Colorectal cancer includes an invasive stem-like/mesenchymal subtype, but its genetic drivers, functional, and clinical relevance are uncharacterized. Here we report the definition of an altered miRNA signature defining this subtype that includes a major genomic loss of miR-508. Mechanistic investigations showed that this miRNA affected the expression of cadherin CDH1 and the transcription factors ZEB1, SALL4, and BMI1. Loss of miR-508 in colorectal cancer was associated with upregulation of the novel hypoxia-induced long noncoding RNA AK000053. Ectopic expression of miR-508 in colorectal cancer cells blunted epithelial-to-mesenchymal transition (EMT), stemness, migration, and invasive capacity in vitro and in vivo. In clinical colorectal cancer specimens, expression of miR-508 negatively correlated with stemness and EMT-associated gene expression and positively correlated with patient survival. Overall, our results showed that miR-508 is a key functional determinant of the stem-like/mesenchymal colorectal cancer subtype and a candidate therapeutic target for its treatment.

Significance: These results define a key functional determinant of a stem-like/mesenchymal subtype of colorectal cancers and a candidate therapeutic target for its treatment. Cancer Res; 78(7); 1751–65. ©2018 AACR.

Epithelial-to-mesenchymal transition (EMT) contributes to cancer progression, particularly tumor cell invasion and metastasis, which enables cancer cells to escape the primary tumor mass and colonize new terrain in the body (1). EMT is thought to be an intrinsic characteristic of a basal-like subtype of breast cancer that has a highly aggressive and stem-like type and a poor clinical outcome (2). Similarly, a stem-like/mesenchymal subtype of colorectal cancer has been identified recently (3, 4). This subtype of colorectal cancer is characterized by a clear upregulation of genes implicated in EMT and stem cell–specific genes (e.g., ZEB1, SALL4, SNAI2, and VIM) and is associated with the activation of the TGF-β signaling pathway. In addition, compared with other subtypes of colorectal cancer, the stem-like/mesenchymal subtype tumors contain more stromal cells in the tumor tissues, and tend to be diagnosed at advanced stages, resulting in worse overall survival and relapse-free survival (4, 5). However, the driving molecular events of this cancer subtype, and their functional and clinical relevance, remain to be determined in patients with colorectal cancer.

Epigenetic alternations of the genomes of colon epithelial cells are involved in the control of colorectal cancer pathogenesis (6). Recent studies have shown that miRNAs might regulate cancer initiation and EMT progression in colorectal cancer (7) and other types of cancer (8) Functional studies and genome-wide bioinformatics analyses suggested that one key miRNA might have numerous target genes (9). Therefore, we hypothesized that certain driving miRNAs might regulate multiple stem-like/EMT-related genes and control the stem-like/mesenchymal subtype of colorectal cancer. To test this hypothesis, in this study, we undertook an integrated approach and analyzed miRNA–mRNA data from colorectal cancers in The Cancer Genome Atlas (TCGA) database and uncovered the potential molecular networks that might control the stem-like/mesenchymal subtype of colorectal cancer. We identified a major genomic loss of four key miR-506 family members, including miR-508, in the stem-like/mesenchymal colorectal cancer subtype. The expressions, functions, and clinical relevance of the miR-506 family members were evaluated in colorectal cancer cells and tissues. We showed that the miR-506 family members, particularly miR-508, functionally and clinically defined the stem-like/mesenchymal colorectal cancer subtype. Furthermore, we found that a novel long noncoding RNA (lncRNA) AK000053 was induced by hypoxia via hypoxia inducing factor alpha (HIF-1α), which mechanistically and pathologically contributed to the miR-508 loss in colorectal cancer.

Clinical specimens

Primary colon cancer cells were reported previously (10). We collected fresh colorectal cancer tissues and normal colorectal tissues adjacent to cancers from 90 patients with colorectal cancer who underwent surgery at Shanghai Renji Hospital between 2010 and 2014 (Renji cohort 1). Similarly, we collected paraffin-fixed colorectal cancer tissues and/or normal colorectal tissues from 100 patients with colorectal cancer between 1999 and 2005 at Shanghai Renji Hospital (Renji cohort 2) and from 128 patients with colorectal cancer between 2007 and 2008 at the Second Department of General Surgery in the Medical University of Lublin (European cohort). All the research was carried out in accordance with the provisions of the Declaration of Helsinki of 1975. This study was approved by the ethics committee of Shanghai Jiao Tong University School of Medicine, Renji Hospital and the Medical University of Lublin. Written informed consents were obtained from all participants in this study.

Identification of miR-508–associated biological pathways

Gene set enrichment analysis (GSEA) was performed using the JAVA program (http://www.broadinstitute.org/gsea) with the Canonical pathways gene set collection (MSigDB C2 CP; 1,320 gene sets available). The GSEA outputs (visualized in Cytoscape, version 2.8.2) and the Enrichment Map software (11) were used to identify the biological processes discriminating the two relevant groups. To simplify the network map, a stringent threshold of gene-set permutations with a false discovery rate (FDR) cutoff of 0.5% and a P value cutoff of 0.005 was used in the Enrichment Map software, as described previously (11).

Cell lines

Human colorectal cancer cell lines SW1116, SW480, Caco2, LoVo, HT29, and HCT116 were purchased from ATCC. All cell lines were genotyped for identity by Beijing Microread Genetics Co., Ltd and tested routinely for Mycoplasma contamination (last date of testing: January 3, 2017).

RNA extraction and quantitative real-time PCR

Total RNA was extracted using the TRIzol reagent (Takara) from the cultured cells and was reverse-transcribed using the PrimeScriptP RT Reagent Kit (Perfect Real Time; Takara). Real-time PCR was performed on an Applied Biosystems 7900 Quantitative PCR System (Applied Biosystems). The amplified transcript level of each specific gene was normalized to that of GAPDH. All primers are listed in Supplementary Table S1.

RNA pull-down assay

To determine whether AK000053 lncRNA is associated with the RNA-induced silencing complex (RISC), we performed an RNA pull-down assay using synthesized biotin-labeled AK000053 lncRNA or an AK000053 lncRNA mutant as a probe and then detected miR-508 by quantitative real-time reverse transcription PCR (qRT-PCR). RNA pull-down was performed as described previously (12).

In vivo tumor formation and metastatic models

Mouse experiments were conducted in accordance with the NIH Guidelines for the Care and Use of Laboratory Animals. The study procedures were approved by the Institutional Animal Care and Use Committee of Renji Hospital, School of Medicine, Shanghai Jiaotong University.

Tumor development.

NOD.SCID/γc−/− (NSG) mice were obtained from the In-Vivo Science Inc. Colorectal cancer cells were infected with control miRNA and miR-508 lentivirus and were subcutaneously injected into nonirradiated 5- to 6-week-old male NSG mice. Tumor incidence was monitored for 3 weeks after tumor inoculation. Tumor incidence was calculated for 3 weeks after tumor inoculation.

Tumor growth.

To demonstrate the effect of miR-508 on tumor growth in vivo, 4-week-old male BALB/c nude mice (Experimental Animal Centre, SIBS) were used in our study. HCT116 (5 × 106) cells were injected subcutaneously into the right flank of these mice to establish the colorectal cancer xenograft model. Once the tumor diameter reached 3 mm, mice were randomly divided into different groups and were injected with PBS, control miRNA, or miR-508 lentivirus through multipoint intratumoral injections every other day for 11 days. Tumor volume (mm3) was estimated by the formula: tumor volume (mm3) = shorter diameter2 × longer diameter/2. The tumor volumes data are presented as means ± SD.

Tumor metastasis.

To investigate the effect of miR-508 on tumor metastasis in vivo, we developed a colorectal cancer metastasis model in nude mice. The HCT116 cells (1 × 107) were subcutaneously injected through the right flank of 4-week-old BALB/c nude mice. Five days after tumor inoculation, the animals were randomly divided into different groups with intratumoral injection of PBS, control miRNA, or miR-508 lentivirus every other day for 12 weeks. For lung metastasis examination, mice were sacrificed at week 12. The numbers of lung metastatic foci were determined in hematoxylin and eosin–stained lung tissue sections under a binocular microscope (Leica, DM 300). All experimental procedures were approved by the Institutional Animal Care and Use Committee.

Statistical analysis

All statistical analyses were performed using R-3.0.2 (http://cran.r-project.org/bin/windows/base/old/3.0.2/). Cytoscape v.2.82 was used to generate the network of miRNAs and their putative target genes. For the clinicopathologic analysis, the χ2 test or Fisher exact test (two-sided) were performed. The Kaplan–Meier method was used to estimate overall survival. The log-rank test was used to evaluate the differences between survival curves. All P values were two-sided unless otherwise specified.

