mediated by the miRNAs. (c) Effects of the RECK expression vectors and miRNA precursors on Matrigel invasion by HT1080 cells. HT1080 cells were either mock-transfected (lanes 1–4), transfected with an irrelevant miRNA precursor (provided by Dharmacon) (lanes 5–8), or transfected with precursors of the five RECK-targeting miRNAs (lanes 9–12) on day 1, subsequently transfected with the indicated expression vectors (see bar legends below panel (d) on day 2, and re-plated onto Matrigel invasion chambers on day 3. Invasive cells were determined after 16 h incubation with 10% FCS in the lower chambers (mean±s.e.m., n=3). **P<0.01. Note that the mutant cDNAs strongly suppressed Matrigel invasion (bars 3, 4) and this effect was insensitive to miRNA precursors (bars 11, 12). (d) Effects of RECK expression vectors and LNAs on Matrigel invasion by HGF-stimulated Caco-2 cells. Caco-2 cells were either mock-transfected (lanes 1–4), transfected with irrelevant LNA (against ath-miR-176) (lane 5–8), or transfected with the five LNAs used in Figure 1f (lanes 9–12) on day 1, subsequently transfected with the indicated expression vector (see the bottom bar legends) on day 2, and re-plated onto Matrigel invasion chambers on day 3. Invasive cells were determined after 16 h incubation with 100 ng/ml HGF in the lower chambers (mean±s.e.m., n=3). **P<0.01. Note that the mutant cDNAs alone could suppress the Matrigel-invasion (bars 3, 4, 7, 8) to levels comparable to that achieved by the wild type cDNA only when the miRNA antagonists were present (bar 10).

Full size image

Elimination of the miRNA-target sites augmented the tumor-suppressor activity of RECK cDNA

To test the activity of mutant cDNAs in vivo, four stable transfectants (that is, vector (V), RECK-wt (wt), RECK-mutABC (ABC), and RECK-Δ3′UTR (Δ)) were prepared using luciferase-tagged SW620 cells (Figure 4a). After rectal transplantation into nude mice (n=4–6), all four samples formed tumors of comparable sizes within 4 weeks (for example, Figure 4b, top panels). The vector-transfected cells formed tumors that continuously grew and gave rise to metastatic nodules in the para-aortic lymph nodes and lungs at high frequencies (Figures 2b–d, V). Tumor sizes and frequency of metastases were slightly lower when wild-type RECK was expressed (Figures 2b–d, wt). When either of the mutant RECK mRNAs was expressed, the rate of tumor growth progressively declined, the tumors eventually regressed and the frequency of metastases was low (Figures 2b–d, ABC and Δ). Histological examination at 6 weeks after inoculation indicated that cell density and blood vessel density were greatly reduced when a mutant RECK mRNA was expressed (Figure 4e, right two rows) as compared with that in the vector-transfected or RECK-wt-transfected cells (Figure 4e, left two rows). Immumoblot assay indicated the retention of high RECK expression in tumors expressing a mutant RECK mRNA (Figure 4f, Protein, lanes 3 and 4) and loss of RECK expression in tumors expressing the wild-type RECK mRNA (Figure 4f, Protein, lanes 2). Among these samples, the levels of RECK expression correlated inversely with the amount of MMP-2 and MMP-9 present in the tissues (Figure 4f, Gelatin Zymo.). Importantly, the levels of human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA in the peripheral blood (that is, human tumor cells in the circulating blood) were low in tumors expressing a mutant RECK mRNA (Figure 4g, bar 3 and 4). Other findings (Supplementary Figure 2) suggest that miRNA-mediated RECK downregulation results in increased intravasation and lung colonization, probably due to MMP-dependent reductions in vascular endothelial integrity. Hence, when the miRNA-mediated suppression of RECK expression is abrogated (by eliminating miRNA-target sites), the abilities of RECK to inhibit tumor growth, invasion, intravasation and metastasis are enhanced.

