Our newly developed OMCD is available at www.oncomir.umn.edu/omcd. It features 4 types of search functions (Fig. 1a). For example, it currently includes miRNA expression data from 8 control colon tissue samples and 272 colon cancer (COAD) tumor samples. When we search for miR-21 in COAD samples (Fig. 1a, b), we obtain a heatmap showing the absolute expression level of miR-21 for all COAD samples (Fig. 1c). We can also obtain the numeric expression data (Fig. 1d; not completely shown, because of space limitations) and relative expression data (Fig. 1e). Clicking on hsa-miR-21 from the heatmap page, we are taken to a page showing links to additional analysis (Fig. 1f). These links provide detailed information about the chromosomal location of miR-21 and the names of colocalized miRNAs (miRNA clusters), as well as additional internal links to the expression data of miR-21 in other cancer types and to further statistical analysis (Fig. 1h).
In our COAD example, each miRNA specific OMCD webpage provides external links to the miRDB website for target prediction (www.mirdb.org) and to Google Scholar for literature searches [9]. From this webpage, we generate a link that allows the visualization of colocalized miRNA expression levels in a heatmap showing absolute expression (Fig. 1g). Expression levels of colocalized miRNAs can be displayed for all cancer types (not shown) and can be visualized in absolute and relative heatmaps as well as in the form of numeric data.
The 3 statistical analyses that we performed—using normal controls vs. tumor samples for each tumor type where available; tissue control samples vs. all other patients’ control samples; and each tumor sample type vs. all other tumor sample types—allowed us to visualize the expression patterns of miR-21 across different cancer types (Fig. 1h).
Further demonstrating OMCD’s utility, we were able to identify miRNAs that were recurrently differentially expressed between tumors and control samples. The difference was highly significant (P < 0.000001). In 5 such comparisons, the mean fold-change in the tumor samples was greater than 2 (Fig. 2). Many miRNAs are functionally well characterized and have been reported to be differentially expressed (between tumor and control samples) in a wide range of cancer types. For example, miR-21 is consistently upregulated in most cancer types [10]. Thus, it could potentially serve as cancer biomarker, but it may not be a suitable for identification of a specific cancer type. We were also able to observe decreases in miR-1/miR-133 in a number of cancers as well as gains in the miR-96/miR-182/miR-183 cluster in a number of other cancers.
In our OMCD testing, we also found that the COAD cluster and rectal cancer (READ) cluster had a very similar miRNA expression pattern, as compared with other cancer types. In COAD miR-101 showed higher expression levelsthen normal tissue and this increase was also observable in READ although not at the statistical power available for COAD.(Fig. 2).
Additionally, because the miR-101 expression was not significantly higher in other cancer types, it is reasonable to hypothesize that this miRNA is a biomarker for COAD. Similarly, we found that miR-10b expression was uniquely higher in hepatocellular carcinoma (LIHC), but not in other cancer types. These are a few examples of the testable hypotheses that OMCD can generate. To more thoroughly investigate the function of miR-21, mir-96/miR-182/miR-183 cluster in cancer, miR-101 in COAD, and miR-10b in LIHC, further experimental validation is warranted.