Differential miRNA and mRNA profiles in the stem-like/mesenchymal subtype of colorectal cancer in TCGA

An invasive stem-like/mesenchymal cancer subtype has been reported in colorectal cancer patients (4). To explore the driving molecular events of this colorectal cancer subtype, we hypothesized that specific miRNA signatures target the stem-like/mesenchymal gene expression and control the phenotype of this colorectal cancer subtype. To test this hypothesis, we analyzed the mRNA and miRNA data in colorectal cancer in the TCGA Data Portal (https://gdc.cancer.gov/). On the basis of 220 RNA arrays (Fig. 1A), we found that 2,235 genes were differentially expressed in the mesenchymal subtype of colorectal cancer compared with the non-stem-like subtypes, including enterocyte, inflammatory, goblet like, and transit-amplifying subtypes (Fig. 1B). Among these genes, 80% (1,781/2,235) were upregulated by more than one fold in the stem-like/mesenchymal subtype relative to the non-stem-like subtypes (Fig. 1A; Supplementary Table S2). On the basis of 198 miRNA-sequence data analyses from the TCGA repository, we identified that four miRNAs, miR-506, miR-508, miR-509, and miR-514, were downregulated by more than 2-fold, with an FDR < 0.0015 in the stem-like/mesenchymal subtype colorectal cancers compared with other subtypes (Fig. 1C; Supplementary Table S3). These four miRNAs belong to the miR-506 family. We used FindTar3 (http://www.tsinghua.edu.cn/publish/newthu/index.html) and Miranda microRNA (www.microrna.org) software to predict the target genes of the differentially expressed miRNAs. A set of 216 mesenchymal genes (12%, 216/1,781 genes) was upregulated in the stem-like/mesenchymal subtype colorectal cancer and were predicted to be targeted by the four miRNAs (Supplementary Fig. S1; Supplementary Table S4). Thus, our bioinformatic approach identified differential mRNA and miRNAs and potential target genes in the stem-like/mesenchymal subtype of colorectal cancer in the TCGA.

Figure 1.

Differential miRNA and mRNA profiles in the stem-like/EMT subtype of colorectal cancer (CRC) in TCGA. A, Identification of the differentiated expression miRNA and mRNA signature from the stem-like/EMT colorectal cancer subtype, compared with other colorectal cancer subtypes from TCGA patients cohort. FDR < 0.01. B, Heatmaps showing the differentiated expression genes between stem-like/EMT and other colorectal cancer subtypes in tumors from TCGA dataset. C, The rank list of significance and fold change in key miRNA expression and miRNA putative functional targets in the stem-like/EMT colorectal cancer subtype, compared with other colorectal cancer subtypes (all the miRNAs listed in the table follow the criterion: logFC ≤1,FDR < 0.0015).

Figure 1.

Differential miRNA and mRNA profiles in the stem-like/EMT subtype of colorectal cancer (CRC) in TCGA. A, Identification of the differentiated expression miRNA and mRNA signature from the stem-like/EMT colorectal cancer subtype, compared with other colorectal cancer subtypes from TCGA patients cohort. FDR < 0.01. B, Heatmaps showing the differentiated expression genes between stem-like/EMT and other colorectal cancer subtypes in tumors from TCGA dataset. C, The rank list of significance and fold change in key miRNA expression and miRNA putative functional targets in the stem-like/EMT colorectal cancer subtype, compared with other colorectal cancer subtypes (all the miRNAs listed in the table follow the criterion: logFC ≤1,FDR < 0.0015).

Close modal

Role of miR-506 family in the control of the stem-like/mesenchymal subtype of colorectal cancer

To determine whether the miR-506 family plays a role in the stem-like/mesenchymal subtype of colorectal cancer, we transfected HCT116, a highly invasive colon cancer cell line, with a mixture of miR-506/508/509/514 or control miRNAs, and performed RNA sequencing. The Cytoscape enrichment map network demonstrated that overexpression of the mixed miRNAs dramatically affected the key stem-like/mesenchymal-related gene signatures (Supplementary Fig. S2A–S2C). GSEA enrichment plots showed that the gene signatures of cancer stem cells and the EMT pathways were enriched in colorectal cancer cells with control miRNA transfection (Fig. 2A). Real-time PCR confirmed that the mixed miRNAs increased the expression of epithelial-related genes (CDH1, ZO1) and decreased the expression of mesenchymal and stemness genes (VIM, CDH2, c-MYC, Nanog, ZEB1, and SALL4; Fig. 2B; Supplementary Fig. S2D–S2E). To define the role of each individual miRNA within the miR-506 family, we transfected cells with miR-506, miR-508, miR-509, or miR-514, respectively. Real-time PCR showed that transfection with miR-508 increased the expression of epithelial-related genes and decreased the expression of mesenchymal and stemness genes to the greatest extent (Fig. 2C and D; Supplementary Fig. S2F–S2K). Analogous results were observed in the Matrigel invasion assay (Fig. 2E). These data suggested that miR-508 mediated the strongest inhibition among the four miRNAs. Thus, the miR-506 family, especially miR-508, may regulate the stem-like/mesenchymal signaling activation in colorectal cancer cells. In further support of this possibility, the integrated analysis demonstrated that miR-508 was the most downregulated miRNA in the stem-like/mesenchymal subtype of colorectal cancer (Fig. 1C; Supplementary Table S3). Thus, we focused our studies on miR-508.

Figure 2.

Role of miR-506 family in the control of the stem-like/EMT subtype of colorectal cancer (CRC). A, The representative pathways from GSEA analysis for comparison between RNA-seq data from HCT116 cells transfected with miR-506/508/509/514 mixture and control microRNA. B, Real-time PCR was performed to examine EMT/stemness markers expression in colorectal cancer cell HCT116 after transfection of miR-506/508/509/514 mixture. Results are shown from three independent experiments. C and D, Real-time PCR was performed to examine EMT/stemness markers expression in colorectal cancer cell HCT116 (C) and primary colorectal cancer cells (D), transfected with miR-508. E, Matrigel invasion assay was performed in colorectal cancer cells after transfection of control miRNA, miR-506, miR-508, miR-509, and miR-514, respectively. The invading cell numbers on each filter were counted. Data were plotted in fold change by defining the percentage from control miRNA transfected cells as 100%. N = 3, P < 0.01. Scale bars, 10 μm.

Figure 2.

Role of miR-506 family in the control of the stem-like/EMT subtype of colorectal cancer (CRC). A, The representative pathways from GSEA analysis for comparison between RNA-seq data from HCT116 cells transfected with miR-506/508/509/514 mixture and control microRNA. B, Real-time PCR was performed to examine EMT/stemness markers expression in colorectal cancer cell HCT116 after transfection of miR-506/508/509/514 mixture. Results are shown from three independent experiments. C and D, Real-time PCR was performed to examine EMT/stemness markers expression in colorectal cancer cell HCT116 (C) and primary colorectal cancer cells (D), transfected with miR-508. E, Matrigel invasion assay was performed in colorectal cancer cells after transfection of control miRNA, miR-506, miR-508, miR-509, and miR-514, respectively. The invading cell numbers on each filter were counted. Data were plotted in fold change by defining the percentage from control miRNA transfected cells as 100%. N = 3, P < 0.01. Scale bars, 10 μm.

Close modal

Target genes and role of miR-508 in colorectal cancer stem-like/mesenchymal signal activation

To specifically define the target genes of miR-508, we assessed the Miranda microRNA database and sorted all potential target genes of miR-508. After overlapping these target genes with the 2,235 genes that were related with the stem-like/mesenchymal subtype of colorectal cancer, we found 75 potential target genes for miR-508. In addition, another top 10 genes that might regulate stem-like and EMT signaling activation were incorporated into the candidate target genes using MiRDB and Broad Institute databases. (Fig. 3A). Next, we selected the top 25 genes among the 75 potential target genes according to their miRSVR score (13), and those 10 additional putative miR-508 stem-like/mesenchymal-related target genes for further validation (Fig. 3A). We found that ZEB1, BMI1, and SALL4 were the most significantly downregulated genes by miR-508 in colorectal cancer cells (Supplementary Fig. S3A). Target prediction programs indicated that there were potential specific targets for miR-508 in the seed regions within the 3′ untranslated (UTR) region of ZEB1, BMI1, and SALL4 (Fig. 3B). Luciferase reporter assays demonstrated that miR-508 suppressed the luciferase activity in cells transfected with wild-type ZEB1, BMI1, and SALL4 reporter plasmids, but not with the mutant-type reporter plasmids (Fig. 3C; Supplementary Fig. S3B). ZEB1 can induce EMT by directly inhibiting CDH1 expression (14, 15) or upregulating BMI1 (16, 17). BMI1 promotes invasion and EMT process by downregulating CDH1 (18, 19). SALL4 promotes EMT by activating ZEB1 and BMI1 (20). In line with these observations, we found that miR-508 decreased ZEB1, BMI1, and SALL4 expression, and increased CDH1 expression (Fig. 3D, Fig. 2C; Supplementary Fig. S3C–S3G). These data indicated that miR-508 may inhibit ZEB1, BMI1, and SALL4 expression, and stimulate CDH1 expression, which in turn prevents the colorectal cancer EMT phenotype transition and stemness progression.