Figure 4
figure 4

Elimination of miRNA-target sites augments the activities of RECK to suppress the growth and metastasis of the human colon cancer cell line SW620 in nude mice. (a) Immunoblot detection of RECK protein in SW620-Luc cells stably transfected with the vacant expression vector (V) or the vector containing wild type RECK cDNA (wt), RECK cDNA carrying mutations at three miRNA target sites (ABC), or RECK cDNA lacking the entire 3′UTR (Δ). (b) Bioluminescense imaging of nude mice after orthotopic inoculation (in the colon submucosa) of the SW620-Luc transfectants. (c) Time course of tumor size. Photon flux data at different time points after inoculation are plotted (mean±s.e.m.; n=6). (d) Incidence of metastatic foci in the para-aortic lymph nodes and the lungs at week 16. The data represent the number of animals with metastatic foci/total number of animals examined. (e) Tumor histology. At 5 to 6 weeks after inoculation, the tumors were resected, and the sections stained with either hematoxylin-eosin (HE) (top panels), anti-CD31 antibodies (red) plus DAPI (nucleus; blue) (second row), or anti-VEGFR2 antibodies (green) plus DAPI (bottom row). Scale bar, 100 μm. (f) RECK and gelatinolytic activities in the tumor tissues resected at 5 to 6 weeks after inoculation. Tissue lysates were subjected to immunoblot assay for RECK (top panel) or α-tubulin (second panel) and to gelatin zymography (bottom panel). (g) Extravasation of human tumor cells. RNA extracted from blood samples of mice bearing tumors was subjected to qRT–PCR to quantify human GAPDH mRNA. The data are normalized against β-actin mRNA (internal control). Bar represents mean+s.e.m. (n=3).

Full size image

Metastatic progression of colon carcinoma cells is associated with EMT and TWIST1-mediated upregulation of miR-372/373

To test whether miRNA upregulation has a role in tumor progression, we established several cell clones (for example, C15–17) from metastatic nodules formed after inoculation of SW620-luc cells into nude mice (Figures 5a and b). Interestingly, all these clones expressed elevated levels of miR-372/373 as compared with that in the parental cells (Figure 5c, right panel) and showed features of EMT such as increased spreading, E-cadherin downregulation and vimentin/TWIST1 upregulation (Figures 5b and d).

Figure 5
figure 5

Sub-clones of SW620 cells isolated from lung metastasis show TWIST1-dependnent over-expression of miR-372/373. (a) Outline of sub-clone isolation. (b) Morphology of the control cell lines (panels 1–3) and three metastasis-derived sub-clones of V/SW620-Luc (panels 4–6) stained with anti-vimentin antibodies plus DAPI. (c) Relative levels of mature form (upper panels) or the primary transcripts (lower panels) of miR-15b/16 (left), miR-21 (center), and miR-372/373 (right), as determined by qRT–PCR and agarose gel electrophoresis of RT–PCR products, respectively. (d) Immunoblot detection of E-cadherin, vimentin, and TWIST1. The sample numbering is the same as in B and C. (e) Structure of the promoter region of the miR-372/373 cluster. Three miRNAs, miR-371, miR-372, and mir-373 (blue boxes) are expressed as a single primary transcript (Houbaviy et al., 2005) (red box). There are four predicted TWIST1 binding motifs (that is, E-boxes) (black bars) near the transcriptional start site (arrow): three in the up-stream region and one down-stream of the start site. (f) Effects of TWIST1 on miR-372/373 expression. SW620 cells were stably transfected with either a vacant expression vector (V) or the vector expressing either of the two TWIST1 cDNA clones, BC033434 (#1) or BC083139 (#2). Both encode an identical protein; the only difference is that #2 has a longer 5′UTR. Cell lysates prepared from transfected and untransfected (−) cells were subjected to immunoblot assay to monitor TWIST1 expression (Control: α-tubulin) (Top panels). Total RNAs was subjected to qRT–PCR to assess the levels of mature miR-372/373 (bar graph) or to RT–PCR followed by agarose gel electrophoresis (control: GAPDH) to visualize the primary transcript harboring miR-372/373 (bottom panels). (g) Chromatin immunoprecipitation assay for TWIST binding sites. Nuclear lysates of C15 cells were precipitated using no (−), non-specific (N; rabbit IgG), or anti-TWIST1 (T) antibodies, and the DNA associated with the precipitates was used to amplify a region of approximately 120 bp containing each E-box (see Methods in SI for the primers used). (h) Luciferase reporter assay with deletion mutants of the miR-372/373 promoter. SW620 cells were co-transfected with a TWIST1-expression vector and a promoter-luciferase reporter construct containing either all four E-boxes (bars 2, 3) or only 2 (bars 5, 6) or 3 E-boxes (bars 7, 8). Luciferase activity was determined 48 h after transfection (n=3). *P<0.05.

