Microarray Analysis of Colorectal Cancer Stromal Tissue Reveals Upregulation of Two Oncogenic miRNA Clusters

Cancer tissues consist of cancer cells and surrounding stromal cells, including inflammatory cells, immunocompetent cells, endothelial cells, and fibroblasts. Cancer stroma interacts with cancer tissues directly or indirectly through cytokines, creating a niche for the cancer cells. Recent studies have focused on altered expression of oncogenes and tumor suppressor genes in stromal tissues. Kurose and colleagues reported that the downregulation of PTEN and p53 is a key step in breast cancer progression, and other reports have indicated that ablation of TGFBR2 in fibroblasts can lead to carcinogenesis in vivo (1–3).

With regard to clinical aspects, a number of studies have revealed the gene expression status of cancer stroma and its correlation to prognosis as well as clinicopathologic factors (4, 5). Particularly, Finak and colleagues analyzed global gene expression patterns in breast cancer stromal tissues and identified gene sets that potently influenced prognosis (6). They revealed that the aggressiveness of cancer could be defined by gene expression patterns in stromal tissue. Moreover, Fukino and colleagues revealed that cancer-specific LOH or allelic imbalance in stromal cells is more highly correlated with clinicopathologic features than that in epithelial cells (7). These findings suggest that cancer stromal tissues are actively involved in cancer progression.

miRNAs constitute a class of small (19–25 nucleotides) noncoding RNAs that function as posttranscriptional gene regulators by binding to their target mRNAs (8). Alterations in miRNA expression are reported in various kinds of human cancers, suggesting miRNAs function both as tumor suppressors and oncogenes in cancer development (9). Genetic alterations in cancer tissues are likely dependent largely on the expression status of miRNAs (9). We considered the possibility that gene expression in cancer stroma is also regulated by miRNAs expressed in cancer stroma.

Conventional gene expression analysis using bulk tumor samples could not reveal the gene and miRNA expression profiles in cancer stroma. In this study, using a laser microdissection (LMD) method, we collected epithelium-specific and stoma-specific RNAs, and investigated the miRNA and gene expression profiles. By analyzing many kinds of microarray data, including that of epithelium, stroma, normal, and cancer, we show how miRNAs in cancer stroma are involved in cancer progression.

Tissues from 13 cases of colorectal cancer and 4 normal colorectal tissues (located more than 5 cm from the colorectal cancer) were obtained during surgery. All patients underwent resection of the primary tumor at Kyushu University Hospital at Beppu and affiliated hospitals between 1993 and 2006. Written informed consent was obtained from all patients, and the study protocol was approved by the local ethics committee. Detailed information is described in Supplementary Data.

Tissue samples were microdissected using the LMD system LMD6000 (Leica Laser Microdissection System; Leica Microsystems) as previously described (10). Detailed protocols are described in Supplementary Information.

To show the accuracy of LMD and RNA separation, we presented tissue section before (left) and after (right) LMD in 6 representative samples of colorectal cancer (Supplementary Fig. S1).

Total RNAs from epithelial and stromal tissues of cancer and normal samples were analyzed by miRNA microarray. Total RNA was extracted from tissue using the miRNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. Concentrations and purities of the total RNAs were assessed with a spectrophotometer and RNA integrity was verified using an Agilent 2100 Bioanalyzer (Agilent Technologies). OD₂₆₀/OD₂₈₀ ratios of 1.8 to 2.1 were accepted to be adequate for microarray. Total RNA (100 ng) was directly labeled with cyanine 3-CTP (Cy3), without fractionation or amplification, using an Agilent protocol that produces precise and accurate measurements spanning a linear dynamic range from 0.2 amol to 2 fmol of input miRNA. Each total RNA sample (100 ng) was competitively hybridized to a miRNA array (Agilent Microarray Design ID = 014947, Early Access version) containing 455 miRNAs [version 15 of the Sanger miRNA database (http://www.mirbase.org/)], according to the manufacturer's protocol (11). The intensity of each hybridization signal was evaluated using Extraction Software Version A.7.5.1 (Agilent Technologies), which used the locally weighted linear regression curve fit (LOWESS) normalization method (12).The robust multiarray average algorithm normalization method was also used for analysis in downregulated miRNAs in cancer stroma (13, 14). miRNA arrays have been deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database with accession code GSE35602.