Figure 3.

Target genes and role of miR-508 in colorectal cancer (CRC) stem-like/EMT-signal activation. A, The flowchart for miRNA target prediction and experimental validation. B, The predictive interaction position between miR-508 and the seed regions within the 3′UTR region and mutation region of the target genes. C, Luciferase reporter vectors were generated by inserting the wild-type or mutated 3′UTRs of ZEB1, BMI1, and SALL4 into pmirGLO plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with either miR-508 or control miRNA. Cells were harvested for luciferase activity assay. D, Real-time PCR was performed to examine BMI1 expression in HCT116 transfected with miR-508 and control miRNA, respectively. E, Real-time PCR was performed to examine CDH1 expression by overexpression of ZEB1, BMI1, or SALL4 with 3′UTR and combined with miR-508. F, HCT116 cells were transfected with miR-508 or control miRNA. Sphere assay was performed with 400 cells. Scale bars, 100 μm. G and H, Representative data of tumors is shown in NSG model after inoculation with 1 × 105 HCT116 cells transfected with control miRNA lentivirus or miR-508 lentivirus. n = 10. Tumor incidence is concluded in the table (H). I, Representative data of tumors is shown in xenograft model after inoculation with HCT116 cells treated with PBS (control), control miRNA lentivirus, or miR-508 lentivirus. n = 8. J, Tumor volume was measured after lenti-miR-508 virus treatments. The tumor volumes are shown as the means ± SD. n = 8, *, P < 0.05, ANOVA, compared with PBS and HCT116-control lentivirus group. K, The xenograft tumors were separated and weighted at the end of experiments. n = 8, P < 0.01, ANOVA, compared with control group. L, Nude mice were inoculated with HCT116-miR-508 lentivirus, HCT116-control lentivirus, or HCT116-mock cells. The survival curves were calculated in three groups. M, Representative hematoxylin and eosin images of lungs of nude mice at 12 weeks in three groups inoculated with HCT116-miR-508 lentivirus, HCT116-control lentivirus, or HCT116-mock cells, respectively. Scale bars, 100 μm.

Figure 3.

Target genes and role of miR-508 in colorectal cancer (CRC) stem-like/EMT-signal activation. A, The flowchart for miRNA target prediction and experimental validation. B, The predictive interaction position between miR-508 and the seed regions within the 3′UTR region and mutation region of the target genes. C, Luciferase reporter vectors were generated by inserting the wild-type or mutated 3′UTRs of ZEB1, BMI1, and SALL4 into pmirGLO plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with either miR-508 or control miRNA. Cells were harvested for luciferase activity assay. D, Real-time PCR was performed to examine BMI1 expression in HCT116 transfected with miR-508 and control miRNA, respectively. E, Real-time PCR was performed to examine CDH1 expression by overexpression of ZEB1, BMI1, or SALL4 with 3′UTR and combined with miR-508. F, HCT116 cells were transfected with miR-508 or control miRNA. Sphere assay was performed with 400 cells. Scale bars, 100 μm. G and H, Representative data of tumors is shown in NSG model after inoculation with 1 × 105 HCT116 cells transfected with control miRNA lentivirus or miR-508 lentivirus. n = 10. Tumor incidence is concluded in the table (H). I, Representative data of tumors is shown in xenograft model after inoculation with HCT116 cells treated with PBS (control), control miRNA lentivirus, or miR-508 lentivirus. n = 8. J, Tumor volume was measured after lenti-miR-508 virus treatments. The tumor volumes are shown as the means ± SD. n = 8, *, P < 0.05, ANOVA, compared with PBS and HCT116-control lentivirus group. K, The xenograft tumors were separated and weighted at the end of experiments. n = 8, P < 0.01, ANOVA, compared with control group. L, Nude mice were inoculated with HCT116-miR-508 lentivirus, HCT116-control lentivirus, or HCT116-mock cells. The survival curves were calculated in three groups. M, Representative hematoxylin and eosin images of lungs of nude mice at 12 weeks in three groups inoculated with HCT116-miR-508 lentivirus, HCT116-control lentivirus, or HCT116-mock cells, respectively. Scale bars, 100 μm.

Close modal

To determine whether miR-508 functionally targets ZEB1, SALL4, and BMI1, we inserted the 3′-UTRs of ZEB1, SALL4, and BMI1 genes in these genes expression vectors. Overexpression of these vectors blocked miR-508–induced CDH1 upregulation (Fig. 3E; Supplementary Fig. S3H–S3I), and reduced the response to chemotherapy (Supplementary Fig. S3J–S3K). Thus, miR-508 specifically targets the 3′-UTRs of ZEB1, SALL4, and BMI1 gene expression, leading to CDH1 upregulation. Next, we examined the role of miR-508 in colorectal cancer stem-like/EMT signal activation. The results showed that transfection with miR-508 increased epithelial-related gene expression and decreased stemness/EMT gene expression (Supplementary Fig. S3L–S3N). Transfection of antagomiR-508 had the opposite effect (Supplementary Fig. S3O, S3P). In addition, GSEA analysis demonstrated that the gene signatures of cancer stem cells and EMT pathways were enriched in colorectal cancer cells with control miRNA transfection, but not in cells expressing miR-508 (Supplementary Fig. S3Q–S3R). These data suggested that miR-508 modulates stem-like/EMT-signals in colorectal cancer cells. In support of this, Western blotting revealed that miR-508 decreased mesenchymal markers and stemness proteins, and increased epithelial markers (Supplementary Fig. S3S–S3U). In addition, treatment with antago-miR-508 had the opposite effect (Supplementary Fig. S3V–S3W). Furthermore, transfection with miR-508 impaired colorectal cancer sphere formation (Fig. 3F; Supplementary Fig. S3X), while transfection of antagomiR-508 had the opposite effect (Supplementary Fig. S3Y–S3Z). To demonstrate the role of miR-508 in tumorigenic potential in vivo, we injected different amounts of colorectal cancer cells into NOD/Shi-scid/IL-2Rγnull (NSG) mice. miR-508 administration abrogated tumor formation (Fig. 3G–H; Supplementary Fig. S3A). In addition, overexpression of miR-508 reduced tumor growth (Fig. 3I and J) and tumor weight (Fig. 3K) in nude mouse models. Furthermore, in a colorectal cancer metastatic model, comprising mice inoculated with miR-508–expressing tumor cells, showed longer survival than the mice that received parental tumor cells (Fig. 3L). There were fewer metastatic foci in the lungs of the nude mice that received the miR-508 lentivirus than in the mice that received the control virus (Fig. 3M). These data suggested that miR-508 might be a tumor repressive gene in colorectal cancer that controls cancer stemness and EMT-related pathways.

Associations of miR-508 expression with the stem-like/EMT and stromal genes in colorectal cancer

To investigate the clinical correlation between miR-508 and stem-like/EMT signatures in colorectal cancer, we collected colorectal cancer tissues and normal colorectal tissues adjacent to cancer lesions from 90 patients in Renji Hospital (Supplementary Table S5). We performed in situ hybridization (ISH) for miR-508 and IHC staining for ZEB1, SALL4, BMI1, CDH1, and VIM in these tissues. We found that the levels of miR-508 and CDH1 were lower in colorectal cancer tissues compared with that in normal tissues, while the levels of ZEB1, SALL4, and BMI1 were higher in colorectal cancer tissues (Fig. 4A; Supplementary Fig. S4A). To confirm these findings, we quantified the expression of miR-508 and EMT/Stem genes (CDH1, VIM, ZEB1, SALL4, and BMI1) in Renji cohort 1 tissues. Consistent with our model, we found that the samples expressing high miR-508 tended to have high expression of the epithelial marker, CDH1, and low expression of the EMT marker, VIM, and stemness markers ZEB1 and SALL4 (Fig. 4B). miR-508 expression was negatively correlated with ZEB1, SALL4, BMI1, and VIM expression and was positively correlated with CDH1 expression (Fig. 4C). These data indicated that miR-508 expression correlates with stem-like/EMT signatures in colorectal cancer.

