Total RNA was isolated from frozen tissue with Isogene (Nippon Gene, Tokyo, Japan). Short-strand RNA was purified from total RNA with the RNeasy MinElute Cleanup Kit (Qiagen, Hilden, Germany). For the oligonucleotide array we used Genopal-MICH07 DNA chips (Mitsubishi Rayon, Tokyo, Japan) comprising 188 oligonucleotide DNA probes. Details are described in Data S1.

Biotin 3′-labeled locked nucleic acid-incorporated miRNA probe (miR-CURY LNA detection probe; Exiqon, Woburn, MA, USA) was used for visualization of miR-148a. Details are described in Data S1.

Nine cell lines derived from human GC were used. All cell lines were maintained as described previously.(11) Details are described in Data S1.

The oligonucleotide array, Genopal (Mitsubishi Rayon), was prepared as described previously.(12) Details of oligonucleotide array probes, total RNA isolation, RNA quantification and assessment of integrity, hybridization, detection, and data analysis were carried out as described previously.(13)

Quantification of MMP7 mRNA levels was carried out using real-time fluorescence detection as described previously.(14) For analysis of miR-148a and U6B, expression levels of miR-148a and RNU6B were measured and quantified as described previously.(15) Western blot analysis was carried out as described previously.(16) Details are described in Data S1.

A 20-kb genomic DNA fragment encompassing the miRNA coding sequences was analyzed with CpG plot software (http://bioweb2.pasteur.fr). We used the parameter values to report a CpG island as described previously.(17) Thermal cycling conditions were as described previously.(18) Methylation status of PCR products were confirmed as described previously.(19) Details are described in Data S1.

For constitutive expression of the MMP7 gene, commercially available MMP7 cDNA construct (Invitrogen, Carlsbad, CA, USA) was purchased. pCMV-MIR (Origene, Rockville, MD, USA) was used for constitutive expression of miR-148a. Details are described in Data S1.

Transfection of cells was carried out with Lipofectamine RNAiMAX Reagent (Invitrogen) according to the manufacturer's instructions. Details are described in Data S1.

Reporter gene assay was carried out as described previously.(20) Details are described in Data S1.

A total of 81 primary tumor samples were collected from patients diagnosed with GC. Details are described in Data S1.

For the MTT assay, cells were seeded at a density of 2000 cells per well in 96-well plates. Cell growth was monitored after 1, 2, and 4 days by MTT assay.(21) Modified Boyden chamber assays were used to examine invasiveness as described previously.(20)

Statistical differences between miRNA levels in non-neoplastic gastric mucosa and GC were evaluated using the Mann–Whitney U-test and Wilcoxon matched pairs test. The correlation between expression levels of miR-148a and clinicopathologic parameters was analyzed with Fisher's exact test. Univariate and multivariate Cox regression was used to evaluate the associations between clinical covariates and cancer-specific mortality. Hazard ratios and 95% confidence intervals were estimated from Cox proportional hazard models. P-values < 0.05 were considered to be statistically significant.

To identify GC-related miRNA, we carried out miRNA microarray profiling to compare between 20 cancer tissue and five corresponding non-neoplastic gastric mucosa. Through the comparison between these two profiles, six miRNAs showed significantly higher expression in corresponding non-neoplastic gastric mucosa than in cancer tissue (Table 1). Among six miRNAs, we focused on miR-148a because its expression was most downregulated in GC tissue. Further analysis of the same GC tissues by qRT-PCR also showed a significant downregulation of miR-148a (Fig. S1).

To confirm the localization of miR-148a expression, in situ hybridization (ISH) was carried out using GC tissues. We could detect the miR-148a expression only in fundic glands and pyloric glands. However, neither cancer cells nor stromal tissues expressed miR-148a (Fig. 1a). Then, to further understand the relationship between miR-148a expression and clinicopathologic parameters, we examined the miR-148a expression levels in 61 formalin-fixed paraffin-embedded samples of primary GC and their corresponding non-neoplastic gastric mucosa by qRT-PCR. In this sample set, miR-148a expression levels were also significantly downregulated in GC tissue compared with corresponding non-neoplastic mucosa (Fig. S2). The expression level of miR-148a was evaluated using the tumor : normal ratio of miR-148a. The relationship between the miR-148a expression level and clinicopathologic parameters of GC is summarized in Table S1. We also evaluated the association between miR-148a expression levels and cancer-specific mortality. In all cases and in 41 advanced GC patients, the GC tissues with lower expression levels of miR-148a showed significantly worse prognosis than those with high expression (Fig. 1b,c). In order to further evaluate the association of miR-148a expression with cancer-specific mortality, we used both univariate and multivariate Cox proportional hazards analyses (Table 2). According to the univariate analysis, T grade, N grade, stage, and miR-148a expression were significantly associated with cancer-specific mortality. According to the multivariate model, expression of miR-148a was an independent prognostic classifier of cancer-specific mortality. These results indicate that loss of miR-148a may play important roles in GC progression.

To pursue the cause of miR-148a downregulation, we examined the methylation status in CpG islands of the MIR148a promoter region in GC cell lines. Expression levels of miR-148a measured by qRT-PCR are shown in Figure S3(a), and we carried out bisulfite sequencing of genomic DNAs from two GC cell lines with high miR-148a expression and two cell lines with low miR-148a expression. Results of bisulfite sequencing are shown in Figure S3(b). An inverse correlation was observed between frequency of methylation in MIR148a promoter-region CpG islands and expression levels of miR-148a. In addition, we ascertained that miR-148a expression was restored in five aza-dc-treated MKN-45 and MKN-74 cells (Fig. S3c). These results suggest that DNA methylation of CpG islands of MIR148A play an important role in transcriptional inactivation in GC cell lines.