Full size image

Database search located four potential TWIST-binding sites (E-boxes) around the predicted transcription start site of the miR-372/373 primary transcription unit (Houbaviy et al., 2005): three in the upstream region and one in a region downstream from the transcriptional start site (Figure 5e). Overexpression of TWIST1 in SW620 cells resulted in upregulation of miR-372/373 transcripts (Figure 5f). Chromatin immunoprecipitation assay showed association of TWIST1 with E-boxes 1 and 2 (Figure 5g). Luciferase-reporter assay showed that the upstream segment containing E-boxes 1 and 2 could drive twofold induction of the reporter gene upon TWIST1 co-transfection (Figure 5h). These data suggest that miR-372/373 are directly upregulated by TWIST1.

Hypoxia upregulates miR-372/373 while RAS/ERK signaling upregulates miR-21

Hypoxia is known to upregulate TWIST1 through HIF1α (Yang et al., 2008). When we treated RECK-wt-transfected SW620 cells with deferoxamine (DFX), a chemical known to stabilize HIFα (Wang and Semenza, 1993), we could observe accumulation of HIF1α followed by upregulation of TWIST1 and miR-372/373, and concomitant downregulation of RECK (Figure 6a). Similar phenomena were observed when the cells were exposed to a hypoxic environment (1% O2, 5% CO2, 94% N2) for 24 h (Figures 6b and h). The putative miR-372/373 promoter fragment was responsive to hypoxia and to DFX in luciferase-reporter assays (Figure 6c). Depletion of TWIST1 using siRNA abrogated the DFX-induced downregulation of RECK (Figure 6d). RECK expression became insensitive to DFX when the mutant RECK cDNAs were used (Figure 6e, lanes 4 and 6). These data support the idea that hypoxia-induced stabilization of HIF1α leads to upregulation of TWIST1 and subsequent induction of miR-372/373 expression. Our histological examination of colorectal tumors indicated colocalization of TWIST1 and miR-372 (Supplementary Figure 3), supporting the in vivo relevance of these findings.

Figure 6
figure 6

Effects of hypoxia and RAS/ERK signaling on miRNA expression. (a, b) Effects of DFX and hypoxia on the levels of HIF1α, TWIST1, RECK, and miR-372/373. SW620 cells were incubated in the presence of 100 μM DFX for the indicated period of time (a), or the cells were kept for 24 h under normoxic atmosphere (20% O2, 5% CO2, 75% N2) (N), hypoxic atmosphere (1% O2, 5% CO2, 94% N2) (H), or in medium containing 100 μM DFX (D) (b). Lysates of these cells were subjected to immunoblot assay to estimate the levels of HIF1α, TWIST1, RECK, and the control α-tubulin (α-Tub) (top 4 panels). Total RNA extracted from these cells was subjected to qRT–PCR to determine the levels of mature miR-372/373 (bar graphs) or to RT–PCR followed by agarose gel electrophoresis to detect the primary transcript harboring miR-372/373 (bottom 2 panels). (c) Effects of hypoxia or DFX on miR-372/373 promoter activity. SW620 cells were transfected with the luciferase reporter plasmid carrying the up-stream region of the miR-372/373 promoter (containing E-box 1, 2, 3). Twenty four hours later, the cells were placed under normoxic atmosphere (N), hypoxic atmosphere (H), or in medium containing DFX (D), incubated for another 24 h, and then subjected to a luciferase assay (n=3). *P<0.05. (d) Effects of TWIST1 siRNA on DFX-induced RECK down-regulation. SW620 cells stably expressing wild type RECK mRNA were transfected with control (C) or TWIST1 (T) siRNA. Twenty four hours later, DFX was added to the cells and incubated for another 24 h before the cells were lysed and subjected to immunoblot assays to estimate the levels of the indicated proteins. (e) Effects of mutations in RECK 3′UTR on DFX-induced RECK down-regulation. SW620 cells stably expressing wild type RECK mRNA (wt), the RECK mRNA carrying triple mutations at miRNA-target sites (ABC), or the RECK mRNA lacking the entire 3′UTR (Δ) were incubated in the absence (−) or presence (+) of DFX for 24 h and subjected to immunoblot assays.

Full size image

We also obtained some evidence indicating that miR-21 could be upregulated by RAS/extracellular signal-regulated kinase (ERK) signaling (Supplementary Table 1 and Supplementary Figure 4). Taken together, our findings indicate that the miR-21-mediated growth factor signaling and miR-372/373-mediated hypoxia signaling converge on RECK, leading to cooperative downregulation of this tumor suppressor (Figure 7).

Figure 7
figure 7

A model summarizing our findings.