Figure 4.

Associations of miR-508 expression with the stem-like/EMT and stroma genes in colorectal cancer (CRC). A, Expression of miR-508 in colorectal cancer tissues and paired normal tissues using ISH analysis. IHC staining showed the expression of ZEB1, BMI1, SALL4, VIM, and CDH1 in the colorectal tissues in Renji cohort 1, n = 90. Scale bars, 100 μm. B, miR-508, ZEB1, BMI1, SALL4, VIM, and CDH1 scores in colorectal cancer tissues (n = 90) and paired normal colorectal tissues (n = 90) in Renji cohort 1. C, The correlation between miR-508 expression and EMT/stemness markers (CDH1, VIM, ZEB1, SALL4, and BMI1) of paired normal colorectal tissues and colorectal cancer tissues in Renji cohort 1 (n = 90). The shaded squares in the top right show Spearman correlation values between the indicated genes. Blue and red colors, negative or positive correlation, respectively. The lower left squares show the scatter plot matrix and fitted trend lines for the same comparisons. **, correlation is significant at the 0.01 level. D, Representative images of miR-508, FAP, and CALD1 expression in mesenchymal and nonmesenchymal colorectal cancer tissues using ISH analysis in Renji cohort 2. n = 100. Scale bars, 100 μm. E–F, Comparison of the miR-508 ISH scores and the CALD1 and FAP IHC scores in colorectal cancer tissues. The correlations are shown in European cohort (n = 128; E) and Renji cohort 2 (n = 100; Fisher exact test; F).

Figure 4.

Associations of miR-508 expression with the stem-like/EMT and stroma genes in colorectal cancer (CRC). A, Expression of miR-508 in colorectal cancer tissues and paired normal tissues using ISH analysis. IHC staining showed the expression of ZEB1, BMI1, SALL4, VIM, and CDH1 in the colorectal tissues in Renji cohort 1, n = 90. Scale bars, 100 μm. B, miR-508, ZEB1, BMI1, SALL4, VIM, and CDH1 scores in colorectal cancer tissues (n = 90) and paired normal colorectal tissues (n = 90) in Renji cohort 1. C, The correlation between miR-508 expression and EMT/stemness markers (CDH1, VIM, ZEB1, SALL4, and BMI1) of paired normal colorectal tissues and colorectal cancer tissues in Renji cohort 1 (n = 90). The shaded squares in the top right show Spearman correlation values between the indicated genes. Blue and red colors, negative or positive correlation, respectively. The lower left squares show the scatter plot matrix and fitted trend lines for the same comparisons. **, correlation is significant at the 0.01 level. D, Representative images of miR-508, FAP, and CALD1 expression in mesenchymal and nonmesenchymal colorectal cancer tissues using ISH analysis in Renji cohort 2. n = 100. Scale bars, 100 μm. E–F, Comparison of the miR-508 ISH scores and the CALD1 and FAP IHC scores in colorectal cancer tissues. The correlations are shown in European cohort (n = 128; E) and Renji cohort 2 (n = 100; Fisher exact test; F).

Close modal

High expression of the stromal signatures is abundant in the stem-like/mesenchymal colorectal cancer subtype compared with that in other colorectal cancer subtypes (21). We next evaluated the correlation between miR-508 and the stromal signatures in our stem-like/mesenchymal colorectal cancer subtype cohort. Caldesmon 1 (CALD1) and fibroblast activation protein (FAP) are the two key stromal signature genes (5). We performed ISH for miR-508 and IHC for CALD1 and FAP in two independent colorectal cancer cohorts, Renji cohort 2 (Supplementary Table S6) and the European cohort (Supplementary Table S7). ISH showed that miR-508 expression was detectable in tumor tissues with low-stromal contents, but not with high-stromal contents (Fig. 4D). CALD1 and FAP were exclusively expressed in stromal tumor tissues of Renji cohort 2 and the European cohort (Fig. 4D). Contingency tables demonstrated a negative correlation between the miR-508 ISH scores and the CALD1 and FAP IHC scores in colorectal cancer tissues (Fig. 4E and F). Thus, miR-508 expression is highly associated with the stem-like/EMT and stromal gene signatures in colorectal cancer.

Involvement of miR-508 in the regulation of TGFβ-mediated EMT in colorectal cancer

TGFβ is a robust inducer of EMT in multiple cell types (22). We wondered whether miR-508 was involved in the regulation of TGFβ-mediated EMT. Our bioinformatics pathway analysis revealed that the TGFβ pathway was activated in the stem-like/mesenchymal colorectal cancer subtype (Fig. 5A). Colorectal cancer cells were transfected with miR-508 or a control miRNA and were subsequently treated with TGFβ. In the control group, TGFβ treatment increased ZEB1, BMI1, and SALL4 mRNA levels and decreased CDH1 mRNA expression. In contrast, miR-508 overexpression abolished this effect (Fig. 5B and C; Supplementary Fig. S5A) and blocked TGFβ-induced invasion (Fig. 5D and E). These data indicated that miR-508 might participate in the prevention of TGFβ-induced EMT in colorectal cancer cells. Consistently, immunofluorescence staining showed that colorectal cancer cells transfected with the control miRNA underwent a morphologic change characterized by elongated cells in response to TGFβ treatment (Fig. 5F and G). In contrast, miR-508 transfection blocked the TGFβ-induced EMT change and promoted an epithelial phenotype characterized by CDH1 localization at cell–cell junctions and cortical F-actin staining (Fig. 5F and G).

Figure 5.

Involvement of miR-508 in the regulation of TGFβ-mediated EMT in colorectal cancer. A, Heatmap showing the expression of major components of the TGFβ pathway. The color bar at the top of the panel indicates mesenchymal (red) and other (blue) subtypes. The color of the square corresponds to the relative expression among cases (red, higher 10% expression; green, lower 10% expression). B and C, Real-time PCR of ZEB1, BMI1, SALL4, and CDH1 in HCT116 (B) and SW480 (C) cells transfected with miR-508 or control (control miRNA). The cells were cultured in serum-free medium containing TGFβ for 48 hours. Data are shown as the means of triplicate PCR results relative to the mRNA levels in colorectal cancer cells. D and E, Matrigel invasion assay was performed in TGFβ-treated HCT116 (D) and SW480 (E) cells, transfected with miR-508 or control miRNA. The invading cell numbers on each filter were counted. Data are shown as the means of triplicate results. F–G, Inverse phase microscopy (left; scale bars, 100 μm) and immunofluorescence staining of CDH1 and F-actin (right; scale bars, 10 μm) in HCT116 (F) and SW480 cells (G), respectively, transfected with miR-508 or control (control miRNA) for 72 hours. Cell nuclei were stained with DAPI. Data are from three independent experiments. H, IHC staining showing the expression of p-SMAD2 in high-stromal content/EMT and non-EMT colorectal cancer tissues. Scale bars, 100 μm. I and J, Comparison of the miR-508 ISH scores and the p-SMAD2 IHC scores in colorectal cancer tissues. The correlations are shown in European cohort (n = 128; I) and Renji cohort 2 (n = 100; Fisher exact test; J).

Figure 5.

Involvement of miR-508 in the regulation of TGFβ-mediated EMT in colorectal cancer. A, Heatmap showing the expression of major components of the TGFβ pathway. The color bar at the top of the panel indicates mesenchymal (red) and other (blue) subtypes. The color of the square corresponds to the relative expression among cases (red, higher 10% expression; green, lower 10% expression). B and C, Real-time PCR of ZEB1, BMI1, SALL4, and CDH1 in HCT116 (B) and SW480 (C) cells transfected with miR-508 or control (control miRNA). The cells were cultured in serum-free medium containing TGFβ for 48 hours. Data are shown as the means of triplicate PCR results relative to the mRNA levels in colorectal cancer cells. D and E, Matrigel invasion assay was performed in TGFβ-treated HCT116 (D) and SW480 (E) cells, transfected with miR-508 or control miRNA. The invading cell numbers on each filter were counted. Data are shown as the means of triplicate results. F–G, Inverse phase microscopy (left; scale bars, 100 μm) and immunofluorescence staining of CDH1 and F-actin (right; scale bars, 10 μm) in HCT116 (F) and SW480 cells (G), respectively, transfected with miR-508 or control (control miRNA) for 72 hours. Cell nuclei were stained with DAPI. Data are from three independent experiments. H, IHC staining showing the expression of p-SMAD2 in high-stromal content/EMT and non-EMT colorectal cancer tissues. Scale bars, 100 μm. I and J, Comparison of the miR-508 ISH scores and the p-SMAD2 IHC scores in colorectal cancer tissues. The correlations are shown in European cohort (n = 128; I) and Renji cohort 2 (n = 100; Fisher exact test; J).