We used the commercially available Human Whole Genome Oligo DNA Microarray Kit (Agilent Technologies). A list of genes on this cDNA microarray is available from http://www.chem.agilent.com. Cyanine (Cy)-labeled cRNA was prepared using T7 linear amplification as described in the Agilent Low RNA Input Fluorescent Linear Amplification Kit Manual (Agilent Technologies). Labeled cRNA was fragmented and hybridized to an oligonucleotide microarray (Whole Human Genome 4 × 44 K Agilent G4112F). Fluorescence intensities were determined with an Agilent DNA Microarray Scanner and were analyzed using G2567AA Feature Extraction Software Version A.7.5.1 (Agilent Technologies), which used the LOWESS normalization method (12). This microarray study followed MIAME guidelines issued by the Microarray Gene Expression Data group (15). Gene expression arrays have been deposited in the NCBI GEO database with accession code GSE35602.

A total of 1,939 Gene set enrichment analyses of differentially expressed genes were carried out using Gene Codis version 2.0 (16, 17). The Kyoto Encyclopedia of Genes and Genomes (KEGG) database (18) was used for systematic analysis of gene functions. TargetScan (19, 20) Version 5.1 algorithm was used to predict putative targets of each miRNA. TargetScan focuses on the exact match to 7 bases or more (hexamer match in positions 2–7, plus an adenosine at a position on the 3′ side) of the miRNA seed sequence in the 3′-untranslated region of target messenger RNA. In our analysis, we included all targets that met TargetScan criteria without taking into account evolutionary conservation.

For miR-25 and miR-92a quantitative real-time reverse transcriptase PCR (qRT-PCR), cDNA was synthesized from 10 ng of total RNA using TaqMan miRNA hsa-miR-25- or 92a-specific primers (Applied Biosystems) and a TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems). RT-PCR protocols are described in Supplementary Information.

Differences between groups were estimated using the χ² test and Student t test. After expression signals were calculated by log₂ transformation of the normalized data, differentially expressed miRNAs and genes were detected by using the fold-change value and Q value. We used the significance analysis of microarrays method in the “samr” package of the R language (http://www.r-project.org/). RT-PCR data were analyzed using JMP software (SAS Institute Inc.). All differences were considered statistically significant at the level of P < 0.05 or Q < 0.05.

To investigate the functions of miRNAs in cancer stroma, we carried out miRNA microarray analyses with samples of 13 cancer stromal tissue and 4 samples of normal stromal tissue. Those miRNAs that were significantly upregulated in cancer stromal tissues compared with normal stromal tissues are listed in Table 1 (fold change >1.5, and Q < 0.05). For example, oncogenic miRNAs, including miR-21, miR-221, and almost all components of the oncogenic miR-17-92a cluster and miR-106b-25 cluster except miR-19b were upregulated in cancer stromal tissue compared with normal stromal tissue (Table 1). Unsupervised hierarchical clustering of normal and cancerous stromal samples revealed that all components of the miR-17-92a cluster and the miR-106b-25 cluster (except for miR-92a) were classified in the same cluster (Supplementary Fig. S2A and S2B). MiR-92a was classified in another cluster, perhaps because miR-92a has a homolog (miR-92a-2) on chromosome X, besides miR-92a-1 in the miRNA cluster on chromosome 13 (21). Therefore, its expression could have diverged from the expression of other miRNAs in the cluster.

The correlation coefficient ratio was significantly higher within miRNA clusters (Supplementary Table S1). These clusters are transcribed from another chromosome. The miR-17-92a cluster was located on chromosome 13 in the host gene C13ORF25, and its homolog miR-106b-25 cluster was located in the intron of the host gene MCM7 on chromosome 7 (Fig. 1, top part). Some components shared the same seed sequence (Fig. 1, bottom part), suggesting they regulated the same targets simultaneously.

In 4 of the 13 samples of cancer tissue and the 4 normal tissues, both epithelial and stromal samples were available. Thus, we carried out combined analyses in epithelial and stromal tissues in these samples. The expression status of epithelial tissues was highly correlated with that of stromal tissues, and the expression of miRNA in the epithelium was always higher than that in the stroma in all components of the miR-17-92a cluster and the miR-106b-25 cluster (Fig. 2).

We also carried out gene expression array analyses using the same samples used in the miRNA microarray. We identified significantly downregulated genes (fold change < 0.5 and Q < 0.05), which were putative targets of the miR-17-92a and miR-106b-25 clusters. The number of genes that matched these criteria was 1,939. Gene set enrichment analysis using Gene Codis version 2.0 (16, 17) and KEGG revealed that the following molecules were enriched: KEGG 04060: cytokine–cytokine receptor interaction [chemokines, hematopoietins, platelet-derived growth factor (PDGF) family, TNF family, and TGFβ family], KEGG 04340: hedgehog signaling pathway, KEGG 05200: pathways in cancer, and KEGG 04514: cell adhesion molecules (CAM; Table 2, Supplementary Table S2A). These indicated the possibility that the miR-17-92a and miR-25-106b clusters were involved in those important pathways in colorectal cancer stroma.

