The specimens of 113 patients with primary CRC, who had undergone surgical resection in Yamaguchi University Medical Hospital (Yamaguchi, Japan) between January 2008 and December 2013, were used in the present study. All patients had undergone resection of the primary tumor, and among the 31 patients with distant metastases, three patients had undergone hepatectomy. None of the patients had received preoperative treatments, for example, chemotherapy and/or radiation.

For screening in the miRNA array analysis, the frozen specimens of primary CRC from 12 patients were used. For RT-qPCR analysis, the formalin-fixed paraffin embedded tissue (FFPE) specimens of 101 cases of primary CRC were used. Of these 101 CRC FFPE samples, 40 samples (20 CRCwLM and 20 CRCwoLM) were used for reverse transcription-quantitative polymerase chain reaction (RT-qPCR) validation of the array analysis (Fig. 1). Survival analysis was performed for all 101 of the cases from which samples were used for RT-qPCR analysis. All samples were obtained with informed consent from the patients. The study protocol was approved by the Institutional Review Board for the Use of Human Subjects at the Yamaguchi University School of Medicine (H17-83).

For the miRNA array analysis, frozen cancer tissue sections were immediately cut into 5-mm cubes and embedded in Tissue-Tek OCT compound medium (Sakura Finetech, Co., Ltd., Tokyo, Japan) following resection. The sections then were fixed in liquid nitrogen and stored at −80°C. Frozen specimens were cut into 10-µm-thick slices using a cryostat, and these sections were mounted onto foil-coated glass membrane slides (Leica Microsystems GmbH, Wezlar, Germany). The tissue sections were stained with toluidine blue prior to air drying. The FFPE specimens were cut into 10-µm-thick sections using a microtome, and these sections were mounted onto foil-coated glass membrane slides. The tissue sections were fixed in 70% ethanol for 30 sec and stained with hematoxylin and eosin prior to dehydration (5 min each in 70, 95 and 100% ethanol). The stained frozen or FFPE sections were microdissected using an LMD7000 Laser Microdissection system (Leica Microsystems GmbH). The cancer cells and cancer stromal tissues were visualized under a bright-field microscope (magnification, ×100) and selectively separated by activation of the laser. From each slide, 1×10⁷−2×10⁷ µm² epithelial or stromal cells were captured.

Total RNA was extracted from dissected tissue using the miRNeasy FFPE kit (Qiagen GmbH, Hilden, Germany) according to the manufacturer's protocol. The extracted total RNA (250 ng) was labeled with Cy5 using the 3D-Gene miRNA labeling kit (Toray Industries, Inc., Kanagawa, Japan). The labeled RNAs were hybridized onto 3D-Gene Human miRNA Oligo chips containing 2,555 miRNAs (Human_miRNA_Ver20; Toray Industries, Inc.). The annotation and oligonucleotide sequences of the probes conformed to the miRBase miRNA database (http://microrna.sanger.ac.uk/sequences/). Following stringent washes (1st wash: 0.5X saline-sodium citrate (SSC)/0.1% sodium dodecyl sulfate (SDS) solution; 2nd wash: 0.2X SSC/0.1%SDS solution; 3rd wash: 0.05X SSC solution), fluorescent signals were scanned with the 3D-Gene Scanner (Toray Industries, Inc.) and analyzed using 3D-Gene Extraction software (Ver 2.0.0.7; Toray Industries, Inc.). The raw data of each spot were normalized by substitution with a mean intensity of the background signal determined by all blank spots' signal intensities of 95% confidence intervals. Measurements of spots with signal intensities differing from the background signal intensity by greater than two standard deviations (SDs) were considered to be valid. The relative expression level of a given miRNA was calculated by comparing the signal intensities of the valid spots throughout the microarray experiments. The normalized data were globally normalized per array, such that the median of the signal intensity was adjusted to 100.

