Genome-Wide miRNA Analysis Identifies miR-188-3p as a Novel Prognostic Marker and Molecular Factor Involved in Colorectal Carcinogenesis

In 2014, an estimated 71,830 men and 65,000 women were diagnosed with colorectal cancer; in the United States alone, 26,270 men and 24,040 women have been estimated to perish due to the disease (1). The introduction and approval of several novel drugs, including inhibitors of angiogenesis and EGFR signaling, have substantially prolonged overall survival rates of metastatic colorectal cancer (mCRC) patients. Still, mCRC remains an incurable disease in most patients (2). In general, prognostic factors might be helpful for better individual risk stratification, patient counseling, and clinical outcome prediction. Traditional prognostic factors mainly rely on clinical and pathologic characteristics [i.e., mainly the tumor–node–metastasis (TNM) classification]. However, the clinical course of patients with the same clinical stage can vary significantly (3). The discovery of novel prognostic biomarkers that improve the risk stratification of patients is considered to be one unmet clinical need in colorectal cancer (4). In the last 15 years, the classical concept of oncogenic and tumor-suppressive protein-coding genes has been expanded by the discovery of small non–protein-coding RNA molecules, miRNAs (5). Different miRNA expression signatures are found in all cancerous tissues, and profiling them on a global scale can provide novel diagnostic and prognostic insights (6–8). Given their chemical stability in formalin-fixed tissue (9), they appear to be optimal biomarker candidates and consequently might overcome at least some of the analytic problems adherent to previously established protein-coding gene expression tools (10). Until today, several miRNAs have been described as differentially expressed and relevant for colorectal cancer prognosis (11). However, many previously published studies tested either single miRNAs as prognostic markers or used genome-wide expression studies in rather small patient cohorts or highly selected patient's cohorts. In addition, many studies lack validation in independent cohorts (12). In this study, we used publicly available genome-wide miRNA expression data from a large cohort of colorectal cancer patients to conduct an unbiased approach for identification of potentially prognostic relevant miRNAs. Following stringent criteria, we externally validated the strongest prognostic miRNAs and definitively confirmed one, miR-188-3p, to represent an independent prognostic factor in colorectal cancer patients. Finally, cellular assays, xenografts, and molecular characterization further determined the biological role of this miRNA in colorectal cancer progression.

To perform an unbiased analysis of the whole miRNA transcriptome in colorectal cancer tissue, we downloaded and analyzed data of 228 publicly available patients with complete follow-up information from The Cancer Genome Atlas Project (TCGA; http://tcga-data.nci.nih.gov/) for colorectal cancer patients (download date, December 2014). In detail, level 3 Illumina miRNASeq (Illumina Sequencing Technology: Genome Analyzer) was used for miRNA expression analysis. We derived the “reads_per_million_miRNA_mapped” values for mature forms for each miRNA from the “isoform_quantification” files. For narrowing the window of identification of potentially prognostic relevant miRNAs, we applied the following strategy: (i) miRNAs that were not expressed (read count = 0) in more than one third of the patient samples were excluded from further analyses; (ii) subsequently, HRs and corresponding 95% confidence intervals (CI) for the clinical endpoint overall survival were calculated by univariate Cox proportional models for each of the miRNAs; and (iii) only miRNAs with an HR 95% CIs <0.9 or >1.1 (and P < 0.05) were included for confirmation in the following validation cohort.