We used different approaches to identify putative targets of the miR-17-92a and miR-106b-25 clusters. Combining the microarray data of miRNAs and genes in stromal samples, we identified highly inversely correlated miRNA gene pairs, which were putative pathways predicted by TargetScan (refs. 19, 20; correlation coefficient >0.65 and P < 0.05; Table 3). These included a number of crucial regulators of cellular function, such as the apoptosis-related miR-17 death–associated protein kinase 2 (DAPK) pairing, the miR-18a–caspase 7 pairing, and the miR-17 cancer-related transcription factor 7 (TCF7) pairing.

The miRNA array analysis revealed that 2 oncogenic miRNA clusters were upregulated in colorectal cancer stroma. Therefore, we confirmed the expression status of 2 representative miRNAs from the 2 clusters, miR-25, and miR-92a in clinical samples. MiR-25 and miR-92a are selected because of relative high expression in cancer stroma and because they have the same seed sequence and share the same targets and could work coordinately. In samples of normal epithelial tissue (n = 4), cancer epithelial tissue (n = 10), normal stromal tissue (n = 4), and cancer stromal tissue (n = 26), which included 13 cancer samples and 4 normal samples used for microarray analysis, we carried out qRT-PCR to investigate the expression of miR-25 and miR-92a. The data confirmed the upregulation of these miRNAs in cancer stroma compared with normal stroma in accordance with upregulation in cancer epithelium (Fig. 3).

For the 26 colorectal cancer stromal samples for which we used RT-PCR analysis, clinicopathologic data were available in 24 cases. Clinicopathologic analysis revealed that the high miR-25 expression group (values > the 0.25 quartile; 0.54, normalized to RNU6B) had more advanced venous invasion compared with the low expression group (values < the 0.25 quartile; P = 0.046, Table 4). In the high miR-92a expression group (values > the 0.75 quartile; 1.16, normalized to RNU6B), there was greater lymphatic invasion (P = 0.005), venous invasion (P = 0.016), and liver metastasis (P = 0.018) compared with the low miR-92a expression group (values < the 0.75 quartile; Table 5). However, no significant differences were observed regarding age, gender, histology, lymphatic invasion, venous invasion, lymph node metastasis, peritoneal dissemination, or distant metastasis.

We also investigated downregulated miRNAs in cancer stroma compared with normal stroma. As a result, previously reported tumor-suppressive miRNAs, such as the miR-192-miR-194 cluster, miR-215, miR-29c, miR-26b, and let-7g were all downregulated in cancer stromal tissues compared with normal stromal tissues (Supplementary Table S3) (22, 23). Downregulation of tumor-suppressive miRNAs in cancer stroma as well as upregulation of oncogenic miRNAs in cancer stroma indicated that alterations of miRNA expression in stromal tissues are similar to that in epithelial tissues.

Significantly upregulated or downregulated miRNAs in cancer stroma compared with cancer epithelium are listed in Supplementary Table S4A and S4B.

In this study, we analyzed expression levels of miRNAs expressed in cancer stroma and revealed that many oncogenic miRNAs including miR-21 (24), miR-221 (25), and almost all components of the miR-17-92a cluster and its homolog, and the miR-106b-25 cluster were upregulated in cancer stroma compared with normal stroma (Table 1). Previous reports showed that the miR-17-92a cluster is upregulated in lung cancer, colorectal cancer, lymphoma, multiple myeloma, and medulloblastoma, whereas the miR-106b-25 cluster is upregulated in gastric, colon, and prostate cancer, neuroblastoma, and multiple myeloma (9, 26, 27). Importantly, the expression of these miRNAs in stromal tissues had not been explored. We focused on these well-characterized miRNAs which are upregulated in cancer stroma as well as epithelial tissues.