Close modal

The TGFβ pathway is activated in the stem-like/mesenchymal colorectal cancer subtype and SMAD2 is a key TGFβ signaling mediator (5). We performed IHC with anti-phosphorylated (p)-SMAD2 antibodies on colorectal cancer tissues. The IHC data showed that p-SMAD2 was mainly expressed in epithelial cells. Interestingly, the expression of p-SMAD2 was higher in high-stromal tumors than in low-stromal tumors (Fig. 5H). Contingency tables demonstrated an inverse correlation between the miR-508 ISH scores and the p-SMAD2 IHC scores in colorectal cancer samples (Fig. 5I and J). Thus, miR-508 regulates TGFβ-mediated EMT in colorectal cancer.

Correlation between miR-508 and survival in colorectal cancer

To assess the impact of miR-508 on the outcome of patients with colorectal cancer, we analyzed the effect of miR-508 on patient survival in three independent cohorts. In Renji cohort 1, the cumulative survival was longer in patients with tumors expressing high levels of miR-508 than those with low levels of miR-508 (Fig. 6A). Multivariate analysis revealed that high miR-508 expression independently predicted improved outcomes in patients with colorectal cancer (Fig. 6B; Supplementary Fig. S6A). Similar results were observed in the European cohort (Fig. 6C) and Renji cohort 2 (Fig. 6D). These data indicated that miR-508 expression might affect patient outcome and represent a new prognostic factor in patients with colorectal cancer.

Figure 6.

Association between miR-508 and survival in colorectal cancer. A, Survival was analyzed and compared between patients with high and low miR-508 expression in Renji cohort 1 (n = 90; P = 0.002; HR = 0.37; 95% CI: 0.19–0.71; Mantel–Cox test). B, Multivariable analysis was performed in Renji cohort 1 (n = 90). C and D, Survival was analyzed and compared between patients with high and low miR-508 expression in European cohort (n = 128; P = 1.1e−06; HR = 0.22; 95% CI: 0.12 − 0.43; Log-rank test) and Renji cohort 2 (n = 100; P = 0.0018; HR = 0.42; 95% CI: 0.24 − 0.74; Log-rank test).

Figure 6.

Association between miR-508 and survival in colorectal cancer. A, Survival was analyzed and compared between patients with high and low miR-508 expression in Renji cohort 1 (n = 90; P = 0.002; HR = 0.37; 95% CI: 0.19–0.71; Mantel–Cox test). B, Multivariable analysis was performed in Renji cohort 1 (n = 90). C and D, Survival was analyzed and compared between patients with high and low miR-508 expression in European cohort (n = 128; P = 1.1e−06; HR = 0.22; 95% CI: 0.12 − 0.43; Log-rank test) and Renji cohort 2 (n = 100; P = 0.0018; HR = 0.42; 95% CI: 0.24 − 0.74; Log-rank test).

Close modal

Hypoxia-mediated cross-talk between AK000053 and miR-508 in colorectal cancer

Many tumors display an anaerobic metabolism (23, 24). We hypothesized that hypoxia caused major genomic loss of miR-508 in the stem-like/mesenchymal subtype of colorectal cancer. To test this hypothesis, colorectal cancer cells were exposed to hypoxia culture conditions. Real-time PCR showed that mature miR-508 (Fig. 7A), but not pri-miR-508 (Supplementary Fig. S7A), was significantly downregulated in response to hypoxia. This result suggested that hypoxia might induce the downregulation of miR-508 via posttranscriptional regulation. We then examined whether hypoxia caused miR-508 loss in a HIF-1α–dependent manner. Real-time PCR data showed that knockdown of HIF-1α dramatically increased miR-508 levels (Fig. 7B), but not pri-miR-508 levels (Supplementary Fig. S7B–S7C). Accordingly, treatment with 2-Methoxyestradiol (2-ME), an HIF-1a inhibitor, induced miR-508 expression (Fig. 7C).

Figure 7.

Hypoxia-mediated cross-talk between AK000053 and miR-508 in colorectal cancer. A, Real-time PCR of mature miR-508 was measured in HCT116 and SW480 cells. The cells were cultured in hypoxia condition for 24 hours. B and C, Real-time PCR was performed to examine mature miR-508 expression in HCT116 cells transfected with HIF1α siRNA (B) or treated with HIF1α inhibitor 2-ME (10 μmol/L; C). D, Real-time PCR was conducted to confirm consequence expression of AK000053 and ENST00000437308.1 transfected with HIF1α siRNA. E, Luciferase reporter vectors were generated by inserting the promoter region of AK000053 lncRNA into pGL3-basic plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with HIF1α siRNA. Cells were harvested for luciferase activity assay. F, Real-time PCR of mature miR-508 in HCT116 cells treated with control siRNA, HIF1α siRNA, or cotransfected HIF1α siRNA and AK000053 lncRNA overexpression plasmid. G, Real-time PCR of mature miR-508 in HCT116 cells transfected with AK000053 lncRNA siRNA or overexpression plasmid. H, Luciferase reporter vectors were generated by inserting the wild-type or mutated interaction region of AK000053 lncRNA with miR-508 into pmirGLO plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with either control miRNA or miR-508. Cells were harvested for luciferase activity assay. I, Real-time PCR of miR-508 in HCT116 cells from WT-AK000053 lncRNA or MU-AK000053 lncRNA pull-down assays. J, RIP-qPCR showing interaction between AGO2 and AK000053 lncRNA in HCT116 miR-508KO cells transfected with miR-508. n = 3 samples; data are shown with SEM. K and L, Survival was analyzed and compared between patients with high and low AK000053 expression in Renji cohort 2 (n = 100; P = 5.6e-13; HR = 7.57; 95% CI: 4.03 – 14.22; log-rank test) and European cohort [n = 128; P = 6.7e–06; HR = 4.56 (2.24–9.31); log-rank test]. M, Representative images of miR-508 and AK000053 expression in mesenchymal and nonmesenchymal colorectal cancer tissues using ISH analysis in Renji cohort 2. n = 100. Scale bars, 100 μm. N and O, Comparison of the miR-508 ISH scores and the AK000053 lncRNA scores in colorectal cancer tissues. The correlations are shown in Renji cohort 2 (n = 100; N) and European cohort (n = 128; Fisher exact test; O). P, Schematic representation for the mechanism of AK000053 lncRNA/miR-508 axis as a switch that regulates EMT/stemness in human colorectal cancer progression by targeting the ZEB1/BMI1/SALL4 network.

Figure 7.

Hypoxia-mediated cross-talk between AK000053 and miR-508 in colorectal cancer. A, Real-time PCR of mature miR-508 was measured in HCT116 and SW480 cells. The cells were cultured in hypoxia condition for 24 hours. B and C, Real-time PCR was performed to examine mature miR-508 expression in HCT116 cells transfected with HIF1α siRNA (B) or treated with HIF1α inhibitor 2-ME (10 μmol/L; C). D, Real-time PCR was conducted to confirm consequence expression of AK000053 and ENST00000437308.1 transfected with HIF1α siRNA. E, Luciferase reporter vectors were generated by inserting the promoter region of AK000053 lncRNA into pGL3-basic plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with HIF1α siRNA. Cells were harvested for luciferase activity assay. F, Real-time PCR of mature miR-508 in HCT116 cells treated with control siRNA, HIF1α siRNA, or cotransfected HIF1α siRNA and AK000053 lncRNA overexpression plasmid. G, Real-time PCR of mature miR-508 in HCT116 cells transfected with AK000053 lncRNA siRNA or overexpression plasmid. H, Luciferase reporter vectors were generated by inserting the wild-type or mutated interaction region of AK000053 lncRNA with miR-508 into pmirGLO plasmid. The reporter vectors were then cotransfected into colorectal cancer cells with either control miRNA or miR-508. Cells were harvested for luciferase activity assay. I, Real-time PCR of miR-508 in HCT116 cells from WT-AK000053 lncRNA or MU-AK000053 lncRNA pull-down assays. J, RIP-qPCR showing interaction between AGO2 and AK000053 lncRNA in HCT116 miR-508KO cells transfected with miR-508. n = 3 samples; data are shown with SEM. K and L, Survival was analyzed and compared between patients with high and low AK000053 expression in Renji cohort 2 (n = 100; P = 5.6e-13; HR = 7.57; 95% CI: 4.03 – 14.22; log-rank test) and European cohort [n = 128; P = 6.7e–06; HR = 4.56 (2.24–9.31); log-rank test]. M, Representative images of miR-508 and AK000053 expression in mesenchymal and nonmesenchymal colorectal cancer tissues using ISH analysis in Renji cohort 2. n = 100. Scale bars, 100 μm. N and O, Comparison of the miR-508 ISH scores and the AK000053 lncRNA scores in colorectal cancer tissues. The correlations are shown in Renji cohort 2 (n = 100; N) and European cohort (n = 128; Fisher exact test; O). P, Schematic representation for the mechanism of AK000053 lncRNA/miR-508 axis as a switch that regulates EMT/stemness in human colorectal cancer progression by targeting the ZEB1/BMI1/SALL4 network.

