The prognostic significance of RUNX2 and miR-10a/10b and their inter-relationship in breast cancer

Ninety five of the 108 breast cancer patients examined in this study had clinicopathologically confirmed primary ductal carcinoma of the breast, and the remaining 13 patients had non-ductal carcinoma of the breast. All of them were diagnosed at the Tri-Service General Hospital, Taipei, Taiwan between October 1994 and February 2013. Patients’ clinical information, including cancer stage, tumor grade, estrogen receptor (ER), progesterone receptor (PR), HER-2/neu, recurrence and survival status, were also noted. Recurrent breast tumors were subjected to pathological confirmation to exclude the possibility of second primary tumors. Moreover, the cause of death was verified from death certificate; patients whose death was clearly documented to be due to breast cancer were considered to have died of breast cancer, whereas other causes of deaths were considered censored events. Tumor and adjacent non-tumor breast tissues from patients were obtained at the time of initial surgery, and were fully encoded to protect patient confidentiality. Clinical specimens were utilized under a protocol approved by the Institutional Review Board of the Human Subjects Research Ethics Committee of Academia Sinica and Tri-Service General Hospital, Taipei, Taiwan.

MCF7 breast cancer cell line was cultured in Modified Eagle Medium (MEM) supplemented with 10% fetal bovine serum (FBS). MDA-MB-231 and MB157 breast cancer cell lines, U2OS osteosarcoma cell line and 293 T embryonic kidney were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% FBS. BT483 breast cancer cell line was cultured in Roswell Park Memorial Institute medium (RPMI) supplemented with 10% FBS. All cell lines were obtained from American Type Culture Collection (ATCC, Manassas, VA, USA).

Total RNA from the osteosarcoma cell line U2OS was extracted using Trizol reagent (Invitrogen, Carlsbad, CA, USA), and 1 μg of RNA was reverse transcribed using the SuperScript III RT–PCR kit (Invitrogen). For cDNA cloning, full-length RUNX2wt was amplified from cDNA of the U2OS cell line using the PCR primer pair (RUNX2wt-forward: 5′-GCTAGCATGGACTACAAAGACGATGACGACAAGGTGATGCGTATTCCCGTAGATCCGAG-3′) and (RUNX2wt-reverse:5′-CTCGAGATATGGTCGCCAAACAGATTCATCC-3′) in an amplification mixture containing 1X Pfu polymerase buffer (Stratagene, La Jolla, CA, USA), 100 μM 4dNTPs, 1.5 mM MgCl₂ and 1U Pfu polymerase (Stratagene). The samples were amplified for 35 cycles of 95°C for 1 min 30 sec, 60°C for 35 sec, and 72°C for 2 min each, preceded by a 5 min denaturation at 95°C and followed by a 10 min extension at 72°C. The amplified cDNAs of RUNX2wt were then cloned into pcDNA3.1 (+) vector (Invitrogen) by ligation of the NheI – XhoI fragments to generate the pcDNA-RUNX2 expression constructs. The sequence of pcDNA-RUNX2 was confirmed by DNA sequencing.

RUNX2 siRNA (Cat. # 1027020) and siRNA control (Cat. # 1027280) were guaranteed products by manufacturers and purchased from Qiagen (Valencia, CA, USA). Precursor miRNA oligos (pre-miR control # 1: Cat. # AM17110, pre-has-miR-10a: Cat. # PM10787 and pre-has-miR-10b: Cat. # PM11108) and anti-miRNA oligos (anti-miR control # 1: Cat. # AM17010, anti-has-miR-10a: Cat. # AM10787 and anti-has-miR-10b-5p: Cat. # AM11108) were guaranteed products by manufacturers and purchased from Ambion (Austin, TX, USA).

For plasmid transfection, 4 g of pcDNA-RUNX2 construct plasmid per well with 4 μl Lipofectamine 2000 (Invitrogen) was transfected into cell lines. For siRNA transfection, 25 nM RUNX2 siRNA per well with 2 μl Lipofectamine RNAiMax (Invitrogen) was transfected into cell lines. For miRNA oligos transfection, 25 nM miRNA oligos per well with 2 μl RNAiMax were transfected into cell lines. For anti-miRNA oligo transfection, 15 nM anti-miRNA oligos per well with 2 μl RNAiMax were transfected into cell lines. The experiments were repeated three times using six-well plates.