To confirm the prognostic value of the miRNAs derived from the screening set, we measured all the six potentially prognostic miRNAs in the cancerous tissue of 332 histologically confirmed colorectal cancer patients that were diagnosed between 2005 and 2012. In this bicenter study cohort, patient's formalin-fixed paraffin-embedded tissue was derived from the Institute of Pathology, Medical University of Graz (Graz, Austria) and the Department of Comprehensive Cancer Care, Masaryk Memorial Cancer Institute (Brno, Czech Republic). The patients' clinicopathologic data were retrieved from medical records at the same institutions. All cases were reviewed on the basis of pathology reports and histological slides for pTNM categories. The patients were not age, gender, or stage matched with cohort 1. Patients were treated by standard surgical procedures and received adjuvant treatment when appropriate (stage II with risk factors or stage III). In case of advanced disease at the date of diagnosis, the patients then received medical treatment according to the last version of European ESMO guidelines (13). Posttreatment surveillance included routine clinical and laboratory examination. Regarding imaging methods, CT was predominantly used. Dates of death were obtained from the medical history, central registry of the Austrian or Czechia Bureau of Statistics or by telephone calls to their relatives as previously reported. miRNA measurement by qRT-PCR was performed as follows: two to eight 10-μm thick tissue sections were used for microdissection to obtain areas with at least 70% tumor cell content. miRNAs were isolated using the miRNeasy FFPE Kit 50 (Qiagen) according to the manufacturer's instructions. cDNA was synthesized from 500 ng of total RNA using a miScript Reverse Transcription Kit (Qiagen). Quantification of miRNAs was performed using the miScript SYBR Green PCR Kit (Qiagen) and the specific miScript Primer Assay Hs_miR-92b_2 (MIMAT0003218: 5′UAUUGCACUCGUCCCGGCCUCC-3′), Hs_miR-221*_1(MIMAT0004568: 5′ACCUGGCAUACAAUGUAGAUUU-3′), Hs_miR-331_1(MIMAT0000760: 5′GCCCCUGGGCCUAUCCUAGAA-3′), Hs_miR-425-3p_1(MIMAT0001343: 5′AUCGGGAAUGUCGUGUCCGCCC-3′), Hs_miR-188-3p_1(MIMAT0004613: 5′CUCCCACAUGCAGGGUUUGCA-3′), Hs_miR-497_1(MIMAT0002820: 5′CAGCAGCACACUGUGGUUUGU-3′), and Hs_RNU6-2_11 (all miScript primer assays from Qiagen) according to the manufacturer's recommendations on a Light Cycler 480 Real-Time PCR Device (Roche). Expression values were calculated using normalization to RNU6B (after the formula 2^{−(target gene – RNU6B)}) as previously published (14). The experimenters were blinded to the results of the cohort 1 analyses as well as to the results of the functional studies below.

The frozen tissue sections of three colorectal cancer tumors and adjacent normal colon mucosa were first digested with 5 μg/mL proteinase K for 5 minutes at room temperature and then were loaded onto Ventana Discovery Ultra for in situ hybridization analysis. The tissue slides were incubated with double DIG–labeled mercury LNA miR-188-3p probe (cat #: 38532-15, Exiqon) or control U6 snRNA probe (Exiqon) for 2 hours at 45°C. The digoxigenin label was then detected with a polyclonal anti-DIG antibody conjugated with alkaline phosphatase using NBT-BCIP as the substrate. The signal intensities of miR-188-3p and U6 expression were quantified by using the intensity measurement tools of the Image-Pro Plus software package (Media Cybernetics) as reported previously (7).

For functional studies, we transfected the colorectal cancer cell lines either with miR-188-3p mimics/inhibitors (50 nmol/L) or MLLT4 siRNA (20 nmol/L), and for studying in vivo effects, we generated stable miR-188-3p–overexpressing cell lines by lentiviral transduction, as it is described in detail in the Supplementary Material. The colorectal cancer cell lines HCT116, HRT18, and RKO were authenticated at the cell bank of the Core Facility of the Medical University of Graz, Austria, by performing an STR profiling analysis (Kit: Promega; PowerPlex 16HS System; cat. no. DC2101; last date of testing, March 3, 2016).

For quantification of mRNA, miRNAs, and proteins, we applied standard methods of quantitative PCR and Western blot analysis as is described in detail in the Supplementary Material.

Cell lines used in this study were maintained under standard conditions and media as recommended by the supplier. Cell lines were validated by the Core Facility of Cell Culture at the Medical University of Graz using STR DNA fingerprinting. All cellular assays were started after 24 hours of transient ectopic transfection of miR-188-3p mimics or inhibitor (Qiagen). Cellular growth was determined every 24 hours using the WST-1 assay for 96 hours (Roche). The same assay was used to test the sensitivity of cell lines to different drugs. For cancer cell migration testing, real-time monitoring by xCELLigence (Roche) and scratch assay (Ibidi Systems), according to recommended standard conditions, were used. For each assay, at least three biological repeats have been performed. In more detail, cell migration of transiently transfected colorectal cancer cell lines was assayed using the xCELLigence Real-Time Cell Analyzer (CIM plates, RTCA; Roche Diagnostics). A total of 6 × 10⁴ cells were plated in each well in serum-free media. The lower chamber contained growth medium with 10% FBS. Cells were allowed to settle for 30 minutes at room temperature before being placed in RTCA in a humidified incubator at 37°C with 5% CO₂. The rate of migration was measured every 15 minutes for 48 hours (“real-time migration”). Data acquisition and analyses were performed using the RTCA software (version 1.2, Roche Diagnostics). The cell index is derived from electrical impedance changes as the cells interact with interdigitated microelectrodes on the bottom of the E-plate. Three replicates of each cell line were performed.