Full size image

Discussion

The main conclusion of this study is that at least three groups of miRNAs (that is, miR-15/16, miR-21 and miR-372/373) negatively and cooperatively regulate the expression of RECK protein. While this study was in progress, three papers proposing RECK as a potential target of miR-21 were published (Gabriely et al., 2008; Hu et al., 2008; Zhang et al., 2008). Conversely, the links between RECK and the other two groups of miRNAs (miR-15/16 and miR-372/373) have not been described. Among the three, miR-21 and miR-372/373 have been classified as oncogenic miRNAs (Cheng et al., 2005; Iorio et al., 2005; Volinia et al., 2006; Voorhoeve et al., 2006). It is therefore reasonable that these miRNAs target the tumor suppressor RECK.

Our findings also support the idea that a single miRNA can regulate a cohort of functionally related genes. Previously identified or proposed miR-21 targets include tropomyosin-1 (an inhibitor of anchorage-independent cell proliferation) (Zhu et al., 2008), PTEN (a negative regulator of two MMP genes, MMP2 and MMP9, and of FAK phosphorylation) (Meng et al., 2006), programmed cell death-4 (PDCD4) (Asangani et al., 2008; Frankel et al., 2008) and maspin (two proteins implicated in suppression of invasion and metastasis) (Zhu et al., 2008), and MARCKS (a protein kinase-C-substrate regulating cell motility) (Li et al., 2009). RECK also regulates proteases and suppresses growth, invasion and metastasis of tumor cells (Takahashi et al., 1998; Oh et al., 2001; Morioka et al., 2009).

Recent reports indicate that miR-21 can be upregulated by two transcription factors, AP-1 and ETS1 (Fujita et al., 2008; Huang et al., 2009). This is consistent with our findings (Supplementary Figure 4) as both factors are known to be activated by the RAS/ERK-signaling pathway (Karin, 1996; Paumelle et al., 2002). We and Hsu et al. (2006) have previously found that RECK is repressed by RAS or HER2 at the transcription level (Sasahara et al., 1999). Thus, RECK expression is regulated by these oncogenes at multiple levels (that is, transcription and translation). Interestingly, miR-21 is also upregulated by transforming growth factor-β signaling (Davis et al., 2008), a potent inducer of EMT (see below).

miR-372/373 are known to help testicular tumor cells overcome senescence by targeting LATS2 (Voorhoeve et al., 2006), to promote tumor metastasis by targeting CD44 (Huang et al., 2008; Mani et al., 2008) and to be induced by hypoxia in a HIF1α-dependent manner to target a nucleotide excision repair gene, RAD23B (Crosby et al., 2009). Our findings show that TWIST1 is a mediator of HIF1α-induced miR-372/373 expression and that RECK is an important target of this signaling pathway (see Figure 7).

This study showed the convergence of two signaling pathways (that is, one activated by growth factors and the other by hypoxia) on RECK through miRNAs, leading to cooperative downregulation of this tumor suppressor (Figure 7). Intriguingly, both pathways are known to induce EMT, a phenomenon closely associated with tumor invasion and metastasis (Polyak and Weinberg, 2009). Recent papers also point to a link between EMT and cancer stem cells (Mani et al., 2008; Hill et al., 2009; Polyak and Weinberg, 2009). Reck has been implicated in angiogenesis (Oh et al., 2001), directional cell migration (Morioka et al., 2009) and suppression of neuronal differentiation (Muraguchi et al., 2007). Thus, the roles of RECK in hypoxia response, EMT and cancer stem cells are interesting subjects to be explored in future studies.

miR-15a and miR-16-1 were originally identified as tumor-suppressor miRNAs that are frequently inactivated in chronic lymphocytic leukemia and suppress the expression of the BCL2 oncoprotein (Cimmino et al., 2005). A recent study indicates the involvement of p53 in the maturation of several miRNAs, including miR-16-1 (Suzuki et al., 2009). Hence, it seems counterintuitive that these miRNAs target RECK, a tumor suppressor. In this study, moderate expression of miR-15/16 was detectable in all the cancer-derived cell lines as well as the non-malignant cells we tested (Figure 2). The roles of miR-15/16 in solid tissues, the stimuli that affect miR-15/16 expression and the reason why RECK has to be regulated by miR-15/16 are among the interesting questions to be addressed in future studies.

Our findings may have several clinical implications. First, for tumors expressing RECK mRNA, the miRNA antagonists may be effective in activating RECK protein synthesis. Second, for tumors without RECK expression, gene therapy using a RECK cDNA lacking 3′UTR may be effective. Third, when certain chemicals activating the endogenous RECK gene become available, they may be more effective when combined with the miRNA antagonists. Finally, when effective therapies for a given solid tumor become available, RECK may serve as a useful molecular marker to predict the aggressiveness of the tumors as well as the effectiveness of such ‘RECK-activation’ therapies.