Total RNAs were extracted from tumor and non-tumor breast tissues for each patient or breast cancer cell lines using Trizol reagent according to the manufacturer’s specifications. 4 ng of total RNAs were used for quantification of miRNAs expression, including miRNA-10a/miR-10b and RNU6B (U6) RNA, by TaqMan RT-qPCR kit (miRNA-10a: Cat. # 000387, miRNA-10b: Cat. # 002218, and RNU6B: Cat. # 001093; Applied Biosystems, Foster City, CA, USA) according to the manufacturer’s instructions. For the quantification of mRNA of RUNX2 and GAPDH, 1 g of total RNAs was reverse transcribed into cDNA using the SuperScript III kit and the specific mRNAs were detected by Power SYBR Green PCR Master Mix (Applied Biosystems) according to the manufacturer’s instructions. The primer sequences of RUNX2 and GAPDH qPCR were: RUNX2 5′-AGCAAGGTTCAACGATCTGAGATT-3′ (forward) and 5′-AAGACGGTTATGGTCAAGGTGAAA-3′ (reverse); GAPDH 5′-CTGCTCCTCCTGTT CGACAGT-3′ (forward) and 5′-ACCTTCCCCATGGTGTCTGA-3′ (reverse). The gene expression was normalized based on threshold cycle (Ct) values of common internal control for miRNA quantification assays, U6 RNA, and RUNX2 quantification assays, GAPDH. The level of RUNX2 and miR-10a/b expression were measured using the 2^-△△Ct method [27]. The results were presented as fold change of each gene in the tumor tissue relative to the non-tumor tissue. The RT-qPCR was performed on the 7300 Real-Time PCR System (Applied Biosystems). The procedures were repeated three times for all samples, and mean values were obtained for statistical analysis.

Cell lysates were extracted in 0.1 M NaCl, 0.01 M TriszHCl (pH 7.6), 1 MM EDTA (pH 8.0), 1 Mg/ml aprotinin, and 100 mg/ml PMSF, while protein concentration was determined by a Bio-Rad protein assay. Protein (50 μg) was boiled at 95°C in sodium dodecyl sulfate (SDS) sample buffer for 5 min, electrophoresed on 7–12% SDS-PAGE gels, and transferred to polyvinyldifluoridine membranes. These were then incubated overnight at 4°C with (i) rabbit monoclonal antibody against human RUNX2 (Cell Signaling Technology, Danvers, MA, USA; #8486; 1:250), (ii) goat polyclonal antibody against human HOXA1 (R&D Systems Inc., Minneapolis, MN, USA; AF5014; 1:500), (iii) goat polyclonal antibody against human HOXD10 (Santa Cruz Biotechnology, Dallas, TX, USA; sc-33005; 1:500), or (vi) rabbit polyclonal antibody against human Actin (Santa Cruz Biotechnology; sc-1615-R; 1:3000). Membranes were washed three times for 15 min with phosphate buffered saline (PBS) containing 0.1% Tween20 (PBST), incubated with anti-goat or anti-rabbit secondary antibody (both Santa Cruz Biotechnology; 1:5000) at room temperature for 60 min, and then washed three times for 15 min with PBST. The signals of protein bands were visualized using ECF western blotting kit (Amersham Biosciences, Piscataway, NJ, USA) and detected by Typhoon 9400 imager (Amersham Biosciences).

Cell migration and invasion were assayed using a transwell chamber (Millipore, Temecula, CA, USA) with or without Matrigel (BD, Franklin Lakes, NJ, USA). 1 Ml serum free medium was added into a 24-well plate, and a transwell chamber was placed into the well in both assays. The chamber was coated with 50 μl Matrigel and incubated for 2 hours at 37°C for the invasion assay. After transfection for 72 hours, in migration assay: cells were trypsinized and seeded into chambers at the density of 1 × 10⁴ cells for MDA-MB-231 and 3 × 10⁴ cells for MCF7 per well; and in invasion assay: cells were 2 × 10⁴ cells for MDA-MB-231 and 6 × 10⁴ cells for MCF7. Eight hours later for migration and 24 hours later for invasion in MDA-MB-231, and 12 hours later for migration and 32 hours later for invasion in MCF7, migrated cells remaining in the basal chamber were fixed and stained with 10% ethanol in 1% crystal violet for 30 min and washed three times with PBS. Non-migrated cells were removed by cotton swabs. Images of the stained cells were obtained under a phase-contrast microscope (Olympus, Tokyo, Japan). To quantify the number of invading cells, crystal violet-stained cells in visual fields were counted. The experiments were repeated three times and mean values were obtained for statistical analysis. Detailed materials and methods can be found in Additional file 1.