To investigate the biological significance of miR-148a in GC, we carried out an MTT assay 4 days after altering miR-148a expression. In this experiment, we used MKN-1 cells transfected with miR-148a precursor and HSC-57 cells transfected with miR-148a inhibitor. MKN-1 and HSC-57 were selected as they possessed the lowest and highest endogenous miR-148a expression among nine GC cell lines, respectively. Successful overexpression and suppression of miR-148a in each treated cell line were confirmed by qRT-PCR (Fig. 2a). Cell growth of GC cells with deregulated miR-148a expression did not differ from that of cells transfected with control miRNA up to day 4 (data not shown). Next, a Transwell invasion assay was carried out. The invasiveness of miR-148a-overexpressing MKN-1 cells was reduced compared with the negative control miRNA-transfected MKN-1 cells (Fig. 2b). In contrast, the invasiveness of miR-148a-suppressed HSC-57 cells was greater than that of the negative control miRNA-transfected HSC-57 cells (Fig. 2c). These results are consistent with those of a previous report,(22) and indicate that miR-148a regulates some genes which in turn affect tumor invasion in GC cells.

Although some reports have identified the target gene of miR-148a, these studies used online databases to derive predicted targets.(22–25) To identify novel GC invasion-related genes whose expression was directly regulated by miR-148a, we analyzed gene expression profiles from four GC cell lines showing deregulation of miR-148a alongside corresponding non-treated cells, using a custom oligonucleotide array. In each cell line, either overexpression or suppression of miR-148a was induced, according to their endogenous miR-148a expression level. The list of genes with expression altered more than twofold in miR-148a-deregulated cells is summarized in Table 3. Among these candidates, MMP7 was markedly downregulated by miR-148a overexpression both in MKN-45 and MKN-74 cell lines. Although target prediction for miR-148a was carried out using online databases including TargetScan, PicTar, and miRanda, these databases did not list MMP7 as a candidate miR148a target gene (data not shown). Therefore, further study of the correlation between miR-148a and MMP7 was undertaken.

To confirm the regulation of MMP7 by miR-148a, MKN-45 and MKN-74 cells were transfected to induce miR-148a overexpression, then examined for MMP7 expression level by Western blotting. In both cell lines, MMP7 expression was significantly downregulated; levels were similar to that of MMP7-specific siRNA treated cells (Fig. 3a). The MMP7 mRNA levels determined by qRT-PCR were consistent with the MMP7 protein levels determined by Western blotting (Fig. 3b). It was also found that upregulation of MMP7 was inversely associated with miR-148a expression in GC cases analyzed in Figure 1(b) (Table S2). In addition, we examined the effect of miR-148a deregulation on the expression of epithelial–mesenchymal transition-related molecules. However, no significant alterations of these molecules were detected in miR-148a deregulated cells (Fig. S4a). As shown in Figure 2(b), overexpression of miR-148a repressed invasion of MKN-1 cells. However, MMP7 expression was not detected in MKN-1 cells (data not shown). It was reported that ROCK1 is a direct target of miR-148a and could be involved in GC invasion.(22) We checked expression of ROCK1 in MKN-1 and the alteration of its expression in miR-148a-deregulated cells. Actually, we detected that expression of ROCK1 was downregulated by miR-148a overexpression in MKN-1 cells (Fig. S4b). Furthermore, to prove that MMP7 is a direct target of miR-148a, we used a 3′-UTR sequence of MMP7 cloned into a reporter vector downstream of the luciferase complementary DNA (Fig. 3c). Transfection of this construct into three GC cell lines with high endogenous miR-148a expression levels led to suppression of MMP7 reporter activity, and mutation of the miR-148a binding site abolished the inhibitory effect of miR-148a on reporter activity (Fig. 3d). Conversely, transfection of this construct into three GC cell lines with low endogenous miR-148a expression levels did not lead to suppression of MMP7 reporter activity. However, cotransfection of this construct with miR-148a precursor into these cells showed that miR-148a suppressed the MMP7 reporter activity (Fig. S5). These results imply that miR-148a downregulates MMP7 expression by directly targeting its 3′-UTR.

Several reports show that miR-148a regulates cancer cell invasion through suppressing its target genes.(22,23) To confirm the cellular level at which MMP7 contributes to GC cell invasion, the MKN-45 cell line was stably transfected with vector expressing miR-148a alone, and cotransfected with vectors expressing miR-148a and MMP7. MKN-45 cells were selected because they express high levels of MMP7 and low levels of miR-148a. In MKN-45-miR-148a, which stably expressed miR-148a, MMP7 expression was significantly suppressed, and in MKN-45-miR-148a/-MMP7, which stably expressed miR-148a and MMP7, the expression of MMP7 was 50% less than that of empty vector-transfected MKN-45 cells (Fig. 3e). Next, the invasive properties of these cells were examined by Transwell invasion assay. As shown in Figure 3(f), the invasiveness of MKN-45-miR-148a was approximately 50% less than that of MKN-45-empty vector at all endpoints. However, the invasiveness of MKN-45-miR-148a/-MMP7 was only approximately 10–20% less than that of the MKN-45-empty vector. These observations suggest that MMP7 plays a central role in GC invasion among miR-148a target genes.