To monitor cell migration with a second independent method, an in vitro wound-healing assay was performed (in quadruplicates). After transient transfection of colorectal cancer cell lines, 70 μL of serum-free medium containing 5 × 10⁴ cells were seeded in an Ibidi culture insert (Ibidi GmbH) in a 6-well plate. After 24 hours of incubation, the culture insert was detached to form a defined gap in the cell monolayer. After washing the cells with PBS, medium containing 10% FBS was added and cell migration toward the gap area was documented 24 and 48 hours later using a microscope at ×10 magnification.

This experiment has been performed by a commercially external facility (EPO, Berlin-Buch GmbH), which was completely blinded to the other results of this study. A total of 2 × 10⁶ HCT116 cells were injected intrasplenically in a volume of 20 μL to NMRI:nu/nu mice (Janvier Labs). Per group, 7 mice were inoculated with cells (HCT116 cells with either a miR-188 stable overexpression or control construct) and received afterwards tramadol for 6 days for pain suppression. All animal experiments were performed under the guidelines of the German Animal Protection Law and with approval by the local responsible authorities. Mice were observed daily for their health status. Body weight was measured twice per week as a parameter of health condition. Mice were sacrificed 33 days after cell inoculation, and the number and size of liver metastases and intra-abdominal metastases as well as the presence of ascites were assessed. Bioluminescence measurement was performed to monitor the engraftment of metastases. Bioluminescence imaging of all mice was performed from dorsal up to five times. Luciferin (5 mg/kg; 30 mg/mL distilled water, AAT Bioquest) was injected intraperitoneally and inoculated for 5 minutes before mice were anesthetized by inhalation of isoflurane. The measurement took place in a VisiLuxx Imager (Visitron Systems GmbH). The VisiView Software made it possible to take a bright field picture first. Afterwards, 10 pictures with an exposition time of 30 seconds were performed to measure the luminescence signal. The analysis of the signals was performed with the MetaMorphOffice Software. Single luminescence signals were summed up and an overlay emerged from the luminescence signal and the bright field picture.

To identify possible molecular mechanisms of action of miR-188-3p in relation to migration, we retrieved putative miRNA–mRNA interactions by nine prediction algorithms from miRWalk 2.0 database (http://zmf.umm.uni-heidelberg.de/apps/zmf/mirwalk2). We also retrieved experimentally validated miRNA–mRNA interactions from miRWalk and miRTarBase (http://mirtarbase.mbc.nctu.edu.tw). The results indicated that there are numerous putative miRNA recognition sites for miR-188-3p. To generate and narrow the list of targets for further analyses, we decided on the following threshold: at least five (more than a half of the total number of programs checked) algorithms predicted miRNA–mRNA interaction, or there was strong experimental evidence of miRNA–mRNA interaction. Integrated function and pathway analysis was performed using DAVID bioinformatics resources (http://david.abcc.ncifcrf.gov/) and significant features were clustered. From the pathway analysis, we focused our literature-based research on three pathway terms: Wnt pathway, Colorectal Cancer, and Adherens junctions. Following the hypothesis that a miR-188-3p target should ideally negatively regulate migration process, we selected three genes that fulfilled this criterion previously published in any cancer type.

To test the hypothesis that miR-188-3p directly interacts with the MLLT4 mRNA, we generated luciferase reporter constructs containing the wild-type (5′-TggttcagtgattgcttaaatggcatgtggaccGTGGGAagcagtaggagcgtagtaaga-3′; CS-HmiT011246-MT06-01; GeneCopoeia) and a sequence-modified (5′-TggttcagtgattgcttaaatggcatgtggaccATAAACgcagtaggagcgtagtaaga-3′; CS-HmiT011246-MT06-02; GeneCopoeia) MLLT4 3′ untranslated region (UTR). For this purpose, we ordered the pEZX-MT06 target reporter vectors containing the putative miR-188-3p–binding site (from GeneCopoeia) and transiently transfected them into HEK cells by using the Lipofectamine 2000 reagents according to the manufacturer's recommendations (Invitrogen). Briefly, HEK cells were grown up in 24-well plates to a confluency of about 80%, followed by a cotransfection of miR-188-3p mimetic (50 nmol/L, Qiagen) together with 200 ng of pEZX-MT06 (CS-HmiT011246-MT06-01 and CS-HmiT011246-MT06-02) constructs. For obtaining 100% reference value, we cotransfected the negative control (AllStars negative control, Qiagen) and plasmids (CS-HmiT011246-MT06-01 and CS-HmiT011246-MT06-02). Twenty-four hours after transfection, cells were lysed in 100 μL of passive lysis buffer according to the Luc-Pair Luciferase Assay Kit 2.0 (GeneCopoeia); 20 μL of the lysate was used for the luciferase activity measurements following the instructions of the Luc-Pair Luciferase Assay Kit 2.0. An empty control plasmid pEZX-MT06 (CmiT000001-MT06, GeneCopoeia) was cotransfected either with miR-188-3p mimics or the miRNA negative control (not both in one approach), respectively, to normalize the luciferase activity. Luciferase assays were run on a LUMIstar Luminometer (BMG Labtech) in three independent biological replicates.