Close modal

We next explored the potential mechanism by which hypoxia resulted in miR-508 loss. LncRNAs may function as "miRNA sponges" to compete and degrade their targeted miRNAs (25) at the posttranscriptional level. To dissect whether hypoxia modulates miR-508 expression in a "miRNA sponge"-dependent manner, we analyzed and compared the lncRNA expression profiles in HCT116 cells under normoxic and hypoxic culture conditions. Stringent filtering criteria were applied to identify high levels of lncRNA changes under hypoxic conditions (Supplementary Fig. S7D). This filter screen identified 35 upregulated and 29 downregulated lncRNAs (Supplementary Fig. S7E). To determine which lncRNAs interact with miR-508, we searched for interaction targets between miR-508 with the 35 upregulated lncRNAs in the FINDTAR3 database (http://www.tsinghua.edu.cn/publish/newthu/index.html). We found four candidate lncRNAs, which contained at least two interaction targets with miR-508 (Supplementary Fig. S7F–S7I). Real-time PCR revealed that ENST00000437308.1 and AK000053 were significantly increased in hypoxic compared with normoxic conditions (Supplementary Fig. S7J). AK000053 was the most increased lncRNA candidate in response to hypoxia treatment and there were more predicted binding sites for miR-508 in AK000053 than in ENST00000437308.1. Therefore, we focused our research on AK000053.

To determine whether AK000053.1 is a novel lncRNA, we first analyzed its sequence. Northern blotting revealed that the size of AK000053.1 was approximately 2 kb in colorectal cancer cell lines, which is consistent with the predicted lncRNA length in the UCSC genome database (Supplementary Fig. S7K). Analysis of the AK000053.1 sequences using ORF Finder from the National Center for Biotechnology Information failed to predict any known human protein (Supplementary Fig. S7L). Furthermore, we performed a codon substitution frequency analysis using PhyloCSF. AK000053.1 had a very low codon substitution frequency score (−905.7; Supplementary Fig. S7M). Thus, AK000053.1 is unlikely to encode a protein product. In further support of this, the nuclear and cytoplasm fractions of HCT116 and SW480 colorectal cancer cells were separated to perform real-time PCR. AK000053.1 was mainly located in the cytoplasm of colorectal cancer cells (Supplementary Fig. S7N). Collectively, AK000053.1 is a novel lncRNA that is regulated by hypoxia in colorectal cancer cells.

As hypoxia induced AK000053 expression, we wondered whether this regulation is mediated by the HIF-1α pathway. To address this question, colorectal cancer cells were exposed to hypoxia for 24 hours after HIF-1α siRNA transfection. Knockdown of HIF-1α (Fig. 7D) or treatment with the HIF-1α inhibitor 2-ME (Supplementary Fig. S7O) significantly reduced AK000053 expression. The dual luciferase reporter assay further confirmed that HIF-1α could regulate AK000053 at the transcriptional level (Fig. 7E; Supplementary Fig. S7P). Given that HIF-1α oppositely regulates AK000053 and miR-508, we manipulated HIF-1α and observed that transfection with an AK000053 overexpression plasmid attenuated the high levels of miR-508 expression induced by the HIF-1α siRNA (Fig. 7F). Taken together, these data demonstrated that hypoxia downregulates the expression of miR-508 and upregulates the expression of a novel lncRNA, AK000053, in colorectal cancer cells in a HIF-1α–dependent manner.

We hypothesized that AK000053 lncRNA target miR-508, resulting in its loss in colorectal cancers. Knockdown of AK000053 increased mature miR-508 expression and its overexpression had the opposite effect (Fig. 7G). To demonstrate the direct interaction between AK000053 and miR-508, we constructed a luciferase reporter plasmid encoding the two excellent binding sites of AK000053 in the 3′ UTR of the luciferase gene. Luciferase assays showed that the miR-508 significantly reduced the luciferase activity in comparison with the control miRNA (Fig. 7H). To determine the direct physical association between AK000053 and miR-508, we performed RNA pull-down experiments and found that the wild-type AK000053 was enriched with more miR-508 than the mutant sequence (Fig. 7I). Given that lncRNA–miRNA target pairs can be purified by immunoprecipitation of the RISC component AGO2 (26), we tested whether AGO2 would coprecipitate with AK000053 in HCT116miR-508KO and HT29miR-508KO cells. RNA immunoprecipitation (RIP) followed by quantitative PCR (qPCR) confirmed the interaction between AK000053 and AGO2 in the presence of miR-508 (Fig. 7J; Supplementary Fig. S7Q–S7S). Collectively, these data indicated that mature miR-508 binds directly to and may guide RISC to its target AK000053. We further found that AK000053 can competitively interact with miR-508 and negatively regulate the expression of miR-508 and miR-508–targeted genes (Supplementary Fig. S7T–S7X).

miR-508 affects colorectal cancer pathology and patient outcome (Fig. 6A–D). Analogously, we assessed the impact of AK000053 on the outcome of patients with colorectal cancer. Cumulative survival was poorer in patients with high levels of AK000053 (Fig. 7K and L). ISH data further showed that miR-508 expression was detectable in tumor tissues with low-stromal contents, but not with high-stromal contents (Fig. 7M). Interestingly, AK000053 was exclusively expressed in stromal tumor tissues (Fig. 7M). Contingency tables demonstrated a negative correlation between the miR-508 and the AK000053 ISH scores (Fig. 7N and O). These data indicated that AK000053 negatively correlates with miR-508 and predicts poor clinical outcome in patients with colorectal cancer.

Using integrated bioinformatics, genetic, functional, and clinical approaches, we revealed a crucial and novel lncRNA–miRNA–mRNA regulatory network that controls the phenotype of the stem-like/mesenchymal subtype of colorectal cancer.

To date, pathologic classification is the most practical tool to assess prognosis and manage the treatment of colorectal cancer. Cancer gene signatures may be important for colorectal cancer diagnosis (27). Efforts from systematic studies have allowed comprehensive molecular classification of cancer by integrated analyses of multidimensional data from a large cohort of clinical samples, at both the genomic and epigenetic levels (3, 4, 28, 29). At the basis of the molecular classification, colorectal cancer is grouped into the stem-like/mesenchymal subtype and other subtypes (4). However, the driving force of the stemness and EMT subtype of colorectal cancer remains poorly understood. We performed an integrated analysis and compared the miRNA and mRNA expression patterns in the stem-like/mesenchymal and other subtypes of colorectal cancer. Our bioinformatics analysis suggested a role for the miR-506 family, consisting of miR-506, miR-508, miR-509, and miR-514, in defining the phenotype of the stem-like/mesenchymal subtype of colorectal cancer. These four miRNAs are located in Xq27.3, a fragile site of the human X chromosome (30). Our genetic, biological, and clinical studies have determined the importance of the miR-508 family in shaping the colorectal cancer phenotype.

We focused our functional studies on miR-508. miR-508 is the most significantly reduced miRNA in the miRNA networks in the stem-like/mesenchymal subtype compared with other subtypes of colorectal cancer. Furthermore, miR-508 is the most efficient miRNA in terms of inhibiting colorectal cancer invasion and stemness. In line with our observations, reduced miR-508 expression has been reported in prostate cancer (31) and ovarian serous carcinoma (32). Furthermore, miR-508 may regulate multidrug resistance by targeting ABCB1 and ZNRD1 in gastric cancer (33). These observations suggest that miR-508 may be broadly involved in cancer biology.

We have observed that miR-508 simultaneously and directly targets multiple EMT and stemness genes including ZEB1, BMI1, and SALL4, and inhibits the colorectal cancer EMT and stemness process. ZEB1 is one of the two zinc finger transcription factors of the ZEB family and is a major regulator of EMT (34). The SNAI and ZEB families can repress CDH1 expression (35, 36). TGFβ induces an EMT-like process (37). Interestingly, ZEB1 participates in TGFβ-induced CDH1 repression (38). The Polycomb-group protein, BMI1, induces EMT in human nasopharyngeal epithelial cells and the expression of BMI1 is positively regulated by ZEB1 (39, 40). BMI1 may regulate ZEB1 expression in pancreatic cancer (16). In vivo RNAi screening indicated that BMI1 is important in the cellular response to the TGFβ and BMP pathways in neural-and glioblastoma stem-like cells (41). In addition, the zinc finger transcription factor SALL4 is a biomarker and therapeutic target for aggressive cancer (42). SALL4 may act as a cell dispersion factor to promote breast cancer EMT by upregulating ZEB1 and BMI1 expression (20). Thus, there is a potential interactive network among SALL4, BMI1, and ZEB1 in the regulation of cancer stemness and EMT. Our data showed that miR-508 directly targets these genes and indirectly regulates their expression and function with TGFβ. Thus, miR-508 is a key node in this network in colorectal cancer.