The reason for the upregulation of such oncogenic miRNAs in stromal tissue is still unclear. However, previous studies have revealed genetic alterations such as LOH of cancer stromal cells (28) or epigenetic modification of cancer-associated fibroblasts (CAF; ref. 29). Hence, genetic or epigenetic changes of cancer stromal cells could be one explanation for this, although it remains controversial. Another possibility is intercellular transfer or penetration of miRNAs. It has been shown that secreted miRNAs from donor cells are transferred to and function in recipient cells. Some reports have shown that miRNAs are transferred in exosomes, vesicles of endocytic origin (30). If this occurred between cancer cells and stromal cells, profiles of miRNAs could become similar in both. We revealed that oncogenic miRNAs such as miR-135b, miR-221, and the miR-17-92a cluster were expressed at relatively higher levels in cancer epithelium than in stroma (Supplementary Table S4B). These findings suggest that upregulation of such oncogenic miRNAs in cancer stroma follows the upregulation of oncogenic miRNA in epithelium during carcinogenesis and cancer progression. However, further studies are required to confirm the interaction between cancer and its stroma.

Recently, Mestdagh and colleagues reported that the miR-17-92 miRNA cluster regulated key components of the TGFβ pathway such as SMAD2, SMAD4, and TGFBR2 in neuroblastoma (31). With regard to epithelial–mesenchymal interactions, Bhowmick and colleagues showed that ablation of TGFBR2 in stromal tissue can lead to carcinogenesis and cancer progression in vivo (3). In our data, TGFBR2, SMAD2, and SMAD3, which are putative targets of the miR-17-92a and miR-106b-25 cluster, were shown to be significantly repressed in cancer stroma (Table 2 and Table 3). These results suggest that, at least in part, inhibition of the TGFβ pathway in stromal tissue is associated with tumorigenesis and cancer progression. However, the putative miRNA target gene pathways we showed in this study have not been experimentally established, therefore, further exploration is needed to confirm that they function in stromal tissue.

Clinicopathologic analysis showed that miRNAs expression in stromal tissue is associated with the malignant potential of cancer (Table 4 and Table 5). High expression of miR-25 was associated with venous invasion, and miR-92a with lymphatic and venous invasion and liver metastasis. These data indicate that miRNA expression in stroma as well as in the epithelium can influence tumor aggressiveness in colorectal cancer. Lymphatic or venous invasion occurs dominantly in stromal tissue; therefore these results are entirely reasonable. Although validation in a larger number of samples is required in future studies, our data indicate the potential clinical significance of miRNAs in cancer stroma.

Stromal tissues are constituted by various kinds of cells, including immune cells, endothelial cells, and fibroblasts. Previous reports suggested CAF or tumor-associated macrophage could enhance tumor progression and metastasis (4, 32), and Vermeulen and colleagues showed that myofibroblasts play a role in the maintenance of cancer stem cell properties (33). In this study, we analyzed unsorted samples of stromal tissue, therefore the miRNA profiling for each cell type was not available. Sorting according to surface antigen, such as CD31 for endothelial cells or CD14 for macrophage, will be required for cell type–specific investigations in future studies.

Our findings suggest the possibility that oncogenic miRNAs including the miR-17-92a and miR-25-106b clusters in colorectal cancer stromal tissues are functionally associated with cancer progression. Because our results are based on microarray data derived from a small sample size, further validation is required.

No potential conflicts of interest were disclosed.

Conception and design: N. Nishida, M. Nagahara, K. Mimori, H. Ishii, M. Mori

Development of methodology: N. Nishida, M. Nagahara, F. Tanaka, H. Ishii, M. Mori

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Nishida, H. Ishii, M. Mori

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Nishida, T. Sato, H. Ishii, M. Mori

Writing, review, and/or revision of the manuscript: N. Nishida, K. Mimori, T. Sudo, F. Tanaka, H. Ishii, M. Mori

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Nishida, T. Sudo, H. Ishii, M. Mori

Study supervision: N. Nishida, K. Mimori, F. Tanaka, K. Shibata, K. Sugihara, Y. Doki, M. Mori

The authors thank T. Shimooka, K. Ogata, M. Kasagi, and T. Kawano for their excellent technical assistance.

This work was supported in part by the following grants and foundations: CREST, Japan Science and Technology Agency (JST); Japan Society for the Promotion of Science (JSPS) grant-in-aid for Scientific Research: 21679006, 20390360, 20590313, 20591547, 21591644, 21592014, 20790960, 21791297, 21229015, 20659209, and 20012039; NEDO (New Energy and Industrial Technology Development Organization) Technological Development for Chromosome Analysis; The Ministry of Education, Culture, Sports, Science and Technology of Japan for Scientific Research on Priority Areas, Cancer Translational Research Project, Japan; and LS094, Bureau of Science, Technology and Innovation Policy, Cabinet Office, Government Of Japan.

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.