All statistical analyses were performed using SPSS version 20 software (SPSS Inc.), MedCalc software (version 13.1.2.0) or R software (version 3.0.1; http://www.r-project.org/). We have listed Recommendations for Tumor Marker Prognostic Studies (REMARK criteria) in the Supplementary Material in relation to our reported results in ref. 15. Unpaired or paired Student t test, Fisher exact test, χ² test, Mann–Whitney, and Kruskal–Wallis test were applied where appropriate to analyze the association between miR-188-3p expression and clinicopathologic parameters. Data of gene expression were log transformed. We decided to use the median value of miR-188-3p expression as the cut-off value for all analyses, as this is an unbiased approach without favoring a mathematically optimized cut-off value. Overall survival was defined as the time from date of diagnosis to the date of death by any cause, and it was assessed using the Kaplan–Meier method. The log-rank test was performed to compare the survival curves of individual groups. Univariate and multivariate Cox proportional hazards models, including age, gender, tumor location, tumor stage (according to the AJCC/UICC 2010 TNM classification), and miRNA expression levels. To test the proportional hazard assumption in Cox models, Schoenfeld residuals test was used. The reported results included HR and 95% confidence intervals (CI). The predictive accuracy of the multivariate models was quantified using the Harrel concordance index (c-index). The interpretation of the c-index is similar to the interpretation of the area under an ROC curve. A value of 1.0 indicates that the features of the model perfectly separate patients with different outcomes, whereas a value of 0.5 indicates the features contain prognostic information equal to that obtained by chance alone (16). A two-sided P < 0.05 was considered statistically significant.

Overall, we obtained the expression data of 971 miRNAs in at least one tumor sample of the 228 colorectal cancer patients derived from the TCGA dataset. Supplementary Table S1 lists the clinicopathologic characteristics of the screening (n = 228) and the validation (n = 332) cohort. Applying the criteria mentioned in the Materials and Methods section, we identified 13 miRNAs whose expression levels were significantly (P < 0.05, univariate Cox model) associated with prognosis in colorectal cancer patients (two of them were associated with lower risk and 11 of them were associated with higher risk of death; Supplementary Table S2). In addition, when performing the same analyses on 186 samples of colon cancer, excluding the rectum cancer cases, we identified two miRNAs significantly (P < 0.05, univariate Cox model) associated with lower risk and 13 miRNAs significantly associated with higher risk of death (Supplementary Table S3). To further validate the potentially prognostic miRNAs in a second independent colorectal cancer cohort, we selected six miRNAs from the whole colorectal cancer cohort and colon only cases according to the prespecified criteria (i.e., with an upper limit <0.9 or a lower limit >1.1 of their respective HR 95% CI; yellow marked miRNAs in Supplementary Tables S2 and S3). On the basis of the data of the screening set, we measured miR-92b-3p, miR-188-3p, miR-221-5p, miR-331-3p, miR-425-3p, and miR-497-5p (labeling according to miRBase version 21) in 332 patients of the validation cohort by qRT-PCR and calculated the HR and corresponding 95% CI by univariate Cox models. Clinicopathologic characteristics of this cohort are summarized and compared with the screening cohort in Supplementary Table S1. Notably, there were significant differences in terms of clinicopathologic variables between the two cohorts, including age, gender, stage, and location (P < 0.05; for details, see Supplementary Table S1). Importantly, the patients in the validation cohort had a significantly worse 5-year survival, which could mainly be explained by the higher numbers of advanced/metastatic stage IV tumors in the validation cohort (36% in cohort 2 vs. 18% in cohort 1, respectively; also see Supplementary Table S1).