In addition to the stem cell-like/EMT traits, an elevated tumor stromal signature is an additional feature of the stem-like/mesenchymal subtype of colorectal cancer (5, 21). High stromal content might contribute to the cancer stem-like/EMT process and TGFβ signal activation (5). Interestingly, we found that miR-508 expression correlates negatively stromal signature expression and TGFβ activation in the TCGA dataset and in our patient population. Furthermore, we observed that miR-508 is a potent inhibitor of TGFβ-induced EMT by directly targeting ZEB1, BMI1, and SALL4. Hence, our data indicated that miR-508 not only regulates epithelial cancer cell progression, but also participates in the tumor environmental stromal alteration. Therefore, miR-508 is involved in defining the two key features of the stem-like/mesenchymal colorectal cancer subtype: high stem-like/mesenchymal signaling pattern and a potent stromal signature.

We explored the mechanism controlling the major genomic loss of miR-508 in the stem cell-like/mesenchymal colorectal cancer subtype. Hypoxia might regulate and maintain the stem cell-like/mesenchymal cancer phenotype (43). In the course of exploring the potential relationship between hypoxia, miR-508, and stem-like/EMT gene signatures, we identified a novel lncRNA, AK000053. AK000053 fits all the criteria for lncRNA definition. The location, expression, regulation, and biological activity of AK000053 have not been reported to date. Interestingly, we found that hypoxia orchestrates the interaction between AK000053 and miR-508 in an HIF-1α–dependent manner to define the stem-like/mesenchymal colorectal cancer subtype. This hypothesis is supported by several lines of experimental evidence: (i) hypoxia induces AK000053 expression and causes a genomic loss of mature miR-508, but not pri-miR-508; (ii) AK000053 contains five predicted binding sites for miR-508. RIP followed by qPCR assays confirmed that mature miR-508 and lncAK000053 might generate target pairs, and directly interact with AGO2. AK000053 can be upregulated then attract miR-508 to generate more target pairs and enter the RISC complex, interacting with AGO2 and triggering the miRNA degradation pathway under hypoxic conditions; (iii) HIF-1α may reduce the expression of miR-508 in a posttranscriptional manner by inducing AK000053 expression, and HIF-1α may bind and increase the promoter transcriptional activity of lncRNA AK000053 under hypoxic condition in colorectal cancer cells; (iv) AK000053 can physically interact with miR-508 and acts as a molecular sponge to determine the availability of miR-508 to support the expression of the stem-like/EMT genes; (vi) manipulation of AK000053 and miR-508 consistently resulted in alteration of the stem-like/EMT properties in vitro and in vivo; and (vii) the network of AK000053 and miR-508 is associated with patient outcome.

In summary, our integrated and comprehensive study identified a novel regulatory network of lncRNA–miRNA–mRNA, namely, AK000053-miR-508-EMT/stem-like genes in defining the stem-like/mesenchymal colorectal cancer subtype (Fig. 7P). Targeting this network may be therapeutically meaningful to treat patients with colorectal cancer.

No potential conflicts of interest were disclosed.

Conception and design: W. Zou, H.-Y. Chen, J. Hong, J.-Y. Fang

Development of methodology: J. Hong

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.-T. Yan, L.-L. Ren, Z.-H. Wang, Y.-N. Yu, J.-Y. Tang, W. Zgodzinski, M. Majewski, P. Radwan, I. Kryczek, M. Zhong, J. Chen, Q. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.-T. Yan, L.-L. Ren, C.-Q. Shen, Q. Liang, J.-Y. Tang, Y.-X. Chen, D.-F. Sun, W. Zgodzinski, M. Majewski, W. Zou, H.-Y. Chen

Writing, review, and/or revision of the manuscript: W. Zgodzinski, M. Majewski, W. Zou, H.-Y. Chen, J. Hong, J.-Y. Fang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.-N. Yu, D.-F. Sun, W. Zou

Study supervision: W. Zou, H.-Y. Chen, J. Hong, J.-Y. Fang

This work was supported by grants from the National Natural Science Foundation (81320108024, 81421001, 81530072, 81522008, 81790632, 31371420, 31371273, 81372267, 81402347); the Shanghai Natural Science Foundation (grant no. 13ZR14244000); the Program for Professor of Special Appointment (Eastern Scholar no. 201268 and 2015 Youth Eastern Scholar no. QD2015003) at Shanghai Institutions of Higher Learning; Shanghai Municipal Education Commission—Gaofeng Clinical Medicine Grant Support (no. 20152512, 20161309) and Chenxing Project of Shanghai Jiao-Tong University (to H.-Y. Chen and J. Hong).

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.