It has been shown that RUNX2 plays an important role in cell migration and invasion in breast cancer and higher expression is associated with poor outcome [6],[8],[14]. Meanwhile, others have reported the contribution of miR-10a/b to cell migration and invasion in breast cancer and miR-10b overexpression correlates with poor prognosis [23]-[26]. However, the inter-relationship between RUNX2 and miR-10a/b and their joint impact on prognosis in breast cancer has yet to be investigated. We examined one hundred and eight patients with histologically proven breast cancers to delineate the relationship between the expression of RUNX2 and miR-10a/b, and clinical outcomes. Their clinical and pathological features were summarized in Table 1. The mean age was 59.2 × 13.2 years and the median follow up for relapsed patients and overall survival was 54.6 and 63.2 months, respectively. The distribution of cancer stages was stage 0 (2.80%), stage I (31.78%), stage II (42.99%) and stage III (22.43%), respectively, and of the molecular subtypes was Luminal A (ER + or PR+, Her2-) (33.33%), Luminal B (ER + or PR+, Her2+) (35.19%), Her2 positive (ER-, PR-, Her2+) (15.74%) and triple negative (ER-, PR-, Her2-) (15.74%), respectively. These clinical and pathological characteristics of the breast cancer subjects across the strata of age, hormone receptor status, and Her2, were similar to those reported in other breast cancer clinics in Taiwan [28]-[30].

The expression levels of RUNX2 and miR-10a/b were determined in tumor tissues and adjacent non-tumor tissues by qPCR. As shown in Table 2, the median expression levels of RUNX2 and miR-10b were significantly higher in tumor tissue than in non-tumor tissues by 2.16 and 1.73 fold, respectively (p < 0.0001, p < 0.0001). There is also a trend for higher expression of miR-10a in tumor tissues than in non-tumor part, but the difference did not reach statistical significance (p = 0.1318). Patients with ER- or triple negative subtypes displayed higher-level expression of these three genes than those with ER + or non-triple negative subtypes. Analysis of clinical outcome revealed that patients who relapsed or died of breast cancer showed higher expression of these three genes than non-relapse or alive patients. On the other hand, no significant association was found between the expression levels of these three genes and gender, histology, stage, regional lymph node involvement or status of PR and Her2, although there was a trend for higher expression in patients with higher histology grade (Table 2; Additional file 1: Table S1).

To evaluate the prognostic significance of expression levels of RUNX2 and miR-10a/b, we used the receiver operating characteristic (ROC) curves based on the expression levels of RUNX2 and miR-10a/b in individual tumor/non-tumor ratios to determine the cutoff values defining “overexpression” of the three genes. According to these cutoff values, the areas under the curves (AUC) values for individual or various combinations of these three genes were found to be significantly lower in non-relapsed patients than in relapsed patients. However, when comparing patients with fatal outcome to those alive, only AUC values for miR-10b and combination of these three genes were significantly lower in alive patients than in those who died (Additional file 1: Table S2). Accordingly, we used the cutoff values to define the status of “overexpression” in subsequent logistic and COX regression analyses to ascertain correlations between gene expression and prognostic features. Under these cutoff values, the frequencies of overexpression of RUNX2 and miR-10a/b genes were significantly higher in patients who relapsed, and that of miR-10b overexpression was higher in patients who died (all p < 0.05). Using the same dichotomized values for logistic regression analysis, the odds ratios (ORs) for relapse were higher in patients with increased mRNA or miRNA levels of these 3 individual genes, while the OR for death was only associated with miR-10b overexpression (all p < 0.05). Patients with no overexpression of these three genes served as the reference group. In terms of the joint effects of overexpression of all three genes on prognosis, the ORs were respectively 20.32 (95% confidence interval (CI) = 1.50 ~ 179.92) and 17.05 (95% CI = 1.76 ~ 165.01) for relapse and death (P < 0.05). There was a trend toward an incremental additive effect on significantly elevated risk of relapse (OR = 2.66; 95% CI = 1.50 ~ 4.75), death (OR = 2.90; 95% CI = 1.22 ~ 6.89) and regional lymph node involvement (OR = 1.51: 95% CI = 1.03 ~ 2.21), respectively, as patients carried a greater number of overexpressed genes (P-trend < 0.05; Table 3; Additional file 1: Table S3).