In univariate Cox analyses, only miR-188-3p complied with the prespecified criteria of the HR and 95% CI borders (i.e., >1.1), and therefore, we focused our following comprehensive prognostic and experimental analyses only on this miRNA. First, in situ hybridization localizes miR-188-3p in epithelial cells and indicates a higher expression level of miR-188-3p in cancerous cells compared with the corresponding normal mucosa (n = 3 paired cancer and adjacent normal colon samples; fold change, 1.91; P = 0.005, paired Student t test; Fig. 1 and Supplementary Fig. S1). Concomitantly, we measured the expression values of miR-188-3p by qRT-PCR in 61 paired colorectal cancer samples (where RNA for both the tumor and the adjacent normal colon mucosa was available in the validation cohort) and observed significant upregulation in cancerous tissue (3.89-fold higher expression, P < 0.001, paired Student t test; Supplementary Fig. S2A). In addition, we explored whether miR-188-3p expression level was significantly associated with other clinicopathologic parameters. In the large validation cohort, miR-188-3p levels significantly increased with higher tumor stage (P < 0.001, ANOVA; Supplementary Fig. S2B). Regarding the association with well-known mutations (KRAS, NRAS, and BRAF genes) in colorectal cancer patients, we analyzed the screening cohort for the possible association of miR-188-3p expression levels and mutational status in one of these cancer genes. For cohort 1, 182 patients (80%) were available for mutational status: KRAS wild type in 107 patients (58.8%), KRAS mutant in 45 (41.2%); NRAS wild type in 164 (90.1%) and NRAS mutated in 18 (9.9%) patients; and BRAF wild type in 162 (89%) and BRAF mutated in 20 (11%) of patients. For KRAS- and NRAS-mutated patients, we could not find any significant association of miR-188-3p expression levels with mutational status, but in patients with a BRAF mutation, we observed significantly lower miR-188-3p expression levels in their tumor tissue (P = 0.01, Mann–Whitney U test). In addition, miR-188-3p expression levels were significantly higher in microsatellite stable tumors (Supplementary Fig. S2C).

In the screening set (cohort 1), univariate analyses identified older age, high miR-188-3p, and advanced tumor stage (stage IV vs. stage I–III) as poor prognosticators for survival (all P values <0.05, univariate Cox model), whereas gender and tumor location (colon vs. rectum) were not significantly associated with clinical outcome (Supplementary Table S4). Figure 2A shows the Kaplan–Meier curve for OS and reveals that high miR-188-3p expression is a significant factor for poor prognosis in colorectal cancer (P = 0.020, log-rank test). In the validation set (cohort 2), univariate analysis identified high miR-188-3p and advanced tumor stage (stage IV vs. stage I–III) as poor prognosticators for survival (all P values <0.05), whereas gender, age, and tumor location were not significantly associated with clinical outcome (Supplementary Table S4). Supplementary Fig. S3 shows the Kaplan–Meier curve for overall survival stratified according to the corresponding quartile (P = 0.001, log-rank test). Figure 2A shows that miR-188-3p expression levels greater than the median levels is a significant factor for poor prognosis in colorectal cancer patients (P < 0.001, log-rank test). Focusing the univariate analyses on patients with stage II/III colorectal cancer, we observed a similar and significant poor prognostic value for miR-188-3p in both cohorts (n = 135 in cohort 1, P = 0.0495; n = 174 in cohort 2, P = 0.0005, respectively). To test whether the prognostic value of high miR-188-3p expression is independent of other well-known prognostic factors, a multivariate analysis was performed using a Cox proportional hazard model for both cohorts. In cohort 1, multivariate analysis, including age, gender, tumor location, tumor stage, and miR-188-3p expression, demonstrated that high miR-188-3p expression represents an independent predictor for poor survival in colorectal cancer patients (HR = 4.137; 95% CI, 1.568–10.917; P = 0.004). Statistically significant results were also obtained for age (P = 0.004) and advanced tumor stage (P < 0.001), whereas all other parameters were not significantly associated with survival (Table 1). Next, we evaluated the impact of miR-188-3p expression on the predictive ability of the model when added to established clinicopathologic factors in the TCGA dataset. Calculation of the Harrel c-index was performed and the predictive accuracies with and without miR-188-3p supplementation were compared. The c-index without miR-188-3p was 0.76 and improved to 0.86 when miR-188-3p was added, but did not reach statistical significance (P = 0.057). In the validation cohort (cohort 2), we confirmed that high miR-188-3p level is significantly associated with poor survival in multivariate Cox analysis (HR = 1.538; 95% CI, 1.107–2.137; P = 0.010; Table 1).