1.
Thiery
JP
,
Acloque
H
,
Huang
RY
,
Nieto
MA
. 
Epithelial-mesenchymal transitions in development and disease
.
Cell
2009
;
139
:
871
90
.
2.
Mani
SA
,
Guo
W
,
Liao
MJ
,
Eaton
EN
,
Ayyanan
A
,
Zhou
AY
, et al
The epithelial-mesenchymal transition generates cells with properties of stem cells
.
Cell
2008
;
133
:
704
15
.
3.
Guinney
J
,
Dienstmann
R
,
Wang
X
,
de Reynies
A
,
Schlicker
A
,
Soneson
C
, et al
The consensus molecular subtypes of colorectal cancer
.
Nat Med
2015
;
21
:
1350
6
.
4.
Sadanandam
A
,
Lyssiotis
CA
,
Homicsko
K
,
Collisson
EA
,
Gibb
WJ
,
Wullschleger
S
, et al
A colorectal cancer classification system that associates cellular phenotype and responses to therapy
.
Nat Med
2013
;
19
:
619
25
.
5.
Calon
A
,
Lonardo
E
,
Berenguer-Llergo
A
,
Espinet
E
,
Hernando-Momblona
X
,
Iglesias
M
, et al
Stromal gene expression defines poor-prognosis subtypes in colorectal cancer
.
Nat Genet
2015
;
47
:
320
9
.
6.
van Engeland
M
,
Derks
S
,
Smits
KM
,
Meijer
GA
,
Herman
JG
. 
Colorectal cancer epigenetics: complex simplicity
.
J Clin Oncol
2011
;
29
:
1382
91
.
7.
Meng
X
,
Wu
J
,
Pan
C
,
Wang
H
,
Ying
X
,
Zhou
Y
, et al
Genetic and epigenetic down-regulation of microRNA-212 promotes colorectal tumor metastasis via dysregulation of MnSOD
.
Gastroenterology
2013
;
145
:
426
36
e1–6.
8.
Song
SJ
,
Poliseno
L
,
Song
MS
,
Ala
U
,
Webster
K
,
Ng
C
, et al
MicroRNA-antagonism regulates breast cancer stemness and metastasis via TET-family-dependent chromatin remodeling
.
Cell
2013
;
154
:
311
24
.
9.
Lim
LP
,
Lau
NC
,
Garrett-Engele
P
,
Grimson
A
,
Schelter
JM
,
Castle
J
, et al
Microarray analysis shows that some microRNAs downregulate large numbers of target mRNAs
.
Nature
2005
;
433
:
769
73
.
10.
Kryczek
I
,
Lin
Y
,
Nagarsheth
N
,
Peng
D
,
Zhao
L
,
Zhao
E
, et al
IL-22(+)CD4(+) T cells promote colorectal cancer stemness via STAT3 transcription factor activation and induction of the methyltransferase DOT1L
.
Immunity
2014
;
40
:
772
84
.
11.
Merico
D
,
Isserlin
R
,
Stueker
O
,
Emili
A
,
Bader
GD
. 
Enrichment map: a network-based method for gene-set enrichment visualization and interpretation
.
PLoS One
2010
;
5
:
e13984
.
12.
Yuan
JH
,
Yang
F
,
Wang
F
,
Ma
JZ
,
Guo
YJ
,
Tao
QF
, et al
A long noncoding RNA activated by TGF-beta promotes the invasion-metastasis cascade in hepatocellular carcinoma
.
Cancer Cell
2014
;
25
:
666
81
.
13.
Betel
D
,
Koppal
A
,
Agius
P
,
Sander
C
,
Leslie
C
. 
Comprehensive modeling of microRNA targets predicts functional non-conserved and non-canonical sites
.
Genome Biol
2010
;
11
:
R90
.
14.
Xiong
H
,
Hong
J
,
Du
W
,
Lin
YW
,
Ren
LL
,
Wang
YC
, et al
Roles of STAT3 and ZEB1 proteins in E-cadherin down-regulation and human colorectal cancer epithelial-mesenchymal transition
.
J Biol Chem
2012
;
287
:
5819
32
.
15.
Lee
JY
,
Park
MK
,
Park
JH
,
Lee
HJ
,
Shin
DH
,
Kang
Y
, et al
Loss of the polycomb protein Mel-18 enhances the epithelial-mesenchymal transition by ZEB1 and ZEB2 expression through the downregulation of miR-205 in breast cancer
.
Oncogene
2014
;
33
:
1325
35
.
16.
Proctor
E
,
Waghray
M
,
Lee
CJ
,
Heidt
DG
,
Yalamanchili
M
,
Li
C
, et al
Bmi1 enhances tumorigenicity and cancer stem cell function in pancreatic adenocarcinoma
.
PLoS One
2013
;
8
:
e55820
.
17.
Liu
Y
,
Sanchez-Tillo
E
,
Lu
X
,
Huang
L
,
Clem
B
,
Telang
S
, et al
The ZEB1 transcription factor acts in a negative feedback loop with miR200 downstream of Ras and Rb1 to regulate Bmi1 expression
.
J Biol Chem
2014
;
289
:
4116
25
.
18.
Zhang
Z
,
Bu
X
,
Chen
H
,
Wang
Q
,
Sha
W
. 
Bmi-1 promotes the invasion and migration of colon cancer stem cells through the downregulation of E-cadherin
.
Int J Mol Med
2016
;
38
:
1199
207
.
19.
Liu
L
,
Qiu
M
,
Tan
G
,
Liang
Z
,
Qin
Y
,
Chen
L
, et al
miR-200c inhibits invasion, migration and proliferation of bladder cancer cells through down-regulation of BMI-1 and E2F3
.
J Transl Med
2014
;
12
:
305
.
20.
Itou
J
,
Matsumoto
Y
,
Yoshikawa
K
,
Toi
M
. 
Sal-like 4 (SALL4) suppresses CDH1 expression and maintains cell dispersion in basal-like breast cancer
.
FEBS Lett
2013
;
587
:
3115
21
.
21.
Isella
C
,
Terrasi
A
,
Bellomo
SE
,
Petti
C
,
Galatola
G
,
Muratore
A
, et al
Stromal contribution to the colorectal cancer transcriptome
.
Nat Genet
2015
;
47
:
312
9
.
22.
Stankic
M
,
Pavlovic
S
,
Chin
Y
,
Brogi
E
,
Padua
D
,
Norton
L
, et al
TGF-beta-Id1 signaling opposes Twist1 and promotes metastatic colonization via a mesenchymal-to-epithelial transition
.
Cell Rep
2013
;
5
:
1228
42
.
23.
Widmer
DS
,
Hoek
KS
,
Cheng
PF
,
Eichhoff
OM
,
Biedermann
T
,
Raaijmakers
MI
, et al
Hypoxia contributes to melanoma heterogeneity by triggering HIF1alpha-dependent phenotype switching
.
J Invest Dermatol
2013
;
133
:
2436
43
.
24.
Rankin
EB
,
Giaccia
AJ
. 
Hypoxic control of metastasis
.
Science
2016
;
352
:
175
80
.
25.
Tay
Y
,
Rinn
J
,
Pandolfi
PP
. 
The multilayered complexity of ceRNA crosstalk and competition
.
Nature
2014
;
505
:
344
52
.
26.
Karginov
FV
,
Conaco
C
,
Xuan
Z
,
Schmidt
BH
,
Parker
JS
,
Mandel
G
, et al
A biochemical approach to identifying microRNA targets
.
Proc Natl Acad Sci U S A
2007
;
104
:
19291
6
.
27.
Ronkainen
J
,
Aro
P
,
Storskrubb
T
,
Johansson
SE
,
Lind
T
,
Bolling-Sternevald
E
, et al
Prevalence of Barrett's esophagus in the general population: an endoscopic study
.
Gastroenterology
2005
;
129
:
1825
31
.
28.
De Sousa
EMF
,
Wang
X
,
Jansen
M
,
Fessler
E
,
Trinh
A
,
de Rooij
LP
, et al
Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions
.
Nat Med
2013
;
19
:
614
8
.
29.
Comprehensive molecular characterization of human colon and rectal cancer
.
Nature
2012
;
487
:
330
7
.
30.
Rapley
EA
,
Crockford
GP
,
Teare
D
,
Biggs
P
,
Seal
S
,
Barfoot
R
, et al
Localization to Xq27 of a susceptibility gene for testicular germ-cell tumours
.
Nat Genet
2000
;
24
:
197
200
.
31.
Peng
X
,
Guo
W
,
Liu
T
,
Wang
X
,
Tu
X
,
Xiong
D
, et al
Identification of miRs-143 and -145 that is associated with bone metastasis of prostate cancer and involved in the regulation of EMT
.
PLoS One
2011
;
6
:
e20341
.
32.
Yu
X
,
Zhang
X
,
Bi
T
,
Ding
Y
,
Zhao
J
,
Wang
C
, et al
MiRNA expression signature for potentially predicting the prognosis of ovarian serous carcinoma
.
Tumour Biol
2013
;
34
:
3501
8
.
33.
Shang
Y
,
Zhang
Z
,
Liu
Z
,
Feng
B
,
Ren
G
,
Li
K
, et al
miR-508–5p regulates multidrug resistance of gastric cancer by targeting ABCB1 and ZNRD1
.
Oncogene
2013
;
33
:
3267
76
.
34.
Peinado
H
,
Olmeda
D
,
Cano
A
. 
Snail, Zeb and bHLH factors in tumour progression: an alliance against the epithelial phenotype?
Nat Rev Cancer
2007
;
7
:
415
28
.
35.
Thiery
JP
. 
Epithelial-mesenchymal transitions in development and pathologies
.
Curr Opin Cell Biol
2003
;
15
:
740
6
.
36.
Arias
AM
. 
Epithelial mesenchymal interactions in cancer and development
.
Cell
2001
;
105
:
425
31
.
37.
Lamouille
S
,
Xu
J
,
Derynck
R
. 
Molecular mechanisms of epithelial-mesenchymal transition
.
Nat Rev Mol Cell Biol
2014
;
15
:
178
96
.
38.
Shirakihara
T
,
Saitoh
M
,
Miyazono
K
. 
Differential regulation of epithelial and mesenchymal markers by deltaEF1 proteins in epithelial mesenchymal transition induced by TGF-beta
.
Mol Biol Cell
2007
;
18
:
3533
44
.
39.
Wellner
U
,
Schubert
J
,
Burk
UC
,
Schmalhofer
O
,
Zhu
F
,
Sonntag
A
, et al
The EMT-activator ZEB1 promotes tumorigenicity by repressing stemness-inhibiting microRNAs
.
Nat Cell Biol
2009
;
11
:
1487
95
.
40.
Song
LB
,
Li
J
,
Liao
WT
,
Feng
Y
,
Yu
CP
,
Hu
LJ
, et al
The polycomb group protein Bmi-1 represses the tumor suppressor PTEN and induces epithelial-mesenchymal transition in human nasopharyngeal epithelial cells
.
J Clin Invest
2009
;
119
:
3626
36
.
41.
Gargiulo
G
,
Cesaroni
M
,
Serresi
M
,
de Vries
N
,
Hulsman
D
,
Bruggeman
SW
, et al
In vivo RNAi screen for BMI1 targets identifies TGF-beta/BMP-ER stress pathways as key regulators of neural- and malignant glioma-stem cell homeostasis
.
Cancer Cell
2013
;
23
:
660
76
.
42.
Masuda
S
,
Suzuki
K
,
Izpisua Belmonte
JC
. 
Oncofetal gene SALL4 in aggressive hepatocellular carcinoma
.
N Engl J Med
2013
;
369
:
1171
.
43.
Carmeliet
P
,
Dor
Y
,
Herbert
JM
,
Fukumura
D
,
Brusselmans
K
,
Dewerchin
M
, et al
Role of HIF-1alpha in hypoxia-mediated apoptosis, cell proliferation and tumour angiogenesis
.
Nature
1998
;
394
:
485
90
.

Supplementary data