miR-221, miR-222, miR-659, miR-4470, miR-4669, miR-5703 and RNU6B (internal control)-specific cDNA synthesis was performed using 10 ng of total RNA and TaqMan miRNA primers (Applied Biosystems; Thermo Fisher Scientific, Inc., Waltham, MA, USA). RT-qPCR analyses were performed using the following TaqMan miRNA assays: miR-221-3p (ID000524), miR-222-3p (ID002276), miR-659 (ID001514), miR-4470 (ID464583_mat), miR-4669 (ID464125_mat), miR5703 (ID472811_mat) and RNU6B (ID001093) with specific primers (Applied Biosystems; Thermo Fisher Scientific, Inc.) as previously described (14). In brief, cDNA was synthesized from total RNA (10 ng) in a 7.5-µl reaction volume (RT primer 1.5 µl, total RNA lysate 2.5 µl, RT mix 3.5 µl) using the TaqMan^® MicroRNA Reverse Transcription kit. The reactions were performed at 16°C for 30 min and then at 42°C for 30 min, and were inactivated at 85°C for 5 min. RT-qPCR analysis was performed using the LightCycler^®480 System II (Roche Diagnostics, Tokyo, Japan). The reactions were performed for 10 min at 95°C, and then for 55 cycles with denaturation for 15 sec at 95°C and annealing/extension for 60 sec at 60°C. The relative expression of miRNA to RNU6B RNA was calculated using the 2^−ΔΔCq method (15). The CRC samples were categorized into miR-221 and miR-222 weak and strong expression groups according to the expression levels of miR-221 and miR-222 as determined by RT-qPCR analysis, using RNU6B as the control. High expression was defined as an expression level above the median value of the 101 CRC samples for each miRNA.

Of the 101 CRC FFPE samples, 20 samples, (CRCwLM, n=12; CRCwoLM, n=8) were used for ISH analysis. The locked nucleic acid (LNA)-ISH system was used to investigate the localization of miR-221 and miR-222. LNA-ISH was performed according to the manufacturer's protocol and relevant LNA-ISH literature (16). Briefly, the FFPE advanced colon cancer tissues were cut into 5-µm-thick sections and deparaffinized. Following treatment with proteinase K and post-fixation with 4% paraformaldehyde (Wako Pure Chemical Industries, Osaka, Japan), the slides were hybridized with the 5′ (digoxigenin-UTP) DIG-labeled miRCURY LNA™ detection probe, miR-221, and miR-222 (18115-01 and 38499–01; Qiagen GmbH), for 8 h at 52°C using the Ventana Discovery Ultra instrument (Ventana Medical Systems, Inc. Tucson, AZ, USA). The digoxigenin was detected with a polyclonal anti-DIG antibody and an alkaline phosphatase-conjugated secondary antibody (Ventana Medical Systems, Inc.), using nitro blue tetrazolium and 5-bromo-4-chloro-3-indolyl phosphate as the substrate. The LNA U6 snRNA probe was used as a positive control for every specimen. The 5′ DIG-labeled miRCURY LNA™ detection probe for human mature miR-221-3p (5′-GAAAGCCAGCAGACAATGTAGCT-3′), miR-222-3p (5′-ACCCAGTAGCCAGATGTAGCT-3′), U6-positive control, and scrambled negative control were all purchased from Exiqon A/S (Vedbaek, Denmark). Signals from the tumor cells and stromal cells were classified as negative (−) or positive (+), respectively.

All calculations were performed using the SPSS software package version 19.0 (SPSS, Inc., Chicago, IL, USA). Differences between groups were estimated using the Mann-Whitney U test. Categorical variables were compared using Fisher's exact test. Patient outcome survival curves were calculated using the Kaplan-Meier method. Factors shown to be of prognostic significance in univariate models were evaluated in a multivariate Cox proportional hazard model. To validate the association between survival rates and expression of the candidate miRNAs, the following approach was used to divide the patients into two groups according to relative expression levels of candidate miRNAs. For miRNAs, the median values of the expression data were used as the cut-off. Survival was defined as the interval from the date of diagnosis until the date of CRC-associated mortality, date of mortality from other cause, or the end of follow-up (June 30, 2017), whichever came first. Patients lost to follow-up were censored at the date of the final follow-up contact. All statistical analyses assumed a two-sided alternative with a 5% level of significance.

An overview of the experimental design and selection of candidate miRNAs is shown in Fig. 1. To investigate the miRNAs associated with liver metastasis in cancer stromal tissue, array-based miRNA profiling was performed. The stromal tissue of six cases of primary CRCwoLM and that of six cases of primary CRCwLM were profiled on the Toray 3D-Gene Human miRNA Oligo chips ver. 19. The clinicopathological data of the patients are summarized in Table I. miRNAs with intensity levels higher than the negative control +2 SD were selected in all cases. This filtering resulted in the identification of 1,024 miRNAs among a total of 2,555. The Fischer ratio was used to evaluate the potential of the selected miRNAs to discriminate between CRCwLM and CRCwoLM (17). The 1,024 miRNAs were ranked in order of decreasing Fischer ratio. The 27 miRNAs that were significantly dysregulated in cancer stromal tissues, comparing between CRCwoLM and CRCwLM, are listed in Table II (Fischer ratio >4). For the selection of candidate miRNAs among these 27 miRNAs, to permit comparison between CRCwLM and CRCwoLM, a fold change of 2.0 or 0.5 was used as the cut-off. Using this cut-off led to the identification of four miRNAs (miR-659, −4470, −4669 and −5703) that were downregulated and two miRNAs (miR-221 and miR-222) that were upregulated in CRCwLM compared with CRCwoLM.