To find a biological explanation for the possible consequences of upregulation of miR-188-3p in cancer tissue and the association of high tumor stage with high miR-188-3p levels as well as with poor survival, we ectopically overexpressed miR-188-3p in different colorectal cancer cell lines and studied the impact on biological features. Supplementary Fig. S4A shows the expression values of miR-188-3p in different colorectal cancer cell lines. First of all, we explored in a panel of three colorectal cancer cell lines (HCT116, HRT18, and RKO) the influence of miR-188-3p expression on cellular growth, but could not detect any significant changes or trend in this hallmark of cancer (Supplementary Fig. S4B–S4D). In addition, we could not observe any significant difference in sensitivity against any of the tested chemotherapeutic drugs [we explored the three most commonly used drugs for colorectal cancer treatment: 5-fluorouracil (0–40 μmol/L), oxaliplatin (0–30 μmol/L), and irinotecan (0–30 μmol/L); Supplementary Figs. S5–S7] in all three tested cell lines. Finally, we investigated the influence of miR-188-3p on cancer cell migration by using the real-time xCELLigence system. Forced expression of miR-188-3p led to a significantly higher migration rate of HCT116, RKO (Supplementary Fig. S8A and S8B), and HRT18 (Supplementary Fig. S9A) cells in this assay. To substantiate these positive signals for migration, we used another migration assay (scratch assay) to independently confirm and quantify the findings in three colorectal cancer cell lines (HCT116, RKO, and HRT18). In line with the real-time data from the xCELLigence assay, forced miR-188-3p expression led to a significantly higher migration rate in all tested cell lines in vitro (Fig. 3). In contrast, when using a miR-188-3p inhibitor, we could observe an inhibition of migration in all three (HCT116, HRT18, and RKO) cell lines (Supplementary Fig. S10). For in vivo confirmation of this potentially promigratory phenotype, we generated HCT116 cells that stably overexpresses miR-188-3p (Fig. 4A) and confirmed at first in vitro a significantly higher migration rate for these cells (Fig. 4B and C). Consequently, we injected HCT116 miR-188-3p stably overexpressing cells into the spleens of NMRI:nu/nu mice and compared the capacity of metastases formation with HCT116 control cells (n = 7 for each group). In vivo imaging monitoring showed the dissemination of metastases and indicated a high degree of correlation with formation of macrometastases (Fig. 4D–G). After about 5 weeks, pathologic assessment of the liver and abdomen showed metastases formation in the liver in all the mice, which is consistent with findings of previous reports for this cell line (7). However, in the group of mice where the HCT116 miR-188-3p–overexpressing cells had been injected, a significantly higher number of animals had large liver metastases (>1 cm in largest diameter) and/or ascites in combination with multiple intra-abdominal metastases (pancreas or peritoneum) compared with the HCT116 control cells (57% vs. 0%, P < 0.05, Mann–Whitney U test; Fig. 4H).

To identify a possible molecular mechanism of action of miR-188-3p related to cellular migration process, we performed comprehensive bioinformatics analyses of putative miR-188-3p target genes. Among several potential interaction partners (Supplementary Table S5), we selected three of them (based on the criteria mentioned in the methods section), which have been previously described as important factors of cancer cell migration in any type of cancer (i.e., NLK, CTNNA2, and MLLT4). First of all, we transiently transfected two independent colorectal cancer cell lines (HCT116 and HRT18) with miR-188-3p mimetic and measured the expression changes of these three genes. MLLT4 showed a reduction of about 40% to 60% mRNA expression levels in both cell lines (Fig. 5A), corroborated by a decreased protein expression (Fig. 5B). In contrast, NLK mRNA expression levels were not influenced by miR-188-3p manipulation and CNNA2 was not even detectable in colorectal cancer cell lines (data not shown). A direct interaction between miR-188-3p and MLLT4 mRNA was further confirmed by the findings of reduced luciferase activity after 24 hours when the MLLT4 3′UTR wild-type sequence (Fig. 5C) was cotransfected with miR-188-3p (Fig. 5D, left bar chart). Conversely, using the MLLT4 3′UTR mutated construct (Fig. 5C), we observed no changes in luciferase activity after cotransfection with miR-188-3p (Fig. 5D, right bar chart). In contrast, using a miR-188-3p inhibitor, MLLT4 mRNA expression significantly increased about 30% (P < 0.05, unpaired Student t test; Supplementary Fig. S11A). Finally, to determine whether MLLT4 is a mediator of the promigratory effects of miR-188-3p, we successfully knocked down MLLT4 mRNA and protein expression by siRNA (Supplementary Fig. S11B–S11D) and consequently measured the influence on migration process of two independent colorectal cancer cell lines. As shown in Fig. 5E, reduced levels of MLLT4 increased HCT116 and HRT18 colorectal cancer cell motility. Taken together, the phenocopy and the interaction experiments support the hypothesis that MLLT4 is a key effector of miR-188-3p promigratory capacity.

Colorectal cancer patient management experienced major progress in terms of diagnostic and therapeutic procedures within the last two decades. Furthermore, the proof of RAS mutations as predictive biomarkers for anti-EGFR agents brought this type of cancer to the grounds of personalized medicine (17). The traditional clinicopathologic characteristics, especially the tumor size, nodal status, and occurrence of distant metastases (summarized as the TNM classification system over many years), are still considered to be the gold standard with regard to prognostic assessment of colorectal cancer patients at diagnosis (18). However, there is significant variation and heterogeneity in the outcomes of patients within the same tumor stage, leading to intensive evaluation of many other histologic, molecular, and genetic features (19–21).