In conclusion, we found that RECK is a target of at least three groups of miRNAs (miR-15b/16, miR-21 and miR-372/373) and that miR-372/373 are upregulated in response to hypoxia through HIF1α and TWIST1. These and other data indicate that the hypoxia- and RAS-signaling pathways converge on RECK through miRNAs, cooperatively downregulating this tumor suppressor. These findings not only yield important insights into the mechanisms of regulation of RECK expression, but also provide some hints toward effective applications of RECK in cancer therapy.

Materials and methods

RNA analyses

In RNA blot hybridization, total RNA separated by urea/polyacrylamide gel electrophoresis and transferred onto a Genescreen Plus membrane (Perkin-Elmer, Waltham, MA, USA) was probed with antisense oligo-deoxynucleotides end-labeled with T4 polynucleotide kinase and γ-32P-ATP. The hybridization and washing conditions were as described by Lau et al. (2001). Quantitative reverse transcription–PCR (qRT–PCR) for miRNAs was performed using a Mx3000P Real-Time PCR System (Stratagene, La Jolla, CA, USA) using the Ncode SYBR Green miRNA qRT–PCR kit (Invitrogen, Carlsbad, CA, USA). Primer sequences are described in the Supplementary Information.

Plasmid construction

The human RECK 3′-UTR was inserted between the XbaI and NotI sites of the pRL-TK vector (Promega, Madison, WI, USA). Mutations were introduced using the Quick Change site-directed mutagenesis kit (Stratagene) and were confirmed by sequencing. Double and triple mutants were generated by exchanging appropriate fragments from these single mutants. For bioassays, pCXN2 was used to express the RECK-coding sequence linked to the mutant 3′-UTR. The enzymes used are listed in the Supplementary Information.

Transfection and cell culture techniques

Transfection of miRNA precursors (Dharmacon, Lafayette, CO, USA; final concentration, 50 nM) or LNA nucleotides (Exiqon, Vedbaek, Denmark; 40 nM) was performed using the RNAiMax transfection reagent (Invitrogen). Hypoxia response was induced by exposing cells to 100 μM deferoxamine (Sigma, St Louis, MO, USA) or a mixture of gases (1% O2, 5% CO2 and 94% N2 for 24 h). Matrigel invasion assays and gelatin zymography were performed as described previously (Takahashi et al., 1998). For luciferase-reporter assay, the pRL reporter plasmid was co-transfected with pGL3 control vector (Promega, internal control) into the cells using the CalPhos Mammalian Transfection kit (BD Biosciences, San Jose, CA, USA). Luciferase activity was quantified using the Dual Luciferase Assay kit (Promega). Clone isolation from metastatic nodules is described in the Supplementary Information.

Immunological detections

For immunoblot assay, proteins separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis were transferred onto a polyvinylidene difluoride membrane (Millipore, Bedford, MA, USA). The primary antibodies used are as follows: RECK (5B11D12), α-tubulin (Santa Cruz, Santa Cruz, CA, USA, sc-5286), E-cadherin (BD Biosciences, 610181), vimentin (Sigma, V5255), TWIST1 (Santa Cruz, H-81, sc-15393) and HIF1α (BD Biosciences, 610958). For immunohistochemistry, tissue slices were prepared and stained as described previously (Yang et al., 2008). The primary antibodies used are as follows: VEGFR2 (Chemicon, Bedford, MA, USA), CD31 (Chemicon) and TWIST1 (Santa Cruz, N-19, sc-6070). Further details are provided in the Supplementary Information.

Bioluminescence imaging

A cell suspension was inoculated into the rectal sub-mucosa of female BALB/c-nu/nu mice as described by Kawada et al. (2007). Growth of primary tumors was monitored by bioluminiscense after intraperitoneal injection of d-luciferin using the Xenogen IVIS system. The values were normalized to the values obtained on day 0 using the same animal (that is, actual inoculum size). Further details are provided in the Supplementary Information.

Chromatin inmunoprecipitation

We used a Chromatin Immunoprecipitation kit (Upstate, Bedford, MA, USA) with 1 μg of anti-Twist antibody (Santa Cruz, sc-15393) or normal rabbit IgG (Santa Cruz, sc-2027) as negative control.

Statistical analyses

Significance of difference between data sets was evaluated using unpaired Student's t-test. For multiple linear regression analysis, we used relative levels of the studied miRNAs as independent variables and the relative level of RECK protein as a dependent variable. The analysis was performed using the software R.