To confirm the microarray findings, RT-qPCR analysis was used to measure the expression level of these six miRNAs, described as above, in the stromal tissues of 20 cases CRCwLM and 20 cases of CRCwoLM. The expression levels of miR-221 and miR-222 in the cancer stromal tissues were significantly higher in CRCwLM than in CRCwoLM (P=0.003 and P=0.02, respectively), whereas no significant difference was found in the expression of the four downregulated miRNAs (miR-659, −4470, −4669 and −5703) between CRCwLM and CRCwoLM (Fig. 2).

To evaluate whether the expression of miR-221 or miR-222 was associated with clinicopathological factors and prognosis, RT-qPCR analysis was used to analyze the expression of miR-221 and miR-222 in 101 CRC FFPE sample sets of cancer stromal tissues and corresponding cancer cells; RNU6B was used as the control. The clinicopathological analysis showed that the group with a high expression of miR-221 in cancer stromal tissues, that is, cases with specimens in which the expression of miR-221 was above the median value, had more advanced venous invasion (P=0.023), liver metastasis (P=0.028), and distant metastasis (P=0.003) than the low-expression group (Table III). The group with a high expression of miR-222 in cancer stromal tissues had more advanced depth of tumor invasion (P=0.045), liver metastasis (P=0.009), and distant metastasis (P=0.003) than the low-expression group (Table IV). By contrast, unlike in the stromal tissue, the clinicopathological factors and the expression of miR-221 and miR-222 in cancer cells only showed a tendency to be associated with distant metastasis (Tables III and IV).

The overall survival rate of the patients according to the miR-221 and miR-222 expression status of the cancer stroma are shown in Fig. 3A and B. The survival analysis showed that the group with a high expression of miR-221 in cancer stromal tissues had significantly shorter overall survival rates than the group with a low expression (P=0.039). Similarly, the group with a high expression of miR-222 in cancer stromal tissues had significantly shorter overall survival rate than the group with a low expression (P=0.02). However, there was no association between the overall survival rate and the expression level of miR-221 or miR-222 in cancer cells (Fig. 3C and D). In addition, a Cox proportional hazard regression model was used to determine whether the expression of miR-221 and miR-222 in the cancer stroma was an independent risk factor for overall survival rate (Table V). The univariate analysis revealed that high levels of stromal miR-221 (P=0.038), high levels of stromal miR-222 (P=0.024), higher pathological T stage (T4; P=0.001), and lymph node metastasis (P=0.007) were significantly associated with poor overall survival rate. The subsequent multivariate analysis confirmed these results and showed that the stromal expression of miR-222 [hazard ratio (HR)=2.280, 95% confidence interval (CI)=1.097–4.742, P=0.027), higher pathological T stage (T4; HR=2.078, 95% CI, 1.043–4.142, P=0.038), and lymph node metastasis (HR=3.679, 95% CI, 1.497–9.043, P=0.005) were independent markers for poor overall survival rates in patients with CRC (Table V).

To investigate the localization of miR-221 and miR-222 in cancer tissue, ISH was performed. The negative control groups are shown in Fig. 4A and B. In CRC with liver metastasis, miR-221 and miR-222 were localized in the cytoplasm of fibroblasts in stromal tissue and cancer cells (Fig. 4C-F). By contrast, in primary CRC without liver metastasis, there was low expression of miR-221 and miR-222 in cancer cells and stromal cells (Fig. 4G and H). In terms of fibroblasts. 9/12 (75%) CRCwLM and 2/8 (25%) CRCwoLM cases (P=0.04) had miR-222-positive fibroblasts. By contrast, 7/12 (58%) CRCwLM and 3/8 (37.5%) CRCwoLM cases (P=0.36) had miR-221-positive fibroblasts. The miR-221 or miR-222 signal was observed in <10% of non-cancerous tissues (Fig. 4I and J).