In this study, we investigated the association of miRNA expression and clinical outcome in colorectal cancer patients. In a comprehensive approach, we analyzed RNA sequencing data of hundreds of miRNAs from more than 200 patients of the TCGA dataset. This approach identified several miRNA candidates, including six miRNAs fulfilling the stringent inclusion criteria (miR-92b-3p, miR-188-3p, miR-221-5p known also as miR-221*, miR-331-3p, miR-425-3p, known also as miR-425* and miR-497-5p). Some of them, for instance miR-221-5p (22), have been previously proposed as prognostic factors in other colorectal cancer studies. miR-331-3p has been linked to progression-free survival in patients with KRAS wild-type mCRC treated with anti-EGFR therapy in two independent studies (23, 24), and miR-497-5p has been involved in the regulation of the endogenous insulin-like growth factor receptor 1 expression, as well as cell survival, proliferation, and invasion (25). Our findings from the TCGA dataset substantiate their involvement in colorectal carcinogenesis. Other miRNAs of this six miRNA prognostic panel derived from the screening cohort, including miR-92b-3p, miR-425-3p, and miR-188-3p, have not been reported in colorectal cancer yet. In general, every novel discovered prognostic marker or prognostic tool has to undergo validation in different datasets before the true value can objectively be assessed (15). In particular, the external validation of such newly identified prognostic risk factors in independent cohorts of patients is paramount prior to the generalization of the applicability of a prognostic marker (26, 27). With this in mind, we made use of a second independent large cohort and explored the association of these six miRNAs with patient prognosis. Applying the stringent prespecified criteria, we could only confirm high miR-188-3p expression as an independent poor prognostic factor for both cohorts. The reasons for these conflicting data between our cohorts and others (22), as well as between the screening cohort and the validation cohort, are not exactly known, but sample size (108/54 vs. 228 vs. >300, small sample size is rather prone for random observations; ref. 22), different tumor stages [20% stage IV in Yuan and colleagues' study (22) vs. 36% in our cohort 2], ethnicity (Asian patients vs. European patients), treatment modalities, and other unknown factors might be responsible.

To further substantiate the clinical significance of miR-188-3p and obtain a biological rationale basis for this observation, we applied several cellular assays after gain and loss of miR-118-3p function experiments. This way, our study further characterizes for the first time a possible biological role of miR-188-3p, which seems to be involved in colorectal cancer cell migration. Migration of cancer cells is central for developing distant metastases. In vitro cellular assays, in vivo metastases formation, and the clinical observation that metastatic stage IV patients in the large validation cohort had higher miR-188-3p levels compared with localized stages underline this assumption. To elucidate possible molecular mechanisms regulated by miR-188-3p, we performed a comprehensive in silico analysis to predict putative interacting partners. One of them, MLLT4 (also called afadin, adherens junction formation factor AF6 or MLL-AF6), has been previously involved as a negative regulator of migration and metastases in breast and pancreatic cancer (28, 29). These data prompted us to focus on MLLT4, and we actually demonstrated that miR-188-3p negatively influences the expression levels of MLLT4 as well as directly interacts with the predicted miRNA–mRNA binding site in the 3′UTR region of the MLLT4 mRNA. Modifying the interaction site in the 3′UTR MLLT4 restores the effects and using miR-188-3p increases the MLLT4 expression. Knockdown experiments of MLLT4 clearly confirmed a phenocopy of the promigratory phenotype after miR-188-3p overexpression and therefore established MLLT4 as a novel regulator of cancer cell migration in colorectal carcinogenesis. In general, our study supports the idea that based on global profiling of miRNA expression, novel candidates and miRNA panels can improve different diagnostic or prognostic steps (11, 30, 31). For instance, miR-21, -93, and -103 upregulation together with miR-566 downregulation defined a colorectal cancer metastatic signature that could discriminate between colorectal recurrences to lymph nodes and liver and between colorectal liver metastasis and primary hepatic tumor (31). Another study showed that the combination of miR-25-3p and miR-339-5p is a valuable independent prognostic factor in stage II and III microsatellite stable tumors, enabling the construction of a multivariate nomogram for metastasis risk prediction (11). Recently, Zhang and colleagues constructed a prognostic classifier based on the six miRNAs: miR-21-5p, miR-20a-5p, miR-103a-3p, miR-106b-5p, miR-143-5p, and miR-215. The authors reported this classifier to be a reliable prognostic and predictive tool for disease recurrence in patients with stage II colon cancer that might be able to predict the benefit of patients from adjuvant chemotherapy (30). In comparison with previously published studies, the strengths of our study are the relatively high numbers of patients, the external validation in a second cohort, and the robust findings despite the use of different technical platforms (RNA sequencing and qRT-PCR). Experimental rationale by showing the positive influence of miR-188-3p on migration of cancer cells substantiates our clinical results.