The correlations between overexpression of RUNX2 and miR-10a/b, individually or jointly, and clinical outcomes were examined by the Kaplan–Meier method of patients. Both RFS (Figure 1) and OS (Additional file 1: Figure S1) were significantly lower in patients with overexpression of RUNX2 or miR-10b individually (Figure 1A, C; Additional file 1: Figure S1 A, C). Compared to the patients with no overexpression of these three genes, those with a greater number of overexpressed genes (1, 2, or 3) in tumors showed lower rates of RFS (log-rank p < 0.0001) and OS (log-rank p = 0.0002) (Figure 1D and Additional file 1: Figure S1D). Multivariate Cox regression analysis showed a significant correlation between higher hazard ratios (HRs) for relapse and greater gene signals for all three genes individually and a correlation between higher HR for death and expression of miR-10b (Table 4; Additional file 1: Table S4). As to the joint effects of overexpression of all three genes compared with no overexpression of any of them on prognosis, the HRs were 12.37 (95% CI = 1.62 ~ 94.55) and 14.59 (95% CI = 1.66 ~ 128.24) for relapse and death, respectively (P = 0.015, 0.016). The multiple gene additive model displayed a statistically significant HR of 2.25 for RFS (95% CI = 1.44 ~ 3.52, P-trend = 0.0004) and 2.88 for OS (95% CI = 1.26 ~ 6.61, P-trend = 0.0125).

In view of our findings of significant correlations among RUNX2 or miR-10a/b and relapse or death, we next investigated the inter-relationship of these 3 genes in breast cancer cell lines in vitro. Using qPCR analysis, we determined the expression levels of RUNX2 and miR-10a/b in 4 breast cancer cell lines. As shown in Figure 2A, MDA-MB-231, the invasive estrogen receptor negative (ER-) cell line, showed the highest expression levels of all these three molecules. Since MDA-MB-231 cells expressed very little Twist [31],[32], which is reported to upregulate miR-10b [26] and since RUNX2 is involved in transcriptional activities of Twist [33], we explored the possible regulation of miR-10a/b by RUNX2. Silencing of RUNX2 by siRNA transfection of MDA-MB-231 cell line to 24% (p < 0.0001) of control resulted in a significant decrease in expression of miR-10a and miR-10b to 52% (p = 0.0349) and 51% (p < 0.0001), respectively (Figure 2B). Consistent with this, HOXA1 and HOXD10 are putative targets for miR-10a/b in Garzon et al. 2006 [34], suppression or overexpression of miR-10a/b increased or decreased HOXA1/HOXD10 protein expression, respectively, in breast cancer cell lines explained in Additional file 1: Figure S2, escalated in the silenced RUNX2 cells by 3.86 and 2.33 fold, respectively (Figure 2C). In line with the increased HOXA1 and HOXD10, migration and invasion ability of MDA-MB-231 cells were significantly reduced to 21% (p = 0.0035) and 29% (p = 0.0029), respectively. (Figure 2D, E). Thus, suppression of RUNX2 reduced miR-10a/b expression, and thereby impairing breast cancer cell motility.

In order to clarify the relationship between RUNX2 and miR-10a/b, RUNX2 was overexpressed in high and low RUNX2 expressing breast cancer cell lines, MDA-MB-231 and MCF7 (Figure 3A, B). RUNX2 increased the transcription of miR-10a by 1.61 fold (p = 0.0079) and that of miR-10b by 1.60 fold (p < 0.0001) in MCF7 cells and increased only the expression of miR-10b by 1.52 fold (p = 0.0413) in MDA-MB-231 cells. Consistently, protein levels of HOXA1 and HOXD10 markedly declined to 0.15 and 0.09 fold, respectively, in RUNX2 overexpressing MCF7 cells, but only slightly decreased to 0.91 and 0.71 fold in MDA-MB-231 cells (Figure 3C). Functional assay showed that cell migration and invasion were significantly enhanced in RUNX2 overexpressing MDA-MB-231 (1.74 fold, p = 0.0315 and 2.51 fold, p = 0.0016, for migration and invasion, respectively) and MCF7 cells (2.96 fold, p = 0.0143 and 4.95 fold, p = 0.0018, respectively) (Figure 3D, E, F, G). Thus, overexpression of RUNX2 upregulated miR-10a/b expression, thereby promoting breast cancer cell motility.

To sum up, these findings suggest that RUNX2 plays an important role in controlling breast cancer cell motility through regulation of miR-10a/b expression, and overexpression of these three genes has poor prognosis.