Regarding the role of miR-188-5p, which is the other miRNA derivative of the common pri-miR-188, a previously published study reported an involvement in rectal cancer and response to neoadjuvant radiochemotherapy (32). In other types of cancer, miR-188-5p has been described as a tumor-suppressive factor that acts through the inhibition of G₁–S-phase cell-cycle transition in nasopharyngeal carcinoma and growth/metastases inhibition in prostate cancer (33, 34). Although these findings are in contrast to the prometastatic properties reported here for miR-188-3p, there are other examples for strand-specific inverse functions of the -5p and -3p miRNA variants (14). To speculate about the clinical implications of our findings, there are two main considerations for miR-188-3p clinical applications. First, the prognostic value, which should be further tested in prospective biomarker studies, might help to better tailor patient´s outcome, stratification in clinical trials, and individual risk stratification. Second, the value of miR-188-3p inhibitors as potential agents to decrease the risk of metastases and the spread of metastases should be further tested in preclinical models. If successful, this strategy might move forward to early clinical trials and this discovery in this study might provide a biological rationale for such trials. In addition, a better understanding of MLLT4 in colorectal cancer might help to develop strategies against this cancer cell migration regulator, a biological role for this protein that has been proposed in this study for the first time in colorectal cancer cells.

Our study is not without limitations. Both cohorts are retrospective in their nature and like all retrospective studies prone to selection bias. The treatment schedule was heterogeneous and we cannot rule out that changes in medical treatment over the years might influence our findings.

In conclusion, this study indicates for the first time that miR-188-3p plays a previously unrecognized role in colorectal carcinogenesis. miR-188-3p represents an independent negative prognostic factor in colorectal cancer patients. Prospective randomized trials are warranted to further test the role of miR-188-3p as a prognostic classifier in different disease settings of colorectal cancer patients.

No potential conflicts of interest were disclosed.

The sponsors of this study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or the decision to submit this manuscript for publication.

Conception and design: M. Pichler, M. Svoboda, G. Hoefler, O. Slaby, G.A. Calin

Development of methodology: M. Pichler, V. Stiegelbauer, X. Zhang, G.A. Calin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Pichler, P. Vychytilova-Faltejskova, H. Ling, X. Zhang, M. Goblirsch, A. Wulf-Goldenberg, J. Haybaeck, M. Svoboda, Y. Okugawa, A. Gerger, A. Goel, O. Slaby

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Pichler, V. Stiegelbauer, C. Ivan, M. Goblirsch, J. Haybaeck, M. Svoboda, A. Gerger, O. Slaby

Writing, review, and/or revision of the manuscript: M. Pichler, V. Stiegelbauer, P. Vychytilova-Faltejskova, H. Ling, M. Goblirsch, J. Haybaeck, A. Gerger, G. Hoefler, A. Goel, G.A. Calin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Pichler, E. Winter, M. Goblirsch, M. Ohtsuka

Study supervision: M. Pichler, G.A. Calin

G.A. Calin is supported by the CLL Global Research Foundation. The work in G.A. Calin's laboratory is supported in part by the NIH/NCI grants 1UH2TR00943-01 and T32CA009599, Developmental Research Awards in Prostate Cancer, Multiple Myeloma, Leukemia (P50 CA100632), and Head and Neck (P50 CA097007) SPOREs, a SINF MDACC_DKFZ grant in CLL, a SINF grant in colon cancer, the Laura and John Arnold Foundation, the RGK Foundation, and the Estate of C. G. Johnson Jr. M. Pichler is supported by an Erwin-Schroedinger Scholarship of the Austrian Science Funds (project no. J3389-B23). This study was also supported in part by funds of the Oesterreichische Nationalbank (Anniversary Fund, project number: 15400; to A. Gerger). The work in A. Goel's laboratory was supported by the grants R01 CA72851 and CA 181572 from the NCI, NIH, and funds from the Baylor Research Institute. This work was also supported by grant IGA NT/13860-4/2012 of the Czech Ministry of Health, by the project “CEITEC—Central European Institute of Technology” (CZ.1.05/1.1.00/02.0068) and the project BBMRI CZ (LM2010004).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.